Coordinated hedgehog signaling induces new hair follicles in adult skin

Hair follicle (HF) development is orchestrated by coordinated signals from adjacent epithelial and mesenchymal cells. In humans this process only occurs during embryogenesis and viable strategies to induce new HFs in adult skin are lacking. Here, we reveal that activation of Hedgehog (Hh) signaling in adjacent epithelial and stromal cells induces new HFs in adult, unwounded dorsal mouse skin. Formation of de novo HFs recapitulated embryonic HF development, and mature follicles produced hair co-occurring with epithelial tumors. In contrast, Hh-pathway activation in epithelial or stromal cells alone resulted in tumor formation or stromal cell condensation respectively, without induction of new HFs. Provocatively, adjacent epithelial-stromal Hh-pathway activation induced de novo HFs also in hairless paw skin, divorced from confounding effects of pre-existing niche signals in haired skin. Altogether, cell-type-specific modulation of a single pathway is sufficient to reactivate embryonic programs in adult tissues, thereby inducing complex epithelial structures even without wounding.


Introduction
The number and pattern of hair follicles (HFs) are specified before birth in humans. In the mouse this is true for most body areas such as back skin (Alonso and Rosenfield, 2003;Millar, 2002;Paus and Cotsarelis, 1999). HF morphogenesis requires Hedgehog (Hh) and Wnt/b-catenin signaling and becomes first morphologically visible at embryonic day 14.5 (E14.5) in mice (Chiang et al., 1999;Gat et al., 1998;Lo Celso et al., 2004;St-Jacques et al., 1998). At this stage, the embryonic hair germ has formed, consisting of an epithelial placode and a dermal condensate, whose epithelialmesenchymal crosstalk is essential for further HF development (Hardy, 1992;Schmidt-Ullrich and Paus, 2005).
De novo HF formation in adult skin has been observed in combination with wounding in rabbits, mice, and humans (Breedis, 1954;Ito et al., 2007;Kligman and Strauss, 1956;Lim et al., 2018), and in unwounded skin as a response to forced epithelial Wnt/b-catenin signaling in mice (Gat et al., 1998;Lo Celso et al., 2004). Two decades after the initial discovery that Wnt/b-catenin-pathway activation results in ectopic HFs (Gat et al., 1998), it is still the only known approach to induce de novo HFs in adult unwounded skin. Accordingly, it remains a major clinical challenge to generate replacement skin with hair, urging the search for new ways to achieve HF formation in adult skin.
In Sonic hedgehog knock out (Shh-/-) mice, HF morphogenesis does not progress beyond the hair germ stage (St-Jacques et al., 1998). Notably, compared to wild type skin, Shh-/-hair germs have normal levels of Wnt/b-catenin signaling but reduced Hh-target activation in both the placode and the dermal condensate (Chiang et al., 1999;St-Jacques et al., 1998). Despite the well-known importance of Hh signaling for HF morphogenesis during embryonic skin development (Botchkarev and Paus, 2003;Mill et al., 2003;St-Jacques et al., 1998) and in wound-induced HF formation (Lim et al., 2018), little is known about its potential for de novo HF induction in adult unwounded skin.
Interestingly, already decades ago it has been observed that basal cell carcinoma (BCC) morphologically mimics HF development until the hair germ stage. BCC develops upon supra-physiological Hh-pathway activation in epithelial cells, with the most prevalent mutations in the inhibitory Hhreceptor gene Ptch1. Like HF formation, Hh-driven BCC is characterized by active Wnt/b-catenin signaling (Yang et al., 2008), and early epithelial BCC buds express typical HF-lineage markers even if they don't originate from HFs (Kasper et al., 2011;Yang et al., 2008;Youssef et al., 2012). Morphologically and molecularly, these early BCC buds are thus very similar to embryonic hair germs with one major difference: BCC buds lack a dermal condensate which serves as a focal point for dermal Hh signaling and is required for HF morphogenesis (Yang et al., 2008).
Based on these observations, we hypothesized that HF development past the hair germ stage in BCC requires Hh signaling at increased levels in both epithelial and stromal cells. To address this hypothesis, we used mouse models to induce supra-physiological Hh signaling via Ptch1 deletion in the epithelium and stroma, which indeed led to the induction of new HFs in adult unwounded skin.

Activated Hh signaling in Gli1-expressing cells induced HF-like structures in touch domes
To test whether activation of epithelial and stromal Hedgehog signaling would result in HF formation, we focused on the touch domes (TDs), which are touch-receptive structures located within the interfollicular epidermis (IFE) ( Figure 1A,B). We reasoned that in adult unwounded skin, TDs were the most likely places to achieve experimentally induced de novo HFs because TDs are hotspots for BCC formation (Peterson et al., 2015), which resembles HF development (Yang et al., 2008).
