Fig. 4: Cancer-myeloid communications stimulate pro-tumor M2-like myeloid differentiation in CDK4/6i -resistant tumors.

A Top-left: UMAP of myeloid phenotypic heterogeneity across discovery/validation cohorts. Cells (points) with similar transcriptomic profiles clustered by ImmClassifier subtype (color). Bottom-left: Major axis of myeloid phenotypic variation reflects polarization from monocyte progenitor (gray) towards M1-like (immune-activating;blue) or M2-like (pro-tumor;red) phenotypes (polarization = pseudotime divergence;see methods). Black curves = pseudotime trajectory branching (M1/M2-like divergence). Right panel: Myeloid cell phenotypes were compared between resistant/sensitive (red/blue) tumors receiving combination ribociclib or letrozole alone in the discovery/validation cohort. Ribociclib-resistant tumors had greater M2-like differentiation (Hierarchical random effects model: Combination ribociclib:estimate = 0.337,se = 0.11, df = 32.66, t  =  3.04, p = 0.0046;Letrozole alone:estimate = −0.30,se = 0.20, df = 20.13, t = −1.52, p = 0.145). B Left: Box plots showing increased M2-like myeloid differentiation in resistant versus sensitive tumors (red/blue) early (day 0-14) in combination ribociclib treatment (top panel)(linear model:estimate = 0.33,se = 0.11, df = 34, t = 2.99, p = 0.005) with no cohort-specific difference (estimate=0.058,se=0.11, df=34, t = 0.52, p = 0.61). Points=mean M2-like differentiation per tumor across early timepoints. No significant (NS) difference between letrozole resistant/sensitive tumors (bottom panel)(linear model:estimate = −0.23,se = 0.21, df = 21, t = 1.11, p = 0.28). Right: Box plots showing lower M1-like (immune-activating) myeloid cell proportion (M1/(M1 + M2)) in ribociclib-resistant tumors early in treatment (linear model:estimate = −0.24,se = 0.12, df = 64, t = −2.008, p = 0.048) with no cohort-specific difference (estimate = 0.21,se = 0.18, df = 64, t = 1.17, p = 0.25). No difference in M1-like myeloid proportion between letrozole-resistant/sensitive tumors (linear model:estimate=0.14, se = 0.14, df = 37, t = 0.97, p = 0.34). Points = mean early M1-like myeloid proportion (60 tumors;multiple myeloid cell to estimate proportion). Box elements = median(center line), upper/lower quantiles(hinges),1.5*inter-quartile range(whiskers). Sample size for A/B:n = 27127 myeloid cells (Discovery:10940+Validation:16187 cells),167/173 biopsies,61/62 tumors (combination ribociclib = 37+letrozole alone = 24). C Schematic of cancer-myeloid communication analysis. For each tumor sample, the average strength of communication (via each LR pathway) sent by heterogeneous cancer populations to myeloid cells was measured. Significant pre-treatment communication differences between resistant/sensitive tumors were verified in the discovery/validation cohorts (log-linear regression+FDR-adjusted ANOVA). D Heatmap showing pre-treatment cancer-myeloid communications targeting M2-like macrophage differentiation (columns) strengthened in tumors resistant (red row annotation) versus sensitive (blue row annotation) to combination ribociclib but not letrozole alone (right-left panels) in the discovery/validation cohorts (top-bottom panels; linear model statistics in Supplementary Data 21–24). Coloration = cancer-myeloid LR-specific communication strength, showing heterogeneous pathways activated across tumors (white = no signaling detected). Sample size:268155 cancer+myeloid cells (Discovery:10940 myeloid+110568 cancer;Validation:16097 myeloid+130550 cancer),21279 genes,1444 LR pathways in 167 biopsy samples (Discovery:biopsies = 86, patients = 34;Validation:biopsies  = 81, patients = 27). All statistical tests two-sided. Source data are provided as a Source Data file.