Fig. 2: Immune-driven transcriptional heterogeneity implicates diverse immune cell populations and highly localized immune activation or suppression.

a Unsupervised hierarchical clustering based on the top 1000 most variant (mean absolute deviation) genes across all samples of the on-PD-1 inhibitor tumor, demonstrating limited associations between clustering of samples and tumor location based on the general transcriptome, but apparent association between high immune infiltrate and location at the tumor margin. b Heatmap of immune signature gene sets (from Rooney et al.¹³) across tumor sub-regions demonstrate dispersed pockets of immune activation or suppression throughout the tumor, wherein immune-high samples (e.g., 4A1/4A2, 6A5, and 8A6) are physically distant from each other within the tumor mass. IHC-based immune infiltrate and ESTIMATE immune scores (top) and IHC-based tumor sample location (bottom) are indicated. c Immunohistochemical marker inter-correlations demonstrating generally diverse representation of immune cell types when infiltrates are present. Data are Spearman’s correlation values (with Benjamini–Hochberg correction; only p < 0.05 are shown) indicated according to the color scale shown. d Correlation of most variably abundant proteins measured by reverse-phase protein array, revealing two main modules of highly correlated molecules. Proteins displaying statistically significant (FDR p < 0.10) correlation with immune infiltrate are indicated by * and color (blue = anti-correlated with immune infiltrate, red = directly correlated with immune infiltrate). Data are Spearman’s correlation values (with Benjamini–Hochberg correction; only p < 0.05 are shown) indicated according to the color scale shown. e Sample-wide similarity of immune activity was estimated by calculation of the distance matrix between samples using the immune activation signature expression data; lines connect samples in the top quartile of similarity scores, demonstrating global immune signature similarities that are not restricted by intratumoral location. f Three-dimensional spatial mapping of subregion Cytolytic activity signature, Type-I IFN response signature and Co-inhibition, T-cell signature scores derived from transcriptomic data, in the manner of Rooney et al.¹³. Data map the geometric mean of genes included in each gene set onto three-dimensional space representing the tumor slices, with color and height indicating expression value (higher expression = red peaks, lower expression = blue troughs).