ST-CellSeg: Cell segmentation for imaging-based spatial transcriptomics using multi-scale manifold learning.

Spatial transcriptomic technology enables resolving a tumor microenvironment heterogeneity of colorectal cancer (CRC) tissue samples. However, there is limited evidence regarding the relative strengths and weaknesses of different CRC sample types and preservation methods, including formalin-fixed paraffin-embedded (FFPE), fresh frozen (FF) and snap frozen (SF). Here, we examine gene expression profiles in 36 tissue areas, FFPE, FF and SF samples, derived from a patient with stage IV CRC, using spatial analysis of the whole transcriptome. The tissue samples obtained from surgical resection of the cancer were subjected to a detailed histological evaluation and spatial transcriptomics analysis to compare the quality of the sample between different preservation methods. Spatial transcriptomic analysis demonstrated that FFPE tissue samples offered improved resolution of cellular morphology in the tumor microenvironment, facilitating greater detail of gene expression and cell types. In contrast, gene expression analysis of FF and SF tissues embedded in optimal cutting temperature compound revealed a high level of CRC gene expression. FFPE, FF and SF samples shared a significant overlap among the top 300 genes that were strongly associated with key biological processes in CRC. In summary, the spatial analysis of the whole transcriptome across FFPE, FF and SF CRC tissues revealed intratumoral heterogeneity and highlighted key transcriptomic signatures, suggesting the specific value of different tissue preservation methods in the profiling of CRC.

Since its first application in 2016, spatial transcriptomics has become a rapidly evolving technology in recent years. Spatial transcriptomics enables transcriptomic data to be acquired from intact tissue sections and provides spatial distribution information and remedies the disadvantage of single-cell RNA sequencing (scRNA-seq), whose data lack spatially resolved information. Presently, spatial transcriptomics has been widely applied to various tissue types, especially for the study of tumor heterogeneity. In this review, we provide a summary of the research progress in utilizing spatial transcriptomics to investigate tumor heterogeneity and the microenvironment with a focus on solid tumors. We summarize the research breakthroughs in various fields and perspectives due to the application of spatial transcriptomics, including cell clustering and interaction, cellular metabolism, gene expression, immune cell programs and combination with other techniques. As a combination of multiple transcriptomics, single-cell multiomics shows its superiority and validity in single-cell analysis. We also discuss the application prospect of single-cell multiomics, and we believe that with the progress of data integration from various transcriptomics, a multilayered subcellular landscape will be revealed.

Background: The SCRUM-Japan MONSTAR-SCREEN and CIRCULATE-Japan platforms have successfully implemented organ-agnostic liquid biopsy-based precision oncology and molecular residual disease (MRD)-guided therapeutic development, leading to multiple regulatory approvals of therapeutic agents and diagnostics through collection of quality-assured clinical data. Understanding tumor heterogeneity and microenvironment dynamics remains fundamental to advancing precision medicine. Here, we present MONSTAR-SCREEN-3, an innovative large-scale prospective study platform designed to integrate high-resolution spatial transcriptomics with multi-omics analysis across diverse cancer types. In our preliminary analyses, spatial transcriptomic analysis of HER2-amplified metastatic colorectal cancer revealed that treatment responses to anti-HER2 therapy are closely associated with tumor microenvironment characteristics, including increased T and B cell infiltration and interactions between cancer stem-like cells and HER2-expressing cells. Methods: MONSTAR-SCREEN-3 (UMIN000053975) is a large-scale multi-institutional study involving 55 centers across Japan, planning to enroll 3,200 patients stratified into three distinct cohorts: advanced solid tumors undergoing systemic therapy (n=1,700), resectable solid tumors receiving definitive treatment (n=1,100), and hematologic malignancies (n=400). The study implements systematic multi-sampling at strategically defined timepoints. Our comprehensive molecular characterization utilizes 5000Plex spatial transcriptomics integrated with circulating tumor DNA/RNA sequencing, bulk tissue WES/WTS, plasma proteomics, and microbiome analyses. Standardized longitudinal monitoring with WGS-based MRD analysis is implemented in resectable cases, while disease-specific MRD assessment approaches are applied to hematologic malignancies. Following SCRUM-Japan's established quality assurance system, standardized monitoring collects regulatory-grade clinical data, including objective response rate, progression-free survival, and overall survival. All molecular and clinical data will be integrated into our VAPOR CONE supercomputing infrastructure, creating an AI-driven ecosystem. This large-scale implementation of spatial transcriptomics analysis across multiple cancer types represents an unprecedented opportunity to comprehensively map the tumor microenvironment architecture and its dynamics during treatment. The integration of this spatial information with molecular and clinical data will provide unprecedented insights into tumor-immune interactions and resistance mechanisms, establishing an evidence-based framework for next-generation precision oncology. Citation Format: Mitsuho Imai, Shingo Sakashita, Junichiro Yuda, Shin Kobayashi, Riu Yamashita, Tadayoshi Hashimoto, Takao Fujisawa, Taro Shibuki, Shun-Ichiro Kageyama, Yoshikatsu Koga, Michiko Nagamine, Takeshi Kuwata, Hideaki Bando, Yoshiaki Nakamura, Woong-Yang Park, Takayuki Yoshino. SCRUM-Japan MONSTAR-SCREEN-3: A large-scale multi-institutional prospective study platform integrating spatial transcriptomics and multi-omics analysis for comprehensive understanding of tumor microenvironment dynamics [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 1154.

