Epithelial stem cells (EP-SCs) located in the bulge region of a hair follicle (HF) have the potential to give rise to hair follicle stem/progenitor cells that migrate down to regenerate HFs. Bone morphogenetic protein (BMP) signaling has been shown to regulate the HF cycle by inhibiting anagen induction. Here we show that active BMP signaling functions to prevent EP-SC activation and expansion. Dynamic expression of Noggin, a BMP antagonist, releases EP-SCs from BMP-mediated restriction, leading to EP-SC activation and initiation of the anagen phase. Experimentally induced conditional inactivation of the BMP type IA receptor (Bmpr1a) in EP-SCs leads to overproduction of HF stem/progenitor cells and the eventual formation of matricomas. This genetic manipulation of the BMP signaling pathway also reveals unexpected activation of β-catenin, a major mediator of Wnt signaling. We propose that BMP activity controls the HF cycle by antagonizing Wnt/β-catenin activity. This is at least partially achieved by BMP-mediated enhancement of transforming growth factor-β-regulated epithelial cell-specific phosphatase (PTEN) function. Subsequently, PTEN, through phosphatidyl inositol 3-kinase-Akt, inhibits the activity of β-catenin, the convergence point of the BMP and Wnt signaling pathways.
The cycle-dependent postnatal regeneration of hair follicles (HFs) and the underlying well organized HF architecture provide an excellent model for studying the molecular mechanisms that regulate stem cell self-renewal, proliferation, differentiation, and cell fate determination . The HF is composed of a permanent portion, which includes sebaceous glands and the underlying bulge area, and a dynamic portion that continually undergoes cycles of apoptosis-driven retraction (catagen phase), a short period of rest (telogen phase), and then a period of active growth (anagen phase) . The bulge area functions as a niche in which epithelial stem cells (EP-SCs) reside [3, 4] and are maintained in an equilibrium between self-renewal and lineage commitment. EP-SCs are multipotent, giving rise to daughter cells (hair follicle stem/progenitor cells) that migrate downward to regenerate hair follicles [5, 6]. It is proposed that in the early anagen phase of each hair growth cycle, signals from the retracted dermal papilla (DP) activate EP-SCs to divide [1, 3]. The hair follicle stem/progenitor cells subsequently migrate downward to the bottom of the hair bulb adjacent to the DP, where they give rise to precursors of the inner root sheath (IRS) and precortex cells. The precortex cells then differentiate into concentric layers of keratinocytes to form the hair shaft.
The Wnt pathway has been well characterized and shown to play an essential role in regulating EP-SC proliferation and fate determination. Wnt signaling is dynamically changed during the HF cycle with high activity in early anagen . β-Catenin is the key transcription factor activated by Wnt signaling [8, 9]. Constitutive activation of β-catenin results in de novo HF morphogenesis in adult skin (normally an embryonic stage-specific phenomenon), and finally leads to HF tumor formation . Targeted inactivation of β-catenin diverts stem cells toward an epidermal cell fate, at the expense of HF cell fate . Lef1, a partner of β-catenin, is not only required for early induction of the hair germ [12, 13], but is also necessary for hair shaft differentiation. Enforced expression of a mutant Lef1 that lacks the β-catenin binding domain leads to squamous cysts and skin tumors . Tcf3, another β-catenin partner, plays a role in maintenance of EP-SCs . These findings suggest that the Wnt signal plays an important role in the regulation of HF morphogenesis and skin regeneration by favoring HF lineage and progenitor cell differentiation versus epidermal cell fate.
Bone morphogenetic protein (BMP) signaling also plays a role in HF morphogenesis, postnatal regeneration, and control of the HF cycle through regulation of hair matrix precursor cell proliferation and differentiation [16, 17]. Enhanced BMP signal activation by ectopically expressing BMP4  or targeted inactivation of the BMP antagonist Noggin  results in significant retardation of HF induction and progressive baldness. Overexpression of Noggin leads to induction of the anagen phase [19, 20] and disruption of hair shaft differentiation . Recently, several groups reported that blocking BMP signaling by deletion of the BMP type 1a receptor (Bmpr1a) leads to disruption of differentiation and hair loss [22, , –25]; subsequent HF tumor formation occurs in some systems, and this seems to depend on which promoter was used to drive Cre [22, 24].
The canonical BMP signal is mediated by Smad transcriptional factors [26, 27]; however, accumulated evidence shows that BMP signaling also occurs through different pathways [26, 28], including the Pten/Akt cascade [29, 30]. The tumor suppressor PTEN (or transforming growth factor-β-regulated epithelial cell-specific phosphatase) is a dual lipid and protein phosphatase. PTEN functions as an antagonist of phosphatidyl inositol 3-kinase (PI3K) . The serine kinase Akt is downstream of the PI3K signal . PTEN has been shown to inhibit nuclear accumulation of β-catenin through inhibition of Akt activity . Likewise, Akt through glycogen synthase kinase (GSK)3β can induce nuclear accumulation of β-catenin . Recently, Akt was also shown to be directly involved in controlling β-catenin stabilization and activation in coordination with 14-3-3ζ .
