High-dimensional immunotyping of tumors grown in obese and non-obese mice

Abstract

Obesity is a disease characterized by chronic low-grade systemic inflammation and has been causally linked to the development of 13 cancer types. Several studies have been undertaken to determine whether tumors evolving in obese environments adapt differential interactions with immune cells and whether this can be connected to disease outcome. Most of these studies have been limited to single-cell lines and tumor models and analysis of limited immune cell populations. Given the multicellular complexity of the immune system and its dysregulation in obesity, we applied high-dimensional suspension mass cytometry to investigate how obesity affects tumor immunity. We used a 36-marker immune-focused mass cytometry panel to interrogate the immune landscape of orthotopic syngeneic mouse models of pancreatic and breast cancer. Unanchored batch correction was implemented to enable simultaneous analysis of tumor cohorts to uncover the immunotypes of each cancer model and reveal remarkably model-specific immune regulation. In the E0771 breast cancer model, we demonstrate an important link to obesity with an increase in two T-cell-suppressive cell types and a decrease in CD8 T cells.

Keywords: Batch correction; Breast cancer; Obesity; Pancreatic cancer; Suspension mass cytometry; Tumor immunology.

Fig. 1.

Experimental design and analysis pipeline for mass cytometry data for immune infiltrate of seven CyTOF batches from five murine tumor models. (A) Cartoon and timeline of experimental design, data collection, data preprocessing and analysis. For each batch, n=chow/HFD. E0771_1 (n=4/4), E0771_2 (n=5/4), Wnt1 (n=6/6), TeLi (n=5/5), C11_1 (n=5/4), C11_2 (n=5/5), UN-KC (n=5/6). UND, uniform negative distribution. For B and C, each scatter dot plot point is from a different animal (mean±s.d.). Unpaired Student's t-tests were not adjusted, with s.d. assumed. (B) Representative mouse weights for male (n=5/4) and female (n=5/4) C57Bl/6 mice on chow and HFD. Weights were collected at 16 weeks before tumor cell injection. (C) Tumor masses from the seven batches for all experimental tumors run on CyTOF. An open square with ‘X’ means that tumor was removed from further analysis due to poor viability in the CyTOF dataset, fewer than 5000 live CD45⁺ cells. (D) Representative gating strategy for identifying live CD45⁺ tumor-infiltrating leukocytes. NA, nucleic acid; marks DNA and RNA in nucleated cells. Cis, cisplatin; used as a membrane exclusion molecule for the viability assay. (E) Sankey plot visualization of CD45⁺ live cells from total raw events collected for each experimental batch.

Fig. 2.

Batch correction algorithm testing. All plots were generated from normalized arcsinh-transformed live CD45⁺ cells. Transformed datasets were warp and range corrected, resulting in three datasets, including the uncorrected (uncorr.) dataset. (A) Biaxial contour density plots from the nine testing batches with 18 common markers. The files displayed are the first chow/control sample from each batch. The quadrant gate is shown to assist in visual comparison between plots. (B) Cydar batch correction density plots of four representative markers, showing the third file from each of the seven experimental batches with 36 common markers in total. Black arrows indicate a gap in the density near zero created by the warp correction algorithm.

Fig. 3.

Immune infiltrate phenotyping using viSNE. All range-corrected experimental files were run together in the same viSNE run to generate one universal viSNE map. (A) Cell density on viSNE plots for concatenated experimental files for chow and HFD groups in each cohort. E0771 (n=9/8), Wnt1 (n=6/6), TeLi (n=5/5), C11 (n=8/3), UN-KC (n=5/2). (B-F) The total concatenated data were used to generate viSNE plots (n=57). (B) Density plot of total concatenated cells. (C-F) viSNE plots showing marker intensity on a spectrum heat scale. Heat scales are specific to individual markers. (C) Marker heat for key myeloid phenotypic markers. The pink line on the CD11b plot indicates the phenotypic divide between myeloid and lymphoid cells. (D) Marker heat for key lymphocyte phenotypic markers. (E) Marker heat for activation/exhaustion markers. (F) Marker heat for additional phenotyping markers.

Fig. 4.

Immune infiltrate metacluster characterization of murine breast and pancreatic cancers. (A) Hierarchical clustering of MEM scores for 37 curated clusters and 26 markers. The dendrogram on the left was used to create 21 metaclusters. MEM scores and marker heat were used to label the metaclusters (labels in B). n=57: E0771 (n=9/8), Wnt1 (n=6/6), TeLi (n=5/5), C11 (n=8/3), UN-KC (n=5/2). B and C show mean values for immune infiltrate metaclusters form chow-fed mice; E0771 (n=9), Wnt1 (n=6), TeLi (n=5), C11 (n=8), UN-KC (n=5). (B) Bubble graph using area to show the mean percentage abundance out of the total CD45⁺ cells for each metacluster for immune infiltrate of each chow-fed non-obese tumor type. (C) Pie charts showing mean percentage abundance of the 21 metaclusters across the five models for the chow tumors. (D) Annotated viSNE map of concatenated data, showing the metaclusters by number. The black line indicates the divide between myeloid and lymphoid lineage cells/clusters (n=57).

Fig. 5.

Analysis and quantification of metacluster abundance differences between chow and HFD. (A,B) Box and whisker plots with all data points shown (mean, minimum to maximum). (A) Box and whisker plots comparing chow and HFD metaclusters for the breast cancer cohorts. Unpaired Student's t-tests were not adjusted, with s.d. assumed between chow and HFD for individual metacluster comparisons. Significant P-values are shown. The blue ‘T’s indicate P-values less than 0.07 that are trending towards significance. E0771 (n=9/8), Wnt1 (n=6/6), TeLi (n=5/5). (B) Box and whisker plots for pancreatic cancer cohorts. There were too few HFD tumors with live immune cells so statistics could not be performed for those cohorts. C11 (n=8/3), UN-KC (n=5/2).

Fig. 6.

CD8 T cells were decreased in the HFD DIO E0771 model of triple-negative breast cancer, and tumor growth advantage was lost when the T-cell compartment was lost in the TKO model. (A,B) Box and whisker plots with all data points shown (mean, minimum to maximum). Unpaired Student's t-tests were not adjusted, with s.d. assumed between chow and HFD for individual cell subsets. E0771 (n=9/8), Wnt1 (n=6/6), TeLi (n=5/5), C11 (n=8/3), UN-KC (n=5/2). (A) CD4/CD8 ratio for each tumor model. (B) Percentage of CD8 T cells of total T cells. (C) CITRUS SAM results for E0771 model. CITRUS clusters that are significantly different between chow and HFD are not blue and are circled with a gray background (n=9/8). (D) Selected significant CITRUS clusters were plotted back onto the viSNE map. Plotted clusters are color coded to match the CITRUS plot in C. viSNE data shown are concatenated data for the ten CITRUS clusters and total cell numbers for the E0771 model. The black line indicates the divide between myeloid and lymphoid lineage cells/clusters. (E,F) Tumor growth volume over time for E0771 WT (n=4/4; E) and E0771 TKO (n=5/5; F) cohorts (mean±s.d.). Unpaired Student's t-tests were not adjusted, with s.d. assumed. (G) Final tumor masses for C11 tumors grown in TKO mice, showing that the HFD tumor growth difference remains (n=5/5). The appropriate comparison is to the C11 tumor masses for C11_1 and C11_2 batches shown in Fig. 1C. Data are graphed as scatter dot plots (mean±s.d.). Unpaired Student's t-test, with s.d. assumed.