Human single-cell atlas analysis reveals heterogeneous endothelial signaling

Why this preprint is important

Key to this preprint is that it establishes a massive, comprehensive single-cell human endothelial cell (EC) atlas that provides major insights into EC heterogeneity. This atlas is a strong contribution to future single-cell EC research and can be used in many ways. For example, future EC cells can be projected onto this atlas instead of going through the typical scRNA-seq pipeline. An especially innovative feature of this study is the use of topic modeling, a method widely used in natural language processing, to uncover tissue-specific CCC patterns in ECs from biological data.

Background:

Endothelial cells (ECs) line the interior of all blood vessels as well as the lymphatic system in the body. They are crucial in managing the blood vessels and their surrounding tissues [1-4]. While all ECs share the functions of maintaining tissue health, regulating the immune responses, and keeping vessels healthy, their roles can vary depending on vessel type, tissue location, and developmental stage [5]. 

EC heterogeneity is regulated by two major mechanisms: transcriptional control and microenvironmental regulation [6-7]. Transcription factors such as GATA2 and ETV2 direct early EC formation [8-9]. After EC development, microenvironmental factors such as extracellular matrix components, physical forces, and interactions with neighboring cells shape the EC function to support an organ’s specific functions [10-11]. However, existing studies have only examined the influence of microenvironment on EC behavior in single organs or in mice [12-17]. A comprehensive study comparing EC activity across tissues and vessel types has not yet been conducted.

In this preprint, Zhu and team constructed a human single-cell EC atlas by integrating single-cell RNA sequencing (scRNA-seq) datasets containing over 3 million cells from 15 different tissues. Using this atlas, they found that ECs were more different across vessel types compared to organs. They then implemented a machine learning method to investigate how ECs interact with other cells via cell-cell communication (CCC) and discovered that most communication patterns were tissue-specific. These insights could ultimately help scientists identify new therapeutic targets for diseases related to the endothelium.

Highlights

Human single-cell atlas and computational pipeline uncover EC heterogeneity

To investigate EC heterogeneity, Zhu and colleagues first integrated scRNA-seq datasets from multiple sources to build a comprehensive atlas. They corrected for batch effects, ensuring that cell clustering reflected biological differences rather than technical biases. Principal Component Analysis showed that the ECs in heart and kidney are vastly different than in other tissues regardless of vessel type. Using unsupervised hierarchical clustering, the researchers identified two clusters with the biggest EC similarity. In cluster 1, ECs were more similar within the same vessel type. In cluster 2, ECs were more similar within the same tissue type.

The authors noticed that within cluster 1 of the atlas, ECs from different tissues within the same vessel type exhibited high variability. Building on this observation and previous studies, they constructed a computational pipeline to investigate how tissue-specific signals are associated with EC tissue heterogeneity through CCC. This method implements two types of cell-cell communication: metabolite-mediated and protein-mediated. They first implemented advanced bioinformatics tools, such as MEBOCOST and LIANA+, to detect metabolite-mediated CCC and protein-mediated CCC events. They then used an unsupervised machine learning method called ‘topic modeling’ to find common patterns of CCC across all tissues, grouping similar communication events together into “topics”. With this model, the authors were able to determine how likely each CCC event is part of a CCC pattern and how strongly a tissue shows a particular CCC pattern.

Figure 2A from the preprint – A visual representation of the computational pipeline developed to understand EC tissue heterogeneity through CCC using scRNA-seq data. Figure made available under a CC-BY-NC-ND 4.0 International license.

The researchers examined the number of EC-associated CCC events in different tissues. Using their computational method, they found that the CCC number for EC differs across tissues. For instance, the ovary and intestine had many more CCC events than the kidney and adipose tissue. However, they found that both protein-mediated and metabolite-mediated CCC showed the same tissue-specific pattern, suggesting that ECs communicate differently depending on the tissue rather than the CCC levels. The authors further computed the percentage of ECs and the overall cell type count across different tissues. They found that the percentage of ECs in a tissue was weakly correlated with the number of CCC events, whereas the number of cell types in the tissue showed a strong correlation. Even after normalizing CCC counts by the number of cell types, there were still differences across tissues, emphasizing tissue-specific CCC patterns. 

Machine learning method uncovers tissue-specific metabolite- and protein-mediated CCC patterns

The researchers used topic modeling to study tissue-specific EC heterogeneity in the context of metabolite-mediated CCC. Specifically, they used topic modeling to group CCC events identified in each tissue, with each topic representing a pattern of metabolite-mediated communication. They identified 15 metabolite-mediated CCC topics, with 9 topics being strongly associated with specific tissues. Within each topic, the authors looked into metabolite-sensor pairs and saw that each pair was strongly associated with CCC topic specificity. For example, topic 13 was closely linked to the ovary, with five metabolite-sensor pairs showing elevated CCC events and expression in this tissue. These findings are supported by previous studies demonstrating the significance of the identified metabolites and sensors in tissue-specific functions.

Figure 4A from the preprint – This figure visualizes the results of topic modeling on metabolite-mediated CCC networks. The heatmap illustrates the strength of the correlation between the 15 topics identified by the model and the tissues type under study. Figure made available under a CC-BY-NC-ND 4.0 International license.

