Background Immuno-oncology (IO) therapies have demonstrated effective and durable benefits in multiple cancer indications but responses are variable. Discovery and validation of better biomarkers of treatment response, ability to modulate immunological states, selection of optimal drug combinations, and identification of new IO targets are top priorities to maximize the clinical impacts of immunotherapy. While the immune and tumor microenvironment have been well-characterized in primary tumors,¹ large-scale assessments in metastatic disease are lacking. Here, we conducted a multi-modal, pan-cancer analysis comparing immune states across primary and metastatic disease using a real-world database.

Methods Analysis was conducted using the Tempus IO platform, encompassing molecular data and immunophenotyping algorithms layered upon a large, real-world patient database. To date, the platform comprises over 200k de-identified records with pre-computed immunophenotypes (e.g. immune cell fraction, HLA typing and quantification, TCR/BCR diversity, TIL estimation from H&E) across dozens of cancer types, with over 20k cases treated with IO therapy. Somatic DNA alterations and whole-exome transcriptomes were profiled using the Tempus xT and xR NGS assays, respectively, with optimized probes for TCR/BCR profiling.

Results A landscape, multi-modal comparison of tumor immune states was performed on 70k de-identified patient records, comprising a diverse spectrum of cancer types, tissue sites, and prognostic stages. Analyses of bulk tumor RNA showed liver mets with significantly lower immune infiltration, higher macrophage fraction, and lower TCR diversity compared to primary and lung met tumors (P<0.001) (figure 1). TIL quantification from 104 NSCLC digitized H&E slides revealed a significant correlation between intratumoral TIL density and cytotoxic T-cell score (ρ=0.57, P<0.001) (figure 2A), and confirmed lower intratumoral TIL density in liver mets versus primary lung (P<0.01) (figure 2B). In a stratified Cox regression analysis of multiple cancer types treated with first-line immune checkpoint blockade (N=1623), IO-response signatures^2–4 were highly predictive of real-world OS (P<0.001)(figure 3A), were superior to TMB, and exhibited significant differential enrichment across primary and metastatic sites (P<0.001) (figure 3B).

Conclusions In the largest pan-cancer analysis to date, significant differences in tumor immune states were observed by primary and metastatic sites, and by indication, with important implications for IO-treatment strategies. These data provide valuable and timely insights for IO research, where simultaneous assessment of molecular, immune, and clinical traits is required.