Despite intensive studies on BMP signaling in HF development, none of these studies has directly focused on EP-SCs. The mechanism of BMP signaling in EP-SC regulation and HF tumorigenesis remains largely unknown. In this study, we report that dynamic expression of Noggin in EP-SCs during the HF cycle leads to the cyclic inactivation of BMP signaling, which correlates with an expansion of EP-SC numbers during the early anagen phase. We also show that blocking the BMP signal by inducing mutation of Bmpr1a in EP-SCs results in the expansion of EP-SCs and progenitors followed by matricoma formation, whereas mutation of Bmpr1a in HF progenitors/precursors leads to disruption of hair shaft differentiation. Furthermore, we provide evidence that the PTEN-Akt cascade mediates the convergence of the BMP and Wnt pathways in EP-SC regulation through control of β-catenin activity.
Materials and Methods
Generation of Bmpr1a Knockout and BMP4 Transgenic Mice
In the Bmpr1afx/fx mouse line, the second exon of the Bmpr1a gene is flanked by two loxP sites . Using the Bmpr1afx/fx line and the interferon-inducible Cre line Mx1-Cre  (The Jackson Laboratory, Bar Harbor, ME), both the Bmpr1afx/fx line and Mx1-Cre are maintained as C57BL/6J background through backcrossing with C57/BL6J for many generations. MX1Cre+Bmpr1afx/fx homozygous, Mx1Cre+Bmpr1afx/+ heterozygous, and Mx1Cre−Bmpr1afx/fx or Mx1Cre−Bmpr1afx/+ wild-type control mice were generated. To introduce Bmpr1a inactivation, pups were injected intraperitoneally with 250 μg of poly(I:C) (Sigma-Aldrich, St. Louis, http://www.sigma-aldrich.com; P-0913) at specific times, once every other day for a total of three injections. poly(I:C) induces the production of interferon which, in turn, induces Cre expression. Cre in turn mediates loxP-dependent DNA recombination. Tails were collected on postnatal day 21 (P21) or at the time of termination to extract genomic DNA. Genotyping was done by PCR assay with different combinations of primers (Pr). Pr1: GCAGCTGCTGCTGCAGCCTCC, Pr2: TGGCTACAATTTGTCTCATGC, Pr3: GGTTTGGATCTTAACCTTAGG, Pr4: TACCTGGA- AAATGCTTCTGT, Pr5: TGATCTCCGGTATTGAAACT. Pr1 and Pr2 amplify 230- and 150-bp products specific for the floxed and wild-type alleles, respectively. The combination of Pr2 and Pr3 amplifies a 180-bp product specific for the targeted allele after poly(I:C)-induced Cre-mediated recombination. Pr4 and Pr5 were used to identify Mx1-Cre with an 808-bp product in Cre-positive mice. The BMP4 transgenic mouse was backcrossed with C57BL/6J for five to six generations. In this mouse line, human BMP4-cDNA was expressed under the control of a 2.4-kb mouse BMP4 regulatory region. Human BMP4 was expressed in differentiated hair matrix and in the hair shaft derived from the epithelium .
Primary Keratinocyte Isolation and Culture
Primary keratinocytes were isolated from the skin of neonatal C57BL/6J pups as follows. Using a keratinocyte primary isolation kit (Cascade Biologics, Portland, OR, http://www.cascadebio.com; R-014-K), and following the protocol provided by Cascade Biologics, ventral skin was collected from neonatal B6 pups and cut into strips of approximately 0.5 cm × 1.5 cm using a scalpel. The skin strips were digested in a splitting solution (Cascade Biologics, R-008-3) at 4°C for 16–21 hours, at which point the epidermis can easily be separated from the dermis. The epidermal pieces were incubated in a trypsin/EDTA solution for 30 minutes in a 37°C water bath, followed by defined trypsin inhibitor neutralization. The cells were then washed, resuspended, and cultured in EpiLife Medium with HKGS-V2 and PSA (Cascade Biologics) at a concentration of 10,000 cells per square centimeter in standard tissue culture flasks. The medium was changed every other day until the culture reached 70% confluence. The cells were harvested and replated into new tissue culture flasks at a density of 10,000 per milliliter for subsequent subculture. For the reporter activity study, cells were subcultured in 24-well plates in a low calcium concentration (0.05 mM, final). After the cultures reached 70% confluence (48 hours after replating), the cells were transfected with 2 μg of Top-flash (T-cell factor [TCF] reporter)  or 2 μg of Fop-flash plasmids (mimicked control for TCF reporter with mutant TCF binding sites)  combined with 0.05 μg of PRL-TK plasmid (experimental reporter control; Promega, Madison, WI, http://www.promega.com; E2241), respectively. To study the effect of PTEN and Akt on BMP/Wnt crosstalk, cells were also cotransfected with wild-type or dominant negative forms of PTEN and Akt. Twelve hours after transfection, the cells were treated with Wnt3A, BMP2, BMP4, Noggin, or Ly294002 (a PI3K inhibitor) separately or in different combinations for another 12 hours. Luciferase activity was detected using the Dual-Luciferase Reporter Assay System (Promega, E1910).