We analyzed and compared the TDs of two mouse models with induced supra-physiological Hhpathway levels in epithelial cells (Lgr6 mouse model), or combined epithelial and stromal cells of the TD (Gli1 mouse model). Specifically, we used Lgr6-EGFP-IRES-Cre ERT2 ;R26 tdTomato ;Ptch1 fl/fl mice (hereafter: Lgr6 creERT2 ;R26 Tom ;Ptch1 fl/fl ) and Gli1-Cre ERT2 ;R26 tdTomato ;Ptch1 fl/fl mice (hereafter: Gli1-creERT2 ;R26 Tom ;Ptch1 fl/fl ). TD areas were identified by the presence of K8+ Merkel cells, and their palisading epithelial cell morphology. Tamoxifen was administered at 8 weeks of age, resulting in the eLife digest We are born with all the hair follicles that we will ever have in our life. These structures are maintained by different types of cells (such as keratinocytes and fibroblasts) that work together to create hair. Follicles form in the embryo thanks to complex molecular signals, which include a molecular cascade known as the Hedgehog signaling pathway.
After birth however, these molecular signals are shut down to avoid conflicting messagesinappropriate activation of Hedgehog signaling in adult skin, for instance, leads to tumors. This means that our skin loses the ability to make new hair follicles, and if skin is severely damaged it cannot regrow hair or produce the associated sebaceous glands that keep skin moisturized.
Being able to create new hair follicles in adult skin would be both functionally and aesthetically beneficial for patients in need, for example, burn victims. Overall, it would also help to understand if and how it is possible to reactivate developmental programs after birth.
To investigate this question, Sun, Are et al. triggered Hedgehog signaling in the skin cells of genetically modified mice; this was done either in keratinocytes, in fibroblasts, or in both types of cells. The experiments showed that Hedgehog signaling could produce new hair follicles, but only when activated in keratinocytes and fibroblasts together. The process took several weeks, mirrored normal hair follicle development and resulted in new hair shafts. The follicles grew on both the back of mice, where hair normally occurs, and even in paw areas that are usually hairless.
Not unexpectedly the new hair follicles were accompanied with skin tumors. But, promisingly, treatment with Hedgehog-pathway inhibitor Vismodegib restricted tumor growth while keeping the new follicles intact. This suggests that future work on improving "when and where" Hedgehog signaling is activated may allow the formation of new follicles in adult skin with fewer adverse effects. Illustrative cartoon of HF and TD structures in wild type skin. Physiological Hh signaling is present in both TD epithelium and TD stroma. The presence of K8+ Merkel cells is characteristic for TDs. (C) Schematic representation of the experimental timeline and the Lgr6 creERT2 ;R26 Tom ;Ptch1 fl/fl and Gli1 creERT2 ;R26 Tom ;Ptch1 fl/fl mouse models. (D-E) Lgr6 EGFP and Gli1 LacZ expression in dorsal telogen skin. Filled arrowhead: indicates Lgr6-expression in the HF infundibulum. Empty arrowheads: indicate lack of Lgr6-or Gli1-expression in HF infundibula (n = 3 mice per genotype). (F-I) Mice were treated with tamoxifen (TAM) at 8 weeks of age and dorsal skin was analyzed 5 weeks post TAM. For each genotype !3 mice and numerous TDs were analyzed (Supplementary file 1). (F) Illustrative cartoon and experimental Tomato-tracing of Lgr6-expressing cells. TDs of Lgr6 creERT2 ;R26 Tom ;Ptch1 fl/wt control skin were phenotypically normal. (G) Illustrative cartoon and experimental Tomato-tracing of Gli1-expressing cells. TDs of Gli1 creERT2 ;R26 Tom ;Ptch1 fl/wt control skin were phenotypically normal. (H) Tomato-traced TD of Lgr6 creERT2 ; Figure 1 continued on next page constitutive activation of Hh signaling via homozygous inactivation of Ptch1 and simultaneous Tomato-tracing of Lgr6-or Gli1-expressing cells ( Figure 1C). First, we confirmed that