Association between genetic mutations, cell of origin, and abundance of subcellular phenotypes in DLBCL TME. (A) Subphenotypes were created and labeled by reclustering cells using all markers. (Left) heatmap depicts the significant spatial colocalizations are depicted on the using an interaction distance of 15 microns and a permutation test (1000; P < .01) comparing significant spatial colocalizations (cyan), or significant avoidance (black) for each pairwise subpopulation as a total sum of signed interaction scores. (Right) Heatmap depicts the heterogeneity state/inducible marker expression (Z-score). The left-most adjacent graphs include cluster size correspondence, differential abundance association with molecular subtypes (C1-C5), IPI, treatment refractory, and double expressor status (|logFC|<10), and the average nearest neighbor distance to tumor per cluster is included (<15 microns). Statistically significant differences of cluster abundances/states were denoted following adjusted significance legend including shapes such as a triangle denoting BH q < 0.09, a cross shape denoting BH q < 0.05, and a box/cross denoting BH q < 0.01. For hypothesis testing cluster abundances related to Chapuy signatures, the size of a given significance shape represents abundance of a given phenotypic cluster relative to the abundance among other Chapuy signatures (%). Baseline T_REG ( cluster 4) (logFC=17.4, BH q = 3.2e-03) abundance was significantly enriched in C3 coordinate signature, with negative association in C2 (logFC= −20.4, BH q = 1.9e-04). PD-L1⁺ endothelial cell (4) abundance were positively associated with C3 (logFC = 23.4, BH q = 3.2e-03) and REF subjects (logFC = 10.9, BH q = 0.033). Proliferative endothelial cells (2) had increased abundance in C2 (logFC = 20.9, BH q = 5.1e-03) and inflammatory tumor (6) (logFC = 22.7, BH q = 0.024). Exhausted/inflammatory CD8 (6) had increased abundance (logFC = 29.4, BH q = 0.024). The expression intensity for each sample was standardized into a Z-scores, and the subsequent phenotype cluster expression profiles are depicted. (B) The volcano representation of the multivariate linear model identified pSTAT3 as statistically significantly overexpressed on highly suppressive T_REG comparing refractory to responders (BH q = 0.0295), similarly Ki67 on the same phenotype was underexpressed (BH q = 0.0295). The shapes of these 2 markers correspond to the adjusted significance legend (BH q < 0.05) used in cluster abundance hypothesis testing. (C) The multivariate Cox proportional hazards model indicates the x-axis as the log hazards estimate, and the y-axis are the T_reg family included in the model (95% CI). For each patient, the relative case proportions for a given phenotype, and the average pSTAT3/Ki67 expression (Z-score) corresponding to the given phenotype, were fit into the Cox hazards model using these patient level features.