IMMUNE AND MOLECULAR CORRELATES OF RESPONSE TO IMMUNOTHERAPY REVEALED BY BRAIN-METASTATIC MELANOMA MODELS

Abstract

Despite the promising results of immune checkpoint blockade (ICB) therapy, outcomes for patients with brain metastasis (BrM) remain poor. Identifying resistance mechanisms has been hindered by limited access to patient samples and relevant preclinical models. Here, we developed two mouse melanoma BrM models that recapitulate the disparate responses to ICB seen in patients. We demonstrate that these models capture the cellular and molecular complexity of human disease and reveal key factors shaping the tumor microenvironment and influencing ICB response. BR1-responsive tumor cells express inflammatory programs that polarize microglia into reactive states, eliciting robust T cell recruitment. In contrast, BR3-resistant melanoma cells are enriched in neurological programs and exploit tolerance mechanisms to maintain microglia homeostasis and limit T cell infiltration. In humans, BR1 and BR3 expression signatures correlate positively or negatively with T cell infiltration and BrM patient outcomes, respectively. Our study provides clinically relevant models and uncovers mechanistic insights into BrM ICB responses, offering potential biomarkers and therapeutic targets to improve therapy efficacy.

Figure 1.. Generation of immunocompetent melanoma BrM models that exhibit ICB responses similar to brain metastases in patients.

(A) Representative stereomicroscope images of whole brains (left) and zoomed-in image of BrM (inset) from BR1 and BR3 bearing animals. (B) Percentage of mice with BrM per experiment; dots depict independent experiments. (C) BrM count (left) and BrM/extracranial metastases ratio (right) per mouse; dots depict individual animals within an experiment. (D-G) BR1 or BR3 bearing animals injected with PBS (vehicle) or treated with isotype control (Iso), anti-CTLA-4 (a-CTLA-4), anti-PD-L1 (a-PD-L1), or anti-PD-L1 + anti-CTLA-4 combination therapy (Combo). (D) BR1 BrM count per individual brain (left), total brain metastatic area per individual brain (middle), and size of individual BrM lesion (right). (E) Representative brain stereomicroscope images from Iso- and Combo-treated BR1 bearing animals. (F-G) Same as D-E but for BR3 bearing mice. B-C shown as mean ± SEM, data from 9 (B) or 3 (C) experiments combined, n=3–12 mouse/group/experiment. D-G representative of 2 independent experiments for each model, data shown as mean ± SEM, n=9–12/group/experiment. *p <0.05, ***p < 0.001, ****p < 0.0001. See also Figure S1 and Table S1.

Figure 2.. Distinct BrTME composition is associated with different responses to ICB.

High-parametric spectral flow cytometry analysis of brain leukocytes from BR1 and BR3 bearing animals untreated or combo treated. (A) Absolute number of leukocytes (live CD45⁺) per gram of brain tissue. (B) UMAP projection of total BrM leukocytes. Main populations indicated: resident microglia, infiltrating mononuclear phagocytes, neutrophils (Neu), B cells, innate lymphoid cells (ILCs), and T cells. (C) Density plots showing cell distribution in untreated (top) and treated (bottom) mice from BR1 (left) and BR3 (right) models. (D) Principal Component Analysis (PCA) of main immune cell population frequencies in total leukocytes for each sample. (E) Proportion of microglia (Mg; purple) and infiltrating cells (CD45^hi; orange) in total leukocytes. Proportion of indicated populations within total infiltrating cells (inset). (F) Absolute number of infiltrating cells (left) and microglia (right) per gram of brain. (G) Absolute number of total T cells per gram of brain. (H) Absolute number of indicated T cell subsets per gram of brain. A, F, G, and H data shown as mean ± SEM, n = 8–10/group/experiment. *p <0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. See also Figure S2 and Table S2.

Figure 3.. Distinct CD8⁺ T cell subsets infiltrate ICB-responder compared to ICB-resistant BrMs.

(A-E) scRNA-seq analysis of CD8⁺ T cells from untreated BR1, BR3, and PBS brains. (A) Volcano plot of BR1 versus BR3 differentially expressed genes (DEGs). Significant DEGs are indicated (BR1: red, BR3: blue) (B) UMAP projection showing cell distribution in each group. (C) CD8⁺ T cell clustering. UMAP projection (left) and proportion of each cluster per sample (right). (D) Dot plot showing the expression of selected genes among clusters. (E) Projection of CD8⁺ T cell clusters (right) onto the single-cell reference atlas (left) for T cell states ProjecTILs. (F-J) High-parametric spectral flow cytometry analysis of CD8⁺ T cells from indicated samples. (F) Density plots showing cell distribution in untreated (top) and treated (bottom) BR1 (left) and BR3 (right) models. (G) UMAP projection and clustering of indicated CD8⁺ T cell clusters among total CD8⁺ T cells. (H) Proportion of indicated CD8⁺ T cell clusters in total CD8⁺ T cells. (I) Heatmap depicting expression level of indicated surface markers (scaled by marker) in each CD8⁺ T cell cluster. (J) Absolute number of indicated T cell clusters per gram of brain. A-E n = 2–3/group. F-J n = 8–10/group/experiment. J data shown as mean ± SEM, *p <0.05, ****p < 0.0001. See also Figure S3 and Tables S3–S4.

Figure 4.. Unique reactive microglia signatures are found in ICB-responder BrMs.

