Mesenchyme governs hair follicle induction

Inductive tissue interactions constitute a fundamental mechanism regulating embryonic development. As most organs develop from multiple tissue types, cell-cell communication across tissue boundaries is crucial to ensure successful organ induction, patterning, morphogenesis and cell differentiation (Thesleff et al., 1995). The hair follicle is an excellent model to study these interactions as it develops from the embryonic surface ectoderm and the underlying mesenchyme, and reciprocal signalling between these compartments directs its development, as well as the life-long cyclic function of the mature organ (Hardy, 1992; Li and Tumbar, 2021; Morgan, 2014; Sennett and Rendl, 2012). The inductive signalling regulating hair follicle development and renewal has been under meticulous investigation for decades, yet some key aspects remain elusive to date.

The timing of hair follicle induction is specific to the body site. In mice, hair follicles of the dorsal skin are induced in three consecutive waves, the first one commencing at around embryonic day (E) 13.5 (Biggs et al., 2018; Fliniaux et al., 2008) as a response to a yet unidentified primary inductive signal. As a result, epithelial keratinocytes migrate centripetally to form a thickening, the placode (Ahtiainen et al., 2014). Inductive signals from the placode then induce the underlying fibroblasts to migrate and aggregate underneath the placode, forming the mesenchymal component of the hair follicle, the dermal condensate (DC) (Biggs et al., 2018; Glover et al., 2017; Gupta et al., 2019; Mok et al., 2019). Subsequently, secondary inductive signals from the DC promote invagination of the epithelium to the mesenchyme, and the DC is engulfed by the epithelium and remains an integral part of the mature hair follicle, named the dermal papilla (DP) (Sennett and Rendl, 2012). As the embryo grows larger, secondary and tertiary hair placodes form between the existing hair primordia in two waves at around E16.5 and E18.5, respectively (Duverger and Morasso, 2009).

Tissue crosstalk between the follicular epithelium and the underlying mesenchyme is mediated by conserved signalling pathways (Biggs and Mikkola, 2014). Initially, Wnt/β-catenin signalling is focally activated in the prospective placode where it is necessary for placode formation, as suppression of epithelial β-catenin signalling activity blocks hair placode formation altogether (Andl et al., 2002; Huelsken et al., 2001; Zhang et al., 2009). Moreover, forced activation of epithelial Wnt/β-catenin signalling is thus far the only known genetic manipulation to induce precocious formation of hair placodes (Närhi et al., 2008; Zhang et al., 2008). After initial epithelial Wnt activation, the Eda/Edar/NF-κB pathway is necessary to reinforce Wnt activity and for placode morphogenesis to continue (Fliniaux et al., 2008; Schmidt-Ullrich et al., 2006; Zhang et al., 2009). Once placode formation has initiated, signals emanating from the placode, in particular fibroblast growth factor 20 (Fgf20) and sonic hedgehog (Shh), govern DC formation (Huh et al., 2013; Woo et al., 2012). Fgf20 has been shown to induce directional migration of mesenchymal fibroblasts and to induce DC cell fate (Biggs et al., 2018; Mok et al., 2019). Fgf20 is necessary for DC formation, as demonstrated by the absence of all molecular and morphological indications of primary and secondary DCs in Fgf20 null mice (Biggs et al., 2018; Huh et al., 2013; Mok et al., 2019).

Several lines of evidence indicate that suppression of Bmp signalling is essential for hair follicle induction. Deletion of Bmp-inhibitor gene noggin (Nog), expressed in the DC, suppresses secondary and tertiary hair placode formation (Botchkarev et al., 1999, 2002), whereas its ectopic expression under the keratin 14 promoter results in increased hair follicle density and formation of ectopic hair follicles in ventral paws and at the site of the Meibomian glands (Plikus et al., 2004). Furthermore, Noggin treatment increases hair placode density in embryonic skin explants, whereas Bmp4-treatment prevents hair placode formation by suppressing expression of Edar and Lef1, the mediator of the Wnt/β-catenin pathway (Botchkarev et al., 1999; Jamora et al., 2003; Mou et al., 2006).

