Immunofluorescence-based adaptive immune subtypes in an association with long-term disease-free interval in the Carolina Breast Cancer Study Phase III

Adaptive immune subtype defined using bulk RNA expression has been associated with lower risk of breast cancer recurrence. In the context of intralesional heterogeneity, these estimates may not capture relevant localized patterns of immune infiltration. We aimed to 1) compare bulk RNA-based and immunofluorescence (IF)-based immune subtypes to better characterize the tumor immune microenvironment and 2) evaluate associations between spatially-resolved adaptive immunophenotypes defined using IF and breast cancer recurrence. We analyzed 1,665 invasive breast cancer tumors from the population-based Carolina Breast Cancer Study Phase 3, using 5,276 1-mm tumor microarrays cores (up to 4 per tumor). Six adaptive markers (PD-1, CD3, CD4, CD8a, CD8, FOXP3) were quantified by multiplex IF. Core-level adaptive immune subtypes (adaptive-high vs adaptive-low) were defined with k-mean clustering of HALO-estimated percent-positive cells for each marker. Participant-level immune subtypes were defined using k-means clustering on weighted averages of core-level marker values, weighted by total cell count. Tumors were labeled as ‘consistent' if all cores matched the participant-level subtypes. As a comparison, we also performed k-mean clustering using only CD8 percent positive (high vs low) on tumor level. Percent agreement was calculated comparing IF (multi-marker) with bulk RNA-based adaptive classifiers (previously defined). We used Cox models to estimate associations between adaptive subtype and 10-year disease-free interval (DFI) stratified by estrogen receptor (ER) status, and adjusted for age, race, and stage. We observed 65% agreement between participant-level IF and RNA adaptive classifiers. Multi-marker IF-defined adaptive-low tumors had worse 10-year DFI [ER+: Hazard Ratio (HR) 1.54, 95% Confidence Interval (CI) 1.07-2.22; ER-: HR 1.72, 95% CI 1.11-2.68]. CD8-low cluster showed a similar association among ER+ tumors [HR 1.52, 95% CI 1.06-2.19], no association among ER- tumors [HR 1.31, 95% CI 0.83-2.06]. RNA-based associations were stronger than those for IF [HR RNA 2.00, 95% CI 1.48-2.70, non-adaptive vs. adaptive, overall]. Exposure misclassification may have attenuated associations because tumors with inconsistent multi-marker adaptive-high IF subtype were significantly associated with worse 10-year DFI compared to those were consistent [HR 1.83, 95% CI 1.18-2.85]. Overall, adaptive immunity is prognostic in breast cancers, but misclassification of adaptive response is a concern for biomarker development. Use of multiple markers reflecting the adaptive immune community may partially overcome intralesional heterogeneity, but bulk profiling masks spatial characteristics. Ideal immune biomarkers for prognosis must be reproducibly measured and sensitive to detect the most impactful immune characteristics.