EP204 - Exploring the Tumor Microenvironment in Early-Onset Colorectal Cancer: A Digital Pathology Analysis Using HD-Yolo

Image or Table (Do Not add text here):

Purpose/Background: The incidence of colorectal cancer in adults younger than 50 years old (EO-CRC) is growing at an alarming rate, and patients with EO-CRC tend to present with more advanced disease. The factors within the tumor microenvironment (TME) that drive EO-CRC pathogenesis remain poorly understood. Tumors with a high proportion of stroma relative to cancer tissue are associated with worse prognosis in epithelial cancer, including CRC. Also, an inverse relationship between stroma and tumor-infiltrating lymphocytes (TILs) has been described, and a lower density of TILs is associated with worse survival. Understanding these TME dynamics is essential to improving prognostic tools and treatments for EO-CRC. Here, we analyzed the cellular TME on whole-slide hematoxylin and eosin (H&E) CRC tumor images using Histology-based Detection using Yolo (HD-Yolo), a novel deep learning algorithm for nucleus detection, segmentation, classification, and TME-related feature extraction. We hypothesize that EO-CRC tumors will have an increased proportion of non-immune stromal cells and a lower proportion of lymphocytes when compared to average-onset colorectal cancer (AO-CRC).

Methods/Interventions: Sixteen H&E slides from independent CRC specimens (9 EO-CRC, 7 AO-CRC) from our institutional (UT Southwestern) biobank were reviewed by a board-certified gastrointestinal pathologist, confirming presence of adenocarcinoma. Whole-slide images were obtained at 40x using NanoZoomer S60 (Hamamatsu). Images were uploaded onto a digital pathology software (i-viewer), and intratumorally regions of interest were manually selected. HD-Yolo, a validated algorithm for nuclear segmentation and cellular TME classification, was used for whole-slide image analysis (WSI). Student’s t-test was performed to compare differences between EO-CRC and AO-CRC. Statistical significance was defined as p < 0.05.

Results/Outcomes: Tumors from 9 EO-CRC patients (4 Hispanic, 2 non-Hispanic Black, and 3 non-Hispanic White) and 7 AO-CRC patients (3 Hispanic, 2 non-Hispanic Black, and 2 non-Hispanic White) were included in our study. WSI analysis revealed no differences in the proportion of cancer or stromal cells between EO-CRC and AO-CRC. The average cancer-to-stroma ratio was 0.63 and 0.45 for EO-CRC and AO-CRC, respectively (p=0.18). However, a lower proportion of lymphocytes was observed in EO-CRC (18% vs 27%, p=0.007). The average lymphocyte-to-stroma ratio was 0.42 and 0.73 for EO-CRC and AO-CRC, respectively (p=0.02). No differences were observed when comparing the proportion of cancer, stroma, and lymphocytes intratumorally between EO-CRC and AO-CRC.

Conclusion/Discussion: A deeper understanding of tumor microenvironment is critical for improving EO-CRC care. Digital pathology analysis is a powerful tool to gain insight into the complexity of the TME. Using HD-Yolo, which has been proven to be an effective and efficient tool in digital pathology analysis, we report differences in cellular distribution of TME in a small cohort of CRC patients. One of the strengths of this research is the access to a racially diverse EO-CRC cohort. Next steps include expanding the cohort size as well as employing advanced techniques such as spatial transcriptomics to further characterize the TME.