Molecular Signatures and Signaling Interactions of the Hair Follicle Stem Cell Niche

Abstract

Signaling interactions between the dermal papilla (DP) and neighboring stem cells in the hair germ (HGSCs) and bulge (BuSCs) regulate new follicle growth in the hair cycle. To study these interactions, the three populations had been profiled together by now-outdated microarrays or separately by bulk RNA-sequencing. Recent single-cell transcriptomics established signatures of DP, BuSCs and HGSCs, but low detection sensitivity limited the depth of gene expression discovery. Here, we define the transcriptomes of DP, BuSCs, HGSCs, epidermal, follicle and dermal fibroblast cells—after flow-sorting each population from four neighboring mouse back skin regions—to gain deeper insights into the unique gene expression programs of SCs and their instructive niche. With cross-comparisons of 56 whole-transcriptome measurements, we classify cell type-specific molecular signatures of enriched genes with unprecedented sensitivity. Joint analysis with signatures from 15 leading studies published in the last 20 years revealed many previously undescribed DP, BuSC and HGSC genes in mouse and human counterparts. With ligand-receptor mapping and CellChat analyses we then uncover comprehensive cell-cell communication insights. Finally, we provide a new installment of our Hair-Gel repository along with numerous signature and other gene tables for easy exploration of gene expression in hair follicle SCs and their niche.

INTRODUCTION

The hair follicle (HF) provides a dynamic and accessible system for studying stem cell biology and tissue regeneration. HFs undergo life-long repeated hair growth cycles with phases of hair growth (anagen), tissue regression (catagen), relative rest (telogen), and regeneration by new follicle down- growth and hair production (anagen) (Muller-Rover et al., 2001, Schneider et al., 2009). These phases and their transitions are orchestrated by signaling interactions between epithelial stem cells (SCs), transit amplifying progenitor cells (TACs) and their instructive mesenchymal niche, the dermal papilla (DP)(Zhang and Chen, 2024). Hair growth is driven by TAC proliferation and differentiation at the follicle base and guided by DP signals. After hair production ceases, during catagen most of the follicle regresses by apoptosis of epithelial cells in the outer root sheath. The uppermost population of these follicle cells selectively survives to give rise to the quiescent SCs of the next bulge (BuSCs) and the primed SCs of the new hair germ (HGSCs) underneath (Figure 1a) (Hsu et al., 2014, Ito et al., 2002). The DP cell cluster at the HF base also survives and relocates upward through mechanical forces until it arrives next to the HGSCs and BuSCs, entering the telogen phase of the hair cycle (Heitman et al., 2020, Martino et al., 2021). Communication among these three populations is essential for timely re-entry into anagen as well as sustaining HF cycling throughout life (Ito et al., 2005, Morgan, 2014, Rompolas et al., 2012): Signals from the DP activate primed HGSCs, which then proliferate and generate transit amplifying cells (TACs) that drive DP engulfment and HF downgrowth (Hsu et al., 2011). In turn, quiescent BuSCs respond to cues from the newly-formed TACs (Hsu et al., 2014) by migrating downwards and proliferating to contribute to the outer root sheath (Zhang et al., 2009), or by self-renewing to maintain the SC pool (Tumbar et al., 2004). Although it was proposed recently that the DP also directly interacts with BuSCs (Avigad Laron et al., 2018), the extent of this finding remains unclear.

Figure 1. Simultaneous isolation of DP, BuSCs and HGSCs during the hair re-growth wave of the first hair cycle.

(a) Schematic of a first-cycle telogen (~P21-P22) follicle with key cell populations. (b) Labeling and isolation strategy of seven skin and HF populations from K14-H2BGFP;Crabp1- GFP;Lef1-RFP reporter back skin using FACS purification for bulk RNA sequencing. (c) Fluorescence image of a first-cycle telogen follicle from K14-H2BGFP;Crabp1-GFP;Lef1-RFP mice back skin (P22). Follicle shows strong K14-H2BGFP expression in epithelial BuSC, HGSC and Foll cells above and RFP expression in the DP. At high exposure GFP from the Crabp1-GFP reporter is detected in the DP (inset). Scale bars, 25 μm. (d) Immunofluorescence for HGSC marker PCAD and BuSC marker CD34 on WT P22 back skin. Scale bar, 25 μm (e) FACS plots for isolating DF, DP, Foll, HGSC, BuSC and negative cells (Neg) from the dermal fraction of transgenic reporter back skin after immunofluorescence staining for PCAD and CD34. (f) qRT-PCR analysis of various known marker genes for each cell population. Data are mean ± SD. N = 2 mice. (g) Schematic of historic hair cycle propagation wave in the mouse from anterior head region to posterior tail region, divided into four zones. (h) Schematic of first-cycle regeneration wave depicting hair follicles (HFs) in four distinct hair development stages. (i) Immunofluorescence staining for KI67 and activated Caspase-3 (CASP3*) on K14-H2BGFP mouse back skin across the four zones. White arrows highlight positively stained cells. Scale bar, 50 μm. (j) Quantification of relative proportion of hair cycle phases with respect to each zone. Data are mean ± SD. Data points are individual averages from N = 10 mice. Statistical significance calculated by unpaired t-test. (k) Relative proportion of distinct hair stage distributions across four zones based on j. (l) Schematic of the observed first-cycle regeneration wave from Z1-Z4 showing higher levels of anagen II near the anterior, catagen near the posterior, and a mixed distribution of anagen I and telogen across all zones.

To better understand how SC quiescence, activation and the onset of new hair growth are regulated, we need to know the DP- and HFSC-intrinsic controls along with the cell-cell communications between HGSCs, BuSCs and the niche (Fuchs, 2016, 2018, Lee et al., 2021, Rompolas and Greco, 2014, Sennett and Rendl, 2012). Gaining this knowledge may also provide novel insights into pathological mechanisms in hair-loss diseases like androgenetic alopecia, which is characterized by successive miniaturizing hair growth cycles in which HFSCs and DP cells are increasingly lost (Garza et al., 2012). Over the past 25 years, groundbreaking profiling techniques have made important gene discoveries for uncovering the molecular mechanisms that govern HF regeneration. Early foundational microarrays defined telogen BuSC signatures and highlighted critical transcriptional controls and signaling factors (Morris et al., 2004, Tumbar et al., 2004). Subsequent studies further extended these findings, identifying molecular signatures of secondary germ and growth-promoting signals such as FGF7 from the DP (Greco et al., 2009), and profiling HFSCs and TACs (Lien et al., 2011). Bulk RNA- sequencing studies further refined our understanding of DP dynamics (Avigad Laron et al., 2018, Hagner et al., 2020) and BuSC regulation (Flora et al., 2021, Ge et al., 2017) across hair cycle stages, while more recent single-cell RNA-seq efforts have generated skin cell atlases capturing DP, BuSCs, and HGSCs across telogen and anagen hair stages and identified more regulatory factors (Joost et al., 2020, Liu et al., 2022, Yang et al., 2017). These studies have not only identified cell-intrinsic transcriptional controls but also unveiled cell extrinsic cell-cell communication signaling pathways offering many leads for central regulators of hair cycle progression.

Despite these advances, the field remains limited by key technical and conceptual gaps. Early profiling with microarrays restricted the depth of gene discovery and often lacked appropriate fibroblast or epidermal populations for comparison, precluding robust definition of cell-type specific signatures of enriched genes. Subsequent bulk RNA-seq studies improved the depth but, in many cases, still did not profile and compare SCs, niche and related cell types for rigorous comparison. Single-cell RNA-seq studies improved resolution but now suffered from limited sensitivity due to inferior detection of low-abundance transcripts, and in some cases lacked cell-type enrichment, compromising the fidelity of defined signatures. As a result, comprehensive and high-resolution transcriptomic maps of DP, BuSCs and HGSCs remain incomplete, leaving significant opportunities for integrative approaches that combine cell-type enrichment, advanced sequencing technologies, and comparative profiling across the regenerative stages.

In this study, we successfully overcame the cell-type abundance and sensitivity limitations of previous expression profiling efforts by utilizing high-sensitivity bulk RNA-seq of simultaneously isolated and highly enriched DP cells, BuSCs, HGSCs and several related epithelial and mesenchymal skin cells. This approach allowed us to establish all-encompassing and rigorous gene expression signatures during the first murine hair cycle. Repeat transcriptome measurements of all sorted cell types from four consecutive zones along the anterior-posterior axis in mouse back skin and their rigorous cross-comparisons enabled us to define molecular signatures of enriched genes for each cell type, which we classify into functional gene categories. Importantly, multi-way signature comparisons of the key HFSC and DP cell populations with the signatures from a total of 15 previously reported murine and human transcriptomics datasets from the past 20 years revealed many previously undescribed DP, BuSC and HGSC genes, all annotated in comprehensive supplementary tables. Moreover, by mapping signature ligands and receptors and other highly expressed cell surface and secreted factors, we established a detailed atlas of cell-cell communication among BuSCs, HFSCs and the DP. Finally, we feature all gene expression values on the web-accessible Hair-GEL.net online repository for gene expression exploration of HFSCs and their niche.