Immunohistochemical or Immunohistofluorescent Staining
Mouse skin was collected and fixed overnight in zinc formalin (Richard-Allan Scientific, Kalamazoo, MI, http://www.rallansci.com) at room temperature, dehydrated and embedded in wax, and sectioned at 5 μm. After deparaffinization following standard procedures, epitope unmasking was accomplished using 10 mM citrate buffer (pH 6.0) in an electric pressure cooker (Biocare Medical, Concord, CA, http://www.biocare.net) at 120°C and 15 psi for 4 minutes, and then cooled for 20–30 minutes. The sections were rinsed three times with double distilled water, followed by 5 minutes in 3% hydrogen peroxide at room temperature to block the endogenous peroxidase. Endogenous biotin was blocked when applicable using the avidin/biotin blocking kit (Vector Laboratories, Inc., Burlingame, CA, http://www.vectorlabs.com). Nonspecific antibody binding was blocked using a combination of 2% normal mouse serum and 10% normal goat serum in phosphate-buffered saline (PBS) for 30 minutes. For mouse monoclonal antibodies, we used the Dako ARK kit (Dako North America, Carpinteria, CA, http://www.dakousa.com; code number K3954) for biotinylating the antibodies and subsequent streptavidin-HRP incubation. For rabbit antiserum we used Dako Envision+ labeled polymer (HRP) and HRP-anti-rabbit antibody (Dako). For rat anti-mouse serum, biotin-conjugated goat anti-rat secondary antibody was used, followed by incubation with streptavidin-HRP. Finally, the HRP was visualized using AEC+ substrate-Chromogen (Dako) and counterstained with hematoxylin. Immunofluorescent staining was performed by incubating with fluorophore-conjugated secondary antibodies (Invitrogen, Carlsbad, CA, http://www.invitrogen.com) for 1 hour at room temperature, after primary antibody incubation, and mounted with 4′,6-diamidino-2-phenylindole (DAPI) blue fluorescent counterstain (InnoGenex, San Ramon, CA, http://www.innogenex.com; CS-2010-06). Images were taken with a fluorescent microscope (Carl Zeiss Inc., Thornwood, NY; http://www.zeiss.com). The antibodies used in this study were mouse AE13 and AE15 monoclonal antibodies (gifts from Dr. T.T. Sun, New York University Medical Center), mouse monoclonal anti-K-14 (Novocastra Laboratories Ltd, Newcastle, U.K., http://www.novocastra.co.uk; NCL-LL002), monoclonal anti-K15 (Biocare Medical), monoclonal anti-K10 (Sigma-Aldrich), anti-K5 serum (Covance, Princeton, NJ, http://www.covance.com; LN#14430002); anti-β-catenin serum (Sigma-Aldrich, product number C 2206), anti-Tcf3 monoclonal antibodies and an anti-active form of β-catenin (ABC) (Upstate Biotechnology, Lake Placid, NY, http://www.upstate.com; 05-602 and 05-665); anti-Ki67 serum (Dako, code number M7249); and anti-BMPR1A serum (a gift from Dr. P. Dijke, Leiden University Medical Center, Leiden, The Netherlands). Antibodies of PTEN (26H9), phospho-PTEN (Ser380/Thr382/383), Akt (5G3), phospho-Akt (Ser473), GSK-3β, phospho-GSK-3β (Ser9), and phospho-Smad1/5/8 were from Cell Signaling Technology Inc. (Danvers, MA, http://www.cellcignal.com) We would like to point out that phospho-GSK-3β is readily detected in the precortex cells (perhaps due to high levels of Wnt signaling), but it takes much longer for substrate reaction in the immunohistochemical assay to occur in the bulge region.
Bromodeoxyuridine Long-Term Labeling and Pulse Labeling
To study stem cells by bromodeoxyuridine (BrdU) long-term retention, pups were subcutaneously injected with BrdU (10 μg/kg of body weight) twice a day for 7 days starting from the first day after birth. On days 2, 4, and 6, pups were also injected i.p. with poly(I:C) (250 μg per pup) to induce Cre-mediated DNA recombination. Skin was collected on day 80 after BrdU labeling. Skins were processed as described above and sectioned at 4 μm. BrdU in situ staining was performed using a BrdU staining kit (Invitrogen) following the manufacturer's instructions.
Analyses of the Z/EG Reporter Mice and X-Gal Staining
Z/EG, a double reporter mouse , was used in this study to examine the Mx1-Cre-mediated loxp-dependent DNA recombination. Skins were collected on day 15 or day 40 post-poly(I:C) injection. For the green fluorescent protein (GFP) study, skin was fixed overnight in zinc formalin, washed with PBS, and immersed overnight in a solution of 30% sucrose dissolved in PBS at room temperature. The skin was then embedded in an OCT solution, snap-frozen, and sectioned at 8 μm. The slides were allowed to air dry for 1 hour at room temperature, were mounted with DAPI blue fluorescent counterstain (InnoGenex, CS-2010-06), and were then ready for imaging.