Lgr6 expression, and consequently Tomato-tracing, were in the TD restricted to epithelial cells ( Figure 1D,F), and Gli1 expression and Tomato-tracing were present in both epithelial and stromal TD cells ( Figure 1E,G). Next, we analyzed the phenotypes of both Lgr6 and Gli1 mouse models 5 weeks post tamoxifen, a sufficiently long time to allow possible de novo HFs to form (Rendl et al., 2005). Homozygous Ptch1 inactivation in Lgr6-expressing cells resulted as expected in BCC-like lesions in HFs, IFE and TDs ( Figure 1H, Figure 1-figure supplement 1; Peterson et al., 2015). Strikingly, however, homozygous Ptch1 inactivation in Gli1-expressing cells resulted in addition to BCC-like lesions in HFs and TDs, in the formation of structures in TDs that resembled de novo HFs ( Figure 1I, Figure 1-figure supplement 3). These structures had the appearance of typical concentric anagen HF layers and a pigmented hair bulb or mature hair shaft ( Figure 1I, Figure 1-video 1). Importantly, by 9 weeks post tamoxifen, these HF-like structures occurred in every single Gli1 creERT2 ; R26 Tom ;Ptch1 fl/fl TD examined (27/27) but extremely rarely in TDs of Lgr6 creERT2 ;R26 Tom ;Ptch1 fl/fl mice (2/25; please also see Discussion) ( Figure 1J). No such HF-like structures were observed in TDs of wild type mice (with physiological Hh signaling in epithelial and stromal cells), upon heterozygous Ptch1 inactivation (Gli1 creERT2 ;R26 Tom ;Ptch1 fl/wt and Lgr6 creERT2 ;R26 Tom ;Ptch1 fl/wt ) ( Figure 1F,G), or in non-tamoxifen controls (Gli1 creERT2 ;R26 Tom ;Ptch1 fl/fl and Lgr6 creERT2 ;R26 Tom ;Ptch1 fl/fl ) (Supplementary file 1). Therefore, induction of supra-physiological Hh signaling in epithelial and stromal cells (Gli1 model) but not epithelial cells alone (Lgr6 model) was sufficient to induce HF-like structures in TDs of adult mouse skin.

Characterization of de novo HFs in touch domes
Next we investigated whether the observed structures were functional HFs, and indeed de novo induced. Thus, we stained the skin of Gli1 creERT2 ;R26 Tom ;Ptch1 fl/fl mice for Keratin 71 (K71) and Keratin 6 (K6) (Figure 2A), which mark specific layers of the anagen HF (Yang et al., 2017). These Keratin-staining patterns were very similar to those of hair-cycle stage-matched wild type anagen HFs ( Figure 2B). Also, the presence, and specific pattern of hair pigment in these structures were typical for anagen HFs, in the shown picture matching the anagen III hair-cycle stage (Figure 2A . This demonstrates that de novo HFs enter the hair cycle after their first anagen (Paus and Cotsarelis, 1999).
To verify that these HFs were de novo induced, thorough examination of the Tomato lineage tracing pattern was of key importance. The Gli1-positive HF and TD populations self-renew    Figure 1I. Figure 1-video 1. Video containing all recorded z-levels of the touch dome area presented in Figure 1I. https://elifesciences.org/articles/46756#fig1video1  Figure 1E; Xiao et al., 2015). Cell crossover only occurs for example in response to full-thickness wounding or TPA treatment (Brownell et al., 2011;Kasper et al., 2011). Importantly, Ptch1 deletion did not trigger HF cell migration towards the IFE or TD as we never observed Tomato-tracing spanning from a pre-existing HF -via the infundibulum -to the IFE or TD in any control or Gli1 creERT2 ;R26 Tom ;Ptch1 fl/fl skin ( An additional characteristic of de novo HFs was that their hair shafts were considerably shorter and thinner than hair shafts from both pre-existing HFs in the same mice (Gli1 creERT2 ;R26 Tom ;Ptch1 fl/ fl ) and HFs in control mice with wild type phenotype (Gli1 creERT2 ;R26 Tom ;Ptch1 fl/wt ) ( Figure 2C and In conclusion, the continuous lineage-tracing from HF to TD, and the characteristic hair shaft measurements, demonstrated that combined epithelial and stromal Hh-pathway activation in the Gli1 mouse model via homozygous Ptch1 inactivation resulted in de novo HFs, within TDs.