RESULTS

Simultaneous isolation of DP cells, BuSCs and HGSCs during the hair re-growth wave of the first hair cycle

To systematically dissect SC and niche gene expression patterns and SC-niche communication pathways, we sought to simultaneously isolate and transcriptionally profile DP cells, neighboring HGSCs, and BuSCs, along with related dermal fibroblasts (DF) as well as epidermal (Epi) and remaining epithelial HF cells (Foll) (Figure 1a). For this, we established a multicolor-fluorescence cell labeling system that enabled high enrichment of these cell types from the skin during the telogen phase of the hair growth cycle (Figure 1b). We generated K14-H2BGFP;Crabp1-GFP;Lef1-RFP triple transgenic reporter mice and digested postnatal day P22 back skins to obtain single cells from epidermis and the HF-containing dermis. Crabp1-GFP and Lef1-RFP together mark DP cells (Rezza et al., 2016) (Figure 1c, Supplementary Figure S1a and b). As Crabp1 expression can vary during hair cycle phases (Joost et al., 2020), we confirmed stable GFP expression in the DPs of P22 telogen and early anagen HFs (Supplementary Figure S1c). All epidermal and HF epithelial cells were highlighted by nuclear K14-H2BGFP (Rezza et al., 2016, Tumbar et al., 2004) (Figure 1c). After additional dual immunofluorescence of the dermal preparation for BuSC surface marker CD34 (Blanpain et al., 2004, Morris et al., 2004) and P-cadherin (PCAD), which enriches HGSCs during telogen (Greco et al., 2009) (Figure 1d), all cell populations were purified by fluorescence-activated-cell-sorting (FACS) for bulk RNA-seq (Figure 1b). Five-color FACS analysis of the dermal preparation distinguished single GFP+ HF epithelial cells from dual GFP+ RFP+ DP cells (Figure 1e). RFP+ GFP- cells comprised the DF population and arrector pili muscle cells. BuSCs and HGSCs were isolated as GFP+ RFP- CD34+ and GFP+ RFP- PCAD+ cells, respectively (Blanpain et al., 2004, Greco et al., 2009, Hsu et al., 2011), while remaining HF epithelial cells (Foll) that included infundibulum and sebaceous gland cells were GFP+ RFP- CD34- PCAD-. A marker-negative remaining dermal population (Neg) of adipocytes, vascular smooth muscle, immune and endothelial cells was collected as well. Finally, GFP+ epidermal epithelial cells (Epi) were sorted from a separately digested single- cell preparation of the epidermis (Supplementary Figure S1d). To validate the correct enrichment of the isolated populations, we performed qRT-PCR analysis for established marker genes (Figure 1f). Epithelial marker K14 was high in Epi cells, while Pcad and Shh were enriched in HGSCs (Greco et al., 2009, Hsu et al., 2011). Although ubiquitously present in DFs, Cd34 was enriched in BuSCs, along with BuSC genes Sox9 and Fgf18 (Blanpain et al., 2004). Lhx2 marked both BuSCs and HGSCs. Sox2, Enpp2, Hhip and Wif1 were highly enriched in DP cells (Driskell et al., 2009, Grisanti et al., 2013, Rendl et al., 2005) confirming the correct purification of the niche cell cluster. Col1a1 confirmed enrichment of DFs. Together, these data demonstrate the successful isolation of DP cells, BuSCs, HGSCs and related skin cell populations for cell type-enriched expression profiling during the telogen phase of the first hair cycle.

To capture the molecular signatures of DP cells and neighboring HGSCs and BuSCs during late regression, rest and early re-growth phases of the hair cycle, we sought to exploit the previously reported regenerative wave of hair re-growth that progresses uniformly from head to tail on mouse back skin during the first hair growth cycle (Figure 1g and h), a phenomenon described in studies dating back nearly a century (Butcher, 1934, Chase, 1954, Dry, 1926, Muller-Rover et al., 2001, Wang et al., 2023). Due to the short telogen phase, HFs along the back skin were reported to already be in early anagen closer to the head, in telogen in the middle, and still in late catagen near the tail, with the re-growth wave advancing towards the tail. To evaluate the hair cycle wave, we analyzed the hair cycle phases in four consecutive zones across the back skins of K14-H2BGFP mice.

Immunofluorescence for active caspase-3 (CASP3*) detected cell death during late catagen and KI67 marked cell proliferation in early anagen (Figure 1i, Supplementary Figure S1e). According to well- established criteria (Muller-Rover et al., 2001), Anagen I phase HFs had proliferating hair germs but no morphological signs of growth, while Anagen II HFs showed downgrowth and incipient engulfment of DP. Telogen phase HFs were negative for CASP3* or KI67. Quantification of hair cycle phases in the four zones showed a significant increase of late Catagen VIII HFs from head Zone 1 to posterior Zone 4 (Figure 1j). In contrast, Anagen I HFs significantly decreased from Zone 1 to Zone 4, and while Anagen II HFs also had a decreasing trend from upper to lower zones, this change did not reach significance. Although these findings supported the regenerative wave paradigm, both Anagen I and Telogen phase HFs were largely present in all four zones, suggesting widespread mixing of these hair cycle phases across mouse back skin. The summary of all phase counts confirmed the presence of a head-to-tail wave of hair cycle re-growth (Figure 1k), though only a fraction of HFs appeared to participate at that timepoint. Overall, these data supported an updated hair re-growth wave model (Figure 1l) that incorporates a broad, albeit heterogenous, head-to-tail wave with widespread co-existence of anagen- and telogen-phase HFs across mouse back skin zones.

Molecular signatures of DP cells, BuSCs and HGSCs during the first cycle hair re-growth wave

Given that catagen HFs were significantly more frequent in the posterior zones and anagen HFs predominated in the anterior zones, we proceeded with FACS purification of all cell populations from each of the four consecutive skin zones and from two biological replicates for high-sensitivity bulk RNA-seq (Figure 2a). The central back skin region from two K14-H2BGFP;Crabp1-GFP;Lef1-RFP littermates was used for cell isolations, while hair cycle phases were determined in lateral skin sections of each zone (Supplementary Figure S2a). Quantification of hair cycle phases again validated catagen and anagen enrichment in Zones 4 and 1, respectively, with similar distributions in both replicates (Supplementary Figure S2b). Comparable FACS profiles for all seven cell populations in the four zones confirmed consistent immunofluorescence performance and cell isolations (Supplementary Figure S2c). We then proceeded with population-based transcriptome analysis of all 56 sorted samples by multiplexed bulk RNA-seq using unique barcodes and deep sequencing (~75 million reads/sample), as previously described (Heitman et al., 2020, Martino et al., 2023, Mok et al., 2019, Rezza et al., 2016, Sennett et al., 2015) (Figure 2a, Supplementary Table S1). Principal component analysis (PCA) of all mapped genes showed clear separation between epithelial (BuSC, HGSC, Foll and Epi) and mesenchymal (DP, DF, Neg) populations across all zones, with DP cells displaying a distinct profile driven by the third principal component (Figure 2b). Notably, the cell populations did not sub-cluster further based on their zone of origin, suggesting a high similarity for each cell type across the four skin zones (Figure 2c). Next, ANOVA analysis identified a total of 10,566 differentially expressed genes (DEGs) across all populations and zones (false discovery rate FDR ≤ 0.05) providing a rich foundation to investigate gene expression patterns. Hierarchical clustering of all DEGs highlighted the differences between cell types and the epithelial and mesenchymal populations (Figure 2d). Interestingly, Epi and Foll populations appeared to mix, indicating a high similarity between these populations. Further zone-specific clustering within cell types was not observed.

Figure 2. Molecular signatures of DP, BuSCs and HGSCs during the first-cycle hair re-growth wave.

(a) Schematic of cell isolations from K14-H2BGFP;Crabp1-GFP;Lef1-RFP mouse back skin from four zones of two biological replicates. (b) Principal Component Analysis. All 7 cell types clearly segregate across 3 principal components (PCs). (c) Principal Component Analysis shown by zone origins. Segregation by anatomical zones was absent. (d) Hierarchical clustering of all cell populations using all 10,556 differentially expressed genes (DEGs). N = 2. (e) Table of signature gene counts for all cell populations across all zones. (f) Venn diagrams of gene signatures for Zone 1–4 highlighting shared genes of related cell types. (g) RPKM bar plots of previously known marker genes for DP, BuSC and HGSC. Data are mean ± SD. N = 2. Gene lists of all signatures and overlaps are provided in Supplementary Table S2.

As enriched expression of genes in specific cell types may predict functional importance, we next defined cell type-specific molecular signatures for each isolated cell population, and within each zone. For this, we cross-compared the Reads Per Kilobase of transcript per Million mapped reads (RPKMs) of all significant DEGs and defined signature gene status as expression average of RPKM ≥ 1 and a fold change ≥ 2 compared to all other cell populations (Figure 2e, Supplementary Table S2). For the BuSC vs. HGSC comparison, we applied a 1.5-fold change cutoff due to their highly overlapping gene expression profiles. DP populations of Zones 1, 2, 3, and 4 revealed 583, 575, 592, and 525 signature genes, respectively, when compared to all other populations in each zone (Figure 2e and f, non-overlapping sections; Supplementary Table S2). Similar levels of signature gene numbers were obtained for DF, Epi and Neg, while HGSCs, BuSCs and Foll had reduced gene numbers in each zone signature, suggesting a higher similarity compared to other cell populations. Additionally, we classified shared enriched genes among related cell types, in comparison to all other populations.

