We are pleased to share our recently published manuscript in Clinical Cancer Research, “Spatial Configuration of Pancreatic Cancer Is Associated with Disease Recurrence after Neoadjuvant Therapy and Curative-Intent Resection.” In this study, we asked whether the spatial organization of residual pancreatic ductal adenocarcinoma, rather than residual tumor burden alone, contains prognostic and biological information after neoadjuvant therapy.
The central premise of the work is ecological. Tumors are not simply collections of malignant cells. They are structured ecosystems composed of cancer, stroma, immune cells, vasculature, matrix, and other microenvironmental components. These elements form spatial patterns: patches, boundaries, interfaces, gradients, and regions of aggregation or intermixing. In ecology, these patterns can shape resource flow, organismal movement, competition, predation, and habitat selection. In pancreatic cancer, analogous spatial features may influence nutrient access, immune exclusion, mechanical constraints, invasion, and therapeutic persistence.
To study this, we applied an AI-enabled digital pathology workflow to routine H&E whole-slide images from 203 patients with resected PDAC who had received neoadjuvant therapy and had only a minor pathologic response. Cancer and cancer-associated stroma were segmented, converted into two-dimensional tissue mosaics, and analyzed using Tumor Landscape Analysis, a framework adapted from landscape ecology. This allowed us to quantify spatial composition and configuration across the tumor microenvironment, including patch area, patch density, edge density, shape complexity, interspersion, aggregation, and diversity.
Our main finding was that residual tumor-stroma topology was associated with disease recurrence. High-risk tumors were characterized by a more fragmented, interface-rich, and intermixed cancer-stroma landscape. In contrast, conventional measures such as residual cancer area and CAP treatment response category did not reliably stratify disease-free survival in this minor-response cohort.
Two complementary spatial risk models emerged. One model identified a high-risk configuration defined by cancer shape index and variability in stromal shape index. A second model identified mean stromal patch area and edge density as prognostic features. Both models remained associated with disease-free survival after adjustment for standard clinicopathologic factors, suggesting that the architecture of the residual tumor microenvironment captures information not contained in stage, nodal status, margin status, or conventional pathologic response grading.
This is where the landscape ecology analogy becomes especially useful. Edge density, fragmentation, patch complexity, and reduced homotypic aggregation are not just image features. They describe how potential tissue habitats are arranged. A highly intermixed tumor-stroma landscape may create more edge habitats where cancer cells interact with stromal, immune, and matrix components. These interfaces may represent biologically important niches for persistence after therapy, invasion, immune evasion, or occult disease survival.
We also found that high-risk spatial configurations were associated with altered lymphocyte localization. Rather than showing increased intratumoral lymphocyte infiltration, high-risk landscapes demonstrated reduced intratumoral TIL density and relative lymphocyte accumulation at the cancer periphery and within stroma. This pattern is consistent with an immune-excluded phenotype and suggests that tumor-stroma topology may provide a structural correlate of immune access within the PDAC microenvironment.
More broadly, this work supports a shift from measuring residual pancreatic cancer only by quantity to also measuring its spatial organization. The topology of residual disease may reflect the ecological state of the post-treatment tumor microenvironment, including how cancer and stroma are arranged, where immune cells are excluded or retained, and where biologically relevant interfaces persist.
Because this approach can be applied to routinely generated H&E slides, it has potential translational relevance for post-neoadjuvant risk stratification. Future work will focus on external and prospective validation, higher-resolution characterization of the cellular states occupying these edge habitats, and development of spatially informed approaches to adjuvant therapy. Our longer-term goal is to convert descriptive histopathology into quantitative tumor ecology that can help guide risk-adapted treatment strategies for patients with pancreatic cancer.
We are grateful to the patients and families who made this research possible, and to our collaborators across oncology, surgery, pathology, quantitative health sciences, computational biology, and evolutionary ecology.
Leave a Reply