A signature of enhanced proliferation associated with response and survival to anti-PD-L1 therapy in early-stage non-small cell lung cancer

Abstract

In early-stage non-small cell lung cancer, the combination of neoadjuvant anti-PD-L1 and subablative stereotactic body radiation therapy (SBRT) is associated with higher rates of major pathologic response compared to anti-PD-L1 alone. Here, we identify a 140-gene set, enriched in genes characteristic of highly proliferating cells, associated with response to the dual therapy. Analysis of on-treatment transcriptome data indicate roles for T and B cells in response. The 140-gene set is associated with disease-free survival when applied to the combined trial arms. This 140-gene set identifies a subclass of tumors in all 7 of The Cancer Genome Atlas tumor types examined. Worse survival is associated with the 140-gene signature in 5 of these tumor types. Collectively, our data support that this 140-gene set, discovered in association with response to combined anti-PD-L1 and SBRT, identifies a clinically aggressive subclass of solid tumors that may be more likely to respond to immunotherapies.

Keywords: NSCLC; biomarker immunotherapy response; combination immunotherapy radiation; immunotherapy; radiation therapy; rapidly proliferating tumors.

Graphical abstract

Figure 1

Pretreatment tumor characteristics of responders (MPR) to dual ICB and SBRT therapy (arm 2) (A) Time to surgery following last neoadjuvant treatment for arm 2 subjects: MPR (n = 10); no MPR (n = 6). Mean ± SEM. Unpaired, 2-sided Mann-Whitney test. (B) PD-L1⁺ cancer cells as percentage of total cancer cells determined by immunohistochemistry of pretreatment biopsies. No MPR (n = 6); MPR (n = 10). Mean ± SEM. Unpaired, 2-sided Mann-Whitney test. (C) Heatmap of pretreatment differentially expressed genes arm 2 tumors with (n = 10) and without MPR (n = 6). Data are scaled to mean expression per row. (D) Pretreatment proliferation indexes (sum Z scores) determined by the expression of 113 genes and by a modified set of 73 genes excluding the 40 genes that are part of the 140-gene set used for clustering (Table S3). No MPR (n = 6); MPR (n = 10). Mean ± SEM. Unpaired, 2-sided Mann-Whitney test. (E) Pretreatment tumor cellularity. No MPR (n = 6); MPR (n = 10). Mean ± SEM. Unpaired, 2-sided Mann-Whitney test. (F) Gene set enrichment. MPR (n = 10) to no MPR (n = 6) tumor groups, with no MPR as reference. NES, normalized enrichment score. (G) CD4⁺ Th2 enrichments determined by xCell deconvolution. No MPR (n = 6); MPR (n = 10). Mean ± SEM. Unpaired, 2-sided Mann-Whitney test.

Figure 2

Characteristics of proliferation and cancer cell PD-L1 expression establish 2 groups of pretreatment NSCLC tumors (A) Hierarchical clustering (Pearson’s 1-r) of all of the pretreatment samples (n = 32) based on expressions of the 140-gene set (Table S2). (B) Pretreatment PI's of the 2 clusters, low PI (n = 14) and high PI (n = 18). Filled symbols are arm 2 tumors. Drv33, the sole monotherapy tumor that achieved MPR, is the green-filled symbol. PI (sum Z scores) determined based on the expression of 113 genes (Table S3). Mean ± SEM. Unpaired, 2-sided Mann-Whitney test. (C) Modified PI (sum Z scores) generated excluding the 40 genes that are also part of the 140 genes used for clustering (Table S3). Low PI (n = 14) and high PI (n = 18). Drv33 is the green-filled symbol. Mean ± SEM. Unpaired, 2-sided Mann-Whitney test. (D) Pretreatment Ki67 gene (mki67) expression (fragments per kilobase of transcript per million mapped reads [FPKM]). Filled symbols are arm 2 tumors. Drv33 is the green-filled symbol. Low PI (n = 14) and high PI (n = 18). Mean ± SEM. Unpaired, 2-sided Mann-Whitney test. (E) Pretreatment tumor cellularity. Low PI (n = 14) and high PI (n = 18). Drv33 is the green-filled symbol. Mean ± SEM. Unpaired, 2-sided Mann-Whitney test. (F) Pretreatment Ki67⁺ cancer cells as a percentage of cancer cells, determined by immunofluorescence of a subset of samples. Low PI (n = 5) and high PI (n = 5). Mean ± SEM. Unpaired, 2-sided Mann-Whitney test. (G) Pretreatment PD-L1⁺ cancer cells as a percentage of cancer cells determined by immunohistochemistry. Low PI (n = 14) and high PI (n = 18). Mean ± SEM. Unpaired, 2-sided Mann-Whitney test. (H) Pretreatment PD-L1⁺ gene (CD274) expression (FPKM) from RNA-seq data. Low PI (n = 14) and high PI (n = 18). Mean ± SEM. Unpaired, 2-sided Mann-Whitney test.