PD09-11 SPATIAL TRANSCRIPTOME ANALYSIS REVEALS THE TUMOR HETEROGENEITY AND NICHE FACTORS IN PROGRESSIVE PROSTATE CANCER

Single-cell spatial transcriptomics such as in-situ hybridization or sequencing technologies can provide subcellular resolution that enables the identification of individual cell identities, locations, and a deep understanding of subcellular mechanisms. However, accurate segmentation and annotation that allows individual cell boundaries to be determined remains a major challenge that limits all the above and downstream insights. Current machine learning methods heavily rely on nuclei or cell body staining, resulting in the significant loss of both transcriptome depth and the limited ability to learn latent representations of spatial colocalization relationships. Here, we propose Bering, a graph deep learning model that leverages transcript colocalization relationships for joint noise-aware cell segmentation and molecular annotation in 2D and 3D spatial transcriptomics data. Graph embeddings for the cell annotation are transferred as a component of multi-modal input for cell segmentation, which is employed to enrich gene relationships throughout the process. To evaluate performance, we benchmarked Bering with state-of-the-art methods and observed significant improvement in cell segmentation accuracies and numbers of detected transcripts across various spatial technologies and tissues. To streamline segmentation processes, we constructed expansive pre-trained models, which yield high segmentation accuracy in new data through transfer learning and self-distillation, demonstrating the generalizability of Bering.

Spatial transcriptomics is an emerging technology that allows for analysis of cellular and molecular heterogeneity at spatial resolution. The accurate identification of pathological regions in spatial transcriptomics data is essential for understanding tissue heterogeneity and disease progression. We introduce UPSST, a comprehensive framework that integrates tissue morphology, imputes gene expression, and clusters spatial regions using a graph attention neural network (GAT). UPSST was evaluated across multiple spatial transcriptomics datasets, achieving high performance, such as achieving an Adjusted Rand Index (ARI) of 0.737 and a Fowlkes-Mallows Index (FMI) of 0.818 on slice 151671 of the LIBD human dorsolateral prefrontal cortex (DLPFC) dataset. These results highlight the robustness and precision of our approach in identifying pathology domains. Additionally, UPSST facilitates downstream analyses such as differential and enrichment analysis, which are crucial for deriving biological insights. In conclusion, UPSST offers a powerful and reliable tool for spatial transcriptomics analysis, advancing the identification of pathological regions with high accuracy.