X-gal staining was performed as described previously . Briefly, unfixed skin was embedded in OCT and snap-frozen, then sectioned at 8 μm and allowed to air dry for 1 hour at room temperature. The sections were placed in fixative solution (1% formaldehyde, 0.2% glutaraldehyde, 2 mM MgCl2, 5 mM EGTA, and 0.02% Nonidet P40 in PBS) for 5 minutes, washed with PBS three times, and stained with X-gal staining solution (5 mM K3Fe(CN)6, 5 mM K4Fe(CN)6 · 3H2O; 2 mM MgCl, 0.01% sodium deoxycholate, 0.02% Nonidet P40, and 1% X-gal in PBS) overnight at 4°C, protected from light. On the second day, sections were washed three times with PBS and counterstained with Nuclear Fast Red.
Dynamic Expression of Noggin in the Bulge Results in Phase-Specific Alteration of BMP Activity During the Hair Follicle Cycle
To better understand the role of BMP signaling in postnatal HF regeneration, we first examined Noggin and Bmp4 expression patterns in the HF by X-gal staining of skin sections (Fig. 1A, 1C, 1F–1H) derived from BMP4-LacZ  and Noggin-LacZ  knock-in mice. In these mice, the expression of LacZ reflects the expression patterns of BMP4 or Noggin. Bmpr1a expression was examined by anti-BMPR1A immunochemical staining (Fig. 1E). Consistent with previous reports [20, 21], we detected constitutive BMP4 expression in the DP mesenchymal cells, cells in the lower part of the outer root sheath, IRS cells, and precortex cells (Fig. 1H), constitutive Bmpr1a expression in all the HF epithelial cells including EP-SCs (Fig. 1E and data not shown), and variable Noggin expression in DP mesenchymal and dermal sheath cells, the highest level being in the full anagen phase (Fig. 1C, 1F). Thus, in full anagen phase, hair matrix (HM) cells adjacent to the DP and the dermal sheath are highly proliferative, as shown by Ki67-positive staining (Fig. 1I), which correlates with low BMP activity as measured by the absence of p-Smad1/5/8, downstream effectors of the BMP pathway (Fig. 1I, 1J). However, high BMP activity, shown by intensive p-Smad1/5/8 staining, was detected in the precortex region (Fig. 1I, 1J). In response to Wnt signaling, β-catenin accumulates in the cytoplasm and then translocates into the nucleus of cells to form a complex with either Tcf or Lef1, regulating downstream gene expression [10, 43]. This can be monitored by β-catenin nuclear staining. Consistent with previous reports that signaling by both BMP and Wnt is required for inducing precortex cells to differentiate [15, 44], p-Smad1/5/8-positive cells were found to accumulate β-catenin in the nucleus (Fig. 1K).
By systematically analyzing the bulge area where EP-SCs reside, we found that BMP4 is constitutively expressed in mesenchymal cells surrounding the bulge epithelial cells (Fig. 1G). We also found that Noggin is expressed by EP-SCs since its expression coincides with the EP-SC marker, K15 (Fig. 1B, 1B″, 1C, 1C″, 1D). Intriguingly, the Noggin expression level changes during the HF growth cycle, the highest level being in the early anagen phase (Fig. 1A–1D). This dynamic change in Noggin expression leads to a dynamic activation/inactivation of BMP signaling in K15-positive EP-SCs (Fig. 2B, 2E, 2I, 2C′) as shown by p-Smad1/5/8 staining (Fig. 2B, 2E, 2I). Active BMP signaling as measured by Smad1/5/8 correlates with inactivation of Wnt signaling as shown by loss of nuclear β-catenin in most of the bulge cells in the telogen phase, in which EP-SCs are in the quiescent state (2A–Fig. 2C). The inactivation of BMP signaling in the correlation of Noggin expression, as shown by negative staining of p-Smad1/5/8, is evident in the majority of cells in the bulge area (Fig. 2E, white arrowhead) in early anagen phase, in which EP-SCs are in the active state. However a few cells in bulge and/or sebaceous gland co-express both active (or nuclear) ABC and p-Smad1/5/8, reflecting a range of states in which a transition between BMP-dominant and Wnt-dominant signaling occurs (Fig. 2D–2F). The majority of cells in the bulge showing nuclear β-catenin are consistent with the proposed Wnt function in HF anagen induction . Thus, Noggin expression by EP-SCs overrides BMP activity and coincides with activation of Wnt/β-catenin signaling and the anagen phase of the HF growth cycle [20, 21].
Mx1-Cre-Mediated loxp-Dependent DNA Recombination in EP-SCs
To investigate the role of BMP signaling in EP-SC regulation and postnatal HF regeneration, we elected to block BMP signaling by inactivation of Bmpr1a, the only BMP receptor known to be expressed in the HF . Because null Bmpr1a mutation leads to embryonic lethality , we generated Bmpr1a conditional mutant mice by using the Cre-loxp system. The most common Cre mouse lines used in HF studies are K14 and K5; however, due to expression and an effect on other tissues, mice with K14-Cre-mediated Bmpr1a mutation do not survive for more than 3 weeks . This obviously impairs investigation of the long-term effects of the loss of BMP activity on HF regulation.