This highlighted joint signatures of DF and DP mesenchymal cells, Foll and Epi cells, and BuSCs and HGSCs (Figure 2f, intersections; Supplementary Table S2). Finally, initial inspection of the signatures revealed a robust presence of previously known DP genes such as Corin, Enpp2, Hhip and Wif1, the BuSC genes Sfrp1 and Sox9, and the HGSC genes Shh and Msx2 (Figure 2g). Overall, our comprehensive transcriptome analysis in four distinct skin zones established cell type-specific signatures in each zone for in-depth gene expression analyses and further exploration of SC-niche interactions during the first hair growth cycle.

Comparative gene expression analysis across skin zones

To explore gene expression changes that could play a role in regulating the rest-to-growth transition during the hair regeneration wave, we calculated significant DEGs across the DP transcriptomes of skin zones. For DEGs to be considered, we employed an average expression cutoff of RPKM ≥ 1. Of all 24,062 measured genes, a total of 10,501 genes (43.7%) were expressed in the DP of all four zones (Supplementary Figure S3a) and 13,237 genes were expressed in at least one zone (Supplementary Figure S3b). Despite the large number of DP-expressed genes, the pairwise DEG analyses (FDR ≤ 0.05, FC ≥ 2) across all four zones yielded only 211 DP genes (1.6% of expressed DP genes) that were significantly changed between any two zones (Supplementary Figure S3b). Moreover, many DEG calls were low quality with highly variable expression values among biological replicates. Hierarchical clustering failed to reveal clear patterns of co-regulated genes (Supplementary Figure S3c). Among the few high-quality DEGs were members of the homeobox family. Known to specify the body plan along the head-to-tail axis (Wellik, 2007, Yu et al., 2018), Hoxa9, Hoxc9 and Hoxc10 were increased in the lower back skin zones in the DP but also in other cell populations, suggesting broad regulation according to body axis position rather than hair cycle phases (Supplementary Figure S3d). Other strongly regulated DEGs were highly enriched DP signature genes such as solute carrier family genes, Slc12a1 and Slc5a3. Slc genes are known to be involved in ion transport across cell membranes (Hebert et al., 2004) and were upregulated in the lower back skin zones of DP (Supplementary Figure S3e). Fluorescence in situ hybridization (FISH) verified increased Slc12a1 and Slc5a3 expression in DPs of Zone 3 and 4 HFs (Supplementary Figure S3f and g). Slc12a1 and Slc5a3 expression was highly similar in both telogen- and anagen I-phase HFs suggesting that the zone-enriched expression is linked to body position and unrelated to hair cycle phases (Supplementary Figure S3h and i). Other signature genes among the 211 DEGs included Fgf10, Corin, and Sox11, previously reported in DP or its dermal condensate precursor cells during development (Enshell-Seijffers et al., 2008, Greco et al., 2009, Lee et al., 2024). Here, expression was high in all zones and fold changes between some zones were barely above 2-fold cutoffs (Supplementary Figure S3j). Overall, our DEG analysis showed that DP gene expression was highly consistent across all skin zones, with only a few DEGs displaying strong expression differences across zones; these reflected back skin position rather than hair cycle phase-specific expression. We conclude that likely due to the widespread mixing of telogen and anagen I phases, our across-zone comparisons did not reveal hair cycle phase-specific DP gene regulation.

Beyond the known: Defining Core and Expanded DP molecular signatures

While zone-comparison analyses failed to reveal hair cycle phase-specific gene expression regulation, the multiple repeat transcriptome measurements across the back skin offer the unique opportunity to classify, with high rigor, a definitive molecular signature of the adult DP. For this, we first defined a “Core” DP signature across all four zones by overlapping the individual DP zone signatures, each itself determined by two biological replicates. Of the 998 genes with DP signature call in at least one zone, 273 qualified as the highly stringent Core signature by receiving signature status in all four zones (4-Zones signature) (Figure 3a, Supplementary Table S3). To test the potential impact of the Core signature, we first explored a custom-curated list of 42 “known” DP genes that were previously expression-validated throughout the years by in situ hybridization or immunostaining (Figure 3b, Supplementary Text S1, Supplementary Table S3). Among those, a significant majority of 30 genes (71.4%) was present in the Core signature (Figure 3b, Supplementary Table S3), with an average 22-fold increased expression compared to DF (Figure 3c, Supplementary Figure S3). The Core DP genes included well-known ligands, enzymes and transcription factors such as Corin, Enpp2, Fgf7, Lef1, Lepr and many others (Clavel et al., 2012, Enshell-Seijffers et al., 2008, Grisanti et al., 2013, Rendl et al., 2005, Rosenquist and Martin, 1996). Four additional genes received signature status in three of four zones (3-Zones), and two genes were in any 1-Zone signature (Figure 3c, Supplementary Figure S5a and b). Only six validated genes lacked DP signature status due to absence of detection in DP or high expression in other populations (Figure 3c, Supplementary Figure S5c). Overall, with this analysis we established a useful collection of validated DP genes and demonstrated that the Core signature reliably detected those. Importantly, the 273 Core signature revealed many additional robust DP genes for future studies.

Figure 3. Defining Core and Expanded DP molecular signatures.

(a) Venn diagram of DP signatures from all four zones results in a Core DP signature of 273 genes out of a total of 998 signature genes present in any zone. (b) Venn diagram of the Core DP and genes present in 3-zones, 2-zones and 1-zone signatures with 42 DP marker genes that were previously experimentally validated. Note a large majority of previously known genes was present in the Core DP signature. *p = 1.27e-9). (c) Relative expression of 42 previously validated DP markers in the DP compared to lineage-related DF. Fold change levels are color-coded for DP signature status in zones. Individual zone replicate data are color-coded in shades of grey. Data are mean ± SD. N = 4 Zones from 2 mice. (d) Relative expression of Core DP signature genes compared to DF in selected functional categories. Previously validated genes are highlighted in bold. Individual zone replicate data are color-coded in shades of grey. Data are mean ± SD. N = 4 zones from 2 mice. (e) Venn diagram comparing the Expanded DP signature (union of Core, 3-zones, 2-zones signatures) with DP signatures from previously published microarray, bulk- and single-cell RNA-seq studies at telogen. (f) Table of DP signature genes that overlap with DP genes identified in previous studies shown in e, and newly identified genes in this study (Expanded DP only). Selected newly identified Core DP genes are highlighted. Genes are organized in functional categories. (g) RPKM bar plots of selected DP signature genes. Data are mean ± SD. N = 2 mice. (h) Immunofluorescence for AQP5 and DCC on K14-H2BGFP mouse back skin validated expression in telogen DP. Scale bar, 25 μm. (i) Venn diagram comparing the Expanded DP signature with human DP signatures from previous microarray and single-cell RNA-seq profiling studies. Examples of overlapping genes are highlighted in the table on the right. All gene lists are provided in Supplementary Table S3.

To gain mechanistic insights into potential functions of Core DP genes, we analyzed Gene Ontology (GO) categories using EnrichR (Chen et al., 2013). Significantly enriched GO terms included extracellular matrix organization and nervous system development, as well as regulation of macromolecule metabolic process and transcription, signaling pathways and receptor ligand activity, among many others (Supplementary Figure S6a, Supplementary Table S3). We next grouped all 273 genes into 10 main functional themes (Figure 3d and Supplementary Figure S6b and c).