Prostate cancer is one of the most prevalent malignancies among men and remains a major clinical challenge due to the complex tumor microenvironment. Understanding gene expression dynamics at both cellular and spatial levels is essential for improving therapeutic strategies. In this study, we performed an integrated multi-omics analysis using single-cell RNA sequencing and spatial transcriptomics. scRNA-seq data from 15 prostate samples, including 8 normal and 7 tumor tissues, were analyzed to characterize distinct cellular populations. Spatial transcriptomic profiling was conducted on three FFPE prostate tissue sections, including adjacent normal tissue, acinar cell carcinoma, and invasive adenocarcinoma, using the standard 10x Genomics Visium FFPE platform (55 µm capture spots). Single-cell analysis revealed heterogeneity among epithelial, stromal, and immune cell populations, highlighting complex signaling networks in which myeloid cells may contribute to tumor progression through immune suppression and epithelial adaptability. Spatial transcriptomic analysis further identified region-specific expression patterns and spatially restricted tumor niches, including the regional establishment of TXNIP and BIRC3 as genes associated with metabolic stress and inflammatory survival pathways. The spatial colocalization of BIRC3 with tumor vasculature in invasive carcinoma tissue suggests a novel interaction. Our discoveries using an integrated single-cell and spatial transcriptomic approach reveal a high-resolution molecular map of prostate cancer with spatial features that may provide further therapeutic investigation.

Spatial transcriptomics is undergoing rapid advancements and iterations. It is a beneficial tool to significantly enhance our understanding of tissue organization and relationships between cells. Recent technological advancements have achieved subcellular resolution, providing much denser spot placement for downstream analysis. A key challenge for this following analysis is accurate cell segmentation and the assignment of spots to individual cells. The primary objective of this study was to evaluate the effectiveness of a new cell segmentation approach based on subcellular level spatial transcriptomic data by confirming nuclei positions and using Voronoi diagrams, compared to direct clustering with cellbin data. Our findings demonstrate that the Voronoi method not only outperforms traditional methods in providing clearer boundaries and better separation of cell types but also excels in preserving the most transcripts, addressing the issue of low capture efficiency. This integrative methodology presents a substantial advancement in spatial transcriptomics, offering improved cell type classification and spatial pattern recognition. All codes in the paper are available at the GitHub repository: https://github.com/Charlottttttte/Cell_Segmentation_Voronoi

P42-T Feasibility of an international normative database for nerve excitability studies

The progression and prognosis of early-stage lung adenocarcinoma are closely associated with histologic subtypes, yet the presence of mixed histologic patterns often complicates prognostic assessment. Currently, the correlation between molecular and histologic features remains poorly understood. Formalin-fixed paraffin-embedded (FFPE) samples were collected from patients with primary early-stage lung adenocarcinoma, encompassing three histologic subtypes: well-differentiated, moderately differentiated, and poorly differentiated. The GeoMx Digital Spatial Profiler platform was utilized to obtain spatial transcriptomic profiling. Regions of interest were carefully selected and further subdivided into three categories of areas of interest, specifically epithelial cell-enriched regions, macrophage-enriched regions, and other regions. Multiplex immunofluorescence (mIF) assays were employed to validate the obtained results. Distinct molecular characteristics were identified in tumor epithelial- and macrophage-enriched compartments spanning well-differentiated to poorly differentiated tumors. In poorly differentiated tumors, we observed enrichment of pathways related to humoral immune response, complement activation regulation, and extracellular matrix receptor interaction pathways, all of which are significantly associated with poorer prognosis. We integrated these pathways to develop a composite molecular signature that strongly correlate with adverse prognosis. Our results provide new insights into the link between molecular and histologic subtypes in mixed-type lung adenocarcinomas. Specifically, the identified molecular signatures offer potential biomarkers for predicting disease progression and prognosis, thus facilitating more precise and personalized therapeutic approaches. Poorly differentiated components in mixed-type early-stage lung adenocarcinoma (LUAD) are characterized by enrichment of humoral immune response, complementactivation regulation, and extracellular matrix receptor interaction pathways, which areassociated with worse prognosis. A composite molecular signature integrating the keypathways strongly correlates with adverse clinical outcomes, serving as apotential prognostic biomarker. Digital spatial transcriptomics reveals spatially resolvedmolecular heterogeneity in tumor epithelial- and macrophage-enrichedcompartments, bridging the gap between histologic subtypes and molecularmechanisms in LUAD.