(A-E) scRNA-seq analysis of microglia from untreated brains. (A) Volcano plot of BR1 (red) versus BR3 (blue) DEGs. (B) Gene set enrichment analysis (GSEA) showing selected pathways among the top 10 Hallmark (HP), Gene Ontology (GO), and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways enriched in BR1 (red) microglia versus BR3 (blue) (see Methods). NES: Normalized Enrichment Score. (C) UMAP projection of total microglia showing cell distribution among groups. (D) UMAP projection clustering (left) and proportion per sample (right) of indicated microglia clusters in total microglia. (E) GSEA showing selected HP, GO and KEGG top pathways enriched in each microglia cluster. NES: Normalized Enrichment Score in each cluster. (F-J) High-parametric spectral flow cytometry analysis of microglia from untreated or combo-treated BR1 and BR3 brains. (F) UMAP projection and unsupervised clustering. (G) Heatmap representing expression level of indicated surface markers (scaled by marker) for each cluster with hierarchical clustering indicated on left and markers on top. (H) Proportion of indicated microglia states in total microglia per group. (I) Proportion of indicated microglia clusters in total microglia per group. (J) Absolute number of indicated microglia clusters per gram of brain. A-E n = 2–3/group/experiment. F-J n = 8–10/group/experiment. J data shown as mean ± SEM, n = 8–10/group/experiment. *p <0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. See also Figure S4.

Figure 5.. BR1 and BR3 tumor cells differentially express molecular programs that are present in human BrM tumor cells.

scRNA-seq analysis of tumor cells from untreated BR1 and BR3 brains. (A) GSEA showing selected GO, KEGG and REACTOME pathways enriched among the BR1-specific and BR3-specific genes (see Methods). (B) Metascape-associated pathways for BrM metaprograms (MP) identified in BR1 and BR3 tumor cells. (C) Per cell basis analysis of MP activity per group. (D) Correlation-based network analysis between mouse MPs derived in our study (green circles) and human MPs from Biermann et al. 2022 (brown rectangles), using three patient cohorts. Edge thickness and color represent the number of cohorts in which each MP pair is correlated. (E) Spearman correlation between mouse versus human MPs activities in the indicated data sets. x- and y-axis depict spearman correlation coefficient values (F) UMAP projection of tumor cells based on Z-scores of MP activities showing BR1 (red) and BR3 (blue) cell distribution (left) and tumor clusters distribution (right). (G) Proportion of tumor clusters per group. (H) Heatmap depicting Z-score of MP mean activity across malignant cells in each cluster (columns). (I) Per cell basis analysis of MP activity within each cluster from each group. A-C and F-I n = 2–3/group, D-E n = 2–3 mouse/group (our study), n= 25 (Biermann et. al), n=14 (Smalley et. al), n=27 (Alvarez-Breckenridge et. al). See also Figure S5 and Table S5. *p <0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Figure 6:. BrM tumor cells dictate the TME differences between ICB-responder BR1 and ICB-resistant BR3.

(A-C) Cell-cell interaction analysis from scRNA-seq data of untreated BR1 and BR3 brains. (A-B) Circos plot showing the overall predicted interaction strength between indicated cell types (A) and indicated clusters within each cell type (B). The outside border color of the circos represent the sender and receiver cell populations. The width of the lines inside the circos plot indicates the interaction strength. The color of the lines inside the circos plot indicates the sender cell population. (C) Heatmap showing the top and bottom 10% most differential interaction (by strength) of ligand-receptor (L-R) pairs between models for indicated cell type pairs (see methods). Interactions stronger in BR1 versus BR3 are depicted in purple, interactions stronger in BR3 depicted in green. Numbers in parentheses indicate the number of different L-R pairs. No number indicates that only one L-R pair is present. (D) Dot plot showing the expression of selected chemokines in each immune cell population. (E) Cytokine/chemokine protein profile (normalized measurement) from untreated BR1 and BR3 whole brain lysates. (F) Correlation-based network analysis of total brain metastatic area (BrM area; determined by machine learning algorithm), microglia (Mg.A – Mg.J) and CD8⁺ T cell (CD8.F1 – CD8.F5) cluster cell count per gram of brain (quantified by spectral high-parametric flow cytometry analysis) from untreated BR1 and BR3 brains. Lines represent spearman correlations. A-D n = 2–3/group, E n = 5–6/group, K n = 8–10/group/experiment. *p <0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. See also Figure S6.

Figure 7.. BR1 and BR3 signatures are associated with BrM patient CD8⁺ T cell infiltration and survival.

(A-C) scRNA-seq analysis of CD8⁺ T cells from Biermann et al. 2022 untreated melanoma BrM patient biopsies. (A) UMAP projection and clustering (left). Dot plot showing the expression of selected differentially expressed T cell markers (right). (B) Spearman correlation of MP4 (BR1-MP) or MP6 (BR3-MP) activity and CD8⁺ T cell percentage among total cells per sample within each biopsy. (C) Heatmap representing spearman correlations between proportion (% of total cells) of individual CD8⁺ T cell clusters (row) and the indicated MP activity (column) within each biopsy. (D-E) Pseudobulk RNA-seq analysis of untreated and treated patient pan-cancer BrM biopsies from Sun et al. 2023. Spearman correlations of the indicated signatures’ expression (see methods). (F) Expression of “tumor-reactive” human CD8 T cell signatures defined by the indicated studies (row) within each mouse CD8⁺ T cell cluster from our study (column). Heatmap depicts Z-score expression across rows. (G) Progression-free survival of BrM patients stratified by median expression of the indicated signatures. BR1 and BR3 signatures are indicated in red and blue respectively. A-C n= 10 patients. D-E, G n=17 patients. F n = 2–3/group. See also Figure S6 and Tables S5–S6.