De novo HF formation recapitulates embryonic HF development
Next we characterized the stages of de novo HF development in TDs equivalent to the embryonic HF developmental stages of HF placode, hair germ, hair peg and mature follicle (Rendl et al., 2005). We analyzed Gli1 creERT2 ;R26 Tom ;Ptch1 fl/fl and control skin, tamoxifen treated at 8 weeks of age, and collected samples 10, 17, 25, 27, 29, 33, 35 and 36 days after tamoxifen treatment. These sampling time points enabled us to map the entire stereotypical time course of de novo HF formation ( Figure 3A-D, Supplementary file 1). In some TDs, the HF-placode stage could be already detected 10 days post tamoxifen administration, and was accompanied by the appearance of a dermal condensate ( Figure 3A). The dermal condensate and dermal papillae were delineated by denser Tomato-tracing of stromal cells compared to adjacent traced K5+ epithelial cells. Note the lack of continuous pre-existing HF-to-TD tracing at this early developmental stage ( Figure   HFs with a thin hair shaft formed in the TDs of Gli1 creERT2 ;R26 Tom ;Ptch1 fl/fl skin (white arrow). For the quantification, we analyzed hair shafts of de novo HFs from Gli1 creERT2 ;R26 Tom ;Ptch1 fl/fl mice (blue bracket), old/pre-existing Zigzag HFs from the same mice (white bracket), and Zig-zag HFs from wild-type-phenotype control mice (Gli1 creERT2 ; R26 Tom ;Ptch1 fl/wt ) (n = 3 mice for each genotype; 34 de novo, 314 old/pre-existing, and 437 control HFs; Figure 2-source data 1). Hair shaft length was measured in telogen stage hair shafts from the hair club to the HF opening (as indicated by the blue and white brackets). Left panel: Each dot represents a hair shaft and the dots are colored according to the HF type (blue for de novo HFs, gray for pre-existing Zig-zag HFs, orange for control Zig-zag HFs). De novo hair shafts were significantly smaller (p-value<1.10 À6 ) and thinner (p-value<1.10 À6 ) compared to old/pre-existing and control Zig-zag hairs (Mann-Whitney U test). Arrowheads on the x-and y-axis indicate the mean values of hair shaft length and width of de novo, old/pre-existing or control Zig-zag HFs, respectively. TD: touch dome. HF: hair follcile. TAM: tamoxifen. BF: bright field. Scale bars: 100 mm (A), 50 mm (B, C). The online version of this article includes the following source data and figure supplement(s) for figure 2: Source data 1. Quantification of hair width and length.  Interestingly, almost all dermal condensates and dermal papillae of de novo forming HFs were fully Tomato traced ( Figure 3A-D,F), suggesting that either continuous high levels of Hh signaling in stromal cells (Ptch1 fl/fl ) are necessary for all stages of de novo HF development, or that stromal cells with constitutive Hh-pathway activation outcompete dermal condensate and dermal papilla cells with lower Hh-signaling levels. Importantly, Syndecan-1 (SDC1) expression that marks early dermal condensates in wild type embryonic skin ( Figure 4A; Richardson et al., 2009), could already be detected 10 days post tamoxifen in dermal condensates of Gli1 creERT2 ;R26 Tom ;Ptch1 fl/fl skin ( Figure 4B), and was fully established in the dermal papilla at the hair germ stage ( Figure 4C).
Finally, to demonstrate that these de novo HFs do indeed have active Hh signaling, we stained for Gli1 mRNA expression; as a reporter of canonical Hh-pathway activity. The placodes ( Chiang et al., 1999;St-Jacques et al., 1998). Wild type TD epithelial and stromal cells expressed Gli1 mRNA as expected ( Figure 4E and This Gli1 RNA-FISH combined with antibody staining for Tomato-lineage tracing also confirmed that Tomato-tracing (cells with Ptch1 deletion) and Gli1 expression were highly correlated, as expected ( Figure    . In TDs, the clearly identifiable de novo HFs develop alongside epithelial tumor growth which is characterized by palisading cells and lack of HF-like structures. It has been shown previously that BCC-like lesions in dorsal skin dramatically shrink within seven days upon vismodegib treatment (Eberl et al., 2018). Vismodegib is a Hh-pathway inhibitor acting at the level of Smoothened (Smo), and using the optimized treatment scheme from Eberl et al. (2018) we tested whether established de novo HFs would persist or would diminish as the BCC-tumor-growth area does. We treated Gli1-creERT2 ;R26 Tom ;Ptch1 fl/fl mice with tamoxifen at 8 weeks. Five to seven weeks after tamoxifen treatment when de novo HFs were clearly established in TDs, we took a dorsal biopsy prior to vismodegib treatment (untreated biopsy), and then treated the mice daily with vismodegib (50 mg/ kg body weight i.p.) for seven days ( Figure 5A).