Spatial transcriptomic analysis has proposed valuable insights into the behavior of tongue cancer. However, the specific cell types involved in chemically-induced carcinogenesis and the process of tumor development remain elusive. We leveraged artificial intelligence (AI) algorithms and spatial transcriptomic sequencing to meticulously characterize the spatial and temporal evolution of 4-nitroquinoline-1-oxide (4NQO)-induced tongue carcinogenesis and intratumor heterogeneity. An AI classifier effectively categorized dysplastic tongue tissue into 13 distinct groups. Spatial transcriptomics identified 13 corresponding cellular subgroups with unique features within the lesion. Both methods successfully distinguished subtle muscle phenotype and genetic lineage variations induced by 4NQO, despite limited morphological differences. Evolutionary tree analysis revealed the dynamic appearance and disappearance of functionally and genetically diverse cell subgroups during the progression from epithelial dysplasia to in situ carcinoma and invasive cancer. Key findings include the identification of specific switch genes associated with tumor invasion and the revelation of significant intratumor heterogeneity. This spatial transcriptomic analysis of 4NQO-induced tongue cancer provides a detailed characterization of tumor evolution and heterogeneity. It elucidates critical aspects of tongue cancer cell behavior and identifies potential therapeutic targets (switch genes). These findings offer novel insights for improving the diagnosis and treatment of tongue cancer.

Glomerular diseases encompass a group of kidney diseases that may share common gene expression pathways. Here, we analyzed glomerular-specific gene expression profiles across various glomerular diseases. We performed spatial transcriptomic profiling using formalin-fixed paraffin-embedded kidney biopsy specimens of controls and patients with five types of glomerular diseases using the GeoMx Digital Spatial Profiler. Disease-representative glomerular regions of interest (ROIs) were configured, probed with oligonucleotide barcodes linked with target complimentary sequence. The UV-cleaved barcodes were amplified to generate libraries and subsequently sequenced. Common differentially expressed genes across glomerular diseases were identified and Gene Ontology annotation was performed using the ToppGene suite. The mean age of patients with glomerular diseases and kidney donors was 49.5 ± 12.2 and 49.5 ± 9.8years, respectively. A total of 35 differentially expressed genes were consistently downregulated in glomeruli across the disease compared to the control, while none of the differentially expressed genes were consistently upregulated. Twelve of 35 downregulated differentially expressed genes, including the two hub genes JUN and FOS, were annotated with molecular function Gene Ontology terms related to DNA-binding transcription factor activity. The annotated biological process Gene Ontology terms included response to lipid-related (17/35 differentially expressed genes), response to steroid hormone (12/35 differentially expressed genes), or cell cycle regulation (10/35 differentially expressed genes). Xenium and immunofluorescence staining confirmed the reduced expression of JUN, ZFP36, and KLF9 in intraglomerular cells of glomerular diseases. Identifying common differentially expressed genes by spatial transcriptomic analysis provides insights into the underlying molecular mechanisms of glomerular diseases and may lead to novel assessment or therapeutic strategies.

ObjectiveHead and neck squamous cell carcinoma (HNSC) is a common malignant tumor with a 5-year survival rate of less than 60% and lactylation modification plays a key role in its occurrence and progression. This study aims to identify key lactylation-related genes (LRGs) in HNSC and their prognostic impact by combining machine learning methods and spatial transcriptomics analysis.MethodsHigh-throughput gene expression data of HNSC were obtained from the TCGA database. The “Limma” R package was used to screen differentially expressed genes (DEGs), and their biological significance was investigated through GO and KEGG functional enrichment analyses. Subsequently, weighted gene co-expression network analysis (WGCNA) was employed to identify disease trait-associated modules. We used 12 machine learning methods to screen for key genes and validated them in the GSE6631 dataset. Survival curves were plotted to evaluate the prognostic ability of key genes. Transcriptional regulation analysis via the “RcisTarget” package identified transcription factors associated with motifs exceeding NES 3.5. Single-cell and spatial transcriptomic analysis was performed, which involved dimensionality reduction, clustering via UMAP, cell type annotation and the plotting of spatial expression distribution maps of key genes. Finally, for immune infiltration analysis, the CIBERSORT algorithm was used to calculate the proportions of 22 types of immune cells, and then the correlations between key genes and immune-related genes were analyzed.ResultsOur study identified three key LRGs (MSN, KIF2C, and RFC4), which exhibited strong diagnostic performance, with AUC values of 0.986 for KIF2C, 0.964 for RFC4, and 0.89 for MSN in the TCGA-HNSC cohort. These findings were validated in the GSE6631 dataset. Western blotting showed a significant upregulation of the expression of these genes in the HNSC group, which was consistent with the RT-qPCR results. Survival analysis showed that the high expression of key genes was closely associated with a poorer overall survival (p < 0.05). Spatial transcriptomics analysis revealed that KIF2C, RFC4, and MSN were preferentially expressed in tumor core regions, indicating their involvement in the malignant microenvironment. Additionally, we discovered that 10 types of immune cell infiltrations were significantly changed in HNSC and well correlated with key genes.ConclusionMSN, KIF2C and RFC4 may serve as potential biomarkers for diagnosing and predicting the prognosis of HNSC, providing new insights into targeted therapy.Supplementary InformationThe online version contains supplementary material available at 10.1007/s12672-025-03849-6.