Unsurprisingly for a signaling niche, most previously validated and many new Core DP genes were among ligands and signaling factors, all highly expressed compared to DF and all other populations (Figure 3d; validated genes are highlighted in bold). Besides known ligands Bmp4, Dkk2 and Mdk (Chu et al., 2014, Kulessa et al., 2000, Rendl et al., 2005), previously underappreciated ligands included Edil3 (epidermal growth factor-like repeats and discoidin I-like domains 3), Ndp (Norrin) and Pgf (placental growth factor), which are secreted glycoproteins that interact with integrins, frizzled and VEGF receptors, respectively (Niu et al., 2023, Van Bergen et al., 2019, Wang et al., 2012). They affect angiogenesis and may regulate the blood vessel supply of the DP. Tfpi (tissue factor pathway inhibitor) is a secreted regulator of blood coagulation (Mast, 2016). Other DP-enriched signaling factors included Sorbs1 and Spry1. Sorbs1 (sorbin and SH3 domain-containing protein 1) is an adapter protein for insulin signaling (Lin et al., 2001), and Spry1 as a receptor tyrosine kinase pathway inhibitor may affect FGF and EGF signaling (Michos et al., 2010). The Core DP signature also contained many receptors indicating that the DP is receiving signals as well (Figure 3d). Known Hhip and Lepr are joined by previously undocumented receptors such as Ackr4, Dchs1, Lingo1, Mc5r and Npr2. Ackr4 (Atypical Chemokine Receptor 4) regulates immune functions by scavenging chemokines (Nibbs and Graham, 2013). Mc5r (melanocortin receptor 5) is potentially involved in pigmentation, sebum generation and inflammation (Eisinger et al., 2011). Other Core signature genes included enzymes and genes encoding cell structure regulation, metabolism and transport proteins, such as Dio2 (Deiodinase, Iodothyronine, Type II) and Plod2 (Procollagen-lysine, 2-oxoglutarate 5-dioxygenase 2) (Supplementary Figure S6b and c). Dio2 converts inactive thyroid hormone thyroxine into active T3 triiodothyronine (Fonseca et al., 2013) which may regulate local thyroid hormone levels in the DP. Plod2 regulates collagen biosynthesis (Wu et al., 2006) possibly ensuring proper structural integrity of the DP. Presence of uniquely enriched adhesion and extracellular matrix genes, such as Adamts4, Cdh11, Col13a1, Lama2 and Lamc3, suggested that the DP shapes its own extracellular environment. As previously described for morphogenetic anagen after birth (Rendl et al., 2005, Rezza et al., 2016), the adult DP also expressed neuronal-type genes such as Nlgn1 (Neurolignin-1), Ndnf (neuron-derived neurotrophic factor), and Shox2 (SHOX homeobox 2) that may be involved in nerve-like cell communication. The Core DP signature also featured several transcription factors and regulation genes such as known Lef1, Runx3, Sox11 and Twist1 as well as— to our knowledge—previously unreported factors including Cbx6, Ebf4, Flis1, and Glis3, many of which may regulate the DP-specific cell fate. Finally, we explored genes of the 3-Zones and 2-Zones signatures (3x and 2x signature status in four zones) and their functional categories. This identified many additional signaling molecule genes such as Edn3, Epha7, Mc2r, Robo2, Serpine1, Wnt10a and transcription factors Meis1, Meis2, Prrx2, Zic1 that in concert may regulate important DP functions as a SC-activating niche (Supplementary Figure S6d). Together, the plethora of DP signature genes suggests that this large collection of highly enriched DP genes may control many cell type-specific functions.

We next combined the most stringent Core (4-Zones) signature with signature genes found in three of four zones (3-Zones) and in any two zones (2-Zones) to include all DP genes with repeat signature status. We then compared this “Expanded” signature of 584 DP genes with four previously published DP signatures to establish an overarching refined telogen signature. Published DP gene lists were used to honor the original signatures or, if unavailable, the datasets were reanalyzed and DP signatures generated with established standardized methods (see “Signature gene lists for cross-comparison studies” in Supplementary Methods). The second hair cycle signatures came from 1) reanalyzed microarrays (Greco et al., 2009) and 2) a published list of bulk RNA-seq profiles (Hagner et al., 2020), both from FACS-sorted DP cells, as well as from 3) published DP gene lists (Joost et al., 2020) and 4) reanalyzed profiles (Liu et al., 2022) of unenriched single-cell transcriptomics data. Some high-quality DP datasets could not be used for signature generation as no other cell type was measured (Avigad Laron et al., 2018). All five studies significantly shared genes in their overlaps (Supplementary Table S3). 5-way comparisons of our Expanded signature and the four published telogen signatures revealed a remarkable abundance of overlapping and distinct features (Figure 3e and f, Supplementary Table S3). Jointly identified by and significantly enriched in all five studies were 41 DP genes. Shared between our Expanded signature and three of the four previous signatures were additional 48 genes. These repeatedly identified 89 DP genes contained transcription factors Lef1, Prdm1, Shox2, Trps1, Twist1, signaling ligands Bmp4, Dkk2, Gdf10, Mdk and Serpine2, and receptors Bambi, Fgfr1, Lepr, and Npr2 (Figure 3f). 215 more genes were significantly shared with any two or one of the published signatures. Notably, 280 additional DP genes were discovered only in our Expanded signature and not in any previous study (Figure 3e and f, Supplementary Table S3). Importantly, of these 280 DP genes a total of 102 genes were in the DP Core (4-Zones) signature suggesting that our study revealed many previously unidentified high-confidence DP signature genes. Indeed, visualizing expression levels of Aqp5, Daam2, Dcc and Ebf4 highlighted the repeat signature status in the DP (Figure 3g). Aqp5 (Aquaporin 5), a water channel that plays a crucial role in saliva production of salivary glands (Hosoi et al., 2020) likely plays a similar role in water transport in DP cells, and Dcc may play a role in cell adhesion (Driskell et al., 2009).

Immunofluorescence for AQP5 and DCC confirmed DP specific expression at the protein level (Figure 3h). Of note, most if not all 1586 genes identified in previous signatures but absent in our Expanded signature (Figure 3e and Supplementary Table S3) were indeed expressed in the DP but also high in DF and/or other cell types and thus were not considered as truly enriched DP signature genes based on our measurements and stringent criteria. Together, the comparison with published telogen signatures validated many DP genes across five studies and unearthed many—to our knowledge—previously unreported DP genes found only in our signatures underscoring the sensitivity and specificity of our cell enrichment, profiling and analysis approach.

Finally, we explored DP genes that are expression-conserved across development and between mouse and human DP. For this, we first compared the Expanded DP signature across developmental stages ranging from embryonic DP progenitors in dermal condensates at embryonic day E14.5 (Sennett et al., 2015) to DP from morphogenetic anagen at P5 (Rezza et al., 2016) and early adult anagen at P26 (Hagner et al., 2020) (Supplementary Figure 6e, Supplementary Table S3). This analysis revealed a cross-developmental DP signature collection of 15 DP genes detected at all ages including signaling ligands Edn3, Fgf10, Inhba, Rspo3, receptors Hhip, Prlr, Unc5c, and other genes. We next placed the Expanded signature from murine HFs in the context of transcriptome information from human DP (Figure 3i, Supplementary Table S3). We compared the 584 DP genes with microarray-derived human DP signatures (Higgins et al., 2013, Ohyama et al., 2012) and with a DP gene list derived from our reanalysis of a recent human single-cell RNA-seq dataset (Ober-Reynolds et al., 2023). This revealed a significant overlap between all four studies (Supplementary Table S3). Four DP genes, Dio2, Edn3, Lamc3 and Spon1 were conserved, and 13 genes were shared with at least two human datasets. 39 genes were shared with the human single-cell-derived signature alone suggesting a substantial amount of mouse-human DP gene conservation of many ligands, receptors and transcription factors (Figure 3i). Overall, our multiple repeat transcriptome measurements in four distinct skin zones enabled the definition of rigorous Core and Expanded DP signatures, ripe for exploration of SC-niche interactions during the first hair growth cycle. Cross-comparisons with reported or newly generated DP gene lists from published microarray, bulk RNA-seq and single-cell transcriptomic data established a robust telogen signature as well as a developmental and a joint human-mouse DP signature, all available in well-annotated Supplementary Tables for further exploration.

Defining BuSC Core and Expanded molecular signatures

Quiescence and activation of BuSCs is regulated by intrinsic transcriptional controls and extrinsic signaling niche inputs throughout the hair growth cycle. We thus defined a Core molecular signature of BuSC identity with the aim of uncovering new high-probability regulators of their fate and function. For this, we performed comprehensive gene expression analyses of the quadruple BuSC transcriptome measurements and placed the individual gene signatures in the context of previously published gene expression knowledge. The BuSC population contained a total of 425 genes present in at least one of the four zone signatures (Figure 4a and b, Supplementary Table S4). Of those, 22 genes received signature status in all four zones representing a stringent Core BuSC signature. Next, we explored the 425 gene BuSC signature by inspecting overlaps with a curated set of 17 expression-verified BuSC genes at telogen (Figure 4b and c, Supplementary Text S2, Supplementary Figure S4). Wnt signaling inhibitor Sfrp1 (Lim et al., 2016), transcription factor Sox9 (Vidal et al., 2005) and Wnt signaling co-receptor Lgr5 (Jaks et al., 2008) were expressed in the Core and 3-Zones signatures (Figure 4c, Supplementary Figure S7a and b). Fgf18 (Kawano et al., 2005), Krt6a (Blanpain et al., 2004), and Tnc (Tumbar et al., 2004) were expressed in 2-Zones, and Foxp1 (Leishman et al., 2013), Hopx (Takeda et al., 2013), and Lgr4 (Ren et al., 2020) were found in 1-Zone signatures (Figure 4c, Supplementary Figure S7c and d). The remaining eight validated BuSC genes did not acquire signature status due to simultaneous high expression in other cell populations (Figure 4c, Supplementary Figure S7e). We next classified the Core BuSC genes into functional categories, which highlighted the presence of ligands, including known Sfrp1 (Figure 4d). Amongst receptors were Il31ra (Interleukin 31 receptor A), a type I cytokine receptor primarily for IL-31, and Tnfrsf11b (Tumor necrosis factor receptor superfamily 11b), which along with signaling molecules Samd12 and Smad9 may promote HFSC activation and hair growth (Figure 4d). Enzymes including HK2 (Hexokinase 2), involved in glucose metabolism, and Tgm5 (Transglutaminase 5), responsible for protein cross-linking, may regulate energy supply and structural integrity of the bulge. Hmcn1 (Hemicentin 1) and Vwa2 (Von Willebrand Factor A Domain Containing 2) are both extracellular matrix proteins (Dong et al., 2006) that may play a role in BuSC cell-cell adhesion. Additional Core enriched genes included integral membrane protein Ank (Ankyrin), Esyt3 and Trpv4, involved in ion channel transport (Figure 4d). Finally, we inspected the functional categories of BuSC signature genes of the 3- and 2-Zones signatures. Among the 82 signature genes, we identified many additional signaling molecules (Ctgf, Diras1, Diras2, Fgf18, Nppc, Pthlh, Rims1, Shisa2) and transcription factors (Creb5, Dlx4, E2f8, Egr4, Foxm1, Id4, Sox12, Zfp473) that may regulate important BuSC functions such as safeguarding BuSC-intrinsic quiescence and self-renewal mechanisms (Supplementary Figure S8). Overall, our analysis suggests that the Core, 3- and 2-Zones BuSC signatures contain many robustly expressed, yet less well-known BuSC genes that may regulate fate and function in the bulge compartment.