Reassuringly, in pre-existing HFs we found considerable reduction of tumor size ( Figure 5B) as well as absence of Ki67 staining in Tomato-traced areas when comparing 7 day vismodegib treated samples with the untreated biopsies of the same mice ( Figure 5C). This reduction in tumor size demonstrated that vismodegib treatment worked as expected. In TDs, the tumor areas also dramatically diminished in size and Ki67 staining, however the de novo HFs persisted ( Figure 5D and Figure 5figure supplement 1). In conclusion, these experiments confirmed that de novo HFs indeed represent HFs that are independent of tumor structures as they persist upon vismodegib treatment when the surrounding BCC-tumor-growth areas are nearly gone.

Stromal Hh-pathway activation alone is not sufficient to induce HF neogenesis
De novo HFs were induced by strong activation of Hh signaling (Ptch1 fl/fl ) in epithelial and adjacent stromal cells and persisted upon vismodegib treatment when the surrounding BCC-growth areas were nearly gone. As BCC growth merely depends on epithelial Hh-pathway activation, it was tempting to test whether homozygous inactivation of Ptch1 exclusively in stromal cells would be sufficient to induce de novo HFs without tumor development. To that end, we generated the Col1a2 mouse model (Col1a2-Cre ER ;R26 tdTomato ;Ptch1 fl/fl , hereafter: Col1a2 creER ;R26 Tom ;Ptch1 fl/fl ) ( Figure 6A,B), which drives supra-physiological Hh signaling in the stromal compartment only, via the collagen type I alpha two chain promoter. Non-tamoxifen controls (Col1a2 creER ;R26 Tom ;Ptch1 fl/fl mice) showed some tracing in the skin stroma, which however did not result in an adverse skin phenotype except for earlier anagen entry ( Figure 6-figure supplement 1). Administration of tamoxifen at 8 weeks of age resulted in substantial Tomato-tracing that was restricted to the stromal skin compartment (Figure 6-figure supplement 2), and importantly, the stromal cells of the TD were also traced ( Figure 6D). Homozygous Ptch1 inactivation in Col1a2-expressing cells resulted in increased stromal cell density in TDs ( Figure 6E), but did not result in de novo HF formation ( Figure 6C,E), nor did stromal cells stain positive for SDC1 even 9 weeks after tamoxifen treatment   Figure 6F). We also detected fully traced and highly condensed dermal clusters of cells (resembling dermal condensates) in TD-adjacent infundibula and underneath the regular IFE ( Figure 6G, Figure 6-figure supplement 2B), which did not result in de novo HF induction and the dermal cell condensations were entirely negative for SDC1 expression ( Figure 6G). We conclude that stromal activation of Hh signaling (Ptch1 fl/fl ) leads to increased stromal cell density and formation of cell condensates, however it is not sufficient to induce HF neogenesis in TDs nor elsewhere in skin without adjacent epithelial Hh-pathway activation.

Only TDs of the Gli1 mouse model are competent for HF initiation
To directly compare all three different mouse models (Lgr6 creERT2 /, Gli1 creERT2 /, Col1a2 creER ;R26 Tom ; Ptch1 fl/fl ) in their competence to initiate de novo HFs, we analyzed TDs 10 days post tamoxifen administration, the time point when epithelial proliferation became evident via, for example, increased BrdU incorporation in hair-forming TDs before appearance of morphological hair germ formation (Figure 7, Figure 7-figure supplement 1). We investigated the stromal TD compartment using the alkaline phosphatase (AP) assay and SDC1 staining, which are both characteristic for HFinducing dermal condensates (Ito et al., 2007;Richardson et al., 2009). Positive staining for both alkaline phosphatase (ALPL) and SDC1 expression was observed in the Gli1 creERT2 ;R26 Tom ;Ptch1 fl/fl mice, but not in Gli1 creERT2 ;R26 Tom ;Ptch1 fl/wt , Lgr6 creERT2 ;R26 Tom ;Ptch1 fl/fl , or Col1a2 creER ;R26 Tom ; Ptch1 fl/fl mice ( Figure 7A,B). The epithelial TD compartment could not be stained for a comparable marker of early HF induction, as early BCC buds express typical HF-lineage markers. Indeed, we and others have not found a single mRNA/protein stain that would distinguish HF epithelial placode from BCC formation (Kasper et al., 2011;Yang et al., 2008;Youssef et al., 2012). Altogether, staining for early signs of HF formation demonstrated that only the Gli1 (Ptch1 fl/fl ) model bears TDs that are competent for de novo HF formation.

Cells outside TD niches can give rise to de novo HFs
We established that de novo HF formation required close apposition of epithelial and stromal Hh signaling (Ptch1 fl/fl ), and furthermore how to identify de novo HFs based on their morphology and continuous lineage tracing. We next asked whether de novo HFs could also form from non-TD areas with comparable adjacent epithelial-stromal Hh-pathway activation.