BackgroundProstate adenocarcinoma (PRAD) is a biologically heterogeneous disease threatening the health of elderly males worldwide, and some PRAD subtypes with certain molecular landscape are always associated with poor prognosis. A more precise molecular classification system for PRAD is urgently needed.MethodsThrough spatial transcriptome analysis, we identified different malignant cell/spot types in PRAD. Then, Monocle 2 analysis was applied to identify malignant cell fates and differentiation-related genes. Together with the prognosis-related genes identified through Kaplan–Meier analysis and univariate Cox regression in TCGA-PRAD cohort, we defined malignant cell differentiation-related prognostic genes (MDPGs). Based on MDPGs, we constructed a malignant cell differentiation-based PRAD classification (MDPC) using the ConsensusClusterPlus algorithm. Then, we explored multi-omics correlations of MDPC, and constructed the regulation networks of MDPC as well as the prognostic prediction model. Finally, we validated the prognostic prediction value of MDPC through immunohistochemical staining and follow-up of a retrospective cohort.ResultsThree malignant spot types were identified through spatial transcriptome analysis. Then, we defined 33 MDPGs and successfully constructed MDPC with three different subtypes (DPP4+MSMB+ MDPC, NHP2+NVL+ MDPC, COL1A1+MYLK+ MDPC). Apart from the correlations with tumor genomics, immunomics, MDPC also harbored convincing prognostic prediction value. In our cohort, COL1A1+MYLK+ MDPC served as an independent risk factor of OS (hazard ratio (HR) = 20.720, P-value = 0.0018) and PFS (HR = 117.00, P-value = 0.0036), and was closely correlated with Gleason grade, WHO/ISUP grade, radiotherapy, chemotherapy, endocrinotherapy, bone metastasis before treatment, and progression after treatment.ConclusionWe successfully constructed MDPC with validated prognostic prediction value. This classification system provided clinicians with an effective tool to stratify PRAD patients, identifying high-risk individuals, recognizing patients prone to develop bone metastasis, and offering opportunities for early intervention to improve patients’ prognosis.

Immunoglobulin A nephropathy (IgA-N) is a primary glomerulonephritis that is characterized by mesangial cell proliferation and expansion. Although glomerular endothelial cells (GE) are implicated in the pathogenesis of IgA-N, their role remains poorly understood. We conduct single-cell and spatial transcriptomic analysis using human specimens with mild IgA-N compared with normal. We integrate single-cell and spatial transcriptome analyses in human samples of mild IgA-N compared to normal samples, revealing several novel findings: (1) identification of clusters of GE and their expressed gene profiles, (2) identification of novel inflammation-related molecules newly implicated in GE in mild IgA-N, (3) activation of inflammatory pathways characteristic of IgA-N. Therefore, we suggest the importance of GE in the mechanism of IgA-N, as GE initiates more active inflammatory response prior to mesangial cells in patients with mild IgA-N. These findings provide useful information for understanding the pathogenesis and choosing the best drug for IgA nephropathy.