Spatial transcriptomic profiling reveals distinct signatures in acute versus chronic wounds through hypergraph modelling and transcriptomic entropy analysis.

IntroductionThe financial burden of wounds to healthcare systems continues to increase, with little progression on advanced treatment options. Single cell RNA sequencing studies have predominantly focused on phenotypically different cell types and are generally descriptive in their analyses. Coordination within the transcriptome can be modelled using hypergraphs, which quantify higher order interactions (coordination between two or more genes) in transcriptomic data. Hypergraph modelling and analysis of transcriptomic entropy of wound samples will further elucidate differences between healing and non-healing wounds. MethodsSpatial transcriptomic analysis (Visium Spatial Gene Expression) was performed on three acute wounds, three chronic wounds and two healthy, unwounded skin samples. Standard transcriptomic analyses were performed in Python using the packages Scanpy and Squidpy. Hypergraph modelling and transcriptomic entropy analyses were performed in R. ResultsAcross the samples, twenty-nine Leiden clusters were defined, with keratinocytes, fibroblasts and adipocytes subclusters present. Fifteen marker genes for acute and chronic wounds were identified, with pathway analysis identifying immune responses and cellular homeostasis present in acute wounds, whilst in chronic wounds, pathways included immune responses and extracellular matrix structure. Spatial transcriptomics allowed for a descriptive analysis of the Leiden clusters within the samples, with each tissue sample being divided up into upper, middle and lower layer for this purpose. Hypergraph analyses of the whole transcriptome from the pathology groups and the independent transcriptomic Leiden clustering, revealed differences in underlying transcriptomic coordination. The top 1000 highly connected genes from each pathology group were analysed by Over Representation Analysis, with integrated stress response signalling, maintenance of cell number and myeloid and mononuclear cell differentiation pathways present in acute samples (all FDR <0.05) compared to regulation of vasculature development, mononuclear cell differentiation and epithelial cell proliferation (all FDR <0.05) in the chronic samples. Analysis of transcriptomic entropy identified further differences between acute and chronic wounds, with decreasing entropy from control to acute to chronic samples (all p-adjusted <0.001). When comparing Leiden clusters that were present in at least two sample groups, there was a statistically significant difference in entropy, although Leiden clusters varied in high or low entropy between acute and chronic wound samples (all p< 2.2 x 10-16). By mapping high and low entropy values to spatial images of each sample, lower transcriptomic entropy within the wound bed was identified compared to the rest of the tissue sample. ConclusionsSpatial transcriptomic analysis allowed for distinct spatial patterning to be observed which delineated acute and chronic wounds. Analysis of higher order interactions within the wound sample transcriptome identified key pathways that were not identified using traditional analyses. Transcriptomic entropy analysis further highlighted the differences between the wound samples. This work has implications for the future development of biomarkers of chronic wounds.