Figure 4. Defining BuSC Core and Expanded molecular signatures.

(a) Venn diagram of BuSC signatures from all four zones results in a Core BuSC signature of 22 genes out of a total of 425 signature genes present in any zone. (b) Venn diagram of the Core BuSC signature, signature genes present in 3-zones, 2-zones and 1-zone, and of 17 BuSC marker genes that were previously experimentally validated. (c) Relative expression of 17 previously validated BuSC markers in BuSCs compared to lineage-related HGSCs. Fold change levels are color-coded for BuSC signature status in zones. Individual zone replicate data are color-coded in shades of grey. Data are mean ± SD. N = 4 Zones; N = 2 mice. (d) Relative expression of Core BuSC signature genes compared to HGSC in selected functional categories. Previously validated genes are highlighted in bold. Individual zone replicate data are color-coded in shades of grey. Data are mean ± SD. N = 4 zones from 2 mice. (e) Venn diagram comparing an Expanded BuSC signature (union of Core, 3-zones, 2-zones signatures) with BuSC signatures from previously published microarray, bulk- and single-cell RNA-seq studies at telogen. (f) Table of BuSC signature genes that overlap with BuSC genes identified in other studies shown in e, and newly identified genes in this study (Expanded BuSC only). Selected newly identified Core BuSC genes are highlighted. Genes are organized in functional categories. (g) RPKM bar plots of selected BuSC signature genes. Data are mean ± SD. N = 2 mice. (h) Fluorescence in situ hybridization (FISH) images of Tgfb3 and Scnn1g on K14-H2BGFP mouse back validate expression in BuSCs. Scale bar, 25 μm. (i) Venn diagram comparing the Expanded BuSC signature with human BuSC signatures from previous single-cell RNA-seq profiling studies. Overlapping genes are highlighted in the table on the right. All gene lists are provided in Supplementary Table S4.

Next, we placed the Expanded BuSC signature of the combined Core, 3- and 2-Zones signatures (104 genes) into the context of six previously published BuSCs gene signatures at telogen (Figure 4e and f). They included the original and foundational 20-year-old BuSC signature lists from FACS-sorted BuSCs profiled with the earliest iterations of microarrays (Morris et al., 2004, Tumbar et al., 2004); second hair cycle BuSC signatures from reanalyzed microarray (Greco et al., 2009) and bulk RNA-seq (Ge et al., 2017) profiles of FACS-sorted BuSCs; a published signature from unenriched single-cell transcriptomics (Joost et al., 2020); and a first-cycle signature list from single-cell profiling of FACS-enriched BuSCs (Yang et al., 2017). 7-way cross-comparisons of the Expanded signature genes with all six published signatures revealed an extraordinary wealth of overlapping and unique features of BuSC genes (Figure 4e, Supplementary Table S4). All seven studies significantly shared genes in their overlaps (Supplementary Table S4). Common among all seven signatures and shared between our Expanded BuSC signature and at least five, four and three other signatures were 4, 5, 7 and 15 genes, respectively (Figure 4f). This stringent list of 31 BuSC genes contained transcription factors Id4, Sox9, Tbx1, signaling ligands Ccn2, Sema3e, Fgf18, Sfrp1 and Tnc, and receptors Il31ra, Lgr5, and Tnfrs11b. Highly overlapping genes of these and other functional categories were part of our Core signature including Chat, Trpv4, Tgm5 and Vwa2. Importantly, 48 BuSC genes were only found in our Expanded signature (Figure 4e and f). Many of those were Core BuSC genes with previously unrecognized expression in BuSCs such as Dnah7b, Scnn1g, Strip2 and Tgfb3 (Figure 4g). Dnah7b (Dynein Axonemal Heavy Chain 7) regulates microtubule motor activity, Scnn1g encodes a sodium channel crucial for maintaining electrolyte balance across epithelial tissues (Hanukoglu and Hanukoglu, 2016), while Strip2 (Striatin Interacting Protein 2) is potentially involved in cell migration, cytoskeleton organization, and regulation of cell shape (Sabour et al., 2017). Validation by FISH confirmed BuSC expression of Scnn1g and Tgfb3, which appears to be a more specific BuSC marker than previously thought (Lin and Yang, 2013) (Figure 4h). Most genes identified in previous BuSC signatures that were not detected in our Expanded signature (Figure 4e and Supplementary Table S4) were still expressed in BuSCs but were also high in other cell population and therefore did not receive signature status in our analysis. Together, the multi-dataset comparisons verified BuSC signature genes across several studies and uncovered many—to our knowledge—previously unreported BuSCs genes for further exploration of BuSC functions.

Finally, we compared the murine BuSC signatures to their human counterparts from three recently published single-cell transcriptomics efforts (Ober-Reynolds et al., 2023, Takahashi et al., 2020, Wu et al., 2022). This revealed a low but significant overlap between each of the four studies (Supplementary Table S4). Sfrp1 was present in all human signature datasets (Figure 4i, Supplementary Table S4). 12 genes of our Expanded signature were shared with the human bulge signature from Ober-Reynolds et al., suggesting conservation between mouse and human BuSC genes (Figure 4i). Together, the analysis indicated that gene expression profiles of human and mouse BuSCs may be relatively distinct; however, it should be noted that overall low overlap may result from the fact that human BuSC profiles were from anagen HFs. Overall, the BuSC gene expression analyses of our Expanded signature and comparison with multiple previously reported signatures established a comprehensive collection of enriched BuSC genes. Several novel BuSC genes were undetected in previous studies and thus our signatures provide a valuable resource for investigating SC functions in regulating BuSC quiescence and activation.

An updated molecular signature of primed SCs in the hair germ

Primed HGSC remain quiescent during telogen before receiving activating signals from the DP, upon which they respond with a burst of proliferation and new growth activity. As such, HGSCs are the primary communication partner of the neighboring DP, but the underlying molecular controls are incompletely understood. Since isolated and enriched HGSCs have last been profiled in 2009 with now-outdated microarray technology (Greco et al., 2009), we aimed to update the molecular insights into their gene expression by sensitive bulk RNA-seq. We first defined a stringent HGSC Core signature: of 516 signature genes present in at least one back skin zone, 36 HGSC genes were shared across all four zones (Figure 5a, Supplementary Table S5). Analogous to DP cells and BuSCs, we then explored expression-validated HGSC genes in our signatures (Figure 5b and c, Supplementary Figure S9). Ligand Shh and transcription factors Cux1, Hoxc13, Lhx2, Msx2 significantly overlapped with a curated list of 14 known HGSC genes (Supplementary Text S3, Supplementary Figure S9a and Supplementary Table S5). Cdh3 (P-cadherin), Myb and Wnt10b also significantly overlapped with the 3-Zones signature (Figure 5c, Supplementary Figure S9b). Dnmt1 and Runx1, highly expressed in the germ and lower bulge (Osorio et al., 2008), were present in 2- and 1-Zones signatures, respectively (Figure 5c, Supplementary Figure S9c and d). The remaining four known HGSC genes without signature status were also highly expressed in BuSCs, DP or upper follicle (Foll) (Figure 5c, Supplementary Figure S9e). Together, the HGSC signatures contained most previously validated genes and included—to our knowledge—previously unreported genes whose roles in the hair growth cycle have yet to be explored.

Figure 5. An updated molecular signature of primed SCs in the hair germ.

(a) Venn diagram of HGSC signatures from all four zones results in a Core HGSC signature of 36 genes out of a total of 516 signature genes present in any zone. (b) Venn diagram of the Core HGSC signature (*p = 0.0015), signature genes present in 3-zones (*p = 0.041), 2-zones and 1-zone, and of 14 HGSC marker genes that were previously experimentally validated. (c) Relative expression of 14 previously validated HGSC markers in HGSCs compared to lineage-related BuSCs. Fold change levels are color-coded for signature status in zones. Individual zone replicate data are color-coded in shades of grey. Data are mean ± SD. N = 4 Zones; N = 2 mice. (d) Relative expression of Core HGSC signature genes compared to BuSC in selected functional categories. Previously validated genes are highlighted in bold. Individual zone replicate data are color-coded in shades of grey. Data are mean ± SD. N = 4 zones from 2 mice. (e) Venn diagram comparing the Expanded HGSC signature (union of Core, 3-zones, 2-zones signatures) with HGSC signatures from previously published microarray and single-cell RNA-seq studies at telogen. (f) Table of HGSC signature genes that overlap with HGSC genes identified in other studies shown in e, and newly identified genes in this study (Expanded HGSC only). Selected newly identified Core HGSC genes are highlighted. Genes are organized in functional categories. (g) RPKM bar plots of selected HGSC signature genes. Data are mean ± SD. N = 2 mice. (h) Fluorescence in situ hybridization (FISH) images of Basp1 and Slc4a11 on K14-H2BGFP mouse back validate expression in HGSCs. Scale bar, 25 μm. (i) Venn diagram comparing the Expanded HGSC signature with human TAC signatures from recent single-cell RNA-seq profiling studies. Overlapping genes are highlighted in the table on the right. All gene lists are provided in Supplementary Table S5.

