胃肿瘤切片的三维空间多组学揭示超越二维分析的隐藏特征

Single-section spatial-omics frequently misses microanatomical context and is sensitive to sectioning variability. Recent 3D reconstruction studies highlight substantial heterogeneity within the tumor microenvironment obscuring true biology. Emerging platforms (e.g. Singular Genomics G4X) enable in-situ RNA/protein readouts at subcellular resolution on the same slide with fluorescent H&E, creating serialized spatial multi-omics. This work allows us to map transcript abundance at the subcellular level with x-y-z coordinates, integrating H&E, transcripts, and proteins on the thick tissues, enabling true 3D spatial multimodal tumor modeling.We profiled serial FFPE sections using the G4X system to jointly quantify targeted RNAs and proteins per slice. Across slices, we compared cell types, neighborhood enrichment matrices, and gene-set enrichment to measure inter-slice inconsistency. We then registered the serial sections into a common 3D frame, reconstructed a volumetric atlas, and identified 3D hotspots for exhausted T-cell programs. To directly visualize whether these patterns reflected true 3D structure, we acquired label-free 3D holotomography from the same serial sections, aligned them with the corresponding serial G4X data in a shared 3D space, applied AI-based single-cell segmentation, and mapped cell types defined by G4X RNA and protein expression back to their x-y-z coordinates.Serial 2D slices showed discordant cell-type compositions and neighborhood enrichments, indicating strong variability even within a single tumor block. In contrast, 3D reconstruction revealed a coherent exhausted T-cell domain that failed to reach statistical significance in any individual slice. Holotomography confirmed dense, continuous interfaces among T cells, extracellular matrix, and tumor cells across depth, supporting true biological continuity and demonstrating subcellular localization of transcripts and proteins in 3D.Our 3D spatial multi-omic approach shows that single-section analyses may miss clinically relevant immune niches, whereas integrated 3D RNA-protein-holotomography profiling recovers hidden signals. By mapping and combining the transcripts with proteins and the 3D holotomography at subcellular x-y-z coordinates in thin and thick tissues, we provide a practical workflow for volumetric tumor-immune profiling that can accelerate biomarker discovery and therapeutic targeting in gastric cancer and can be generalized to other cancer types."Generative AI was used only to edit this abstract's language. The authors are responsible for the study concept, data interpretation, and conclusions, and have approved the final version."