In the past, hair follicle induction has been studied extensively by transplantation and tissue recombination experiments. It was first demonstrated that the whisker follicle DP contains the capacity to induce new follicles by transplantation of rat vibrissae DPs in the ear skin and subsequent formation of ectopic whisker follicles (Cohen, 1961). Later, tissue recombination of hair-forming dorsal mesenchyme with the glabrous plantar epithelium revealed that the inductive capacity resides in the dorsal skin mesenchyme as early as the DC forms (Kollar, 1970). These and other studies have established that the hair follicle dermal component, the DP and its precursor the DC, contain the capacity to induce de novo hair follicle formation in the skin, as well as other epithelial tissue of ectodermal origin, such as dental and corneal epithelia (Cohen, 1961; Dhouailly, 1973; Ferraris et al., 2000; Hardy, 1992; Higgins et al., 2013; Jahoda et al., 1992; Kollar, 1970; Kollar and Baird, 1970; Oliver, 1970). The extrapolation of these findings has established the current paradigm, which posits that the mesenchyme is the site of the hair-inductive signal. However, experiments investigating the inductive capacity of the mesenchyme before DC formation are scarce and, to our knowledge, no homospecific recombination experiments of mouse back skin and glabrous skin have been performed before DC formation. Heterospecific recombination of mouse back skin mesenchyme before hair follicle induction (E12.5) with glabrous epithelia of chick embryonic skin did not result in appendage induction (Dhouailly, 1973; for nomenclature used in tissue recombination studies, see Fig. S1). In contrast, the reverse recombination of the E12.5 hair-forming back skin epithelium with the glabrous chick skin mesenchyme produced short epithelial ‘nodules’ protruding into the dermis (Dhouailly, 1973). Furthermore, studies showing that DC formation is preceded by a placodal prepattern (Biggs et al., 2018; Glover et al., 2017; Huh et al., 2013; Mok et al., 2019) and that primary and secondary hair placodes form in the absence of DCs in Fgf20 null embryos (Huh et al., 2013) have challenged the prevailing model of mesenchymal induction of hair follicle formation.

Here, we sought to identify the source of the hair follicle inductive cue by performing tissue recombination experiments between heterologous epithelia and mesenchymes using genetically labelled tissues expressing fluorescent reporters allowing unambiguous assessment of tissue purity. Our results reveal that the hair follicle inductive signals reside in the mesenchyme even before DC formation. We show that mesenchymal expression of the Wnt pathway agonist R-spondin 1 (Rspo1) is upregulated at the time of hair follicle induction. This prompted us to assess the impact of Rspo1 on hair follicle induction in ex vivo cultured skin explants. Our results indicate that Rspo1 alone, or in combination with inhibition of Bmp signalling, is insufficient to induce precocious hair follicle formation. However, the latter altered hair placode patterning and the former instigated hair placode-specific transcriptional responses, though unpatterned, even before hair follicle induction. We expect our results will motivate further research to identify the mesenchymal hair-inductive cues.

To determine whether the hair follicle inductive signal arises from the epithelium or the mesenchyme, we performed heterotypic tissue recombination between the hair-forming back skin and the glabrous plantar skin by enzymatically separating the tissue compartments, recombining them and culturing ex vivo (Fig. S1B). To monitor the tissue separation efficiency and retrospectively verify the origin of the cells, we collected epithelia from R26R^mTmG/+ and mesenchymes from R26R^mG/+ littermate embryos, ubiquitously expressing membrane-bound tdTomato (mT) and membrane-bound eGFP (mG), respectively. Alternatively, epithelia were obtained from R26R^{floxedTomato/+} or R26R^{floxedTomato/+};K17-GFP and mesenchymes from wild-type embryos. First, we performed homotypic recombinations of E14.5 tissues that served as technical controls. Dorsal skin mesenchyme and epithelium cultured for 4 days readily formed hair follicle-like appendages, whereas homotypic recombinants of plantar tissues did not generate any appendages (Fig. 1A), validating our experimental set-up. In heterotypic recombinants, the DC-containing dorsal mesenchymal tissue induced hair follicle-like appendage formation in the glabrous plantar epithelium (Fig. 1A), as expected. No appendages were observed in the recombinants consisting of dorsal skin epithelium and plantar mesenchyme (Fig. 1A).

We then asked where the inductive potential resides at the time of hair placode induction at E13.5, when DCs cannot yet be morphologically discerned. Like E14.5 tissues, homotypic recombinations between back skin epithelium and mesenchyme readily produced skin appendages, whereas recombinations between plantar skin compartments did not (Fig. 1B). Interestingly, at this stage too, hair follicle-like appendages were induced upon heterotypic recombination of the dorsal mesenchyme with the plantar epithelium (Fig. 1B), whereas dorsal skin epithelium was not able to induce appendages in the glabrous plantar mesenchyme (Fig. 1B), indicating that the dorsal mesenchyme is inductive even before DC formation.

As DC formation is a gradual process and the first loosely organized Sox2⁺ cells are found below the emerging placodes from E13.5 onward (Biggs et al., 2018; Gupta et al., 2019), small DCs might already exist at E13.5. Therefore, we assessed where the hair follicle inductive potential resides before any known molecular or morphological signs of hair follicle formation by recombining tissues from E12.5 embryos. Although the homotypic control recombination of the back skin formed appendages and the corresponding recombination of plantar tissues did not (Fig. 1C), we again observed appendages in the heterotypic recombinants consisting of the dorsal skin mesenchyme and glabrous plantar epithelium, but not in explants of dorsal skin epithelium and plantar mesenchyme (Fig. 1C). Quantification of all E12.5-E14.5 recombinant explants confirmed that only mesenchymal back skin was inductive (Table 1). These data imply that the hair-inductive potential resides in the mesenchyme, even before DC formation.