Recently, we found that the Mx1-Cre mouse, an interferon-inducible Cre line , is able to target hair follicle epithelial cells including epithelial stem/progenitor cells. To determine the expression pattern and efficiency of the Mx1-Cre-mediated loxP-dependent DNA excision in HFs after injection of poly(I:C), an interferon inducer, we crossed Mx1-Cre mice with Z/EG reporter mice . The hair growth cycle is time-dependent, with the first cycle initiated around birth and the second cycle initiated between 20 (P20) and 30 days (P30) after birth . We injected the pups with poly(I:C) on day 2 (P2) for one group and on P4 for a second group, to compare the efficiency of Mx1-Cre-induced recombination in HFs at these two times. In the P2-induced group, Cre-mediated DNA excision occurred at the stem cell level in most of the ventral HFs as evidenced by permanent GFP expression, not only in HFs but also in the overlying epidermis (Fig. 3A, 3B). Targeting of EP-SCs in P2-induced HFs was confirmed by costaining of GFP-positive cells with K15 during the first (Fig. 3E, 3F) and second (Fig. 3M, 3N) HF cycle. However, in the P4-induced group, Cre-mediated DNA recombination only occurred in some of the progenitor or HM precursor cells of ventral HFs (Fig. 3G–3J), leading to transient expression of GFP only in the first hair cycle (Fig. 3G), but the K15-positive stem cells were GFP-negative (Fig. 3K, 3L). In the second HF cycle, the new hairs, emanating from untargeted EP-SCs, retained LacZ expression (Fig. 3J).
Conditional Inactivation of Bmpr1a in EP-SCs Leads to Permanent Hair Loss and Tumor Formation
Based on the reporter mouse results, we generated a conditional knockout mouse model by crossing a Bmpr1a-loxP (Bmpr1afx) mouse line  with an Mx1-Cre mouse line  as described previously [29, 48]. Mating the Mx1-Cre+Bmpr1afx/+ line with the Bmpr1afx/fx line generated litters with homozygous Mx1-Cre+Bmpr1afx/fx (Bmpr1a mutant hereafter), heterozygous Mx1-Cre+Bmpr1afx/+, and wild-type control Mx1-Cre−Bmpr1afx/fx and Mx1-Cre−Bmpr1afx/+ pups. The pups were randomly divided into two groups. One group was injected with poly(I:C) on P2, P4, and P6 (P2-induced group); the other group was injected on P4, P6, and P8 (P4-induced group). Since Mx1-Cre+Bmpr1afx/+ heterozygotes did not develop any phenotypic changes, we used them as a control in some of our studies. Prior to P10, pups in both groups were grossly normal in development; however, after P11, the Bmpr1a mutant mice showed signs of growth retardation when compared to paired control mice. The most obvious phenotype was hair loss at 2 weeks after birth. In the P2-induced group, there was hair loss in both dorsal and ventral skin, but it was more severe in the ventral skin (Fig. 3O). This phenotypic difference in dorsal and ventral skin hair loss may be due to the time differences in initiation of the HF growth cycle in dorsal versus ventral skin, accounting for the difference in sensitivity to poly(I:C)-induced gene targeting. In the P4-induced group, ventral hair loss was obvious, but dorsal hair development appeared to be less affected (Fig. 3O and data not shown). Between P35 and P40, hair growth recovered in all mice, apart from the ventral skin in the P2-induced mice (Fig. 3P and data not shown). This finding was presumably due to secondary HF growth, the newly formed hair being regenerated from untargeted stem cells. Thus, permanent hair loss in the ventral skin of the P2-induced mice indicates that inactivation of Bmpr1a occurs at the stem cell level (Fig. 3O, 3P). This observation is consistent with the results from the reporter mouse assay (Fig. 3A–3F, 3M–3N). In general, the earlier poly(I:C) is injected, the more severe the hair loss. This may reflect activation of the EP-SCs during the early anagen phase.
Around day P90, the first visible abnormality associated with HFs is the appearance of black speckles under the ventral skin in the P2-induced Bmpr1a mutant mice (Fig. 4A). These speckles were due to an accumulation of melanin deposits, together with incompletely differentiated keratinocyte in disorganized HFs (Fig. 4C, 4H). A small fraction of the disorganized HFs progressed to visible, solid tumors (diameter >1 cm) after 6 months (Fig. 4E).
To investigate further the hair defects in Bmpr1a mutant mice, we examined the morphologic changes in the hair follicles by analyzing hematoxylin and eosin-stained sections. In the P4-induced Bmpr1a mutant mice, wavy and misaligned HF structures and lack of hair shaft formation were apparent between P15 and P25 compared to the HFs of the control mice (data not shown); this disappeared after day P30. This is consistent with the observed temporary interruption of HF growth due to targeting of Bmpr1a in progenitor rather than stem cells. The stem cells with untargeted Bmpr1a then generate completely new HFs in the following HF cycle.
