Abstract 1159: Machine learning integrating spatial omics uncovers humoral immunity patterns in intratumoral tertiary lymphoid structures in pancreatic cancer pathologic responders

Abstract Background: In pancreatic ductal adenocarcinoma (PDAC), rare long-term survivors correlate with high intratumoral tertiary lymphoid structure (TLS) density. This finding prompted our investigation of clinically viable strategies to induce TLS in patients with immune-excluded tumors. We previously reported the induction of intratumoral TLS following administration of a neoadjuvant GM-CSF-secreting allogeneic vaccine (GVAX) to PDAC patients (NCT00727441). However, no clinical benefit was observed, likely owing to immune tolerance mechanisms governing the PDAC tumor microenvironment (TME). We previously observed upregulation of both the PD-1 and 4-1BB pathways with GVAX, and thus in a subsequent neoadjuvant trial combined GVAX with PD-1 blockade and 4-1BB agonism (NCT02451982) which was associated with pathologic responses. We hypothesized this combination strategy induced TLS of higher maturity and anti-tumor activity compared to GVAX alone. Methods: To explore how this therapeutic strategy affected TLS morphology and intercellular crosstalk, we leveraged the Visium spatial transcriptomics platform and a 35-marker customized TLS panel for imaging mass cytometry. We generated cellular and molecular maps of the TME after neoadjuvant treatment in 26 PDAC patients (GVAX n=19, GVAX+aPD1 n=2, GVAX+aPD1+a41BB n=5). To compare TLS maturation in parallel with secondary lymphoid organ-mediated tumor immunity, we also profiled tumor-adjacent lymph nodes. We applied unsupervised learning with non-negative matrix factorization (NMF) and trained AI-enabled image classification models to characterize cellular states within tissue structures of interest. Results: We identified spatial gene expression NMF patterns in PDAC TLS spanning across distinct morphologies and neoadjuvant treatment arms. Intratumoral TLS after GVAX were found to propagate activated B cells expressing immunoglobulins that infiltrated into malignant cellular niches. TLS NMF patterns were also associated with autoimmune disease signatures, such as diabetes, in a subset of patients. We scored TLS using tumor-draining lymph nodes as a reference and found increased maturation of TLS after PD-1 blockade, while addition of 4-1BB agonism significantly boosted the cytotoxic NK/T cell compartment compared to GVAX alone. Conclusions: We present machine learning approaches for spatial multi-omics analysis to characterize the TLS-enriched TME. We mined genome-wide spatial TLS gene expression patterns elucidating the spatial dynamics of humoral immunity of rare immunotherapy pathologic responders in PDAC. Altogether our findings shed light on the plasticity of TLS in neoadjuvant immunotherapy and suggest future immunotherapy approaches should target both humoral and cytotoxic NK/T cell compartments to augment responses in solid tumors. Citation Format: Dimitrios N. Sidiropoulos, Sarah M. Shin, Alexander Girgis, Daniel H. Shu, Janelle Montagne, Atul Deshpande, Jeanette A. Johnson, Lucie Dequiedt, Victoria Jacobs, Aleksandra Ogurtsova, Guanglan Mo, Xuan Yuan, Genevieve Stein-O’Brien, Mark Yarchoan, Qingfeng Zhu, Ashley Kiemen, Elizabeth M. Jaffee, Lei Zheng, Won Jin Ho, Robert Anders, Elana J. Fertig, Luciane T. Kagohara. Machine learning integrating spatial omics uncovers humoral immunity patterns in intratumoral tertiary lymphoid structures in pancreatic cancer pathologic responders abstract. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts) ; 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84 (6Suppl): Abstract nr 1159.