Figure 3

Pretreatment samples clustered by proliferation genes reveal additional characteristics of the clusters (A) Gene expressions of the enzymes and transporters of glycolysis. Low PI (n = 14) and high PI (n = 16): p < 0.0001, 1-way ANOVA. Sidak’s p values corrected for multiple testing shown for individual genes. ∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001; ∗∗∗∗p < 0.0001; ns, not significant. (B) Pretreatment diagnostic SUVmax values. Low PI (n = 14) and high PI (n = 16). Mean ± SEM. Unpaired, 2-sided Mann-Whitney test. (C) Pretreatment expressions of MHC class II genes. FPKM are scaled to the mean value of expression of the individual genes in the low PI group. Significance between low and high PI tumor groups, Kruskal-Wallis p < 0.0001. Dunn’s corrected p values for individual genes are shown. Low PI (n = 14) and high PI (n = 16). Mean ± SEM. (D and E) Th2 CD4⁺ cells and Th1 CD4⁺ cells enrichment scores (xCell deconvolution¹¹) of pretreatment RNA-seq data. Low PI (n = 14) and high PI (n = 16). Mean ± SEM. Unpaired, 2-sided Mann-Whitney test. (F) CD3 cells per square millimeter determined by immunofluorescence imaging analyses of 5 samples of each group. Mean ± SEM. Unpaired, 2-sided Mann-Whitney test. (G) CD3⁺CD8⁺ cells per square millimeter determined by immunofluorescence imaging analyses of 5 samples of each group. Mean ± SEM. Unpaired, 2-sided Mann-Whitney test. (H) CD3⁺CD8⁺GZB⁺ cells as a fraction of cells from immunofluorescence imaging analyses of 5 low PI and 4 high PI tumors. One high PI sample had no CD3⁺CD8⁺ cells. GZB, granzyme B. Mean ± SEM. Unpaired, 2-sided Mann-Whitney test. (I) CD3⁺CD8⁺PD1⁺ cells as a fraction of CD3⁺CD8⁺ cells determined by immunofluorescence imaging analyses of 5 low PI and 4 high PI tumors. One high PI tumor had no CD3⁺CD8⁺ cells. Mean ± SEM. Unpaired, 2-sided Mann-Whitney test. (J) CD3⁺CD4⁺FoxP3⁺ cells (Tregs) as a fraction of CD3⁺CD4⁺ cells determined by immunofluorescence imaging analyses of 5 low PI and 5 high PI samples. Mean ± SEM. Unpaired, 2-sided Mann-Whitney test. (K) TMB from whole-exome data for a subset of low PI (n = 18) and high PI (n = 13) tumors. Filled symbols are the values for arm 2 tumors in the 2 clusters. Mean ± SEM. Unpaired, 2-sided Mann-Whitney test.

Figure 4

MPR associated with adaptive immune responses (A) Some GO pathways enriched among genes increased in expression, posttreatment, in arm 2 tumors with MPR (n = 14) relative to posttreatment arm 2 tumors without MPR (n = 10). (B‒F) Enrichment scores (xCell deconvolution¹¹) of posttreatment RNA-seq data from tumors that did or did not achieve MPR. No MPR (n = 10) and MPR (n = 14). Mean ± SEM. Unpaired, 2-sided Mann-Whitney test. (G) Posttreatment proliferation indexes (sum Z scores) determined by the expression of 113 genes and by a modified gene set of 73 genes that excluded the 40 genes that are part of the 140-gene set used for clustering (Table S3). No MPR (n = 10) and MPR (n = 14). Mean ± SEM. Unpaired, 2-sided Mann-Whitney test. (H) Some GO pathways enriched among genes increased in expression, posttreatment (n = 14) relative to pretreatment (n = 10) for tumors in arm 2 tumors that achieved MPR. (I and J) Enrichment scores (xCell deconvolution¹¹) of pre- (n = 10) and posttreatment (n = 14) RNA-seq data for DCs (I), CD8⁺ cells (J), and M1 and M2 macrophages (K) from tumors that achieved MPR. Pretreatment MPR (n = 10) and MPR (n = 14). Mean ± SEM. Unpaired, 2-sided Mann-Whitney test.

Figure 5

TCGA NSCLC tumors clustered by 140-gene signature are associated with survival in LUAD but not LUSC (A) Consensus clustering of 407 TCGA LUAD by expressions of the 140-gene set. Low PI (n = 196), high PI (n = 211). (B) Consensus clustering of 311 TCGA LUSC by expressions of the 140-gene set. Low PI (n = 132), high PI (n = 198). (C and D) DFS (C) and OS (D) of LUAD low and high PI clusters. (E) Multivariant Cox regression DFS in LUAD using covariates listed. Reference groups for EGFR mutation status are EGFR wild type, for gender is female, and for tumor stage is stage I. (F and G) DFS (F) and OS (G) of LUSC low and high clusters. (H) Multivariant Cox regression DFS in LUSC using covariates listed. Reference groups as in (D) and (E). (I) The 140-gene set enrichment scores. LUAD: low PI (n = 196), high PI (n = 211); LUSC: low PI (n = 132), high PI (n = 198). Wilcoxon signed rank test p values.

Figure 6

The 140-gene set associated with DFS (A‒E) Clustering of TCGA tumors: (A) melanoma; n = 103, low PI (n = 36) and high PI (n = 67); (B) breast; n = 893, low PI (n = 475) and high PI (n = 418); (C) prostate; n = 236, low PI (n = 179) and high PI (n = 57); (D) pancreas; n = 124, low PI (n = 71) and high PI (n = 53); and (E) colon; n = 297, low PI (n = 116) and high PI (n = 181). Percentage of tumors in low and high PI clusters are shown. The distributions of the 140-gene set scores are shown below. Wilcoxon test. (F‒J) DFS of the low and high PI clusters in (F) melanoma, (G) breast, (H) prostate, (I) pancreas, and (J) colon TCGA tumors.

Figure 7

Disease-free survival (A) Subjects grouped by the 140-gene set into low (n = 14) and high (n = 18) PI clusters, regardless of treatment arm. (B) Arm 1 subjects grouped by PI group. High PI (n = 8) and low PI (n = 8). (C) Arm 2 subjects grouped by PI group. High PI (n = 10) and low PI (n = 6). CI, 95% confidence interval.