The heterogeneity of the epidermis reflects cell development and location. The IFE maintains homeostasis by moving the SB cells upwards to form the SS, SG, and SC[8, 23]. The HFs renew from the bulge area. While differentiating, the state and function of the epidermal cells change as do their surface markers. Additionally, heterogeneous stem cell populations have been proven to be present in the basal layer and HFs. Recent studies have indicated the existence of various stem cell populations that maintain their proliferative frequency under steady-state condition in the IFE and HFs[6, 25]. MFC provides an effective and simple tool to detect and sort epidermal cells for downstream application[19, 26, 27]. A valid polychromatic flow cytometric panel could be a useful tool for identifying various populations of epidermal cells.
This panel was developed with specific criteria and guidelines. The primary considerations throughout the process were epitope expression levels, fluorescein brightness, and commercial clone availability. As integrin plays a fundamental role in keratinocyte adhesion and migration, a high expression level of CD49f and CD29 indicates proliferative and transit-amplifying cells in the basal layer and bulge. CD117, a rete ridge marker, is enriched on a mixture of dormant stem cells, melanoblasts, and melanoma cells without expression at the bulge[7, 29]. In healthy adult skin, CD146 is expressed at the epidermal appendage and the external root sheath of HFs[30, 31]. CD24 is detected in postmitotic, nonclonogenic suprabasal keratinocytes located in the stratum spinosum of the IFE and the outer and inner root sheath of HFs[32, 33]. CD24 has been identified as a differentiation marker of keratinocytes. CD34 is a heavily glycosylated 110-kDa transmembrane protein. CD34+ epithelial cells have been confirmed to be present at the outer root sheath of anagenic human HFs. TLR7 is an intracellular Toll-like receptor that is known for its importance in autoimmunity[36, 37]. Recently, TLR7 has been proven to be a surface marker for skin stem cells in mice. In this panel, the antibody cocktail combination was developed to dissect the IFE layer by layer and subsets of these layers, as well as subsets in HFs, which could provide inspiration for studying the structure of the human epidermis. Given the epidermis is a closely arranged barrier, the trypsinization effect was estimated with H&E staining, which showed the epidermal cells detached after enzyme digestion (Fig. S1).
Manual gating in flow cytometric analysis uses biaxial plots to display the expression of multiple markers with the risk of overlooking small changes. In our panel, the manual gating strategy was performed by an experienced expert, with the epidermal cells divided into five subgroups based on CD49f and CD29 expression. Charruyer et al. revealed that there was no significant alteration of the epidermal stem cell frequency during aging through long-term repopulation in vivo and colony formation in vitro, and we found that the frequency of CD49fhi/CD29+ subpopulation between kids and adults showed no difference in flow cytometric analysis (data not shown). The MFC was used to delineate the phenotypic signatures among body sites. The distribution of subgroups among the ear, thorax and abdomen indicated a tendency toward more TLR7+ subsets in the ear, especially the TLR7+ group in PI. CD49fhi/CD29+ subgroups indicated a proliferative subset of basal epidermal cells. Yin et al. demonstrated the stemness of TLR7+ cells. The CD49fhi/CD29+/TLR7+ subset indicated one or more epidermal stem cell subpopulations.
Computational analysis makes it possible to cluster epidermal subsets in an unbiased way and reveals the possibilities for cellular identification in future clinical assessment by dimensional reduction and machine learning algorithms, which relies less on the experience of the operators. viSNE is a common used algorithm for high-dimensional data with a limitation of crowding problem, which could be solved with the combination of clustering method, such as SPADE. There were 25 clusters confirmed by SPADE on viSNE, which decreased the workload to determine the differences between groups compared to manual analysis, which divided the whole epidermal cells into 160 subpopulations. The cluster distributions of the ear, abdomen, and thorax demonstrated distinctive patterns (Fig. 3A). In further quantitative analysis of the cluster cell count, significant differences were confirmed (p < 0.05) by multiple t-tests for each cluster. clusters_15 and _24 expressed the highest levels in the ear, while clusters_05, _18 and _23 expressed the highest levels in the abdomen. The phenotype of each cluster was confirmed with the MFI heatmap, demonstrating negative expression of TLR7 in cluster_05, 18 and 23 and positive expression in cluster_15 and 24. Considering that Yin et al. reported TLR7 as an epidermal stem cell marker, the distinct expression pattern of TLR7 indicates the possibility that TLR7 could be a stem cell surface marker in the human epidermis. The phenotype consisting of cluster_05, 18 and 23 was TLR7−/CD24+, which indicated differentiated epidermal cells. At the same time, the phenotype of cluster_15 and 24 indicated possible stem cells, with a CD49fhi/CD29+/TLR7+ phenotype. Variation in the epidermis at different body sites has been reported with various methods, including reflectance confocal microscopy and flow cytometry[42, 43]. Webb et al. sorted human skin epidermis labeled with CD49f, CD71 and keratins (K14, K10, and K15), analyzed the cell cycle of each subset and found that epidermal stem cells (CD49fhi/CD7neg) were decreased in the sun-exposed area. Our results demonstrated that a lower frequency of the CD49fhi/CD29+ subpopulation in the ear than in the chest and abdomen, which are sun-exposed areas in China.
Several computational clustering algorithms have been established for MFC or mass cytometry, of which SPADE and PhenoGraph are commonly applied to stratify all events into subpopulations. The PhenoGraph analysis divided the whole epidermis into 26 clusters, of which 5 clusters demonstrated significant differences. The phenotypes and percentages of these clusters revealed that these clusters are the same in SPADE analysis, which indicates the consistency of computational analysis. Compared to manual analysis, a supervised method, machine learning, as an unsupervised clustering method, gives a holistic and clear view of the whole panel.