Transcriptomic analysis of glycosylation at single-cell resolution

Introduction:: Glycosylation is a ubiquitous protein post-translational modification that results in the formation of an array of complex carbohydrate structures or glycans [1]. These glycans control or fine-tune a multitude of biological functions during normal physiology and disease [2, 3]. Glycans are formed by the concerted action of ~2% of the expressed proteome that are collectively called ‘glycoEnzymes’ that are encoded ‘glycogenes’. An ontology called “GlycoEnzOnto” has recently been curated to organize glycoEnzymes based on their molecular function, biological pathways, and also physical location [4]. Besides the genes involved in glycan biosynthesis, GlycoEnzOnto also includes the molecules involved in the regulation of nucleotide-sugar metabolism, glycosyl-substrate/donor transport, glycan degradation, and other regulatory components. In this paper, we apply GlycoEnzOnto to analyze single-cell RNA-sequencing (scRNA-seq) data that are part of the Tabula Sapiens (TS) project [5]. This dataset provides transcriptomics data for a total of 483,152 human cells from 24 tissues and >400 cell types. Specifically, the study sheds light on the extent to which single-cell data can inform us about glycosylation patterns in individual cells and tissue types.


Materials and Methods:: We applied bioinformatics analysis workflow summarized in Fig. 1A to the TS dataset, focusing on glycogenes and glycopathways that were recently curated in GlycoEnzOnto [4]. As glycogenes are traditionally known to be lowly expressed, we assessed the ability of single-cell data to inform us about genes and pathways involved in glycosylation. We identified a total of 19,847 protein-coding (PC) genes in the TS dataset, among which 400 are glycogenes. First, statistical analyses were performed to elucidate the number and expression of glycogenes in single cells and to compare these with PC genes. The second type of analysis employed single-cell Variational Inference (scVI) [6] and Uniform Manifold Approximation and Projection (UMAP) [7] to characterize and visualize the manifold of single-cell expression of glycogenes. The third analysis aimed to elucidate the variability of glycopathway activity across tissues. Specifically, the method Model-based Analysis of Single-cell Transcriptomics (MAST) [8] was applied to assess the differential expression of glycopathways across tissues. In the last analysis, transcriptional factors (TFs) of glycogenes and glycopathways were inferred using a combination of TF binding site information from TFLink database and normalized mutual information (NMI) between single-cell expressions of TF-glycogene pair. Here, only TFs from the highest 5^th percentile of NMIs were considered.


Results, Conclusions, and Discussions:: We observed that the number of glycogenes detected in single cells ranged from 2 to 98 at the median UMI depth of 6496 (Fig. 1B). Glycogene detection plateaued at ~220 glycogenes at higher sequencing depth, suggesting that only ~55% of all glycogenes are expressed in a typical cell. Additionally, glycogenes were expressed as commonly as other PC genes, contrary to the common misconception that these are “lowly-expressed genes” (Fig. 1C). Among the glycogenes involved in carbohydrate biosynthesis, enzymes involved in core glycan biosynthesis (‘Core’), were more abundant compared to glycan chain elongation (Extension) and capping (Terminal) enzymes (Fig. 1D). Also, the scVI latent embedding of glycogene expression using UMAP in Fig. 1E showed that epithelial cells (and also other cell types) from different tissue sources have distinct glycosylation patterns as they were segregated into distinct clusters.

We assessed the differential expression of glycopathways in single cells across tissues and cell types. Fig. 1F shows that pathways in the Core and Extension groups have higher logFCs than those in the Terminal group. Finally, we identified potential transcriptional regulators of glycosylation using NMI between single-cell expression of TF and glycogene in the TS dataset. A total of 614 TFs were found to have TF binding site evidence and strong NMI. Fig. 1G shows the top 25 TFs that are most frequently associated with glycopathways.

Overall, our study provides a broad framework to understand glycosylation in humans. The results suggest that glycogenes are more ubiquitously expressed in cells than other PC genes. Among glycogenes, core enzymes are expressed at a higher level than enzymes that modify terminal glycan residues that have more specialized functions. These observations are consistent with the notion that the core enzymes are often unique in mammals with roles in initiating the formation of specific glycan types. Differential expression analysis further points to a higher tissue-to-tissue variability in the Core and Extension glycopathways than the Terminal group. Finally, our study reveals common and unique transcription factors for glycogenes and glycopathways. The findings from the single cell analysis represent the starting point for wet-lab validation underway in our research laboratories.


Acknowledgements (Optional): : This work was supported by start-up funding from University at Buffalo-SUNY and funding from the National Institutes of Health (award #3R01HL103411-10S1).


References (Optional): : 1. Neelamegham, S. and L.K. Mahal, Multi-level regulation of cellular glycosylation: from genes to transcript to enzyme to structure. Curr Opin Struct Biol, 2016. 40: p. 145-152.

2. Varki, A., Biological roles of glycans. Glycobiology, 2017. 27(1): p. 3-49.

3. Reily, C., et al., Glycosylation in health and disease. Nat Rev Nephrol, 2019. 15(6): p. 346-366.

4. Groth, T., et al., GlycoEnzOnto: a GlycoEnzyme pathway and molecular function ontology. Bioinformatics, 2022. 38(24): p. 5413-5420.

5. Tabula Sapiens, C., et al., The Tabula Sapiens: A multiple-organ, single-cell transcriptomic atlas of humans. Science, 2022. 376(6594): p. eabl4896.

6. Gayoso, A., et al., A Python library for probabilistic analysis of single-cell omics data. Nat Biotechnol, 2022. 40(2): p. 163-166.

7. McInnes, L., et al., UMAP: Uniform Manifold Approximation and Projection. Journal of Open Source Software, 2018. 3(29): p. 861.

8. Finak, G., et al., MAST: a flexible statistical framework for assessing transcriptional changes and characterizing heterogeneity in single-cell RNA sequencing data. Genome Biol, 2015. 16: p. 278.