Researchers Hope Single-Cell Tumor Immune Atlas Can Improve Predictions in Precision Oncology

Although tumor immune microenvironments (TMEs) contribute significantly to disease progression and provide opportunities for targeted cancer therapies, a more precise roster of biomarkers is necessary to predict cancer prognosis and immunotherapy response based on the TME. Researchers analyzed data from hundreds of thousands of cells across 13 cancer types to create a single-cell tumor immune atlas that can help group patients by the immune cell composition of their particular cancer.

TMEs differ significantly depending on a patient’s specific cancer, but are promising targets for targeted cancer therapies. By collecting single-cell RNA sequencing (scRNA-seq) data sets and analyzing transcriptome profiles from both normal tissue and diseased tissue, researchers can gain an understanding of disease-specific phenotypes and generate high-resolution landscapes of those various cancer types.

“Phenotyping single cells from the TME has led to the identification of cancer-specific stromal cell states and supported their contribution to tumor progression,” the authors wrote in Genome Research. “Functional and integrative analysis further support dependencies of stromal and cancer cells and their predictive value for patient outcome.”

Cancer-associated fibroblasts and tumor immune cells have both been shown effective biomarkers and targets for therapy, for example.

To produce the tumor immune cell atlas, Nieto et al. collected scRNA-seq data from a total of 526,261 cells from 217 patients with 13 different cancer types, which included:

• breast carcinoma
• basal cellcarcinoma
• squamous cell carcinoma
• endometrial adenocarcinoma
• renal cell carcinoma
• intrahepatic cholangiocarcinoma
• hepatocellular carcinoma
• colorectal cancer
• pancreatic ductal adenocarcinoma
• ovarian cancer
• non-small-cell lung cancer
• cutaneous melanoma
• uveal melanoma

The data were analyzed separately by cancer type first before forming a pan-cancer reference atlas. The cells were grouped by cell identity to allow subsetting of the immune compartment.

They used canonical correlation analysis, which identifies common cellular phenotypes and allows researchers to merge data sets from different studies and technologies for one large analysis, to combine data from 317,111 immune cells. “We hypothesize that joining cells from different cancer types into a single reference data set may define commonalities and harmonize annotations between studies,” the authors wrote.

From the overall group, the immune cells were stratified into 25 types, or clusters, based on major immune cell types. These included 12 T cell clusters; 5 macrophage/monocyte clusters; 3 dendritic cell clusters, 3 B- and plasma B cell clusters, 1 natural killer cluster, and 1 mast cell cluster. The clusters varied in size, yet nearly all cancer types were present within each cluster, confirming that there are common characteristics in immune cell compositions across disease types.

When a random forest classifier was trained to predict cell annotation based on the 25 signatures, a 5-fold cross-valuation was done to assess the clusters’ biases and variance. The mean accuracy was 0.76 and the kappa statistic was 0.75 across the folds, which are 3- and 4.2-fold increases from random signatures and comparable values from other high-quality atlases. Some of the misclassification was within cell types of the same lineage that share several markers, the authors noted, adding that deeper clustering could identify even more relevant cell states.

“Using the classifier, the pan-cancer immune classification system could be extended to additional cancer types and drive the design of basket clinical trials in which a common immune stratification and recruitment framework is applied across cancer types,” the authors wrote.

The distribution of immune cell types is another factor that may play into predicting immunotherapy outcomes. Therefore, the study authors combined the single-cell reference immune profiles with spatial transcriptomics (ST) data utilizing SPOTlight, which identifies type-specific gene expression signatures from scRNA-seq data that can identify ST spots. These data showed clear regionalization of distinct immune cell types and states.

Overall, the authors hope that the atlas they have created will be applied in multiple facets of precision oncology.

“We provide a framework for an immune-based patient stratification, the feasibility to integrate newly generated patient single-cell data, and a toolbox to map immune cells directly in tumor sections,” they wrote. “Following the integration of clinical cohort single-cell studies with patient outcome and response metadata, we expect the atlas to be predictive for patient prognosis and immunotherapy response to a level that greatly exceeds currently applied stratification strategies.”

Reference

Nieto P, Elosua-Bayes M, Trincado L, et al. A single-cell tumor immune atlas for precision oncology. Genome Res. Published online September 21, 2021. doi:10.1101/gr.273300.120