Purpose: Programmed death ligand 1 (PD-L1) is expressed in tumor cells and immune cells, and both have been associated with response to anti-PD-1 axis immunotherapy. Here, we examine the expression of PD-L1 to determine which cell type carries the predictive value of the test. Experimental Design: We measured the expression of PD-L1 in multiple immune cells with two platforms and confocal microscopy on three retrospective Yale NSCLC cohorts (425 nonimmunotherapy-treated cases and 62 pembrolizumab/nivolumab/atezolizumab-treated cases). The PD-L1 level was selectively measured in different immune cell subsets using two multiplexed quantitative immunofluorescence panels, including CD56 for natural killer cells, CD68 for macrophages, and CD8 for cytotoxic T cells. Results: PD-L1 was significantly higher in macrophages in both tumor and stromal compartment compared with other immune cells. Elevated PD-L1 in macrophages was correlated with high PD-L1 level in tumor as well as CD8 and CD68 level (P < 0.0001). High PD-L1 expression in macrophages was correlated with better overall survival (OS; P = 0.036 by cell count/P = 0.019 by molecular colocalization), while high PD-L1 expression in tumor cells was not. Conclusions: In nearly 500 non-small cell lung cancer (NSCLC) cases, the predominant immune cell type that expresses PD-L1 is CD68(+) macrophages. The level of PD- L1 in macrophages is significantly associated with the level of PD-L1 in tumor cells and infiltration by CD8+ T cells, suggesting a connection between high PD-L1 and "hot" tumors. In anti-PD-1 axis therapy-treated patients, high levels of PD-L1 expression in macrophages are associated with longer OS and may be responsible for the predictive effect of the marker.

Assessment of tumor infiltrating lymphocytes (TILs) as a prognostic variable in melanoma has not seen broad adoption due to lack of standardization. Automation could represent a solution. Here, using open source software, we build an algorithm for image-based automated assessment of TILs on hematoxylin-eosin stained sections in melanoma. Using a retrospective collection of 641 melanoma patients comprising four independent cohorts; one training set (N = 227) and three validation cohorts (N = 137, N = 201, N = 76) from 2 institutions, we show that the automated TIL scoring algorithm separates patients into favorable and poor prognosis cohorts, where higher TILs scores were associated with favorable prognosis. In multivariable analyses, automated TIL scores show an independent association with disease-specific overall survival. Therefore, the open source, automated TIL scoring is an independent prognostic marker in melanoma. With further study, we believe that this algorithm could be useful to define a subset of patients that could potentially be spared immunotherapy.

Purpose: In both the IMpassion 130 trial in the metastatic setting and in Keynote 522 in the neoadjuvant setting, patients with triple-negative breast cancer (TNBC) showed benefit from PD-1 axis immunotherapy. Here, we assess PD-L1 expression on both tumor and immune cells using quantitative immunofluorescence to assess association with benefit from neoadjuvant durvalumab concurrent with chemotherapy in TNBC. Experimental Design: Pretreatment core needle biopsies (n = 69) were obtained from patients who participated in a phase I/II clinical trial (NCT02489448). The final analysis included 45 patients [pathologic complete response (pCR) = 18, non-pCR = 27] due to technical issues and insufficient tissue. Slides were stained using a previously validated Ultivue DNA-based Ultimapper kit (CD8, CD68, PD-L1, Cytokeratin/Sox10, and Hoechst counterstain). The PD-L1 expression was analyzed by molecular compartmentalization without segmentation using AQUA software (version 3.2.2.1) in three tissue compartments including tumor (cytokeratin-positive cells), CD68+ cells, and overall stroma. Results: In patients with pCR, PD-L1 expression was significantly higher in tumor cells, in CD68+ cells and in the stroma compared with patients non-pCR. There was no difference in the amount of CD68+ cells in the tumor or stromal compartments between cases with pCR and non-pCR. Conclusions: Expression of PD-L1 in tumor cells, immune cells in stroma, and colocalized with CD68+ cells is associated with higher rates of pCR to durvalumab and chemotherapy in TNBC.

Purpose: Although tumor-infiltrating lymphocytes (TIL) assessment has been acknowledged to have both prognostic and predictive importance in triple-negative breast cancer (TNBC), it is subject to inter and intraobserver variability that has prevented widespread adoption. Here we constructed a machine-learning based breast cancer TIL scoring approach and validated its prognostic potential in multiple TNBC cohorts. Experimental Design: Using the QuPath open-source software, we built a neural-network classifier for tumor cells, lymphocytes, fibroblasts, and “other” cells on hematoxylin–eosin (H&E)–stained sections. We analyzed the classifier-derived TIL measurements with five unique constructed TIL variables. A retrospective collection of 171 TNBC cases was used as the discovery set to identify the optimal association of machine-read TIL variables with patient outcome. For validation, we evaluated a retrospective collection of 749 TNBC patients comprised of four independent validation subsets. Results: We found that all five machine TIL variables had significant prognostic association with outcomes (P ≤ 0.01 for all comparisons) but showed cell-specific variation in validation sets. Cox regression analysis demonstrated that all five TIL variables were independently associated with improved overall survival after adjusting for clinicopathologic factors including stage, age, and histologic grade (P ≤ 0.0003 for all analyses). Conclusions: Neural net-driven cell classifier-defined TIL variables were robust and independent prognostic factors in several independent validation cohorts of TNBC patients. These objective, open-source TIL variables are freely available to download and can now be considered for testing in a prospective setting to assess clinical utility. See related commentary by Symmans, p. 5446