Background
The H&E stromal tumor-infiltrating lymphocyte (sTIL) score and programmed death ligand 1 (PD-L1) SP142 immunohistochemistry assay are prognostic and predictive in early-stage breast cancer, but are operator-dependent and may have insufficient precision to characterize dynamic changes in sTILs/PD-L1 in the context of clinical research. We illustrate how multiplex immunofluorescence (mIF) combined with statistical modeling can be used to precisely estimate dynamic changes in sTIL score, PD-L1 expression, and other immune variables from a single paraffin-embedded slide, thus enabling comprehensive characterization of activity of novel immunotherapy agents.
Methods
Serial tissue was obtained from a recent clinical trial evaluating loco-regional cytokine delivery as a strategy to promote immune cell infiltration and activation in breast tumors. Pre-treatment biopsies and post-treatment tumor resections were analyzed by mIF (PerkinElmer Vectra) using an antibody panel that characterized tumor cells (cytokeratin-positive), immune cells (CD3, CD8, CD163, FoxP3), and PD-L1 expression. mIF estimates of sTIL score and PD-L1 expression were compared to the H&E/SP142 clinical assays. Hierarchical linear modeling was utilized to compare pre- and post-treatment immune cell expression, account for correlation of time-dependent measurement, variation across high-powered magnification views within each subject, and variation between subjects. Simulation methods (Monte Carlo, bootstrapping) were used to evaluate the impact of model and tissue sample size on statistical power.
Results
mIF estimates of sTIL and PD-L1 expression were strongly correlated with their respective clinical assays (p < .001). Hierarchical linear modeling resulted in more precise estimates of treatment-related increases in sTIL, PD-L1, and other metrics such as CD8+ tumor nest infiltration. Statistical precision was dependent on adequate tissue sampling, with at least 15 high-powered fields recommended per specimen. Compared to conventional t-testing of means, hierarchical linear modeling was associated with substantial reductions in enrollment size required (n = 25➔n = 13) to detect the observed increases in sTIL/PD-L1.
Conclusion
mIF is useful for quantifying treatment-related dynamic changes in sTILs/PD-L1 and is concordant with clinical assays, but with greater precision. Hierarchical linear modeling can mitigate the effects of intratumoral heterogeneity on immune cell count estimations, allowing for more efficient detection of treatment-related pharmocodynamic effects in the context of clinical trials.
Trial registration
NCT02950259.