Spatial clustering reveals differences in tumor topology that associate with COO and TME abundance. (A) Neighborhood analysis of cells describes the local arrangement of the TME within the tumor area. The metric is calculated for each cell by locating the 5 nearest cells belonging to the immune TME, locating the centroid of those nearest neighbors, and measuring the distance from the centroid to the original cell. A smaller distance metric indicates that a cell is embedded within the immune TME; a longer distance denotes exclusion of tumor cells from immune cells. (B) The tumor neighborhoods were constructed relative to their distance to the TME, which can identify themes of immune activity or nonactivity within distinct tumor-centric regions. The histogram showing the ordered average distance from each tumor topology class to its nearest immune cells. Tumor topology classes were ordered by their distance/proximity to the TME (microns). Tumor_d (average centroid distance to immune [dist] = 47.9 µm) was defined as the “core” based on its furthest distance to immune cells. Tumor_f (dist = 30.9 µm), followed by tumor_i (dist = 24.1 µ) were next nearest clusters to immune cells and were labeled as the “mantle.” Tumor_c (dist = 18.9 µm) and tumor_h (dist = 18.9 µm) represented the boundary to the immune interface and subsequently labeled as “crust” because they were adjacent to both the TME and the mantle regions. Tumor_g (dist = 13.8 µm), tumor_e (dist = 13.7 µm), tumor_a (dist = 13.2 µm), and tumor_b (dist = 12.2 µm) had the shortest distance to nearest to TME but formed unorganized clusters and were labeled as “dispersed.” The x-axis denotes the topology cluster, and the y-axis denotes the average centroid distance to the TME (microns). (C) Intratumor spatial heterogeneity is depicted in a representative annotated image from case 26. The first inset shows tumor classes that are more intermixed with the immune cells. The second inset shows tumor spatial arrangement structures similar to the geological topography, with tumor_d situated at the “core,” tumor_f at the inner “mantle,” tumor_i at the outer “mantle,” and tumor_c at the “crust” of tumor clusters. (D) Nine classes of tumor topology were identified and were well distributed across cases, ordered from immune-cold to immune-hot. The y-axis denotes case proportion (%), and x-axis denote the cases. All cases were included that had less than 79% immune cell sample proportion present; case 20 was excluded in this analysis because it had approximately 97.5% immune cells with very sparse tumor cells present indicating that this topological model describes cases with at most 79% TME proportion. (E) Within the tumor-centric zones, the total significant attractions/repulsions (P < .01) of each major cell component (CD4/CD8/MAC/T_REG) were summed in each tumor topology. The x-axis denotes the topological regions ordered by distance to the TME from closest (tumor_b) to furthest (tumor_d). The y-axis denotes the total sum of the significant signed interactions (P < .01). Positive values indicate attraction, whereas negative values indicate significant repulsions. The CD4 were enriched within tumor core and dispersed regions but depleted in the mantle neighborhoods, whereas CD8 were depleted in the core and mantle regions but enriched in the crust and dispersed regions. Macrophages were found in the dispersed and crust regions but decreased in the mantle and core areas. The total signed interactions of each TME compartment significantly differed across tumor topologies (ANOVA P = 1.14e-02, F = 1.71, DF = 32) indicated significant heterogeneity of TME interaction across each topological zone. The line graph summarizes the trends of the observed spatial estimates depicted, with 95% CIs in Supplemental Figure 11A,F. Within each tumor-centric neighborhood, the total case proportion per subphenotype with significant neighborhood interaction (radius of interaction = 15 micron, P < .01) was computed. The x-axis denotes interactions for each tumor topology such that “a”/“b”/“e”/“g” were summarized into as “dispersed” topology, the “c”/“h”/“I”/“f” were summarized into as “mantle/crust” topology, and tumor “d” as the “core.” The y-axis denotes the case proportion of statistically significant interactions from the tumor topology (reference) to the TME phenotype (P < .01) and corresponding 95% CI. Supplemental Figure 11B depicts the T_REG subtypes. There was a statistically significant association between the total CD4/CD8/macrophage/T_REG spatial interactions per the mantle/core and crust/dispersed regions (P = .019, χ² = 9.3, df = 3). CD8 T cells and macrophages were 0.33 (95% CI, 0.15-0.61; P = .001) and 0.44 (95% CI, 0.19-0.91; P = .037) times as likely to form neighborhoods in the core/mantle compared with the crust/dispersed regions. (G) The tumor core had the fewest significant interactions with the immune phenotypes (radius of interaction = 15 microns). Averaging the number of significant attractions between the CD4 family level phenotypes and the tumor core (P < .01), there is a significant association with corresponding CD4 phenotype average CXCR3 (Z-score) expression level (P = .00198, adjusted R² = 0.9941, CXR3 estimate = 2.94; 95% CI, 2.37-3.50). The x-axis denotes the average CXCR3 (Z-score) expression corresponding to CD4 family phenotypes. The y-axis denotes the average number of significant interactions between the CD4 family and the tumor core (P < .01). Both the reference CD4 interaction with the tumor core, and the reference tumor core interaction with the CD4 phenotype directions of spatial interactions were averaged.