In addition to TDs, the HF isthmus also harbors adjacent epithelial and stromal Gli1-Tomato traced cells; the latter evident through Tomato/PDGFRb co-staining of stromal cells ( Figure 8A,B). Thus, we re-examined the HF isthmus areas of Gli1 creERT2 ;R26 Tom ;Ptch1 fl/fl skin for potential de novo HFs. Indeed, at low frequency, we observed de novo HFs in the isthmus of pre-existing HFs that were most likely newly formed ( Figure 8C,D). Although it was not possible to unequivocally determine de novo formation through continuous Tomato-tracing of the infundibulum (as these de novo HFs seem to merge directly into the isthmus area of pre-existing HFs), based on their morphology, positioning and the co-occurrence of four hair shafts instead of normally three (in dorsal skin of 17 week-old mice), these HFs likely formed newly from the isthmus of pre-existing HFs ( Figure 8C). Such HFs have not been observed in phenotypically normal control skin (Gli1 creERT2 ;R26 Tom ;Ptch1 fl/wt and Lgr6 creERT2 ;R26 Tom ;Ptch1 fl/wt ) or in skin with HF tumors (Lgr6 creERT2 ;R26 Tom ;Ptch1 fl/fl )   Figure 8D). Reassuringly, in the HF isthmus Lgr6 expression is restricted to epithelial cells only, whereas Gli1 is expressed in epithelial and adjacent stromal cells (Füllgrabe et al., 2015); suggesting that adjacent epithelial-stromal Hh signaling in areas outside of the TD may form de novo HFs.

Epithelial and adjacent stromal Hh-pathway activation induces de novo HF formation in hairless paw skin
The mouse hindpaw (plantar) epidermis is a skin region devoid of hair follicles and sweat glands, and is therefore ideal to test whether epithelial and stromal Hh-pathway activation can induce de novo HFs divorced from any confounding effects of nearby HF-or TD-niche signals ( Figure 9A). When we probed for Gli1 expression in the plantar skin using Gli1 LacZ reporter mice, we consistently found small Gli1-BGAL expressing clusters of epithelial and adjacent stromal cells in the plantar skin ( Figure 9B, Figure 9-figure supplement 1A), which we confirmed with short-term lineage tracing in Gli1 creERT2 ;R26 Tom mice (tamoxifen at P8w; sample collection 7 days later) ( Figure 9C, Figure 9figure supplement 1B). Therefore, the plantar skin was a suitable area for studying if de novo HFs can form in the Gli1 mouse model upon homozygous Ptch1 inactivation.
Testing for de novo HF induction, we analyzed the hindpaws of Gli1 creERT2 ;R26 Tom ;Ptch1 fl/fl mice and control littermates (Gli1 creERT2 ;R26 Tom ;Ptch1 fl/wt ) using the same treatment scheme and sample collection times as for dorsal skin (tamoxifen at P8w; sample collection 5 and 9 weeks post tamoxifen; additional time points see Supplementary file 1). Indeed, we found numerous de novo HFs with hair shafts in the normally hairless region. These HFs were fully Tomato-traced and showed normal inner-layer differentiation based on morphology and K6 staining ( Figure 9D-E, Figure 9-figure supplement 2).
Taken together, the analysis of dorsal HF-isthmus as well as hairless paw skin demonstrated that combined epithelial and stromal Hh-pathway activation can induce de novo HFs independently of TD niches.

Discussion
Hitherto, hair follicle neogenesis in adult skin had only been observed under exceptional circumstances, such as upon repair of large wounds (Breedis, 1954;Ito et al., 2007). A recent study found that it is possible to induce HFs even in small wounds upon supra-physiological Hh-pathway activation in the wound stroma (Lim et al., 2018). This report and our present study recognize modulation of Hh signaling as a new approach to induce de novo HFs in adult skin, and define activation in the stromal skin compartment as critical. As wounding of skin initiates a major reorganization of the epithelial and mesenchymal tissue including the activation and differentiation of a large number of cell types (Arwert et al., 2012;Joost et al., 2018;Schäfer and Werner, 2008), and may even provide an embryonic-like environment , the precise (e.g. minimal) molecular signals that are required for HF induction in adult skin remain elusive. Here we revealed that in unwounded skin, experimentally elevated Hh signaling in epithelial and adjacent stromal cells was sufficient to induce Figure 6 continued the TDs of Col1a2 creER ;R26 Tom ;Ptch1 fl/fl mice treated with tamoxifen (TAM) at 8 weeks. Dorsal skin was analyzed 9 weeks post TAM treatment (n = 3 mice). No de novo HFs were observed. (D-G) Col1a2 creER ;R26 Tom ;Ptch1 fl/fl and Col1a2 creER ;R26 Tom ;Ptch1 fl/wt control mice were treated with TAM at 8 weeks and dorsal skin was analyzed 10 days, 5 weeks, and 9 weeks after TAM treatment (n = 3 mice per genotype and time point). (D) TDs of mice with heterozygous Ptch1 deletion were phenotypically normal. (E) TDs of mice with homozygous Ptch1 inactivation did not develop de novo HFs. Frequently, a higher cell density in stroma was observed. (F-G) Syndecan-1 (SDC1) staining was negative in the condensed stroma of TDs (F) as well as in dermal cell condensations (arrows) underneath the IFE and HF infundibula (G) in Col1a2 creER ;R26 Tom    de novo HFs, extending our understanding of how Hh signaling can be modulated to induce HFs in adult skin.
