Of relevance for drug development and MC-related therapies, we next investigated the transcriptomic heterogeneity of human MC populations across different organs. We aggregated the databases of 24 different organs from the Tabula Sapiens, which is already decontaminated from ambient RNA with Decontx (Yang et al., 2020), as part of the processing guidelines (https://tabula-sapiens-portal.ds.czbiohub.org/), in a unique UMAP composed of 264,009 single cells (Fig. 5 A). We next identified a unique cluster composed of 2,690 MCs based on the signature of the cardinal MC genes KIT, CPA3, TPSB2, and CMA1 (Fig. 5, B and C). We then projected all of the identified MCs on the same UMAP, reaching a total number of 2,690 single MCs from 12 different organs. The human datasets appear to be more complex and heterogeneous than the mouse datasets, and we could not observe an obvious CTMC/MMC transcriptomic dichotomy in humans based on the expression of the classical histochemical markers CMA1 and TPSB2 reported in the literature (Derakhshan et al., 2022).

We, therefore, decided to adopt an unbiased approach to better understand human MC heterogeneity and performed an unsupervised nearest-neighbor analysis that identified the presence of 12 potential clusters (Fig. 5 D). To decipher the number of MC subsets present among these 12 clusters, we generated a correlation heatmap (Fig. 5 E) followed by a cluster dendrogram (Fig. 5 F) to directly visualize the strength of relationships between the different clusters. We could identify the presence of seven potential MCs states (Fig. 5 G) with a total of 4,564 statistically significant DEGs that defined the transcriptomic heterogeneity between each state of MCs (Fig. 5 H). The total list of DEGs characteristic to each of the human MC subsets, hereafter named MC1–7, is provided in Table S3.

Among the many DEGs, MC1 notably were characterized by the expression of genes encoding cytokines and chemokines (IL13, CXCL8, CCL2), the pleckstrin homology-like domain family A member 1 (PHLDA1), the early growth response 3 gene (EGR3), prostaglandin-endoperoxide synthase 2 (PTGS2), and the cluster differentiation 83 (CD83). MC2 notably expressed genes encoding the TNF superfamily member 12 (TNFSF12), the leukotriene C4 synthase (LTC4S), the fructose-1,6-bisphosphatase 1 (FPB1), and the GRB2-related adaptor protein 2 (GRAP2). MC3 expressed, among others, the vascular endothelial growth factor A (VEGFA), the cytoskeleton component utronin (UTRN), a chemokine receptor (CXCR4), the aryl hydrocarbon receptor (AHR), and the interleukin 1 receptor associated kinase 3 (IRAK3). MC4 expressed genes encoding the procathepsin L (CTSL), the interleukin 7 receptor (IL7R), the granzyme B (GZMB), the neuronal calcium sensor 1 (NCS1), the arginase 2 (ARG2), and the suppressor of cytokine signaling 1 (SOCS1). MC5 was preferentially enriched in genes encoding the tryptase delta 1 (TPSD1), SIGLEC8, CMA1, the Janus kinase 3 (JAK3), the microsomal triglyceride transfer protein (MTTP), and cathepsins (CTSB, CTSG). MC6 was enriched in genes encoding the neuronal growth factor neugrin (NGRN), laminin subunits (LAMA2, LAMA5), the interleukin 1 receptor 1 (IL1R1), the interleukin 1 receptor accessory protein (IL1RAP), the interleukin 1 receptor antagonist (IL1RN), interleukin 10 receptor subunits (IL10R1, IL10RB), the retinoic acid receptor α 1 (RARA), and the adrenomedullin (ADM). MC7 expressed a signature of genes encoding the Ras-related glycolysis inhibitor and calcium channel regulator (RRAD), the programmed death-ligand 1 (CD274 or PD-L1), a proto-oncogene serine/threonine protein kinase (PIM2), a sphingomyelin synthase (SAMD8), IL7R, and RARA. Interestingly, MRGPRX2 was found heterogeneously expressed among clusters with a tendency for enrichment in MC5 but, MRGPRX1, another receptor reported to be the human ortholog of Mrgprb2, could not be found in any dataset.

We then isolated each MC’s DEGs signature (Table S4) to create seven cluster scores that we projected on the aggregated UMAP of MCs (Fig. 5 I). Using this approach, we could confirm that each identified set of DEGs enabled the precise identification of the corresponding MCs in the UMAP.

Finally, we extracted scRNAseq transcriptomic signatures from the states of MCs identified in chronic rhinosinusitis with nasal polyposis patients by Dwyer et al. (2021) and projected them on our aggregated UMAP of MCs (Fig. 5 J). We could show that both MC_1 and MC_3 populations identified by Dwyer et al. (2021) matched several of our populations whereas the discrete MC_4 population did not match any. These data demonstrate that at least seven MC populations/states with distinct transcriptomic signatures exist across organs in humans and strongly suggest that the heterogeneity of human MCs might extend far beyond the classical CTMC/MMC dichotomy.