Single molecule spatial transcriptomics demonstrates spatial heterogeneity of pre-miRNA and classifies β-cells in type 1 diabetes.

Abstract

Recently developed spatial transcriptomics approaches combine RNA sequencing (RNA-Seq) with cell localization to reveal the spatial heterogeneity of transcriptomes in tissue samples. However, the interrogation of transcriptomic profiles in subcellular locations has not been reported. Here we propose a quantitative approach to measure the spatial distribution of single-cell transcriptomes, and we use this technique to study pre-miRNA expression in single β-cells obtained from human pancreatic tissues. A multi-dimensional quantitative model was established to describe the spatial distribution of RNAs as “features”, including RNA expression, location, and degree of clustering/dispersion. These features were analyzed using statistical distribution modeling and supervised machine learning to enable the classification of β-cells from the pancreas of non-diabetic control organ donors, autoantibody-positive (AAb+) organ donors, and organ donors with type 1 diabetes (T1D). This approach provided a quantitative evaluation of the distinctive features contributing to the classification and phenotyping and showed that expression levels of pre-miR-155 in β cells differ significantly among the groups, with more pre-miR-155 in the nucleus than cytoplasm of β-cells in the T1D and AAb+ samples compared to β-cells from control donors (p<0.0001). Interestingly, pre-miRNA distribution directly correlated with β-cell phenotype and T1D disease status. Results show higher degree of clustering for nuclear pre-miR-155 at short distances and greater amount of aggregation at the nuclear boundary for T1D samples (p<0.0001). These findings suggest that the spatial heterogeneity of the transcriptome of β-cells can classify these cells into different pathological conditions. Taken together, data from this study reveal fundamental mechanisms associated with RNA distribution and the β-cell phenotype during the pathogenesis of T1D. This approach could lead to the discovery of important subcellular spatial transcriptomic features that may have clinical relevance in stratifying T1D phenotypes.