Next, we classified the 36 Core HGSC genes in different functional categories (Figure 5d and Supplementary Figure S10a). HGSC-expressed enzymes, such as metallopeptidase Adamts17 and hyaluronan synthase Has3 may regulate ECM turnover (Apte, 2009, Malaisse et al., 2014). Signaling molecules Sema4g, Wnt7a and Stra6 may regulate HGSC proliferation and differentiation. A plethora of transcription factors such as Bach2, Dlx1, Lbh, Sox5 and Sox13 were also part of the Core signature (Figure 5d). Lbh, a developmental transcription activator (Young et al., 2023) and SoxD family members Sox5 and Sox13 involved in neuronal and immune cell fate specification (Lefebvre, 2010), may regulate HGSC quiescence and activation. Moreover, we found several genes associated with neuronal development (Chrna4, Igsf9, Shroom3), extracellular matrix and membrane structure (Fras1, Frem2, Tspan18), as well as transmembrane transport (Slc4a11, Slc40a1, Slc7a5) as part of the Core signature (Supplementary Figure S10a). Many additional transcription factors (Ascl4, Dlx2, Etv4, Hoxc12, Snai3), signaling molecules (Rasd2, Srgap1), ligands (Efnb1, Kitl, Npy, Pdgfa) and receptors (Cntfr, Nrp2, Ptch2) were found in the 3- and 2-Zones signatures (Supplementary Figure S10b), supporting the paradigm of HGSCs serving as the key signaling partner of the DP in orchestrating HF cycling.

We subsequently explored the Expanded HGSC signature (Core, 3-Zones, 2-Zones) in the context of previous profiling studies of the secondary hair germ during telogen (Figure 5e). Greco et al. were the first to isolate HGSCs by FACS and establish signatures by microarrays (Greco et al., 2009). Yang et al. and Joost et al. applied single-cell RNA-seq to FACS-purified and unenriched HGSCs, respectively (Joost et al., 2020, Yang et al., 2017). All four studies significantly shared genes in their overlaps (Supplementary Table S5). A 4-way comparison of the Expanded HGSC signature (140 genes) with the three previously defined signatures revealed 4 common genes and 13 that overlapped with at least two published signatures (Figure 5e and f, Supplementary Table S5). Highly overlapping genes included transcription factors Msx2, Hoxc13, Lbh, receptors Nrp2, Ptch2 as well as less well-known genes such as Abi3bp, Arg2, Mtmr7, Srgap1. A total of 80 genes were discovered only in our Expanded HGSC signatures that included transcription factors, ligands and many other factors such as Bach2, Basp1, Krt28 and Slc4a11 (Figure 5f and g). Visualizing their expression levels showcased the Core signature status in HGSCs. Basp1 is a membrane-bound signaling protein with neurogenic roles (Mac Donald and Iulianella, 2022) and Slc4a11 regulates water transport with a potential impact on cell growth and proliferation (Romero et al., 2013). FISH validated their expression in HGSCs close to the DP junction (Figure 5h). Together, our analysis uncovered many—to our knowledge—previously unreported HGSC genes for future functional studies of HGSC quiescence and hair cycle progression regulation.

Finally, we compared murine HGSCs to their human counterparts from recent single-cell RNA-seq datasets (Ober-Reynolds et al., 2023, Takahashi et al., 2020, Wu et al., 2022) (Figure 5i, Supplementary Table S5). We observed a rather limited, yet significant, overlap between the four studies, potentially because all human samples were from anagen TACs, or because human and mouse HGSC/TAC expression may be less conserved. Nevertheless Dlx3, Hoxc13 and Krt35 were shared between all human populations (Figure 5i). Although subtle, these findings suggest a physiological relevance of these genes in both mouse and human HF biology. Overall, the characterization of our HGSC signatures is valuable for investigating SC-niche interactions.

Unraveling cell-cell communication between DP cells, HGSCs and BuSCs

Upon receiving activating signals from its neighboring DP, primed HGSCs in the hair germ fuel new downgrowth from telogen HFs. HGSCs signal back to the DP and proliferate to give rise to the transit amplifying cells of growing HFs, but the extent of this mutual cell-cell communication is poorly understood. HGSCs also activate the quiescent SCs of the bulge to self-renew or migrate and proliferate to supply cells to the outer root sheath of the downgrowing HF (Hsu et al., 2014). It is less clear whether DP signals can also reach and activate BuSCs in this process directly, although DP-BuSC interactions were proposed in the context of SHH signaling during hair growth (Avigad Laron et al., 2018). Equally unknown is whether cell-cell communication between DP and HGSCs or even between DP and BuSCs plays a role in keeping SCs quiescent before growth activation. To better understand signaling and other cell-cell interactions between DP, HGSCs and BuSCs during the first hair cycle, we performed a ligand-receptor analysis of our signatures and an unbiased cell-cell communication exploration of all expressed genes using CellChat (Jin et al., 2025).

We first mapped the DP, HGSC and BuSC signatures onto a comprehensive ligand-receptor interaction database (Ramilowski et al., 2015). The curated collection of 708 ligands and 691 receptors feature 2,557 unique interaction pairs of potential cell-cell communication. For ligand- receptor matches, genes within the Expanded DP, HGSC and BuSC signatures and with RPKM ≥ 5 expression levels in all zones were considered. This stringent analysis of only 78 enriched ligands and receptors yielded 10 paracrine and 10 autocrine ligand-receptor pairs (Figure 6a and b, Supplementary Table S6). Among the DP-HGSC pairs were several well-known signaling factors. Dkk2→Kremen2 interaction inhibits WNT activity by internalizing co-receptor Lrp5/6 (Mao et al., 2001) suggesting tight control of WNT activity in the germ (Figure 6a and b). Pgf binding to Nrp2 has suggested roles in angiogenesis (Escudero, 2009). Interactions of HGSC-expressed ligands and DP receptors include Shh→Hhip, which inhibits Hedgehog signaling by SHH sequestering (Griffiths et al., 2021). Conversely, in a Shh→Gpc5 interaction, Gpc5 may promote Hedgehog signaling by stabilizing Ptch1 binding (Li et al., 2011). Potential interaction calls from DP to BuSCs included all four R- spondins, highly expressed in the DP, and may reach Lgr5, expressed by BuSCs. Besides their proposed roles in HGSCs (Jaks et al., 2008), Rspo→Lgr interaction may function in BuSCs to promote WNT signaling activity (Carmon et al., 2011). Similarly, in the reverse direction, bulge-derived extracellular glycoprotein Tenascin-C (Tnc) may reach and bind to DP-expressed Contactin (Cntn1), an interaction known to regulate axonal outgrowth in neurons (Michele and Faissner, 2009). Additionally, Tgfbr1 in the DP might receive bulge-secreted Tgfb3, potentially influencing TGF-β signaling. Finally, considerable autocrine signaling may also take place in the DP and to some extent also in the HG. DP-derived Edn3, normally involved with pigmentation regulation in melanocytes (Yaar and Park, 2012), may signal to its receptor Ednra. Likewise, several FGF family members including Fgf7 and Fgf10 (Greco et al., 2009), but also Fgf2 and Fgf11 may bind to DP-expressed Fgfr1 receptor and initiate autocrine signaling, as may TGF-β family member Inhba (Inhibin-A) at the Bambi receptor. The hedgehog pathway could also be activated with HGSC-expressed Shh signaling to Ptch2 receptor on the same cells.

Figure 6. Ligand-receptor and other cell-cell communication between DP, HGSCs and BuSCs.

(a) Gene expression heatmap and pairings of 41 and 37 DP, BuSC and HGSC ligands and receptors, respectively, present in the Expanded signatures. Ligands, receptors and pairings were matched from the Ramilowski database (Ramilowski et al, 2015) (Supplementary Table S6). Arrows indicate interaction pairs and are color-coded by source for paracrine signaling. Arrows of autocrine signals are in black. (b) Summary of ligand-receptor interactions and downstream signaling pathways. (c-h) Chord diagrams of CellChat analysis of paracrine communication pathways between DP, BuSC and HGSC compartments. Directionality of communication is indicated above each chord diagram and by the arrows. Each plot shows a selection of 30 ligand-receptor pairs. A complete list of all interactions is provided in Supplementary Table S6.