To characterize the appendages that formed in the E12.5 tissue recombinants, epithelia were collected from R26R^{floxedTomato/+};K17-GFP embryos, in which tdTomato (tdT) is ubiquitously expressed and the nascent hair placodes are marked by K17-GFP reporter expression (Bianchi et al., 2005), and the mesenchyme from wild-type littermates. Homotypic back skin recombinations served as a positive control revealing strong K17-GFP expression in developing hair follicles (Fig. 2A). Likewise, K17-GFP expression marked the appendages formed in recombinants between back skin mesenchyme and plantar epithelium (Fig. 2A). Immunostaining with the early hair follicle marker Lhx2 (Rhee et al., 2006) not expressed in sweat gland placodes (Cao et al., 2021) and the DC-marker Sox2 (Driskell et al., 2009) verified that the formed appendages were hair follicles (Fig. 2A; Fig. S2A,B).

To ascertain that the mesenchyme harbours hair inductive capacity even in the absence of DC formation, we performed tissue recombination experiments using E12.5 back skin mesenchyme isolated from Fgf20 null embryos that lack primary and secondary hair follicle DCs (Huh et al., 2013). Indeed, like the wild-type mesenchyme, the Fgf20 null mesenchyme induced formation of appendages when recombined with wild-type back skin epithelium, or wild-type plantar epithelium (Fig. 2B). These appendages, too, were associated with Sox2⁺ DCs and Lhx2⁺ epithelial cells (Fig. 2B; Fig. S2C,D) confirming that the hair follicle inductive capacity of the back skin mesenchyme is independent of DC formation.

Lastly, we asked whether the capacity of the naïve dorsal mesenchyme to induce appendages is restricted to epithelia of epidermal origin by assessing whether hair follicles could form in the oral epithelium. To this end, we recombined non-fluorescent E12.5 back skin mesenchyme with the epithelium of R26R^{floxedTomato/+};K17-GFP embryos, isolated from the E12.5 lower jaw diastema, a toothless region of the oral cavity between the incisor and the molar tooth primordia (Fig. S3A). Back skin epithelium from K17-GFP;R26R^{floxedTomato/+} embryos was used as a control. Again, K17-GFP⁺ hair follicles readily formed in homotypic back skin control recombinants after 5 days of culture (Fig. 3A). Strikingly, hair follicle-like K17-GFP⁺ appendages formed also in explants consisting of the dorsal mesenchyme and the oral epithelium (Fig. 3B). Like the control explants, these appendages expressed Lhx2 and Sox2 in the epithelium and the mesenchyme, respectively (Fig. 3A,B; Fig. S3B,C). Furthermore, the recombinants generated visible hair shafts after 2 weeks of culture (Fig. 3A,B) confirming their hair follicle identity.

Next, we asked how the mesenchymal primary inductive signal is generated in the mesenchyme. As induction of feather placodes is associated with an increase of mesenchymal cell density (Wessells, 1965), we hypothesized that perhaps a similar increase of mesenchymal cell density occurs also during hair follicle induction. Given that Wnt/β-catenin signalling activity is essential for mesenchymal cells to gain DC identity (Atit et al., 2006; Gupta et al., 2019), we analyzed Wnt⁺ cells.

To assess mesenchymal cell density, we produced confocal 3D image stacks of transversal vibratome sections from mouse embryos before hair follicle induction at E12.5 and at the time of hair induction at E13.5 (Fig. 4A). We quantified the density of mesenchymal nuclei within 80 µm distance from the epithelium and calculated nuclear densities as average rolling densities in 20 µm windows at 1 µm intervals (Fig. S4A). For statistical analysis, the data were binned at 20 µm intervals. Although cell densities were significantly higher closest to the epithelium at both stages analyzed, no difference was observed between E12.5 and E13.5 (Fig. 4B).

To assess the distribution of Wnt-activated cells, we took advantage of the TCF/Lef1:H2B-GFP Wnt signalling reporter mouse (Ferrer-Vaquer et al., 2010) and calculated the average rolling density of GFP⁺ dermal cells at both stages (Fig. S4B). Highest density of TCF/Lef1:H2B-GFP⁺ cells was observed in the proximity of the epithelium, confirmed by statistical analysis of the binned data, yet there was no difference between the two stages (Fig. 4C). These results suggest that the signal initiating the formation of hair follicles likely arises through changes in gene expression of the mesenchymal fibroblasts rather than via an increase in their density or change in the proportion of Wnt/β-catenin-activated cells.

We next aimed to gain insights on the molecular composition of the mesenchymally-derived hair follicle inductive signal using a candidate approach. The evidence showing that Bmp signalling inhibits hair placode formation (Botchkarev et al., 1999; Mou et al., 2006; Plikus et al., 2004; Pummila et al., 2007) led us to hypothesize that downregulation of BMP signalling activity could be sufficient to initiate formation of hair follicles. However, in addition to Nog, numerous other Bmp antagonists are expressed in the embryonic skin mesenchyme at the time of hair follicle induction (Jacob et al., 2023; Sennett et al., 2015; http://kasperlab.org/embryonicskin) making a genetic approach challenging. Therefore, to test our hypothesis, we inhibited BMP signalling with LDN193189, a small molecule inhibitor of BMP type I receptors (Cuny et al., 2008). Back skin of E12.5 Fgf20^βGal/+ embryos were cultured with 1 µM LDN193189 for 12 and 24 h and hair placode induction was assessed by analyzing the expression of hair placode markers Fgf20 (using anti-β-Gal staining as a surrogate) and Edar by whole-mount immunostaining (WMIF). Unsatisfyingly, suppression of Bmp activity did not induce precocious hair follicle formation at 12 h (Fig. 5A). However, when placodes started to emerge after 24 h culture, they were larger and less regular in size in LDN193189-treated explants compared with controls (Fig. 5B).