The research team also used topic modeling to study tissue-specific EC heterogeneity in the context of protein-mediated CCC. Using the same pattern-detecting method as for metabolite-mediated CCC, they identified 15 protein-mediated CCC topics. 13 out of these topics are strongly associated with tissue types. An analysis into the ligand-receptor pairs showed that each pair was specific to a topic. They further investigated topic 15, which is closely associated with the heart, and found that six ligand-receptor pairs stood out. Published literature reveals that these ligand-receptor pairs have crucial roles in the heart. These results indicate that different tissues have unique ligand-receptor pairs involving ECs and offer insights into why ECs vary across tissues.

Conclusion

Overall, Zhu and team assembled an atlas composed of over 3 million cells from 15 tissues and established a computational pipeline to elucidate the atlas. Using this pipeline of bioinformatics and machine learning approaches, they uncovered how CCC-mediated microenvironments affect EC heterogeneity at the metabolite and protein level. Specifically, they found that EC-related CCC is highly specific to each tissue across both modes of communication. Moreover, the identified CCC patterns grouped in each topic provide a foundation for future studies on how metabolite- and protein-mediated CCC contribute to tissue-specific EC differences.

Questions for the authors:

1. How do these tissue-specific EC CCC patterns change in diseased states such as cancer?
2. Are sex differences or age-dependent shifts detectable in EC heterogeneity?
3. Could integrating spatial transcriptomics or proteomics data validate and refine the CCC topics identified in this study?

References

[1] Krüger-Genge A, Blocki A, Franke R-P, Jung F. Vascular endothelial cell biology: An update. Int J Mol Sci 2019;20:4411.
[2] Pi X, Xie L, Patterson C. Emerging roles of vascular endothelium in metabolic homeostasis. Circ Res 2018;123:477–94.
[3] Jalkanen S, Salmi M. Lymphatic endothelial cells of the lymph node. Nat Rev Immunol
2020;20:566–78.
[4] Iglesias MJ, Kruse LD, Sanchez-Rivera L, Enge L, Dusart P, Hong M-G, et al. Identification of endothelial proteins in plasma associated with cardiovascular risk factors. Arterioscler Thromb Vasc Biol 2021;41:2990–3004.
[5] Aird WC. Phenotypic heterogeneity of the endothelium: I. Structure, function, and mechanisms. Circ Res 2007;100:158–73.
[6] De Val S, Black BL. Transcriptional control of endothelial cell development. Dev Cell
2009;16:180–95.
[7] Potente M, Mäkinen T. Vascular heterogeneity and specialization in development and disease. Nat Rev Mol Cell Biol 2017;18:477–94.
[8] Lugus JJ, Chung YS, Mills JC, Kim S-I, Grass J, Kyba M, et al. GATA2 functions at multiple steps in hemangioblast development and differentiation. Development 2007;134:393–405.
[9] Shi X, Richard J, Zirbes KM, Gong W, Lin G, Kyba M, et al. Cooperative interaction of Etv2 and Gata2 regulates the development of endothelial and hematopoietic lineages. Dev Biol 2014;389:208–18.
[10] Majewska A, Wilkus K, Brodaczewska K, Kieda C. Endothelial cells as tools to model tissue microenvironment in hypoxia-dependent pathologies. Int J Mol Sci 2021;22:520.
[11] Leone P, Malerba E, Susca N, Favoino E, Perosa F, Brunori G, et al. Endothelial cells in tumor microenvironment: insights and perspectives. Front Immunol 2024;15:1367875.
[12] Kalucka J, de Rooij LPMH, Goveia J, Rohlenova K, Dumas SJ, Meta E, et al. Single-cell transcriptome atlas of Murine endothelial cells. Cell 2020;180:764–79.e20.
[13] Paik DT, Tian L, Williams IM, Rhee S, Zhang H, Liu C, et al. Single-cell RNA sequencing unveils unique transcriptomic signatures of organ-specific endothelial cells. Circulation 2020;142:1848–62.
[14] Bondareva O, Rodríguez-Aguilera JR, Oliveira F, Liao L, Rose A, Gupta A, et al. Single-cell profiling of vascular endothelial cells reveals progressive organ-specific vulnerabilities during obesity. Nat Metab 2022;4:1591–610.
[15] Schupp JC, Adams TS, Cosme C Jr, Raredon MSB, Yuan Y, Omote N, et al. Integrated single-cell atlas of endothelial cells of the human lung. Circulation 2021;144:286–302.
[16] Li Q, Zhu Z, Wang L, Lin Y, Fang H, Lei J, et al. Single-cell transcriptome profiling reveals vascular endothelial cell heterogeneity in human skin. Theranostics 2021;11:6461–76.
[17] Geldhof V, de Rooij LPMH, Sokol L, Amersfoort J, De Schepper M, Rohlenova K, et al. Single cell atlas identifies lipid-processing and immunomodulatory endothelial cells in healthy and malignant breast. Nat Commun 2022;13:5511.

Tags: cell-cell communication, endothelial cell, machine learning, scrna-seq

doi: https://doi.org/10.1242/prelights.42083

Read preprint (No Ratings Yet)