Distinct spatial immune microlandscapes are independently associated with outcomes in triple-negative breast cancer

Abstract

The utility of spatial immunobiomarker quantitation in prognostication and therapeutic prediction is actively being investigated in triple-negative breast cancer (TNBC). Here, with high-plex quantitative digital spatial profiling, we map and quantitate intraepithelial and adjacent stromal tumor immune protein microenvironments in systemic treatment-naïve (female only) TNBC to assess the spatial context in immunobiomarker-based prediction of outcome. Immune protein profiles of CD45-rich and CD68-rich stromal microenvironments differ significantly. While they typically mirror adjacent, intraepithelial microenvironments, this is not uniformly true. In two TNBC cohorts, intraepithelial CD40 or HLA-DR enrichment associates with better outcomes, independently of stromal immune protein profiles or stromal TILs and other established prognostic variables. In contrast, intraepithelial or stromal microenvironment enrichment with IDO1 associates with improved survival irrespective of its spatial location. Antigen-presenting and T-cell activation states are inferred from eigenprotein scores. Such scores within the intraepithelial compartment interact with PD-L1 and IDO1 in ways that suggest prognostic and/or therapeutic potential. This characterization of the intrinsic spatial immunobiology of treatment-naïve TNBC highlights the importance of spatial microenvironments for biomarker quantitation to resolve intrinsic prognostic and predictive immune features and ultimately inform therapeutic strategies for clinically actionable immune biomarkers.

Fig. 1. Distribution of key immune proteins in intraepithelial and stromal segments as a function of recurrence.

Distribution of tumor-average log2 transformed normalized counts from 10 immune proteins (differentially expressed at BYadj p < 0.05) for both intraepithelial and stromal segments from FinXX samples. Tumors that did not recur (RFS = YES, n = 22) are shown in light gray, whereas tumors that did recur (RFS = NO, n = 22) are shown in blue. Two-sided p-values were calculated using the Kolmogorov–Smirnov (KS) model to test the null hypothesis that the distribution of protein expression in the two groups of samples was the same. GZMB granzyme B, B2M beta-2 microglobulin. Source data are given in Supplementary Data 1.

Fig. 2. Kaplan–Meier analysis of recurrence as a function of HLA-DR or IDO1 abundance in All Intraepithelial and All Stromal segments.

Tumor-average protein counts were used to stratify samples into tertiles (high = black, mid = blue, low = orange) based on the abundance of HLA-DR or IDO1. Kaplan–Meier analysis was carried out using protein abundance and time to event as continuous variables. Log-rank p-values are given, with the number of patients “at risk” shown below each curve. Source data are given in Supplementary Data 2. KM analysis of recurrence as a function of PD-L1 or IDO1 in CD45EnR Intraepithelial, CD68EnR Intraepithelial, TumorEnR Intraepithelial, CD45EnR Stroma, and CD68EnR Stoma segment classes are shown in Supplementary Figs. 6 (HLA-DR) and 7 (IDO1). HLA-DR abundance is given in All Segments (a), Intraepithelial segments (b), and Stromal segments (c); whereas IDO1 abundance is given in All Segments (d), Intraepithelial segments (e), and Stromal segments (f).

Fig. 3. Recurrence in FinXX and Mayo Clinic TNBC_TMA cohorts as a function of PD-L1 protein abundance and the APC or TCA eigenprotein scores.

Each dot or square represents tumor-average values for each feature in each tumor, including tumors that recurred (yellow squares) and tumors that did not recur (blue circles). The X axis contains numerical scores for the APC or TCA eigenproteins, and the Y axis contains PD-L1 protein log2 normalized counts for intraepithelial segments. Odds ratios (ORs) for 5-year recurrence were calculated as dichotomous variables (recurrence YES or NO). The reference lines indicate the population median values for each feature, and ORs were calculated in reference to samples that were <median for both features (e.g., APClow/TCAlow). Significance was defined as p < 0.05, and “quadrants” with p > 0.05 for ORs are designated as “ORns”. ORs for FinXX APClow/TCAhigh and APChigh/TCAlow are not given in Fig. 3, considering the low number of samples in these quadrants. However, ORs for all quadrants are given in Supplementary Table 3. Source data are given in Supplementary Data 2 (FinXX tumor-average abundance), Supplementary Data 7 (TNBC_TMA), and Supplementary Data 5 (eigenprotein scores). Patient numbers: FinXX, 22 recurred and 22 did not recur; TNBC_TMA, 81 recurred and 181 did not recur.

Fig. 4. Recurrence in FinXX (left panel) and Mayo Clinic TNBC_TMA (right panel) cohorts as a function of IDO1 protein abundance and APC or TCA eigenprotein scores.

As in Fig. 3, each dot or square represents an individual tumor, including tumors that recurred (yellow squares), and tumors that did not recur (blue circles). Numerical values for APC or TCA eigenprotein scores are given on the X axis, and normalized log2 counts for IDO1 protein are given on the Y axis. ORs were calculated for 5-year recurrence. Reference lines indicate the median values for each feature, and ORs were calculated in reference to samples that were <median for IDO1 and APC/TCA scores. ORs with p < 0.05 are designated “ORns”, and ORs for all segments are given in Supplementary Table 8. Source data for eigenprotein scores are given in Supplementary Data 5. Source data for protein abundance are given in Supplementary Data 2 (FinXX) and Supplementary Data 7 (TNBC_TMA). Patient numbers: FinXX, 22 recurred and 22 did not recur; TNBC_TMA, 81 recurred and 181 did not recur.