Given the essential role of epithelial Wnt/β-cat signalling in hair follicle induction (Huelsken et al., 2001; Närhi et al., 2008; Zhang et al., 2008, 2009), we next focused our attention to potential activators of this pathway. As dermal deletion of Wls, a gene indispensable for Wnt ligand secretion, does not impair hair follicle induction (Chen et al., 2012), we hypothesized that the inductive signal could be another type of soluble Wnt agonist. R-spondins 1-4 (Rspo1-4) act synergistically with Wnt ligands to enhance Wnt/β-catenin signalling activity via leucine-rich repeat-containing G protein-coupled receptors 4-6 (Lgr4-6) (Nusse and Clevers, 2017; Steinhart and Angers, 2018). Intriguingly, Lgr4 and Lgr6 are expressed in the nascent primary hair placodes (Sennett et al., 2015; Snippert et al., 2010; Tomann et al., 2016) and loss of Lgr4 leads to near total absence of primary hair placodes and greatly reduced total number of hair follicles in newborn mice (Mohri et al., 2008) implying an important role for R-spondins in hair follicle induction.

Previous RNA-seq profiling of E14.5 back skin has indicated that Rspo1 and Rspo3, but not Rspo2 and Rspo4, are expressed in the mesenchyme (Mok et al., 2019; Sennett et al., 2015; Snippert et al., 2010). We first analyzed the expression pattern of Rspo1 and Rspo3 in the back skin before hair follicle induction (E12.5), during hair follicle induction (E13.5) and at placode stage (E14.5) by RNA in situ hybridization. We did not observe focal expression of Rspo1 at any stage, instead the signal appeared to be homogenous in the entire mesenchyme (Fig. 6A). Expression of Rspo3 was unpatterned at E12.5 and E13.5 but, at E14.5, a punctate expression in the DCs was evident, confirmed by wholemount in situ hybridization (Fig. S5A), in line with RNA-seq profiling of DCs (Sennett et al., 2015). To quantify the expression levels of Rspo1 and Rspo3, we performed qRT-PCR on E12.5-E14.5 mesenchymes. Interestingly, Rspo1 expression was significantly upregulated between E12.5 and E13.5 (Fig. 6B), whereas the expression of Rspo3 did not significantly alter between these stages (Fig. 6B). These data suggest that Rspo1 might be associated with hair follicle induction.

In adult mice, exogenous recombinant Rspo1 injection leads to precocious activation of hair follicle stem cells during the resting phase of the hair cycle (Li et al., 2016). To examine whether Rspo1 is sufficient to induce hair placode formation in the embryonic skin, we cultured E12.5 back skin explants for 12 and 24 h with 100 ng/ml Rspo1. At 12 h, hair placodes had not formed in control or Rspo1-treated skins (Fig. 6C), whereas after 24 h of culture, focal expression of placode markers was evident in both samples, indicative of placode induction (Fig. 6D). A higher Rspo1 concentration (500 ng/ml) did not induce precocious placodes either (Fig. S5B).

We also tested whether simultaneous activation of Wnt/β-catenin signalling via supplementation of Rspo1 and inhibition of BMP signalling together would be sufficient to advance hair placode formation. To this end, we cultured E12.5 back skin for 12 h in media supplemented with 100 ng/ml Rspo1 and 1 µM LDN193189, or bovine serum albumin (BSA) and DMSO as vehicle controls, but observed no formation of hair placodes in either condition (Fig. 6E). When hair placodes emerged after 24 h of culture, they were more irregular and even stripe-like in samples supplemented with Rspo1 and LDN193189 (Fig. 6F), similar to those treated with LDN193189 alone (Fig. 5B). Taken together, ex vivo manipulation of the developing skin with Rspo1 and LDN193189, alone or together, was not sufficient to induce precocious hair placode formation.

These results raised the question whether exogenous Rspo1 can induce a Wnt signalling response in the naïve epidermis, before placode formation. To assess this, we treated E12.5 embryonic skin with 100 ng/ml recombinant Rspo1 in hanging drops for 4 h and measured the signalling response by analyzing expression of placode markers and Wnt/β-catenin target genes Fgf20 and Dkk4 (Bazzi et al., 2007; Fliniaux et al., 2008; Huh et al., 2013) by qRT-PCR. We observed a significant increase in the expression of both Dkk4 (53.7-fold±29.3) and Fgf20 (4.9-fold±1.6) upon Rspo1 treatment (Fig. 6G), indicative of a positive Wnt signalling response. To assess whether the Wnt response was patterned or not, we divided back skin explants into two halves along the dorsal midline, treated them with BSA or Rspo1 and analyzed Dkk4 expression by in situ hybridization. Dkk4 expression was consistently much stronger in Rspo1-treated skin halves compared with controls, induction observed typically adjacent to the mammary line where the first hair placodes normally form at E13.5 (Fliniaux et al., 2008). However, the signal was ubiquitous rather than focal (Fig. 6H).

