Tumor cell phenotypes predict response to treatment. (A) UMAP of reclustered tumor cells from the scRNA-seq data set. (B) Strategy used to identify baseline biomarkers. Genes differentially expressed between tumor cells from patients with higher (n = 6) vs lower (n = 6) distant response at the injected tumor site were selected. Genes whose patient-level averages at both sites correlated with outcome were further evaluated. (C) Correlation between CD47 (left) and CD37 (right) expression in tumor cells at baseline and percentage distant tumor reduction at the injected tumor site. See supplemental Figure 5B for uninjected site data. (D) Bar plot of correlation coefficients between selected pathways at baseline and percentage distant tumor reduction. Bars are colored by negative log10 of the P value. See supplemental Figure 5C for uninjected site data. (E) Comparison of pretreatment (pre Tx) and on-treatment (day 8) scores for the indicated gene sets at the injected tumor site. (F) Tumor cell MHCII gene set score on day 8 at the injected site correlates with percentage distant tumor reduction. Correlations in (C), (D), and (F) were assessed using the Spearman test. P values in (E) were calculated using 2-sided paired Wilcoxon rank-sum tests: ∗P < .05; ∗∗P < .01; ∗∗∗P < .001; ∗∗∗∗P < .0001. P values are only shown if P ≤ .05. BCR, B-cell receptor; FC, fold change; MHC I, antigen presentation via MHCI; MHC II, antigen presentation via MHCII. UMAP colors in (A) and data point colors in (C) and (F) correspond to individual patients as in (A) and in Figure 2(A-B).