While the strength of gene signatures of highly-enriched genes lie in identifying cell type-specific unique functions, any ligands and receptors highly expressed in more than one cell population, and thus with multiple functional roles, will be missed. Therefore, we expanded our cell-cell communication assessment to all highly expressed genes in our transcriptomes. Using the CellChat analysis platform (Jin et al., 2025), we considered genes with RPKM ≥ 20 and that were categorized as either ‘Secreted Signaling’ or ‘Cell-Cell Contact’ in the CellChat database or were part of the ‘Ramilowski’ ligand-receptor interactions catalog (Ramilowski et al., 2015). This approach yielded a refined set of 664 cell-cell interactions with 448 paracrine (Figure 6c–h; Supplemental Figure 11, Supplementary Table S6) and 216 autocrine interaction pairs (Supplemental Figure 12, Supplementary Table S6). Specifically, from a DP-centric view with the DP serving as signaling center, we identified 111 DP→HGSC pairs (Supplemental Figure 11a) and 98 DP→BuSC interactions (Supplemental Figure 11b), with a remarkable overlap of 94 pairs due to highly expressed receptors on both SC types (Figure 6c and d; 30 selected pairs are shown). Dkk2→Kremen2 and Rspo→Lgr validated our earlier guided analysis using signature pairs. Importantly, the signature-independent analysis revealed many potential interactions, including those between DP-derived BMPs (Bmp3, Bmp4, Bmp7) and their receptors, Bmpr1a and Bmpr2, highly expressed by both HGSCs and BuSCs and known to regulate SC quiescence (Botchkarev et al., 2001, Kobielak et al., 2003). Wnt5a or Wnt6 interacting with Frizzled receptors, Fzd1, Fzd2 and Fzd7/Lrp6 co-receptors also play an important role in Wnt signaling crosstalk (Zeng et al., 2008). Less-well known interactions included DP ligands Inhba (Inhibin A) and Gdf10 (Growth differentiation factor 10) that may interact with Acvr1b (Activin receptor type 1B) of the TGF-β superfamily. Rspo1→Znrf3 interaction between DP and HGSCs potentially enhances Wnt signaling by clearing of Wnt-inhibitor Znrf3 (Hao et al., 2012) (Figure 6c). Igf→Igf1r insulin signaling between both DP-HGSCs and DP-BuSCs may play a role in hair cycle control similar to hair growth regulation (Weger and Schlake, 2005) (Figure 6c and d). DP-expressed Vim (Vimentin) or Col14a1 may interact with Cd44, while Vcan (Versican) can bind to Cd44 and Egfr receptors, all known to play roles in adhesion, migration and signaling (Jia et al., 2025).

Many potential interaction pairs can also be found from HGSCs and BuSCs to the DP (Supplementary Figure S11c and d). HGSC-specific ligand Shh, in addition to potential interaction with Hhip and Gpc5, may bind to both Ptch1/Smo and Ptch2/Smo in the DP (Hsu et al., 2014) (Figure 6e). HG-DP pairs Lama5→Sdc1, Lama5→Itgb1, Thbs1→Itgb1 and Thbs1→Lrp1 might regulate DP structural integrity (Hoffman et al., 1998, Resovi et al., 2014). Other pairs included Uba52 potentially interacting with Tgfbr1/2 and both Fbln1 or Npnt interacting with Itgb1 (Fujiwara et al., 2011). Pairs Calr→Lrp1 and Gas6→Axl were both found between HGSC-DP and BuSC-DP (Figure 6e and f). Interestingly, Pdgfa secreted from HGSCs or BuSCs may signal to DP-expressed receptor Pdgfra, potentially affecting the telogen-anagen transition similar to adipocyte-derived Pdgfa (Festa et al., 2011). BuSC-specific factors towards DP cells included Tnc interacting with Itgb1, Col18a1 with Itgb1 and/or Gpc4, Serping1 with Lrp1 and Tgfb3 with Tgfbr1/2 (Figure 6f). Regarding prospective signaling between HGSCs and BuSCs, Shh signals to Ptch1/Smo and Ptch2/Smo, like for the DP, but also may involve co-receptor Boc (Figure 6g), while Jag1/2→Notch1 interactions are directed from BuSC to HGSC (Figure 6h). Calmodulins (Calm1, 2, 3) and Cdh1 interaction with Egfr appears to be bi-directional between BuSCs and HGSCs (Figure 6g and h, Supplementary Figure S11e and f). Other notable HGSC-BuSC interactions included Sema4g→Plxnb2 and Mif→Ackr3. Finally, CellChat analysis also identified many potential autocrine interactions (Supplementary Figure S12). For the DP, expected Fgf ligand and receptors (FGF signaling), but also Ncam1 with Fgfr1/2 or Slit receptor Robo1 were found. Many promising and somewhat unexpected pairings were highly expressed; Inhba→Tfgbr3, Apod→Lepr, Dkk2→Lrp6, Igfbp4→Lrp6, Gnai2→Ednra, Pdgfc→Pdgfra and Tfpi→Sdc4. Besides autocrine Shh signaling in HGSCs, autocrine interactions in HGSCs and BuSCs mostly included pairs involved in cell adhesion such as Cdh3→Cdh3, Flrt3→Flrt3, Cdh1→Cdh1, Dsc3→Dsg3, and Dsg2→Dsc3. Other frequently occurring autocrine interactions for DP cells, HGSCs and BuSCs were in different combinations fibronectin (Fn1) and collagens (Col3a1, Col1a1, Col1a2, Col6a2, Col4a1) pairing with integrins (Itgb1, Itgb4, Itgb6) and Mdk (Midkine), or Ptn (Pleotropin) binding to Ncl (nucleolin) or Sdc1, Sdc4 (Syndecans) suggesting conserved roles in cell-cell and cell-matrix communication across all three HF cell types. Overall, our stringent analysis of enriched signature ligand-receptor pairings revealed molecular interactions between the key cell populations during the hair cycle. The expanded cell-cell communication analysis of highly expressed DP, HGSC and BuSC genes revealed many additional interactions with potential functional implications for the regulation of SC quiescence and activation, including the potential and less-well understood communication between DP and BuSCs.

DISCUSSION

Early insights into the molecular control of HFSCs and their signaling crosstalk with the mesenchymal niche came from functional studies focusing on select candidate genes (Botchkarev et al., 2001, Kobielak et al., 2007, Lowry et al., 2005, Nguyen et al., 2006, Paladini et al., 2005, Sano et al., 2000, Sato et al., 1999, Zhang et al., 2006). Since the advent of transcriptomic profiling methods, a plethora of new candidate molecular regulators of HFSC and DP functions during the hair growth cycle were identified and several of them were functionally validated (Fuchs, 2016, Hsu and Fuchs, 2022, Lee and Tumbar, 2012, Schneider et al., 2009, Sennett and Rendl, 2012, Zhang and Chen, 2024). Among other insights, such profiling efforts revealed cyclically regulated genes in either DP or BuSCs. Many of these and other strongly expressed genes were included even though they could not be assigned as enriched and with signature status due to the absence of comparisons with closely related cell- types (Avigad Laron et al., 2018, Flora et al., 2021, Greco et al., 2009, Lien et al., 2011). Yet, other transcriptomic studies established cell-type enriched comparative gene signatures, but narrowly for either BuSCs or DP cells, thereby limiting insights into gene expression in the partner cell types (Ge et al., 2017, Hagner et al., 2020, Lien et al., 2011, Morris et al., 2004, Tumbar et al., 2004). Only one study simultaneously profiled sorted BuSCs, DP cells and HGSCs, albeit exclusion of other related lineage cells precluded it from establishing SC- and DP-specific signatures (Greco et al., 2009). Conversely, single-cell RNA-seq approaches have the major advantage of simultaneously profiling all cell types in a given tissue (Joost et al., 2020, Liu et al., 2022), but they lack the sensitivity of bulk RNA-seq for discovering low abundance transcripts. In this study, we successfully overcame these previous limitations by cell-sorting and profiling highly enriched BuSCs, HGSCs and DP cells along with three related epithelial and mesenchymal cell types for added rigor. Population-based deep-sequencing transcriptomics then enabled us to define with high sensitivity the cell type-specific molecular signatures of the key SC and niche cell populations.

A notable strength of the gene signature approach employed by us here is the promise of discovering uniquely expressed genes that likely underly specific functions of BuSCs, HGSCs and DP cells in the hair cycle. The rigor of our approach was bolstered by the cell isolation design, which included six defined cell types accounting for the majority of total skin cells, and by the bioinformatic comparison strategy across the continuous physiological process—the first hair cycle. We followed this approach with all biological replicates and based our signature calculations on transcriptome profiles of all seven sorted cell populations from eight independent measurements in four neighboring skin zones from the same animals. By defining Core (4-Zones) and Expanded (Core, 3-Zones, 2-Zones) signatures for DP cells, BuSCs and HGSCs we thereby unearthed a plethora of high-confidence genes in many functional categories. Unsurprisingly, many genes were classified as signaling factors, receptors and transcription factors as well as members of other important categories such as adhesion/ECM, enzymes, and metabolism, fitting for cells with dynamically regulated cell fates and functions. Further, in the process of evaluating the validity of our signatures, we compiled custom-curated lists of BuSC, HGSC and DP genes that were previously expression-confirmed throughout the years. While some genes might have been unintentionally missed, for which we apologize, we feature all validated genes—42 DP, 17 BuSC, 14 HGSC genes—in a comprehensive collection of RPKM expression graphs for each gene in supplementary figures and in additional supplementary tables. Importantly, in addition to encompassing many previously validated genes and several genes from earlier profiling efforts, many more genes with previously undefined roles in hair growth were included for all three signatures. Any new gene identified here as specifically enriched in BuSCs, HGSCs and DP cells could be essential for hair cycle regulation, but the task of their functional validation is beyond the scope of this work. Instead, to facilitate easy access to the signature gene information, we provide expression levels and signature calls in numerous well-annotated supplementary tables to be used as a resource for researchers in the field.