The hair follicle inductive signal has been hypothesized to arise from the mesenchyme, but until now, definitive evidence has been lacking (Hardy, 1992). Here, we show that the hair-forming back skin mesenchyme induces hair development in the normally hairless plantar epithelium at developmental stages after DC formation (E14.5), at the time of epithelial thickening (E13.5), as well as before DC formation (E12.5). Even the E12.5 Fgf20 null mesenchyme that is devoid of primary and secondary hair follicle DCs induced hair follicle formation in the normally glabrous epithelium. These data confirm that the mesenchyme is the source of the inductive cue(s) even before any molecular or morphological signs of hair follicle induction. This contrasts with the observations of the developing tooth where the tooth inductive capacity resides initially in the epithelium, but shifts to the mesenchyme concomitant with the condensation of the dental mesenchyme (Lumsden, 1988; Mina and Kollar, 1987). In our recombination experiments, the back skin epithelium failed to induce hair follicle formation at all stages analyzed, but we cannot exclude the possibility of an even earlier (before E12.5) inductive epithelial signal. But why does mouse E12.5 dorsal mesenchyme fail to induce appendage formation when recombined with the glabrous epithelium from the chick mid-ventral apterium (Dhouailly, 1973) when it is inductive when recombined with mouse plantar or oral epithelium? At this point, we can only speculate the possible causes. One possibility is that the chick mid-ventral epithelium is unable to respond to the primary inductive signal arising from the mouse back skin mesenchyme. If this was the case, the inductive signal(s) arising from the unpatterned mesenchyme and the DC/DP need to be (at least partly) distinct (and perhaps even consecutive), as the glabrous chick epithelium can respond to signals emanating from the hair follicle DC (Dhouailly, 1973). Another possibility, not necessarily mutually exclusive with the previous one, is that the initial (epithelial) patterning process fails in the glabrous chick epithelium. The initial inductive process may also be inherently different in these two species, mechanical cues being more important in feather follicle induction (Shyer et al., 2017).

In developing chicken embryos, a global increase in mesenchymal cell density precedes feather follicle induction (Ho et al., 2019; Wessells, 1965). We found no evidence for a similar change in the upper dermis of mouse embryos at the time of hair placode formation suggesting that alterations in cellular gene expression levels rather than cell density govern hair follicle induction. The identity of the hair follicle inductive signal remains elusive to date. However, all available evidence points to positive and negative roles for the Wnt/β-catenin and Bmp signalling pathways, respectively (Biggs and Mikkola, 2014). Here, we assessed the ability of the Bmp receptor inhibitor LDN193189 and Wnt agonist Rspo1 to induce hair follicle development. Although suppression of Bmp signalling changed the regular arrangement of the primary hair placodes into a stripe-like pattern, in line with its proposed involvement in hair follicle patterning (Mou et al., 2006; Glover et al., 2017), it did not result in precocious placode formation. In vivo, hair follicle formation is suppressed when Nog is ablated (Botchkarev et al., 1999, 2002), whereas its overexpression leads to formation of excess and ectopic follicles (Plikus et al., 2004). In cultured skin explants, exogenous Bmp4 prevents placode formation (Mou et al., 2006; Pummila et al., 2007). Together, these results show that Bmp signalling plays a central role in hair follicle induction and patterning, yet its inhibition alone appears to be insufficient to trigger hair follicle development. In developing skin, downregulation of epithelial Bmp activity might be achieved via concomitant downregulation of pathway agonists (such as Bmp4) and upregulation of soluble antagonists, many of which are expressed in the mesenchyme at the time of hair follicle induction (Jacob et al., 2023; Sennett et al., 2015).

