SOURCE

WHY？

• Immune microenvironments vary profoundly between patients and biomarkers for prognosis and treatment response lack precision
• To pinpoint predictive cellular states of tumor immune cells and their spatial localization

HOW？

• Analyzing >500,000 cells from 217 patients and 13 cancer types
• Data projection: Seurat's anchor-transferring method
• Using SPOTlight to combine single-cell and spatial transcriptomics data and identifying striking spatial immune cell patterns in tumor sections
• ShinyApp (in progress) to project external data and to apply the immune classifier

GET WHAT?

GET 1: Generating a tumor immune cell atlas

• Collected scRNA-seq datasets from 13 different cancer types, 217 patients and 526,261 cells

• Immune cells clustered by cell identity rather than patient origin: integrated 317,111 immune cells using canonical correlation analysis =>  25  clusters

Get 2: Tumor subtype classifier

For Current: to establish a pan-cancer immune classification system

• used immune cell type and state frequencies of the reference atlas as input for similarity assessment across the 13 cancer types
• A hierarchical k-means clustering using immune cell proportions as features defined six clusters with largely different compositions (almost all cancer types were presented in each cluster)

For future: to facilitate the classification of immune profiles

• trained a RF(random forest) classifier with the 25 immune cell population achieving a highly accurate classification
• using the classifier, the pan-cancer immune classification system could be extended to additional cancer types

GET 3: A resource for immune cell annotation

To demonstrate the potential value of the atlas

The applicability of the atlas as reference across different cancer types

• First: Project cells onto atlas using a reference-based projection (Fig. A)

• Third: Check correlation (Fig. C)

The applicability of the atlas as reference across species

• two liver metastases derived from mouse CRC organoids
• main subtypes and specific subpopulations could also be assigned using the human reference

GET 4: Spatial localization of immune cells in tumor sections

Spatial distribution of immune cells is important for ICI (immune checkpoint inhibitors) response

Single-cell reference atlas immune profiles + Spatial transcriptome data

SPOTlight : non-negative matrix factorization (NMF) based spatial deconvolution framework

Analysis  of oropharyngeal squamous cell carcinoma (SCC)

• cluster 1/2 (cancer cells) is surrounded by cluster 0 (stroma) and cluster 3 (immune cells)

• cluster1/2 presented a similar immune infiltration pattern, with an enrichment of proliferative T-cells and SPP1 macrophages

• cluster 3 presented a distinct immune infiltration pattern characterized by an enriched presence of (proliferative) B-cells

  cluster 0 harbored regulatory T-cells and terminally exhausted CD8 T-cells and was specifically enriched in M2 macrophages and naive T-cells.

  

Analysis of ductal breast carcinoma (BC)

• also get a cancer-specific regional distribution:

• subclonal was directly associated with local enrichment of distinct immune cell states

• Foresee the regional distribution of immune cell types to become an important feature for the prediction of immuno-therapy outcome.