New therapies have been designed to stimulate the host’s immune system to fight cancer. Despite these exciting, recent successes, a large proportion of patients still do not respond to anti–programmed cell death-1 (PD-1) or anti–programmed death ligand-1 (PD-L1) therapies, and thus, biomarkers for patient selection are highly desirable. The only U.S. Food and Drug Administration–approved histopathology biomarker tests for anti–PD-1 or anti–PD-L1 therapy is assessment of PD-L1 protein expression by means of immunohistochemistry. This approach is unidimensional and has limitations. Innovative characterization of the tumor microenvironment (TME) with a focus on multidimensional, spatially resolved interactions at the single-cell level will provide critical mechanistic insights into therapeutic responses and potentially identify improved biomarkers for patient selection. Using multispectral approaches to image the TME and substituting cells for stars and galaxies, we applied the methodology and infrastructure developed for astronomy to pathologic analysis of specimens from patients with melanoma.

The next generation of pathologic analyses will require platforms that can characterize the coexpression of key molecules on specific cellular subsets in situ and spatial relationships between tumor cells and multiple immune elements. To that aim, we applied astronomical algorithms for high-quality imaging and the establishment of relational databases to multiplex immunofluorescence (mIF) labeling of pathology specimens, facilitating spatial analyses and immunoarchitectural characterization of the host-tumor interface. In all, we curated and coordinately mapped six markers, both individually and in combination in tumor tissue from 98 patients with melanoma receiving anti–PD-1 therapy. This dataset comprised ~127,400 image mosaics composed of more than 100 million single cells. The data outputs were linked to patient outcomes, informing in a clinically relevant way how cancer evades the immune system and potentiating biomarker assay development for precision immunotherapy.

The imaging protocols used in this study were used to address outstanding questions regarding the impact of high-power field sampling strategies on biomarker performance. This information was then used to develop an approach for operator-independent field selection. The image handling strategies also facilitated the robust assessment of the intensity of PD-1 and PD-L1 expression in situ (negative, low, mid, and high levels) on different cell types. Thus, with only six markers (PD-1, PD-L1, CD8, FoxP3, CD163, and Sox10/S100), we were able to develop 41 combinations of expression patterns for these molecules and map relatively rare cells such as CD8⁺FoxP3⁺ cells to the tumor stromal boundary. Moreover, a high density of CD8⁺FoxP3⁺PD-1^low/mid cells was closely associated with response to PD-1 blockade. Cell types associated with a lack of response to therapy were also identified—for example, CD163⁺ macrophages that were PD-L1^–. This latter phenotype was also found to have a negative effect on long-term survival. When these and other key cell phenotype densities were combined, they were highly predictive of objective response and stratified long-term patient outcomes after anti–PD-1–based therapies in both a discovery cohort and an independent validation cohort.

Here, we present the AstroPath platform, an end-to-end pathology workflow with rigorous quality control for creating quantitative, spatially resolved mIF datasets. Although the current effort focused on a six-plex mIF assay, the …