Augmentation of Wnt signalling via application of Rspo1 alone or together with LDN193189 did not induce precocious hair follicle development either, although we did observe a significant upregulation of hair placode marker genes Dkk4 and Fgf20 upon Rspo1 treatment at E12.5, providing evidence that the naïve epidermis is responsive to Rspo1 before hair follicle induction. Although the failure in hair placode induction may indicate that additional signalling pathways are necessary, alternative explanations are also plausible. The periodic organization of hair follicles is thought to arise via the Turing reaction-diffusion system (Andl et al., 2002; Glover et al., 2017; Mou et al., 2006; Sick et al., 2006). This self-organizing model posits that spatial patterns result from the interaction of diffusible signals that minimally consist of a pair of interacting diffusible chemical signals: a short-range, self-enhancing activator and an activator-induced, longer-range inhibitor (Gierer and Meinhardt, 1972; Kondo et al., 2010; Turing, 1952). This system leads to the focal production of both the activator and the inhibitor. Here, supplying the activator Rspo1 in the culture medium led to a rapid, ubiquitous induction of Dkk4 expression, potentially flattening the Wnt activity pattern and thereby preventing hair follicle induction. It is also possible that Rspo1 alone is not sufficient to increase Wnt signalling activity above a critical threshold. We speculate that to achieve this threshold, combined upregulation of mesenchymally produced agonists (such as Rspo1) and downregulation of antagonists (such as Dkks and secreted frizzled-related proteins) might be necessary. Indeed, a recent transcriptomic profiling study revealed strong downregulation of Dkk2 in the upper dermis at the time of hair follicle induction (Jacob et al., 2023). Dkk2 is also expressed at higher levels in the non-hairy regions of the embryonic skin compared with the hair-forming regions, and its deletion is sufficient to induce ectopic hair follicle formation in the plantar skin (Song et al., 2018) and cornea (Mukhopadhyay et al., 2006), underscoring the significance of diffusible mesenchymal Wnt inhibitors in hair follicle induction. Once hair follicle development has been initiated by chemical cues, mechanical forces come into play and coordinate cell shape and fate transitions in nascent placodes and DCs ensuring continuation of morphogenesis (Villeneuve et al., 2022 preprint).

Taken together, our study complements the tissue recombination studies of inductive signalling in hair follicle development by establishing that the primary inductive cue arises from the mesenchyme. Although the identity of the signal remains elusive, our results from back skin mesenchyme-oral epithelium recombination experiments reveal that the hair inductive cue is readily interpreted by a tissue of non-skin origin, resulting in the formation of functional hair follicles. Recombination of DC-containing whisker pad mesenchyme with the dental epithelium has demonstrated that under the inductive signals of the DC, the dental epithelium switches from the tooth-forming programme to one generating whiskers (Kollar, 1966, 1970). As the oral epithelium used in our experiments was devoid of tooth placodes, our interpretation is that E12.5 back skin mesenchyme induced hair follicles de novo, rather than by transforming an existing tooth primordium into a hair-producing organ. Instead, genetic activation of epithelial Wnt signalling induces ectopic epithelial invaginations and teeth, not hair follicles, in the oral epithelium (Zhang et al., 2008; Wang et al., 2009; Järvinen et al., 2006; Närhi et al., 2008). This, together with our results, implies that hair follicle identity is governed by mesenchymal cues other than those merely intensifying epithelial Wnt activity and placode formation. Importantly, our results demonstrate that hair follicle inductive capacity is an intrinsic property of the naïve skin mesenchyme even before formation of the dermal condensate.

All mouse studies were approved and carried out in accordance with the guidelines of the Finnish national animal experimentation board.

The R26R^mTmG mouse line was obtained from The Jackson Laboratory (stock #007576, mixed background) and crossed to Pgk-Cre (The Jackson Laboratory, stock #020811, maintained on C57Bl/6J background) to obtain mice where R26R^mTmG was recombined to R26R^mG, resulting in constitutive and heritable expression of membrane-bound GFP from the Rosa26 locus in every cell (Lan et al., 2023 preprint). R26R^tdTomato line (The Jackson Laboratory, stock #007914, C57Bl/6 background) was crossed with Pgk-Cre mice to obtain mice where tdTomato is constitutively and heritably expressed from the Rosa26 locus in every cell, hereafter called R26R^floxedTOM. Fgf20^βGal (Huh et al., 2012) and TCF/Lef1:H2B-GFP (The Jackson Laboratory, stock #013752) mice were maintained on C57Bl/6J background, and K17-GFP (Bianchi et al., 2005) on NMRI (Janvier Labs) background. K17-GFP; R26R^floxedTOM embryos used in recombination experiments were of a mixed C57Bl/6;NMRI background. R26R^mTmG/wt and R26R^mG/wt embryos used in recombination experiments were obtained from timed pregnancies of wild-type NMRI females mated with R26R^mTmG/mG males, maintained on a mixed background. Sex of the embryos used in the experiments was not controlled.

The back skins were dissected from embryos of the indicated stages and genotypes. To obtain the plantar skin, the entire distal part of the limb was dissected. To obtain the diastema tissue from lower jaws, the mandible was dissected and trimmed to contain only the lingual surface of the mandible with the incisor and molar tooth primordia to mark borders of the diastema. The dissected tissues were treated with 2.5 U/ml Dispase II (Roche, 04942078001) in PBS at +4°C for the following times: E12.5 dorsal and plantar skin, 20 min; E13.5 back skin, 25-30 min; E14.5 back skin, 40 min and oral tissue, 8 min. The tissues were then allowed to rest in culture medium at room temperature for 30 min. The plantar skin was dissected from the rest of the limb, and all epithelial and mesenchymal tissues separated using forceps and needles. Epithelial and mesenchymal tissues were recombined so that the epithelium was placed on top of the mesenchyme. Tissue separation efficiency was assessed by microscopy. The separation was always efficient, yet it was not possible to avoid the transfer of single mesenchymal cells with the epithelium. Samples where the epithelium and the mesenchyme detached from each other after recombination, or the tissue contracted heavily and died, were excluded from the analyses.

