A gene expression signature of TREM2hi macrophages and γδ T cells predicts immunotherapy response

Identifying factors underlying resistance to immune checkpoint therapy (ICT) is still challenging. Most cancer patients do not respond to ICT and the availability of the predictive biomarkers is limited. Here, we re-analyze a publicly available single-cell RNA sequencing (scRNA-seq) dataset of melanoma samples of patients subjected to ICT and identify a subset of macrophages overexpressing TREM2 and a subset of gammadelta T cells that are both overrepresented in the non-responding tumors. In addition, the percentage of a B cell subset is significantly lower in the non-responders. The presence of these immune cell subtypes is corroborated in other publicly available scRNA-seq datasets. The analyses of bulk RNA-seq datasets of the melanoma samples identify and validate a signature - ImmuneCells.Sig - enriched with the genes characteristic of the above immune cell subsets to predict response to immunotherapy. ImmuneCells.Sig could represent a valuable tool for clinical decision making in patients receiving immunotherapy.

. Cell abundance comparison stratified by treatment schemes. The scRNA-seq dataset -GSE120575 was used in this analysis. The percentages (% of CD45 + cells) of each of the 23 single-cell clusters for the responders (R) and non-responders (NR) groups of melanoma samples collected in the three scenarios. (a) before anti-PD-1 treatment; (b) after anti-PD-1 treatment; (c) after anti-CTLA4+anti-PD-1 treatment. No enough single cells were available for comparision between responders and non-responders for other scenarios. For the after anti-PD-1 treatment melanoma samples, the denominators for the R and the NR groups are 1524 and 6334, respectively; for the after anti-CTLA4 plus anti-PD-1 treatment melanoma samples, the denominators for the R and the NR groups are 1315 and 1190, respectively.
Supplementary Figure 4. Fraction of each macrophage subsets. The scRNA-seq dataset -GSE120575 was used in this analysis. Proportions of inflammatory macrophages (cluster 6), TREM2 hi macrophages (cluster 12), and Immunoregulatory related macrophages (cluster 23), the three macrophage subsets in immune cells from the melanoma tumor samples.

All Macrophages
Immunoregulatory related M (Cluster 23, 2.5%) TREM2 hi M (Cluster 12, 35.9%) Supplementary Figure 5. Pathway analysis for macrophage cluster 6. The scRNA-seq dataset -GSE120575 was used in this analysis. IPA analyses based on the list of differentially expressed genes between cluster 6 and other macrophages. The results revealed that inflammatory response was significantly activated with a large number of overexpressed inflammatory marker genes in cluster 6 macrophages (adjusted P = 3.93E-10, activation Z score = 2.01). Cluster 6 macrophages population was thus identified as the 'Inflammatory M'.
Supplementary Figure 6. Gene ontology enrichment analysis of three macrophages subsets. The scRNA-seq dataset -GSE120575 was used in this analysis. Gene ontology enrichment analysis of reactome pathways in (a) Inflammatory macrophages, (b) TREM2 hi macrophages, and (c) Immunoregulatory related macrophages infiltrating the melanoma tumor samples from patients subjected to ICT. Size of the circles is proportional to the fold difference.  Figure 7. A 40-gene expression signature that can characterize the TREM2 hi macrophage population. (a) The scRNA-seq dataset -GSE120575 was analyzed, which generated the heatmap of the expression of a 40-gene signature representing the TREM2 hi macrophage population. In the boxplots for (b) GSE78220 dataset and (c) GSE91061 dataset, the GSVA scores of the TREM2 hi macrophage geneset were significantly higher in the ICT non-responding tumors than the responding tumors. Center line, median. Box limits, upper and lower quartiles. Whiskers, 1.5 interquartile range. Points beyond whiskers, outliers. For (b) and (c), the two-sided t-tests were performed with no adjustment for multiple comparisons. (d) Violin plot showed that the actitiy of this gene set is higher in the TREM2hi macrophages compared to the other macrophages in the GSE120575 dataset.  Figure 10. The enrichment of the ImmuneCells.Sig signature for the characteristic genes of the immune cell subpopulations. The dataset -GSE78220 was used in this analysis. This ICT outcome signature was positively enriched for the characteristic genes of the (a) TREM2 hi M, (b) Tgd_c21, and negatively enriched for the (c) B_c22.

Regulation of expression of SLITs and ROBOs
Supplementary Figure 11. Evaluation of the ImmuneCells.Sig signature. The dataset -PRJEB23709 was used in this analysis. The performance of the ImmuneCells.Sig signature in predicting ICT responders based on the pre-treatment melanoma biopies from patients subjected to different ICT regimen. ImmuneCells.Sig can accurately distinguish responders from non-responders in both Pre_anti-PD-1 and Pre_Combo subgroups (anti-PD-1 plus anti-CTLA-4) as can be seen in the ROC (receiver operating characteristic) curves of the (a) PRJEB23709_Pre_anti-PD-1 subset and (b) PRJEB23709_Pre_Combo subset.