An additional main aim of our project—besides gene signature considerations—was to share the gene expression information of all measured transcriptomes as a new data tab on our companion Hair-GEL (Gene Expression Library) web resource (https://hair-gel.net/), similar to our previous iterations for embryonic HF development and early hair growth after birth (Rezza et al., 2016, Sennett et al., 2015). With a layout similar to RPKM expression graphs featured in this paper, one can look up the expression of any gene simultaneously in all cell populations, learn about the signature status (denoted with *), and download the individual graph and its underlying values. This freely available resource will promote a deeper understanding of the gene expression in BuSCs, HGSCs and DP cells and we hope will inspire future functional genetic studies on the molecular controls of HFSCs and their instructive niche.

One particular strength of our signature analysis lies in its contextual integration with the most authoritative transcriptomics studies on the hair cycle from the last 20 years. This was done to establish an overarching account of enriched genes in the HF SCs and the DP niche identified by all studies, but also to tease out many newly identified signature genes present only in our signatures. Multi-dimensional cross-comparisons with a total of 15 signatures of mouse or human BuSCs, HGSCs and/or DP cells, obtained either from published signature lists or derived from our own standardized reanalysis of available datasets, confirmed many previously reported signature genes. For example, 5-way comparison of microarray, bulk and single-cell RNA-seq-derived DP signatures revealed 263 genes overlapping at least once with our signatures and 41 genes present in all studies. Conversely, we also uncovered many more previously undescribed DP genes in critical functional categories that were unique to our Expanded signature. Such genes represent promising new targets for future studies. A similar scenario was observed for the 7-way BuSC and 4-way HGSC comparisons, except here the overlaps among multiple studies were much less pronounced. While the reason for this divergence is ultimately unclear, it could lie within the different isolation methods and resulting impurities or broadly variable sensitivities of different profiling methods. In this regard, our successful detection of many additional DP, BuSC and HGSC signature genes could stem from our deep- sequencing of >75 million reads for each of the PCR-amplified cDNA libraries of a total of 56 samples (Supplementary Table S1). As with our own signatures, we feature all signatures from the past profiling studies and their multi-way overlaps in easily tractable color-coded Supplementary Table S3, S4 and S5 which we hope will serve as a useful resource for the field. We noted that many DP, BuSC and HGSC genes identified as signature genes in previous signatures were not part of our stringently defined Expanded signatures. Interrogation of expression of such genes—easily accessed in the provided supplementary tables and queried on Hair-GEL.net—demonstrates that they were correctly detected in the respective cell types by the previous studies; however, most of those genes exhibited comparable or even higher levels in other cell types—not measured in individual previous studies—and therefore they did not make the signature cut in our hands.

The power of our BuSC, HGSC and DP signatures lies in the implication of distinct functional roles of uniquely enriched genes, identified through contextualized overlaps with profiled related cell populations. On the other hand, collections of only enriched genes likely represent the figurative tip of the iceberg, as an exclusive focus on cell type-specific signatures misses the important functions of more broadly expressed genes that are for example shared among related cell types but do achieve distinctive functions because of a unique combinatorial gene expression and molecular configuration of each population. This proposition became particularly apparent in our analyses of ligand-receptor signaling and other cell-cell communication between BuSCs, HGSCs and DP cells. Focusing on ligand-receptor pairs strictly following stringent signature criteria revealed only 20 paracrine and autocrine interactions pairs largely within WNT, TGFβ, and HH signaling pathways. Expanding the exploration, however, to any indiscriminately highly expressed genes mapped a total of 664 interaction pairs increasing the network complexity of potential cell-cell communication, adding BMP, FGF, NOTCH and many other less well-known signaling pathways and cell-cell interactions. Again, we map all interactions in extensive supplementary figures and tables for all-inclusive presentation. Finally, this signature-unrestricted analysis of highly expressed genes revealed that many receptors are comparably expressed and, thus, shared between BuSCs and HGSCs, which adds the tantalizing possibility that besides DP-HGSC interactions there also exist many somewhat longer-range exchanges between DP cells and BuSCs, as proposed recently during hair growth (Avigad Laron et al., 2018). Future cell type-specific functional studies will be required to confirm direct DP-BuSC interactions during the telogen phase of the hair cycle, beyond ligand and receptor availability.

Limitations of the study

While the octuple transcriptome measurements of DP, BuSC, HGSC and all other cells in four neighboring skin zones allowed us to define high-confidence Core and Expanded gene signatures, they had the downside that cells in each zone came from HFs that were largely in equal parts in telogen and the earliest anagen phase of the hair cycle. This limitation should be kept in mind when interpreting gene expression values as it is possible that some genes in our bulk profiling could be attributed to exclusive expression in either hair cycle phase. Since our across-zone comparisons revealed no meaningful gene expression changes that could be assigned to telogen- or anagen I- preferred expression, however, we conclude that hair cycle phase-specific expression is not a major factor in our dataset. On the other hand, because we profiled only two biological replicates in each zone—powerfully sufficient to define cell-type specific signatures—it is possible that subtle telogen/anagen I expression differences were missed due to limited statistical robustness. In this regard, we failed with our initial goal to utilize the long-reported head-to-tail hair re-growth wave during the first hair cycle for discovering telogen-to-anagen I regulated genes. Although we observed a morphological hair cycle wave, it was largely heterogeneous with broad co-existence of telogen and anagen I HFs—possibly due to mouse strain differences with prior reports—thereby precluding resolving hair cycle phase-specific gene expression with population-based bulk transcriptomics. Single-cell transcriptomics or spatial transcriptomics, although both still restricted by lower gene- detection sensitivity, may make inroads in the future in resolving telogen vs. anagen I gene expression regulation. Finally, the multiway analyses of signatures from 15 previous studies and our new Expanded signatures compared the gene identities and not their underlying expression values derived from a re-analysis of all datasets with a unified statistical framework. While the latter quantitative analysis would be powerful, it is not feasible due to the heterogeneous transcriptomics platforms from which the datasets are derived. Instead, our comparative analyses of signature gene identities honor the longstanding existing signature collections of each previous study and provide a full synthesis of all signature lists reported over the last 20 years.

MATERIALS AND METHODS

Mice

K14-H2BGFP, Lef1-RFP, and Crabp1-GFP mice were described previously (Gong et al., 2003, Rendl et al., 2005, Rezza et al., 2016). All mice were housed in facilities operated by the Center for Comparative Medicine and Surgery (CCMS) at Icahn School of Medicine. All animal experiments were conducted in accordance with the guidelines and approval of the Institutional Animal Care and Use Committee at Icahn School of Medicine at Mount Sinai under protocol number 08–0452.

Immunofluorescence and microscopy

For immunofluorescence staining, back skins from K14-H2BGFP (Tumbar et al., 2004) or CD1 mice at first-cycle telogen were collected and fresh embedded in OCT (Tissue Tek). Back skins from K14-H2BGFP; Crabp1-GFP; Lef1-RFP mice were prefixed in 4% PFA for 1 hour at room temperature to retain the intrinsic fluorescence and sections were cut at a thickness of 10 μm using the Leica cryostat. CD34 (rat, eBiosciences 14034185) and PCAD (goat, R&D Systems BAF761) stainings were performed on CD1 mice to visualize BuSCs and HGSCs respectively at first-cycle telogen. To validate the head-to-tail wave, KI67 (rat 1:200, eBiosciences 13569880) and activated CASP3 (rabbit 1:300; R&D systems AF835) staining was performed on K14-H2BGFP and K14-H2BGFP;Crabp1-GFP;Lef1- RFP mouse back skins. AQP5 (LSBio; LS‑C756566) and DCC (R&D systems; AF844-SP) antibodies were used to validate DP expression. Skin sections were fixed with 4% PFA for 1 hour at room temperature, then washed with PBS, and incubated with primary and secondary antibodies for 1 hour at room temperature or overnight at 4 °C. Nuclei were labeled with Hoechst 33342 (Thermo Fisher, 1:1000), and stained sections were mounted in glycerol-based p-phenylenediamine (Sigma) antifade reagent. Epifluorescence microscopy was performed using the Leica DM5500 widefield microscope, while confocal microscopy was performed with a Leica SP5 DMI confocal microscope equipped with Leica LASAF software. Images were post-processed and adjusted for levels, brightness and contrast using ImageJ/FIJI (NIH).

ETHICS STATEMENT

All experimental animal procedures were approved by the Institutional Animal Care and Use Committee at Icahn School of Medicine at Mount Sinai. The mouse care and handling procedures followed the ethical guidelines of the NIH Guide for the Care and Use of Laboratory Animals.

DATA AVAILABILY STATEMENT

Bulk RNA-sequencing data are available at the NCBI-Gene Expression Omnibus (GEO) under the accession number GSE300176. https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE300176. The online Hair-GEL repository can be accessed at: https://hair-gel.net/