Tissues were cultured in the air-liquid interface in a Trowell-type setup as described earlier (Närhi and Thesleff, 2010) using DMEM (Gibco by Life Technologies) and 10% fetal bovine serum (FBS), except with recombinations including oral tissues when DMEM:F12 (1:1) (Gibco by Life Technologies) supplemented with 10% FBS and 100 ng/ml ascorbic acid was used, both supplemented with 10 U/ml penicillin and 10 µg/ml streptomycin (Life Technologies). Media were exchanged every other day. When indicated, the media were further supplemented with Rspo1 (R&D Systems, 7150-RS-050/CF) and LDN193189 (Stemgent, 04-0074) at concentrations indicated in the text, and BSA and DMSO used as vehicle. At the end of the culture, tissues were fixed in 4% paraformaldehyde (PFA) in PBS for immunofluorescence staining for 1 h at room temperature and washed with PBS, or processed for in situ hybridization (see below).

To make transversal vibratome sections of E12.5 and E13.5 embryonic skin, the embryos were decapitated and fixed in 4% paraformaldehyde in PBS overnight at +4°C. The embryos were then embedded in 2% low-melting point agarose (Thermo Fisher Scientific, R0801) in PBS, sectioned to 250-300 µm sections using a Microm HM 650 V vibratome, and the sections were collected in PBS before immunostaining.

The fixed tissues were washed in PBS and stained with antibodies according to manufacturers' instructions. In brief, samples were blocked and permeabilized using PBS, 10% normal donkey serum and 0.3% Triton X-100 at room temperature for 1 h, followed by overnight incubation at +4°C with the primary antibodies diluted in the blocking solution: Sox2 (1:250, Santa Cruz Biotechnology, SC-17320; or 1:250, Stemgent, 09-0024), LHX2 (1:400, Santa Cruz Biotechnology, SC-19344), EpCAM (1:500, BD-Pharmingen, 552370), β-Galactosidase (1:1000, Abcam, ab38295) and Edar (1:400, R&D Systems, AF745). All samples were co-stained with 1:2000 Hoechst 33342 (Invitrogen). The samples were further washed with several changes of PBS over 6 h at room temperature before incubation with AlexaFluor conjugated secondary antibodies (Invitrogen) diluted 1:400 in PBS overnight at +4°C. The following day, the samples were washed with several changes of PBS over a minimum of 4 h at room temperature and mounted on microscopy slides with Anti-Fade Fluorescence Mounting Medium (Abcam).

The samples were imaged using a Zeiss LSM 700 laser scanning confocal microscope equipped with PMT using Plan-Apochromat 10×/0.45, LD LCI Plan-Apochromat, 25×/0.8 Imm Corr and LCI Plan-Neofluar 63×/1.30 Imm Corr objectives. For visualizing recombinants, overview images were taken with the 10× objective at 5-10 µm intervals, and detailed images of appendages with 63× objective. To analyze dermal cell density, image stacks of the vibratome sections were obtained using the 63× objective at 0.5 µm intervals. To analyze treated skin explants for placode induction, images were obtained with a 10× objective at 5-10 µm intervals.

To quantify mesenchymal cell densities, confocal image stacks of vibratome sections were processed and analyzed using Imaris 10.0 and R softwares. Firstly, the attenuation of signal in the sample was determined by measuring average Hoechst 33342 signal intensities on the top and bottom z-sections of the stack in 5-10 measuring points made on the nuclei. The attenuation correction function of Imaris was then used on the channels of the image. The epithelial surface was created manually based on the anti-EpCAM staining and a distance transformation was made to delimit the 80 µm of the mesenchyme to epithelium. The nuclear Hoechst 33342 signal within this volume was then used to segment the nuclei using the machine learning Labkit plugin (Arzt et al., 2022) by classifying Hoechst 33342 and background pixels. Distance of the cell to epithelium was determined as the intensity of the distance transformation channel at the centre and the TCF/Lef1:H2B-GFP intensity as the mean intensity of the GFP channel of each segmented nucleus. The volumes at each 1 μm distance interval were determined using the auxiliary ImageJ histogram function, by quantifying the number of pixels with respective values of the distance transformation channels within the delimited 80 μm from the epithelium volume. The densities were calculated, and statistical testing was performed using R-studio software. The normality of the data was assessed with Shapiro-Wilk test and the statistical difference was analyzed using unpaired Student's t-test for normally distributed data and Wilcoxon matched-pairs signed rank test for not normally distributed data. For ex vivo cultured samples, the confocal image stacks were visualized as maximal intensity projections and placode induction was determined as patterned upregulation of Fgf20^βGal reporter and Edar expression.