In P2-induced mutant mice, the permanent segment of the HF structure appeared grossly normal. However, the cycling segment of the HF structure was impaired, with an abnormal morphology and the absence of a hair shaft (data not shown). Unlike the P4-induced mice, in P2-induced mice, the normal HF structure did not recover. In fact, when analyzed on P90, many bulb-like structures were found to emerge from a large, central cyst (Fig. 4B–4D). This abnormal structure had a morphology typical of human matricomas—a hamartoma —with multiple abortive HFs opening into a central cyst (Fig. 4B–4C), reflecting ongoing abnormal de novo HF morphogenesis (occurred in 100% of the 15 P2-induced mice). The number of de novo HF bulb-like structures increased as the mice aged. On P180, solid tumors were seen with large numbers of de novo HF bulbs (Fig. 4F–4G), and accumulated melanin deposits were found within the cysts of some tumors (Fig. 4C, 4H).
Mx1Cre+Bmpr1afx/fxZ/EG triple genotypic mice were used to detect the Cre-mediated DNA recombination. These mice were injected with poly(I:C) on P2, and tissue was collected from animals sacrificed on day P90. The absence of X-gal staining in tumor cells and the overlying epidermis demonstrates that Cre-mediated DNA recombination occurred in the precursors of these tissues. However, adjacent normal HFs and the overlying epidermis remained positive for X-gal staining (Fig. 4D). This finding suggested that the tumor arose clonally from a Bmpr1a mutant EP-SC.
BMP Signaling Restricts EP-SC Activation and Expansion
Using K15 immunochemical staining, we studied the EP-SCs in the HF. As expected, K15 predominately stained cells within the HF-bulge region in normal hair follicles (Fig. 5A–5C). To investigate the role of BMP signaling on EP-SC regulation, we compared the number of K15-positive cells in HFs taken from skin at different phases, originating from normal and BMP4 transgenic (BMP4-tg) mice (Fig. 5D), and skin containing tumorous HFs in Bmpr1a mutant mice (Fig. 5E). In normal HFs, the number of K15-positive cells per HF section in the early anagen phase was doubled compared to that in the telogen- and late anagen-phase HFs (Fig. 5A–5C, 5F). There was no substantial difference in the number of K15-positive cells in the telogen- and late anagen-phase HFs (Fig. 5A, 5C, 5F). These data suggest that the EP-SC cell number is expanded during the early anagen phase but is relatively constant in the rest of the HF cycle. The number of K15-positive cells was dramatically increased in the Bmpr1a mutant mouse hair follicles (greater than 12 times that of normal telogen HFs in the P90 tumorous HFs). K15 staining occurred in cells located in the cyst boundary and in cells migrating from the cyst boundary toward the tumorous HFs. However, the EP-SC number was not significantly changed in BMP4-tg mouse HFs. These data indicate that BMP signaling restricts EP-SC activation and expansion, and represents an important function in EP-SC maintenance. Blocking BMP signaling results in uncontrolled overproduction of EP-SCs, which contributes to tumor formation.
BMP Signaling Inhibits β-Catenin Activity in EP-SCs
Active Wnt signaling plays an important role in HF development in normal skin , which can be monitored by β-catenin nuclear staining [10, 43]. In the bulge region of normal HFs, nuclear staining of β-catenin is detected only during the early anagen phase when EP-SCs are activated and migrate downward to generate new HFs (Fig. 2D, 6D) . Overexpression of the activated form of β-catenin in the β-catenin-tg mice leads to overproduction of HFs and eventual formation of HF tumors , a phenotype similar to that seen in the Bmpr1a mutant mice. We therefore investigated β-catenin localization in our mutant animals.
In Bmpr1a mutant tumors, increased nuclear staining of β-catenin is observed in the epithelial cells around the cyst boundary and the cells migrating from the cyst boundary toward the tumorous HFs (Fig. 6H). These epithelial cells, which accumulate nuclear β-catenin, coexpress K15, an EP-SC marker [11, 50, 51] (Fig. 6J–6K). These observations support our interpretation that the number of abnormal HF stem/progenitor cells, which are located around the cyst boundary and display nuclear-localized β-catenin, is substantially increased in the Bmpr1a mutant skin. In contrast, nuclear staining of β-catenin is barely detected in any HFs in the BMP4-tg mouse (Fig. 6L). Therefore, BMP signaling appears to inhibit nuclear localization of β-catenin and its underlying activity, thus affecting the activation and expansion of HF stem/progenitor cells.
BMP Signaling Regulates β-Catenin Activity Inducing EP-SC Activation Partially Through the PTEN-Akt Cascade
To investigate the molecular mechanism by which BMP signaling inhibits the activity of β-catenin, we explored other signals which might coordinate with Wnt signaling to stabilize β-catenin, contributing to the translocation of β-catenin from the cytoplasm into the nuclei of arrested EP-SCs.
Because PTEN, through suppression of Akt, is able to inhibit nuclear accumulation of β-catenin  and is also subject to BMP regulation , we examined whether the PTEN-Akt pathway is involved in EP-SC regulation. We first determined the distribution patterns of the active and inactive forms of the components of this pathway. In the telogen phase of the normal HF, no phosphorylated PTEN (p-PTEN; the inactive form of PTEN) or p-Akt (active form of Akt) staining was detected (data not shown), although a high level of unphosphorylated PTEN was present (data not shown). However, in the early anagen phase, the reduced level of BMP signaling is reflected by reduced p-Smad1/5/8 in the bulge area (Fig. 2 E), and this parallels detection of p-PTEN, p-Akt, and p-GSK3β in the bulge area, where they all have a similar distribution pattern (Fig. 6A–6C). Intriguingly, activation of Akt was detected in stem cells beginning to migrate from the bulge, as evidenced by its presence in the long-term BrdU-retaining cells (Fig. 6B). The retained BrdU staining diminished as the activated stem cells divided and migrated downward (Fig. 6B). This distribution pattern of p-Akt in the downward-migrating cells is strikingly similar to that of β-catenin, which localizes in the nuclei (together with Tcf4) of the downward-migrating hair follicle stem/progenitor cells (Fig. 6D). These results suggest that BMP signaling may regulate β-catenin nuclear localization in the EP-SCs through the PTEN-PI3K-Akt pathway.
