Intratumoral Cell Neighborhoods Coordinate Outcomes in Pancreatic Ductal Adenocarcinoma

Abstract

Background & aims: Pancreatic ductal adenocarcinoma (PDA) is a highly lethal disease characterized by a spatially heterogeneous tumor microenvironment. Within the PDA microenvironment, cells organize into communities where cell fate is influenced by neighboring cells of diverse ontogeny and function. However, it remains unclear how cell neighborhoods in the tumor microenvironment evolve with treatment and impact clinical outcomes.

Methods: Here, using automated chromogenic multiplex immunohistochemistry and unsupervised computational image analysis of human PDA tumors, we investigated cell neighborhoods in surgically resected tumors from patients with chemotherapy-naïve PDA (n = 59) and neoadjuvant chemotherapy-treated PDA (n = 57). Single cells were defined by lineage markers (CD3, CD8, Foxp3, CD68, CK19), proliferation (Ki67), and neighboring cells.

Results: Distinct intratumoral immune and tumor cell subsets were defined by neighboring cells. Higher content of stromal-associated macrophages was seen in chemotherapy-naïve tumors from long-term survivors (overall survival >3 years) compared with short-term survivors (overall survival <1 year), whereas immune-excluded tumor cells were higher in short-term survivors. Chemotherapy-treated vs -naïve tumors showed lower content of tumor-associated T cells and macrophages but similar densities of stromal-associated immune cells. However, proliferating tumor cell subsets with immune-rich neighborhoods were higher in chemotherapy-treated tumors. In a blinded analysis of tumors from patients treated with neoadjuvant chemotherapy, a composite index comprising lower quantities of immune-excluded tumor cells and higher spatially distinct immune cell subsets was associated with prolonged survival.

Conclusions: Together, these data provide new insights into discrete cell communities in PDA and show their clinical relevance.

Keywords: Biomarkers; Immune Cells; Inflammation; Pancreatic Cancer; Tumor Microenvironment.

Figure 1.. Immunologically defined microenvironments in surgically resected chemotherapy naïve PDA.

(A) Schematic for multiplex immunohistochemistry (mIHC) analysis workflow. (B) Representative mIHC images of Panel 1 (top): CD8 (yellow), Foxp3 (purple), CD68 (brown) and CK19 (teal) and Panel 2 (bottom): Ki67 (yellow), CD3 (purple) and CK19 (teal). (C) Heatmap of control chemotherapy naïve PDA resection specimen (n = 51). (D to G). Violin plots of cell counts per area for CD68⁺ cells (D), CD68⁺ cells adjacent to CK19⁺ cells (E), CD8⁺ cells (F) and Foxp3⁺ cells (G) among myeloid (Gr_1), lymphoid (Gr_2) and mixed (Gr_3) PDA resection specimens. (H) Representative hematoxylin and eosin (H&E, top) and Panel 1 (middle) and Panel 2 (bottom) mIHC images. (I) Kaplan-Meier analysis of overall survival from time of. Log-rank test was used. FFPE, formalin fixed paraffin embedded; HRP, horse radish peroxidase; AP, alkaline phosphatase; HQ, hydroxy-quinoxoline; Unadjusted P values: *, P < 0.05; **, P < 0.01, ***, P < 0.001; ****, P < 0.0001.

Figure 2.. Tumor and immune cell neighborhoods associate with survival in surgically resected chemotherapy naïve PDA.

(A) Schematic showing approach to cells based on the abundance of neighboring cells. (B) Heatmap and hierarchical clustering of cell identity (top) and corresponding cell neighborhoods (bottom). (C) Pie chart of the frequency of CK19⁺ cells with the indicated number of immune cells detected within 15 μm. (D) Representative mIHC images with insets showing stromal (A) and tumor (B) regions. Corresponding digitized images depicting annotated cell subsets. Cells stained for hematoxylin only were excluded from the analysis. Scale bar represents 250 μm. (E to G) Ridge plots showing the distance (μm) between cells of neighborhood-defined cell subsets and nearest CK19⁺ cell. Cell subsets classified as tumor-associated (yellow), stromal-associated (blue) or mixed (gray) based on proximity to CK19⁺ cells of < 10 μm or > 210 μm. (H) Stacked bar graph showing the median number of cells in adjacent cell neighborhood (within 15 μm) for each cell subset. (I) Forest plot of selected immune cell subsets in STS and LTS. (J) Forest plot of neighborhood-defined tumor cell subsets in STS and LTS. (I and J) Effect size (Cohen’s D) and 95% confidence intervals are shown. Bold indicates statistically significant, based on the unadjusted P values shown. P1, Panel 1; No., number; mIHC, multiplex immunohistochemistry; STS, short term survivor; LTS, long term survivor.

Figure 3.. Neoadjuvant chemotherapy associates with less tumor-associated immune cells in resectable PDA.