The short-term treatment of tissue samples in the hanging drop culture was performed as previously described (Biggs et al., 2018). Briefly, in two independent experiments E12.5 skins were dissected from NMRI embryos and cut in half along the dorsal midline. The explants were then placed individually in 40 μl droplets of DMEM, 10% FBS, 1% penicillin-streptomycin supplemented with either 100 ng/ml Rspo1 (R&D Systems, 7150-RS-050/CF) or the respective volume of 0.1% BSA in PBS for vehicle control under a culture dish lid that was turned on a culture dish, and maintained for 4 h at 37°C, 5% CO₂. RNA was extracted from the samples using the RNeasy Plus Mini Kit (Qiagen GmbH, 74126) according to the manufacturer's instructions. cDNA was synthesized from 1 µg RNA using iScript gDNA Clear cDNA Synthesis Kit (Bio-Rad, 1725035), according to the manufacturer's instructions and the cDNA was diluted to 5 ng/µl.

For analysis of Rspo1 and Rspo3 gene expression, the back skin was dissected from wild-type NMRI embryos at E12.5 (n=13), E13.5 (n=12) and E14.5 (n=10), treated with Dispase II as described above to separate the mesenchyme from the epithelium. The mesenchyme was collected, the RNA extracted and cDNA translated as above.

For embryonic skin samples treated with Rspo1, qRT-PCR was performed using gene-specific PrimePCR Probe Assay (Bio-Rad) for mouse Dkk4 (10031225), Fgf20 (10031234) and Gapdh (10031231) multiplexed in 20 µl reactions according to the manufacturer's instructions with 25 ng of template. Reactions were performed in triplicate wells with the CFX96 Real-Time System (Bio-Rad) with the protocol described earlier (Biggs et al., 2018). The fold changes were calculated with the ΔΔCT method (Livak and Schmittgen, 2001) by normalizing to the mean of reference gene Gapdh in all controls. The normal distribution of the data was assessed with a Shapiro-Wilk test and the statistical difference was analyzed using a paired Student's t-test for normally distributed data and Wilcoxon matched-pairs signed rank test for not normally distributed data.

Expression of Rspo1 and Rspo3 in the embryonic skin mesenchyme was analyzed with primer-based RT-qPCR using Fast-SYBR Green Master Mix (Thermo Fisher Scientific, 4385612) in 20 μl reactions with 15 ng template and 0.5 µM of each primer. Rspo1 primers were designed using template NM_138683.2 (forward AGAGACAGAGGCGGATCAGTG and reverse CAGAATGAAGAGCTTGGGCG) and Rspo3 primers were designed using template NM_028351.3 (forward GTCAGTATTGTACACTGTGAGGC and reverse CCGTGTTTCAGTCCCCCTTT). Gapdh primers were: forward CTCGTCCCGTAGACAAAATGG and reverse AGATGGTGATGGGCTTCCC. Reactions were run in triplicate as above. The relative expression was calculated using the ΔCT method by normalizing to Gapdh. Normality of the data was assessed with a Shapiro-Wilk test and statistical difference analyzed using an unpaired two-tailed Student's t-test for normally distributed data and unpaired Wilcoxon test for not normally distributed data.

For whole-mount RNA in situ hybridization, E14.5 wild-type NMRI embryos were decapitated and fixed in 4% PFA in PBS overnight at 4°C, and then dehydrated in a methanol series. For treated skin explants, the explants were fixed to the Nucleopore™ filter with 4°C methanol and fixed in 4% PFA in PBS overnight at 4°C, and then dehydrated in a methanol series. The hybridization was performed manually using a protocol described before (Biggs et al., 2018). Antisense RNA probes were designed to target Rspo1 (template NM-138683.2, base pairs 924-1359) and Rspo3 (NM-028351.3, base pairs 975-1448), and Dkk4 as described before (Fliniaux et al., 2008). Probe constructs are available upon request. The samples were imaged using Zeiss AxioZoom.V16 stereomicroscope with PlanZ 1.0×/C objective and Axiocam 305 colour camera.

For radioactive in situ hybridization on paraffin sections, the E12.5 (n=4), E13.5 (n=5) and E14.5 (n=5) embryos were decapitated, fixed in 4% PFA overnight at 4°C, processed into paraffin blocks using standard protocols and cut into 5 µm sagittal sections. The in situ hybridization was performed according to standard protocols (Huh et al., 2013). Imaging and image processing was performed as previously described (Biggs et al., 2018).

We thank Ms Raija Savolainen, Ms Riikka Santalahti and Ms Merja Mäkinen for excellent technical assistance, Ms Verdiana Papagno and Ms Iina Häkkänen for assistance with in situ hybridization, Dr Qiang Lan for crosses of R26R^floxedTOM mice, Dr Anamaria Balic for providing Sox2 antibody, Dr Kimmo Tanhuanpää for expert advice on image analysis, and past and present members of the Mikkola, Jukka Jernvall and Anamaria Balic labs for stimulating discussions. We thank Dr Leah Biggs for critical reading of the manuscript. Mouse studies were carried out with the support of HiLIFE Laboratory Animal Center Core Facility, University of Helsinki. Confocal microscopy was conducted at the Light Microscopy Unit, Institute of Biotechnology, supported by HiLIFE and Biocenter Finland.

The peer review history is available online at https://journals.biologists.com/dev/lookup/doi/10.1242/dev.202140.reviewer-comments.pdf.