To further test this hypothesis we examined the HFs of the Bmpr1a mutant mice. Consistent with our previous results, p-PTEN, p-Akt, and the inactivated form of GSK3β (p-GSK3β) were all detected in multiple cells located in the cyst boundary and in downward-migrating cells of the Bmpr1a mutant HFs (Fig. 6E–6G). The HF-stem/progenitor cells, as evidenced by their K15 expression, also showed nuclear localization of β-catenin (Fig. 6H–6K). In contrast, in the BMP4-tg mouse HFs, there was no detectable p-PTEN (data not shown), but there was a relatively high level of PTEN in the epithelium and HFs (data not shown). Cytoplasmically localized β-catenin was also seen in the epithelial cells and cells surrounding the cyst boundary in the BMP4-tg mice (Fig. 4L). These results further support our hypothesis that regulation of β-catenin activity by BMP signaling is at least partially through the cascade of PTEN-Akt-GSK3β.
The data above are primarily correlative in nature. To determine whether activated BMP signaling affects β-catenin activity as predicted, we introduced the β-catenin-responsive Top-flash reporter construct into undifferentiated primary keratinocytes . We found that Noggin or Wnt3A-containing medium alone enhanced Top-flash reporter activity 3- and 1.5-fold, respectively. Noggin plus Wnt3A-containing medium synergistically enhanced Top-flash activity 5.1-fold, and the addition of Ly294002 (which inhibits PI3K) significantly reduced Noggin-enhanced Top-flash activity. In contrast, BMP2 antagonized Wnt3A-containing medium and reduced Top-flash activity (Fig. 7A). These results are in agreement with a previous report that Noggin with Wnt synergistically enhanced Top-flash activity . Furthermore, we found that Top-flash activity could be induced by cotransfection with the dominant negative form of Bmpr1a (Bmpr1a DN), the phosphorylated mimic of PTEN (PTEN-D3), and the constitutively active form of Akt, but was inhibited by the constitutively active form of Bmpr1a (Bmpr1a-Ca) and wild-type PTEN (PTEN wt) (Fig. 7B). Western blot analysis confirmed that p-PTEN, p-Akt, and p-GSK3β levels were increased in undifferentiated primary keratinocytes following Noggin treatment (Fig. 6M) and in Bmpr1a mutant skin (Fig. 6N). These results support the idea that BMP signaling through BMPR1A inhibits β-catenin activity, which is mediated through the PTEN-PI3K-Akt pathway. Noggin is required as the secondary signal to coordinate with Wnt to induce nuclear localization of β-catenin in EP-SCs through overriding the BMP inhibition.
Taken together, we conclude that BMP signaling plays an essential role in controlling activation and expansion of EP-SCs. We propose that this is achieved, at least in part, through the PTEN-PI3K-Akt cascade to antagonize the positive control of β-catenin activity imposed by Wnt signaling.
We elucidated the roles of BMP signaling in postnatal HF regeneration and EP-SC regulation by inducible inactivation of Bmpr1a. We have shown that BMP signaling correlates with inhibition of the β-catenin activity in the bulge region where EP-SCs reside, and that this is in part regulated through the PTEN-Akt-GSK3β cascade and contributes to the control of activation and expansion of EP-SCs. Thus, crosstalk between the BMP and Wnt pathways ensures appropriate activation of EP-SCs during induction of the anagen phase for regeneration of a new HF, but restricts the EP-SC activation in other phases of the HF cycle. Inactivation of Bmpr1a in EP-SCs releases the restriction of BMP signaling and results in expansion of the EP-SC population which, together with abnormal proliferation and differentiation of the mutant HM cells (data not shown), and possibly increased cell survival, contributes to tumor formation.
Expression of Noggin by EP-SCs and HF Anagen Induction
Noggin, generated from DP, is known to be involved in HF anagen induction through blocking BMP signaling during the early anagen phase [19, –21]. In this study, we have a novel finding that expression of Noggin also occurs in EP-SCs, with the highest level detected in the early anagen phase and the lowest level in the telogen phase. Expression of Noggin by EP-SCs is simultaneous with or even prior to the expression of Noggin in DP during the early anagen phase (Fig. 1A–1C). The highest Noggin expression in DP mesenchymal cells occurs in the mid to late anagen phase, adjacent to the proliferative HM cells. This finding suggests that Noggin expression in EP-SCs and DP cells functions together with other DP signals to play a role in anagen phase induction, whereas Noggin expression in DP cells also plays a role in HM cell proliferation by overriding the BMP inhibition signal. Indeed, during the early anagen phase, expression of Noggin by EP-SCs is accompanied by EP-SC activation, expansion, and downward migration.