To achieve efficient de novo HF induction in unwounded skin, supra-physiological Hh signaling in both compartments, the epithelium and stroma, was necessary. Normal TD maintenance also requires active and balanced Hh signaling in adjacent epithelial and stromal cells ( Figure 4E; Xiao et al., 2015). Increased Hh-signaling levels in TD epithelial cells result in BCC-like tumors, even  (C) Gli1 creERT2 ; R26 Tom mice were treated with tamoxifen (TAM) at 8 weeks of age. The hindpaws were collected 1 week post TAM and immuno-stained with K5 antibody (n = 3 mice). Asterisks mark autofluorescence on the outermost keratinized layer. (D-E) Gli1 creERT2 ;R26 Tom ;Ptch1 fl/fl and control Gli1 creERT2 ;R26 Tom ;Ptch1 fl/wt mice were treated with TAM at 8 weeks of age. Hindpaws were collected 5 or 9 weeks post TAM (n = 3 mice for each genotype; except n = 2 for Gli1 creERT2 ;R26 Tom ;Ptch1 fl/wt 5 weeks post TAM ). (D) Numerous de novo HFs formed in Figure 9 continued on next page though stromal cells have active (albeit physiological) Hh signaling. Only in extremely rare cases (twice) did we detect a de novo HF-like structure in TDs of Lgr6 creERT2 ;R26 Tom ;Ptch1 fl/fl mice based on morphology (as lineage tracing in these mice cannot provide information on de novo HF formation; Figure 1H and Figure 1-figure supplement 2). Importantly therefore, to effectively induce de novo HFs in TDs, high levels of Hh/Gli signaling in both compartments were necessary (Gli1 creERT2 ; R26 Tom ;Ptch1 fl/fl ). This requirement of Hh-signal activation at precise levels and in the right compartments is in agreement with a recent study demonstrating that b-catenin-induced de novo HF formation was not only dependent on stromal Hh signaling, but also required two intact Smo alleles (for a maximal Hh-pathway activation) to enable efficient de novo HF induction (Lichtenberger et al., 2016).
It has been shown more than twenty years ago that the activation of epithelial b-catenin in mouse skin can induce new HFs (Gat et al., 1998;Lo Celso et al., 2004), and more recently that activation of epithelial Wnt/b-catenin signaling increases de novo HF formation within wounds (Ito et al., 2007). However, Wnt signaling has to be blocked in dermal fibroblasts to allow de novo HF induction during wound regeneration (Rognoni et al., 2016). It is known that early stage BCCs resemble early stages of HF development and both are dependent on Wnt-and Hh-pathway activation, with the major morphological difference that BCC lacks a dermal condensate (Yang et al., 2008). Learning from abrogated embryonic HF development (St-Jacques et al., 1998) led us to hypothesize that simultaneously activating supra-physiological Hh signaling in the stroma underneath developing BCC may enable de novo HFs. Indeed, by coordinating the activation of Hh/Gli signaling (cell type specific and high levels) we were able to induce de novo HFs by 'redirecting' some of the BCC buds to HF formation without the need of wounding. Nevertheless, this induction occurred in the presence of oncogenic signal activation (i.e. presence of a tumor environment or tumor-like cellular status of Ptch1 fl/fl HF-inducing epithelial cells) which may to some extent mimic a wounding situation (Dvorak, 1986).
