Evaluation of immunotherapy (I-O)-associated immune cell (IC) dynamics and PD-L1 expression using multispectral immunohistochemistry (mIHC) and single-cell hierarchical regression modeling in early-stage breast cancer (ESBC).
e12070 Background: In triple negative breast cancer (TNBC), anti-PD-L1 is associated with gains in overall survival, however only in tumors with > 1% PD-L1+ ICs. Locoregional cytokine therapy is being evaluated as a method of priming and redirecting ICs to tumors, which may increase I-O benefit particularly in PD-L1 negative tumors. We report a sensitive method of quantifying I-O-related changes in PD-L1 and ICs using mIHC and single-cell hierarchical regression, which controls for within-tumor and across-patient PD-L1/IC heterogeneity. Methods: Pre-treatment and resection tissues from a phase Ib trial of locoregional cytokines (IRX-2, n = 16 subjects) in ESBC were analyzed. IRX-2 contains immunostimulatory cytokines (GM-CSF, IL-2, IFN-α, INF-γ, and IL-12) and was injected in peri-areolar tissue, which communicates directly with tumor-draining lymphatics. Specimens were analyzed for 1) H&E stromal TILs score; 2) clinical PD-L1 IC expression (Ventana SP142, 2 blinded pathologists); and 3) mIHC (PerkinElmer Vectra). InForm software was used to obtain geospatial outputs of tumor cells (CK+) and ICs (CD3, CD8, CD163) across multiple regions of interest (mean:18; range: 9-32). Mixed-effects modeling was used to evaluate for changes in PD-L1, ICs, and IC/tumor distance metrics. Results: PD-L1 and IC quantity by mIHC was highly concordant with clinical PD-L1 IC classification (JT test = 211, p < .001) and TIL score (r = 0.77). Cytokine therapy was associated with higher PD-L1 IC classification in 9/13 tumors, and increased PD-L1 ICs by mIHC in 14/15 (+337%, 95% CI: 120,769%). After adjusting for heterogeneity, cytokine therapy increased CD3+CD8+ ICs (+173%, CI: 93,287%) and CD3+CD8- ICs (+120%, CI: 21,301%). Therapy was also associated with increased effector-helper T-cell clustering (nearest-neighbor distance, -40%, CI: -29,-50%) and effector cell tumor localization (stromal-tumor interface CD8+ density +90%, Cl: 74, 305%). Conclusions: Our method is concordant with clinical PD-L1 and sTILs scores, and can additionally quantify IC and IC distances. This assay may allow for comparative I-O assessments with increased resolution, and will be used in a randomized phase II trial of pembrolizumab, chemo +/- IRX-2 in stage II/III TNBC.