(A) SWOG S1505 study schema. (B to G) Bar graphs showing the density of indicated cells in resected PDA tissues from control and chemotherapy-treated tumors. Log₂ of cells per mm² + 1 is shown. Mann-Whitney U tests were performed. (H to J) Bar graphs showing the density of tumor-associated and stromal-associated lymphoid cells subsets (H), macrophages (I) and Ki67⁺CK19^negCD3^neg cells (J) in control and chemotherapy-treated tumors. Mann-Whitney U tests were performed. (K) Heatmap of control chemotherapy-naïve PDA resection specimens classified by selected neighborhood-defined immune cell subtypes. (L) Kaplan-Meier analysis of overall survival since time of surgery based on classification as high (Group 1: immune^high) or low (Group 2: immune^low) in selected neighborhood-defined immune cell subtypes. Unadjusted P value from a log-rank test shown. (M) Heatmap of chemotherapy-treated PDA resection specimen classified as in (K). (N) Kaplan-Meier analysis of overall survival since time of surgery in patients treated with neoadjuvant chemotherapy and classified as in (M). For two patients who did not undergo resection, date of specimen collection was used in place of date of surgery. Unadjusted P values: *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001.

Figure 4.. Tumor cell neighborhoods define outcomes in neoadjuvant treated resectable PDA.

(A) Quantification of tumor cell subsets in control and chemotherapy treated tumors. Mann-Whitney U tests were performed. (B) Forest plot of selected tumor cell subsets in control and chemotherapy-treated tumors. Effect size (Cohen’s D) and 95% confidence interval is shown. Bold indicates significant. Adjusted P values by Bonferroni method are shown. (C) Heatmap of Panel 1 tumor cell subsets showing number of adjacent cell types. Frequency of cluster as percent of total measured tumor cells is shown. (D) Representative mIHC images of Panel 1 tumor cell subsets. Corresponding digitized images with inset depicting annotated cell subsets are shown. Cells stained for hematoxylin only were excluded from the analysis. Scale bar represents 250 μm. (E) Heatmap of Panel 2 tumor cell subsets showing number of adjacent cells. Frequency of cluster as percent of total measured tumor cells is shown. (F) Representative mIHC images of Panel 2 tumor cell subsets. Corresponding digitized images with inset depicting annotated cell subsets are shown. Scale bar represents 250 μm. (G) Heatmap of control chemotherapy-naïve PDA resection specimen classified by selected neighborhood-defined tumor cell subtypes. (H) Kaplan-Meier analysis of overall survival since time of surgery based on classification as low (Group 1: tumor^low) or high (Group 2: tumor^high) in selected neighborhood-defined tumor cell subtypes. Unadjusted P value from a log-rank test shown. (I) Heatmap of chemotherapy treated PDA resection specimen classified by selected neighborhood-defined tumor cell subtypes. (J) Kaplan-Meier analysis of overall survival since time of surgery based on classification as high (Group 1: tumor^low) or low (Group 2: tumor^high) in selected neighborhood-defined tumor cell subtypes. Unadjusted P value from a log-rank test shown. For two patients who did not undergo resection, date of specimen collection was used in place of date of surgery. CI, confidence interval; adj, adjusted; Unadjusted P values: **, P < 0.01; ****, P < 0.0001.

Figure 5.. Neighborhood-defined tumor cell and immune cell subsets reveal determinants of survival resectable PDA.

(A and B). Correlation matrix of selected tumor and immune cell subsets in control tumors (A) and chemotherapy-treated tumors (B). Spearman correlation coefficient is shown. (C) Heatmap of control chemotherapy naïve PDA resection specimen classified by selected tumor and immune cell subtypes. (D) Kaplan-Meier analysis of overall survival since time of surgery classified as low in tumor cell subsets and high in immune cell subsets (Group 1: immune^high/tumor^low) or high in tumor cell subsets and low in immune cell subsets (Group 2: immune^low/tumor^high). Unadjusted p-value from a log-rank test shown. (E) Heatmap of chemotherapy-treated PDA resection specimen classified by selected tumor and immune cell subtypes. (F) Kaplan-Meier analysis of overall survival since time of surgery based on tumor classification as low in tumor cell subsets and high in immune cell subsets (Group 1: immune^high/tumor^low), low in tumor cell and immune cell subsets (Group 2: immune^low/tumor^low), or high in tumor cell and low in immune cell subsets (Group 3: immune^low/tumor^low). Unadjusted p-value from a log-rank test shown. For two patients who did not undergo resection, date of specimen collection was used in place of date of surgery.

Figure 6.. Associations of spatially defined tumor and immune cell subsets with mFOLFIRINOX and gemcitabine/nab-paclitaxel treatment.

(A) Forest plot of tumor cell subsets in control and mFOLFIRINOX (FFX) or gemcitabine/nab-paclitaxel (GA) treated tumors. (B) Forest plot of immune cell subsets in control and mFOLFIRINOX (FFX, black box) or gemcitabine/nab-paclitaxel (GA, white box) treated tumors. CK19⁺ tumor cell subsets defined as tumor-associated (yellow), stromal-associated (blue) or mixed tumor and stromal-associated (gray). (A and B) Effect size (Cohen’s D) and 95% confidence interval is shown. Adjusted P values by Bonferroni method are shown. CI, confidence interval; Adj, adjusted.