The molecular and cellular similarities of tumorigenesis and wound healing are still unfolding, yet whenever de novo HFs were to be found, either oncogenic signaling or a wound environment was involved. This supports the long-standing recognition of the similarity between tumor and wound healing signaling (tumors as 'wounds that do not heal') -and raises the key question of what exactly is the relationship between tumorigenesis and signals inducing de novo HF formation? It may indeed be the case that in order to overcome inhibitory signals, de novo HF morphogenesis in adult skin requires such major activating signals provided by oncogenesis or wounding. Interestingly however, previous literature suggests that only initial HF placode and/or dermal condensate formation may require such strong signals whereas progression to a mature HF does not require continued tumorigenic or wound signaling (Brown et al., 2017;Ito et al., 2007;Lo Celso et al., 2004;Silva-Vargas et al., 2005). For example, HF tumors require continuous Wnt/b-catenin signaling, whereas transient activation of this pathway is sufficient to induce de novo HFs in adult mouse epidermis (Lo Celso et al., 2004). More recently, intra vital imaging from Wnt/b-catenin induced tumor outgrowths demonstrated that non-mutant cells, remaining from regressed outgrowths, could develop into new functional HFs. Most interestingly, tumor outgrowth depended on the presence of mutated cells, however the new appendages were formed from wild type cells facilitated by their (altered) niche environment (Brown et al., 2017). Taken together, these are promising examples that de novo HF induction without accompanied tumor growth in unwounded adult skin may in principle be  possible, if the right signals at the right time and restricted period, and in the right compartments were provided. Here, we spatiotemporally defined such productive and specific molecular signals.
Lastly, and importantly, we exploited the hairless paw plantar skin to examine de novo HF morphogenesis in the absence of confounding signals from pre-existing HFs or TDs. Strikingly, we observed numerous de novo HFs throughout this nominally hairless skin in Gli1 creERT2 ;R26 Tom ;Ptch1 fl/ fl mice. Crucially, nearly all these de novo HFs developed without attendant BCC-like lesions suggesting that de novo HF morphogenesis may indeed be successfully initiated without a tumor microenvironment; while the vismodegib experiment suggests persistence of such structures when the tumor microenvironment regresses in dorsal skin. Examining these two divergent tissues in molecular detail, one permissive (dorsal) and one suppressive (paw) to dual BCC and HF induction, could therefore be a next step of unraveling the complexity of how these heterogeneous signals interact.
In sum, molecular strategies for the induction of complex epithelial structures in the adult remain a major challenge in regenerative medicine. Our study demonstrates that cell-type specific modulation of a single pathway was sufficient to induce complex epithelial structures in the adult body, a discovery aiding our understanding of adult tissue biology and regenerative medicine. Mouse models and treatments treatment, when de novo HFs were clearly established in TDs, a dorsal biopsy prior to vismodegib treatment was taken. From then, the vismodegib was given daily (50 mg/kg body weight i.p.) for a week and dorsal samples were collected for further analysis (n ! 3 mice). Embryos were collected at E15.5 and E16.5, and wild-type control samples for dorsal tissue were collected at postnatal day 27 and week 9. All mouse experiments were performed in accordance to Swedish legislation and approved by the Stockholm South or Linkö ping Animal Ethics Committees.

Materials and methods
Tissue staining, microscopy and image analysis All antibodies, b-Galactosidase, alkaline phosphatase (AP), and RNA-FISH stainings were performed on either mouse dorsal skin or hindpaw samples as described below (1-5). For nuclear stains, TO-PRO-3, Hoechst 33342 or DAPI (all from Invitrogen) were used in the different applications below (1-4).
(5) LacZ (b-Galactosidase) staining. Freshly obtained skin tissue was fixed (4% paraformaldehyde in PBS) for 30 min at RT. Tissues were washed for 15 min with rinse buffer (2 mM MgCl2, 0.01% Nonidet P-40 in PBS). Subsequently, the b-galactosidase substrate solution (1 mg/mL X-Gal, 5 mM K 3 Fe (CN) 6 , 5 mM K 4 Fe (CN) 6 Á3H 2 O in rinse buffer) was added and the tissues were incubated for 18 hr at 37˚C in the dark. The substrate was removed, and the tissues were washed twice in PBS for 10 min and kept in 70% ethanol until embedding (maximum 48 hr). The stained tissues were processed into paraffin blocks according to standard procedures. Tissue sections (4 mm) were prepared and counterstained with eosin or H&E. Used in Figure 1E; Figure 9B; Figure 9-figure supplement 1A.
Imaging was performed using a Leica (color bright-field images), LSM710-NLO confocal microscope (Zeiss) or Nikon A1R confocal microscope. Image analysis was performed using NIS-Elements software (Nikon), Zen 2009 software (Zeiss), or ImageJ, and images were occasionally optimized for brightness, contrast, and color balance. RNA-FISH images are presented as maximum intensity projections covering 6 mm of depth.