Spatial genomics reveals cholesterol metabolism as a key factor in colorectal cancer immunotherapy resistance

Abstract

Immune checkpoint inhibitors (ICIs) have transformed the treatment landscape across multiple cancer types achieving durable responses for a significant number of patients. Despite their success, many patients still fail to respond to ICIs or develop resistance soon after treatment. We sought to identify early treatment features associated with ICI outcome. We leveraged the MC38 syngeneic tumor model because it has variable response to ICI therapy driven by tumor intrinsic heterogeneity. ICI response was assessed based on the level of immune cell infiltration into the tumor - a well-established clinical hallmark of ICI response. We generated a spatial atlas of 48,636 transcriptome-wide spots across 16 tumors using spatial transcriptomics; given the tumors were difficult to profile, we developed an enhanced transcriptome capture protocol yielding high quality spatial data. In total, we identified 8 tumor cell subsets (e.g., proliferative, inflamed, and vascularized) and 4 stroma subsets (e.g., immune and fibroblast). Each tumor had orthogonal histology and bulk-RNA sequencing data, which served to validate and benchmark observations from the spatial data. Our spatial atlas revealed that increased tumor cell cholesterol regulation, synthesis, and transport were associated with a lack of ICI response. Conversely, inflammation and T cell infiltration were associated with response. We further leveraged spatially aware gene expression analysis, to demonstrate that high cholesterol synthesis by tumor cells was associated with cytotoxic CD8 T cell exclusion. Finally, we demonstrate that bulk RNA-sequencing was able to detect immune correlates of response but lacked the sensitivity to detect cholesterol synthesis as a feature of resistance.

Keywords: MC38; PD-1; Visium; cholesterol; colorectal cancer; immunotherapy resistance; spatial genomics; spatial transcriptomics.

Figure 1

ICI responders have elevated immune infiltration and activation. (A) Experimental overview of ICI treatment and molecular characterizations. (B) Example IHC images of CD8 (red) and CD4 (brown) T cell markers. The columns correspond to treatment and response status: IgG control, aPD-1 Non-Responder (NR), and aPD-1 Responder (R). The top row displays the full tumor section, and the bottom row is a high resolution inset. Inset location is marked as a square in top row. (C) Quantification of CD8 positive cells in tissue compartments identified via digital pathology. IHC images were co-registered to the Visium H&E section, and positive cells in each capture spot were counted and normalized by the total number of spots per compartment and tumor section. (D) Principal components analysis (PCA) plot of bulk RNA-seq data for IgG control, aPD-1 non-responder (NR), and aPD-1 responder (R). (E) Volcano plot of differential gene expression between aPD-1 non-responder and aPD-1 responder tumors. Significant genes are marked in red and select immune related genes are labeled. (F) Gene set analysis (Fisher’s test) of the top significant genes upregulated in responders using Gene Ontology Biological Processes gene sets (109). Point size is proportional to significant genes count for each gene set.

Figure 2

Spatial transcriptomic map of ICI-treated MC38 tumors. (A) Spatial distributions of unique molecular identifiers (UMIs) per spot with the base Visium protocol (top) or optimized protocol with collagenase and dispase permeabilization step (bottom). (B) Gene expression capture metrics of the base and optimized spatial transcriptomics protocols. (C) UMAP of unsupervised clusters from 48,636 ST spots across 21 tumor sections from 16 MC38 tumors. Clusters represent tumor and stroma subsets, named based on differentially expressed genes ( Supplementary Table S2 ). (D) Spatial distribution of unsupervised clusters from (C) for a single tissue section (right) and its corresponding hematoxylin & eosin (H&E) stain (left). (E) Dot plot of differentially expressed biomarkers for each unsupervised cluster in the MC38 spatial atlas. Clusters are colored to match the legend above in (C). Dot size is proportional to the number of spots that express the gene, and color matches the z-scaled gene expression. (F) Treatment and response group composition based on unsupervised clusters present within each tumor. Proportions are first averaged across replicate sections, if applicable, and then treatment group.

Figure 3

Cholesterol pathway associated with aPD-1 non-responders. (A) Transcriptome changes as assessed by number of differentially expressed genes across tumor and stroma subsets for the aPD-1 responders and non-responders. Double slash indicates a scale break used for data visualization. (B) Volcano plot of genes that are differentially expressed between aPD-1 responders and non-responders in the Tumor 1 subset. Selected genes are labeled with color indicating direction of expression change, consistent with (A). (C) Pathway schematic of cholesterol regulation, synthesis, and transport. Genes identified via our spatial atlas as associated with non-responders are indicated (blue ovals). (D) Summary of cholesterol pathway genes that have significant expression changes associated with aPD-1 therapy resistance. Spatial atlas cell subsets are shown along the vertical axis. A dot is present if the gene is significantly upregulated in non-responders versus responders in that given subset. (E) Dot plot of cholesterol gene expression by subset. Dot size indicates the percent of spots in a subset that have expression of the given gene. Color represents the z-scaled average gene expression across the subset. The vertical axis labels from (D) extend to this figure. (F) Spatial gene expression plots of cholesterol pathway genes upregulated in non-responders depicted as a signature of 16 genes in (C-E) and calculated using UCell. Top row is H&E-stained section. Bottom row is the corresponding signature values. The two IgG samples show tumors with either high or low expression of the cholesterol gene signature. (G) Volcano plot of bulk RNA-seq data (from Figure 1E ) depicting only cholesterol signature genes. Dashed lines indicate the significance cut-offs for log fold change and adjusted p-values. Blue dots indicate the genes that pass the threshold for significance by adjusted p-value.

Figure 4

Location of cholesterol synthesis associated with dampened T cell response. (A) Framework to identify location-based changes in spatial gene expression that occur as a consequence of distance from a given feature of interest. Left panels: First, spots expressing high levels of the genes for a feature of interest are identified (indicated by star; Sig^high). Next, spots with low expression of genes for the feature of interest are identified (Sig^low). Non-tumor spots and spots with mid-signature expression are excluded from analysis to reduce confounding variables (grey spots). Distance from each Sig^high spot to the nearest Sig^low spot is calculated. Right panels: Representation of linear model used to identify gene expression of Sig^low spots as a function of distance from Sig^high spots. A coefficient is calculated for every gene to quantify the strength of the trend between expression and distance from a Sig^high spot. A positive coefficient indicates the expression increases with distance from a Sig^high spot, while a negative coefficient means the expression increases with proximity to a Sig^high spot. (B) Scatterplot of tumor spatial gene expression coefficients anchored on CD8 T cell Sig^high spots. Points are colored red if the difference between the coefficient in responders and non-responders is significant, and gray if it is not significant. The dotted lines indicate significance thresholds. (C) Scatterplot of tumor spatial gene expression coefficients anchored on cholesterol pathway Sig^high spots. Points are colored red if the difference between the coefficient in responders and non-responders is significant, and gray if it is not significant. The dotted lines indicate significance thresholds.