Abstract 2434: Preliminary feasibility of co-registering tissue autofluorescence with 3D spatial transcriptomics for multimodal analysis of destructively profiled colorectal tumor sections

Abstract Spatial multiomics, combining spatial transcriptomics (ST), which maps gene expression across tissue sections, with assays like LA-ICPTOF-MS (1-micron resolution elemental profiling), can uncover nuanced biomolecular interactions within the tumor microenvironment (TME) linked to tumor progression. However, many spatial assays are destructive, often requiring co-registering serial or non-adjacent tissue sections, introducing discrepancies that hinder spatial multimodal association studies. To assess the impact of profiling distant tissue layers on data analysis from destructively profiled sections, we co-registered H&E staining, autofluorescence (AF) imaging, and ST data across 16 serial sections from two T4 N0/1 colon cancer patients (A & B). AF data were acquired using DAPI and GFP channels on the EVOS™ M7000 Imaging System. VALIS software aligned AF imaging with 40X H&E whole-slide images (Leica Aperio GT450) to create a 3D model of serial sections, applying transformations to same-section Visium CytAssist ST data. Co-registration accuracy across sections was assessed using target registration error (TRE) from QuPath annotations of matching points across 3D serial section inflammation and vascular TME components. ST data were analyzed using 3D and 2D spatial domain analyses with Leiden clustering to identify biologically relevant gene expression patterns. Enrichr gene set enrichment identified pathways (Reactome) associated with 3D domains present in all slices within patients. Correlations of the top 35 spatially variable genes’ (Moran’s I) expression across sections were assessed, with hierarchical regression modeling used to examine how gene expression correlations changed with increasing section separation (random intercept by gene). AF co-registered with ST across sections with high accuracy, with minor losses due to architectural changes (TRE- A: 9.5±1.0µm, B: 10.4±1.1µm). The number of 2D spatial domains varied across sections (A: 6.4±1.3, B: 5.1±1.5), while 3D ST identified more domains (A: 8, B: 8). Persistent 3D domains were linked to pathways associated with colorectal cancer progression (e.g., A: Extracellular Matrix Degradation, p=0.001, B: CD163 Mediating Anti-Inflammatory Response, p=0.003). Gene expression correlations across sections for spatially variable genes decreased with greater section separation (A: β=-0.03, p<0.001; B: β=-0.02, p<0.001). Our work demonstrates successful 3D AF-ST co-registration providing non-destructive capture of persistent histological features across sections, crucial for destructively profiled samples. The diminishing correlation across sections highlights the need to explore information loss and biases from serial sectioning, including imputing ST data to improve spatial multiomics associations tied to tumor progression. Citation Format: Zarif Azher, Gokul Srinivasan, Minh-Khang Le, Alos Diallo, Harsimran Kaur, Fred Kolling, Laurent Perreard, Scott Palisoul, Keluo Yao, Ken S. Lau, Louis J. Vaickus, Xiaoying Liu, Joshua Levy. Preliminary feasibility of co-registering tissue autofluorescence with 3D spatial transcriptomics for multimodal analysis of destructively profiled colorectal tumor sections [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 2434.