Mouse experiments

All animal experiment protocols were a performed at Yale University in accordance with University guidelines and approved by Institutional Animal Care and Use Committee (IACUC). Male C57BL/6J (RRID:MGI:3028467) were purchased from the Jackson Laboratory (Bar Harbor, ME). For tumor challenge, 5*105 YUMMER1.7-GFP/Luciferase (YUMMER1.7-GL) cells in 100 mL sterile PBS were injected into eight-week old mice. Length, width, and height measurements were recorded starting 7 days post tumor injection, and tumor volume was calculated as 0.5233 x length x width x height. Mice were euthanized when their tumor reached 1000 mm3 or when ulcerated, in accordance with Yale Office of Animal Research Support Committee guidelines. 10mg/kg a-PD-1 (RMP1-14, Bio X Cell, #BE0146, RRID:AB_1094905), 10mg/kg a-CTLA4 (9H10, Bio X Cell, #BE0131, RRID:AB_10950184), 0.32mg/kg DR18 and rat IgG2a isotype control (Bio X Cell, #BE0089, RRID:AB_1107769) were administered via intraperitoneal (IP) injection every 3 days beginning 7 days after initial tumor challenge. 

Single cell RNA sequencing

YUMMER1.7-GL tumors were harvested from mice 12 days after initial tumor challenge, dissociated, and pooled into single cell suspensions with their biological replicates. The successful immune clearance group consisted of tumor treated with DR-18 or combination of antibodies targeting PD-1 and CTLA4 as described above, and the unsuccessful immune clearance group consisted of vehicle treated tumors. The following populations were then purified by sorting: P1: GFP-CD45+CD3+ (T cells), P2: GFP-CD45+CD3- (non-T immune cells), P3: GFP+/-CD45-CD3- (tumor and stromal cells). P1, P2 and P3 for each sample were then mixed back together at a 2:1:1 ratio, respectively. 5000 cells from each of the mixed sorted samples for each condition were loaded onto 10x Genomics Chromium System. Library preparation was performed using 10x Genomics reagents according to the manufacturer’s instructions and was performed by the Yale Center for Genome Analysis (YCGA) and passed QC. Libraries were sequenced using an Illumina HiSeq 4000 (one library/lane) at the YCGA. Samples were processed using the Cellranger software suite commands cellranger mkfastq for processing raw call files into fastq files. Cellranger count was used to align reads to a custom mm10 reference modified to include eGFP (marking tumor cells), to filter reads, and to generate a cell-by-gene matrix for each sample library. Libraries were aggregated using cellranger aggr without normalization to generate a single cell-by-gene matrix. Based on Gapdh expression, the top 14000 ranked by nUMI were retained for analysis. The Seurat package for R v.2.3.4 was used to process the matrix and perform downstream analysis. Expression values were log-normalized with a scaling factor of 104, and the 2509 most variable genes were detected and used for further analysis with the FindVariableGenes function. Values were scaled to number of UMIs and percent mitochondrial genes. The FindClusters command was used to perform a shared nearest neighbor (SNN) modularity optimization-based clustering algorithm using a resolution of 1.0, and tSNE dimensional reduction was calculated on the first 50 principal components to visualize data. Clusters consisting of cells with low/null expression of Gapdh and Eno1 (non-cells), or co-expression of cell type exclusive markers (doublets) such as Cd3d and Cd68 were removed from further analysis by the SubsetData command, and variable genes were re-identified, data were re-scaled and PCA clustering and tSNE were re-run as described. Cell type assignments for each cluster were verified by comparing with ImmGen datasets. T cells were subsetted and re-analyzed separately as described above. Cluster frequencies by library were normalized to number of cells per library and column plots were generated using ggplot2 v.3.2.0. Gene expression and UMAP plots were plotted using ggplot v.3.2.0. For heatmaps, mean scaled expression values of each gene were calculated per cluster and plotted using pheatmap v.1.0.12 with values scaled by row (gene). Cell cycle scoring was performed using the Seurat CellCycleScoring command using mouse gene sets orthologous to previously described human gene sets.