Clinical activity and molecular correlates of response to atezolizumab alone or in combination with bevacizumab versus sunitinib in renal cell carcinoma

Abstract

We describe results from IMmotion150, a randomized phase 2 study of atezolizumab (anti-PD-L1) alone or combined with bevacizumab (anti-VEGF) versus sunitinib in 305 patients with treatment-naive metastatic renal cell carcinoma. Co-primary endpoints were progression-free survival (PFS) in intent-to-treat and PD-L1+ populations. Intent-to-treat PFS hazard ratios for atezolizumab + bevacizumab or atezolizumab monotherapy versus sunitinib were 1.0 (95% confidence interval (CI), 0.69-1.45) and 1.19 (95% CI, 0.82-1.71), respectively; PD-L1+ PFS hazard ratios were 0.64 (95% CI, 0.38-1.08) and 1.03 (95% CI, 0.63-1.67), respectively. Exploratory biomarker analyses indicated that tumor mutation and neoantigen burden were not associated with PFS. Angiogenesis, T-effector/IFN-γ response, and myeloid inflammatory gene expression signatures were strongly and differentially associated with PFS within and across the treatments. These molecular profiles suggest that prediction of outcomes with anti-VEGF and immunotherapy may be possible and offer mechanistic insights into how blocking VEGF may overcome resistance to immune checkpoint blockade.

Figures

Fig. 1|. Positive independent review facility (IRF)-assessed efficacy associated with atezolizumab + bevacizumab in mRCC patients with PD-L1+ IC.

a,b, Kaplan-Meier curves depict IRF-assessed median PFS in the atezolizumab (atezo) + bevacizumab (bev), atezolizumab monotherapy, and sunitinib treatment arms in the (a) ITT population and (b) PD-L1+ (≥1% PD-L1 expression on IC by IHC) population across 33 months. Censored data are indicated by vertical tick marks in Kaplan-Meier curves. Sample numbers per group and timepoint indicated below each graph. HR calculated using stratified Cox proportional hazard regression models, and P values calculated using stratified log-rank test (for details, see Methods). P values are provided for descriptive purposes only and were not adjusted for multiple comparisons. Mo, months. c, ORRs as depicted by PR and CR for the ITT and PD-L1+ populations for each treatment arm (n = 101, 103, and 101 patients for atezolizumab + bevacizumab, atezolizumab monotherapy, and sunitinib treatment arms, respectively, in the ITT population; n = 50, 54, and 60 patients, respectively, in the PD-L1+ population). ORR values are indicated above each bar (with 95% CIs for ORR plotted as error bars). Values within the lighter and darker regions of the bars refer to the PR and CR rates, respectively

Fig. 2|. Baseline tumor gene signature analyses.

a, Heatmap showing expression of genes of interest (rows) in 263 pretreatment tumors (columns). Normalized counts of genes related to angiogenesis (brown), immune and antigen presentation (purple), and myeloid inflammation (gray) were z-score transformed before visualization. Sample annotations include PD-L1 IHC status for tumor-infiltrating ICs, presence of sarcomatoid features, Memorial Sloan Kettering Cancer Center (MSKCC) score, tumor stage, number of mutations (TMB), and mutation status of VHL and PBRM1. b, Mean CD31 IHC staining intensity is higher in AngioHigh than in AngioLow (two-tailed t test, P=4.19×10−21). c,d, Teff signature scores are associated with (c) PD-L1 protein expression levels on IC by IHC (one-sided Wald test, P=3.26×10−20) and (d) intratumoral CD8A protein expression by IHC (two-tailed t test, P = 1.26×10−28). Box plot elements in b-d are defined in Methods. Sample numbers per group indicated above each graph. e, ORR= PR + CR in the AngioHigh and AngioLow populations for each treatment arm. Error bars represent 95% CI for ORR; P values calculated using a two-sided χ2 tests. f, Forest plots of PFS HRs and CIs for AngioHigh vs. AngioLow populations within each treatment arm. g,h, Kaplan-Meier curves showing the probability of PFS across treatment arms in the AngioLow (g) and AngioHigh (h) subgroups; HR calculated vs. sunitinib. i, ORR (PR + CR) in the TeffHigh and TeffLow populations for each treatment arm. Error bars represent 95% CIs for ORR; P values calculated using two-sided χ2 tests. j, Forest plots of PFS HRs and CIs for TeffHigh vs. TeffLow populations within each treatment arm. k,l, Kaplan-Meier curves showing probability of PFS across treatment arms in TeffLow (k) and TeffHigh (l) subgroups; HR calculated vs. sunitinib. m,n, Kaplan-Meier curves showing probability of PFS in TeffHighMyeloidLow (m) and TeffHighMyeloidHigh (n) subgroups; HR calculated vs. atezolizumab monotherapy. Censored data indicated by vertical tick marks in Kaplan-Meier curves. All HR and CI values for PFS were extracted from Cox proportional hazard regression models; median survival time per group is indicated. P values reported are for descriptive purposes only and were not adjusted for multiple comparisons. Sample numbers per group indicated below the graphs in e, g-i, and k-n, and within the graphs in f and j.

Fig. 3|. Association between tumor mutations and clinical outcome.

a, TMB and TNB are plotted by response group (CR and PR vs. SD (stable disease) and PD (progressive disease)) for each treatment arm. No apparent difference was observed between response groups in the sunitinib (two-tailed t test, P = 0.06), atezolizumab + bevacizumab (two-tailed t test, P = 0.14), and atezolizumab monotherapy arms (two-tailed t test, P = 0.93). The violin plots show the kernel probability density of the data. The point indicates the mean TMB or TNB in each group, while the bars represent the 95% CIs of the mean. Sample numbers per group indicated above the violin plots. b, Presence of loss-of-function mutations in genes of interest across 201 tumors. Based on functional prediction of mutation consequence, mutations were categorized as frameshift, splicing, missense, nonsense, or in-frame. The overall prevalence of mutations for each gene is shown on the left as percentages. In addition, TMB is shown. c, Patients were divided into two groups based on the presence of loss-of-function mutations (mutant vs. nonmutant). The probability of PFS for these two groups is shown for each treatment arm. HR and CIs were extracted from Cox proportional hazard regression models, comparing the mutant group with the nonmutant group. Censored data are indicated by vertical tick marks in Kaplan-Meier curves. Median survival time per group is indicated. P values reported are for descriptive purposes only and were not adjusted for multiple comparisons. Sample numbers per group and timepoint indicated below the graphs.