Background: The heterogeneity of mesenchymal stem cells (MSCs) is poorly understood, thus limiting clinical application and basic research reproducibility. Advanced single‐cell RNA sequencing (scRNA‐seq) is a robust tool used to analyse for dissecting cellular heterogeneity. However, the comprehensive single‐cell atlas for human MSCs has not been achieved.Methods: This study used massive parallel multiplexing scRNA‐seq to construct an atlas of > 130 000 single‐MSC transcriptomes across multiple tissues and donors to assess their heterogeneity. The most widely clinically utilised tissue resources for MSCs were collected, including normal bone marrow (n = 3), adipose (n = 3), umbilical cord (n = 2), and dermis (n = 3).Results: Seven tissue‐specific and five conserved MSC subpopulations with distinct gene‐expression signatures were identified from multiple tissue origins based on the high‐quality data, which has not been achieved previously. This study showed that extracellular matrix (ECM) highly contributes to MSC heterogeneity. Notably, tissue‐specific MSC subpopulations were substantially heterogeneous on ECM‐associated immune regulation, antigen processing/presentation, and senescence, thus promoting inter‐donor and intra‐tissue heterogeneity. The variable dynamics of ECM‐associated genes had discrete trajectory patterns across multiple tissues. Additionally, the conserved and tissue‐specific transcriptomic‐regulons and protein‐protein interactions were identified, potentially representing common or tissue‐specific MSC functional roles. Furthermore, the umbilical‐cord‐specific subpopulation possessed advantages in immunosuppressive properties.Conclusion: In summary, this work provides timely and great insights into MSC heterogeneity at multiple levels. This MSC atlas taxonomy also provides a comprehensive understanding of cellular heterogeneity, thus revealing the potential improvements in MSC‐based therapeutic efficacy.

The translation of genetic findings from genome-wide association studies into actionable therapeutics persists as a critical challenge in Alzheimer's disease (AD) research. Here, we present PI4AD, a computational medicine framework that integrates multi-omics data, systems biology, and artificial neural networks for therapeutic discovery. This framework leverages multi-omic and network evidence to deliver three core functionalities: clinical target prioritisation; self-organising prioritisation map construction, distinguishing AD-specific targets from those linked to neuropsychiatric disorders; and pathway crosstalk-informed therapeutic discovery. PI4AD successfully recovers clinically validated targets like APP and ESR1, confirming its prioritisation efficacy. Its artificial neural network component identifies disease-specific molecular signatures, while pathway crosstalk analysis reveals critical nodal genes (e.g., HRAS and MAPK1), drug repurposing candidates, and clinically relevant network modules. By validating targets, elucidating disease-specific therapeutic potentials, and exploring crosstalk mechanisms, PI4AD bridges genetic insights with pathway-level biology, establishing a systems genetics foundation for rational therapeutic development. Importantly, its emphasis on Ras-centred pathways—implicated in synaptic dysfunction and neuroinflammation—provides a strategy to disrupt AD progression, complementing conventional amyloid/tau-focused paradigms, with the future potential to redefine treatment strategies in conjunction with mRNA therapeutics and thereby advance translational medicine in neurodegeneration.

Acute graft-versus-host disease (aGVHD) remains a major life-threatening complication of allogeneic haematopoietic cell transplantation (allo-HSCT), often limiting the therapeutic efficacy of allo-HSCT. Recent studies have suggested that mesenchymal stem cells (MSCs) may be beneficial for the treatment of aGVHD. However, the therapeutic potential of MSCs is often negatively impacted by their heterogeneity. To investigate MSCs heterogeneity, we conducted single-cell transcriptomic analysis of human umbilical cord-derived MSCs (HUC-MSCs) and identified key feature genes that distinguish MSCs subpopulations. The function of the newly discovered biomarker CRISPLD2 was also explored. We engineered human umbilical cord-derived MSCs (HUC-MSCs) to overexpress the CRISPLD2 gene using lentiviral vectors. The downstream regulatory effects of CRISPLD2 overexpression were assessed through bulk RNA sequencing. Additionally, we evaluated its impact on cellular senescence using Western blotting and β-galactosidase (SA-β-gal) staining. The immunoregulatory capability of HUC-MSCs was tested through coculture experiments with T cells and liver organoids in vitro. Mitochondrial function was analysed via flow cytometry and electron microscopy. The in vivo therapeutic effects of HUC-MSCs on aGVHD were evaluated using an aGVHD murine model. The graft-versus-leukaemia (GVL) effect was measured via the inoculation of luciferase-positive A20 cells, and tumour growth was monitored via bioluminescence imaging. Our findings indicated that the CRISPLD2 gene is heterogeneously expressed in HUC-MSCs subsets characterized by stemness and immunosuppressive properties. Transcriptomic analysis revealed that CRISPLD2 overexpression suppressed calcium ion binding and G protein-coupled receptor signalling. In vitro studies demonstrated a marked increase in IL-10 secretion, which enhanced T-cell suppression in CRISPLD2-modified HUC-MSCs. The in vivo results demonstrated that transfusion of CRISPLD2-overexpressing HUC-MSCs ameliorated aGVHD while maintaining GVL activity. Mechanistically, CRISPLD2 overexpression overcomes the mitochondrial damage mediated by extracellular ATP and LPS in HUC-MSCs by inhibiting P2Y11 receptor signalling, thereby preserving their stemness and IL-10-mediated immunosuppressive functions. Our study revealed that CRISPLD2 is a novel marker for identifying HUC-MSCs subpopulation with enhanced immunosuppressive functions. CRISPLD2 overexpression enhances the immunosuppressive function of HUC-MSCs by inhibiting P2Y11 receptor signalling. Targeting CRISPLD2 is a promising strategy to improve the therapeutic efficacy of HUC-MSCs in aGVHD while maintaining GVL activity.

Predicting spatial gene expression from Hematoxylin and Eosin histology images offers a promising approach to significantly reduce the time and cost associated with gene expression sequencing, thereby facilitating a deeper understanding of tissue architecture and disease mechanisms. Achieving accurate gene expression prediction requires the extraction of highly refined features from pathological images; however, existing methods often struggle to effectively capture fine-grained local details and model gene-gene correlations. Moreover, in bimodal contrastive learning, dynamically and efficiently aligning heterogeneous modalities remains a critical challenge. To address these issues, we propose a novel method for predicting gene expression. First, we introduce a dense connective structure that enables efficient feature reuse, thereby enhancing the capturing and mining of local refinement features. Second, we leverage the state space models to uncover underlying patterns and capture dependencies within 1D gene expression data, enabling more accurate modeling of gene-gene correlations. Furthermore, we design the Residual Kolmogorov-Arnold Network (RKAN) that uses a learnable activation function to dynamically adjust bimodal mappings based on input characteristics. Through continuous parameter updates during contrastive training, RKAN progressively refines the alignment between modalities. Extensive experiments conducted on two publicly available datasets, GSE240429 and HER2+, demonstrate the effectiveness of our approach and its significant improvements over existing methods. Source codes are available at https://github.com/202324131016T/DANet.