Because Wnt10a and b are expressed in postnatal hair follicles only at the onset of the anagen phase , and activation of β-catenin induces initiation of the anagen phase in HFs and new HF germ formation [10, 45, 54] in a dosage-dependent manner , Wnt signaling is considered to be an anagen induction signal. However, transgenic overexpression of Wnt3 only affects HF cell differentiation and produces a short-hair phenotype , which does not fully reflect the phenotype of transgenic overexpression of the active form of β-catenin . This suggests that a secondary signal, in addition to Wnt, is required to fully activate β-catenin. Supporting this interpretation, we observed that de novo formation of abnormal HFs in the HF tumors of our Bmpr1a mutant mouse correlates with the hair growth cycle. Our data from both intestinal stem cell  and EP-SC (this study) studies suggest that BMPs function to restrict the signal to the stem cell for activation and expansion. Blocking the BMP signal is required to coordinate with Wnt signaling to fully activate β-catenin. This is achieved by a crosstalk of these two pathways through the Pten/Akt cascade.
Germline mutation of Pten leads to Cowden disease, characterized by hamartomatous in multiple tissues, especially multiple facial trichilemmomas (a HF tumor). The fact that this Cowden-like syndrome has also been found to be caused by mutation in the Bmpr1a gene supports our model that Bmpr1a and Pten are functionally related . However, Pten-specific deficiency in keratinocytes by K5-Cre leads to epidermal hyperplasia and spontaneous papilloma and cutaneous squamous cell carcinomas in the absence of HF tumors . Thus, the role of Pten in EP-SCs and HF tumor formation needs further study.
More recently, studies of constitutively activated β-catenin transgenic mice suggest that a secondary signal from DP combined with activated β-catenin is required for new HF induction . Studies from hematopoietic stem cells (HSCs) have demonstrated that activation of β-catenin, either by transfection of the active form of β-catenin or by Wnt treatment, can stimulate HSC expansion. However, transgenic expression of Bcl-2 (an anti-apoptotic factor) appears to be required in these experiments [58, 59]. Likewise, activation of the PI3K/Akt pathway, which inhibits Bad and Caspase 9, plays an important role in cell survival [60, –62]. It is worthwhile to point out that blocking BMP signaling by either inactivation of Bmpr1a or expression of Noggin can induce the activation of Akt, which may also provide a survival signal during EP-SC activation and expansion.
EP-SCs and HF Tumor Stem Cells
Recent studies in other systems have suggested the existence of tumor stem cells or cancer stem cells, a rare population of cells that plays a role in initiation of tumorigenesis and is also responsible for tumor relapse. Like normal stem cells, tumor stem cells have the ability to self-renew and the potential to be highly proliferative. Tumor stem cells may originate from normal stem cells in a given tissue and retain their self-renewal ability, but are abnormal in their proliferation, differentiation, or survival. However, tumor stem cells may also originate from progenitors which later regain their self-renewal capacity. Identification of tumor stem cells in a given tumor is important for understanding tumorigenesis and designing clinical therapeutic strategies [63, 64].
Our inducible gene target system provides an alternative model for tumor stem cell studies. In this system, we can control the time for induction of mutation of target genes. By parallel studies of Bmpr1a mutant mice and Z/EG reporter mice, we found that P2 induction can mediate gene mutation targeting on EP-SCs, which results in prominent hair loss and tumor formation, whereas P4 induction only targets the progenitors or precursors, leading to temporary hair loss and HF morphology disruption. In addition to EP-SCs, melanocyte stem cells, which produce melanocytes, are also located in the hair follicle bulge [65, –67]. This raised a possibility that the HF tumor seen in Bmpr1a mutant mice may be attributed to targeting Bmpr1a in melanocyte stem cells instead of EP-SCs. However, we have observed that Col3.6-Cre-mediated Bmpr1a deletion in bulge cells only leads to hair color loss, reflecting the idea that targeting Bmpr1a in bulge may affect melanocyte stem/progenitor cells without tumor formation (supplemental Figure). This result strongly argues that the HF tumor seen in Bmpr1a mutant mice is due to targeting Bmpr1a in EP-SCs. In addition, a severe hair loss phenotype observed in Emx1-Cre-mediated Bmpr1a mutant mice was due to the disruption of hair matrix cell differentiation; the mutation may not have been targeted on the EP-SCs. The lack of HF tumor formation in this mutant mouse, even at 10 months of age , supports our conclusion that tumors only arise from EP-SCs that harbor a mutation of Bmpr1a. In a normal situation, BMP signaling through BMPR1A functions as an inhibitory signal that restricts EP-SC activation and expansion. Bmpr1a deficiency in EP-SCs results in tumor stem cells that are no longer subject to BMP-driven inhibition, so that the stem cells remain in an active state. The unlimited expansion of tumor stem cells is the initiator and continuing source of tumor development.
Differential Crosstalk Between the BMP and Wnt Pathways in the Stem Cell and Differentiation Compartments
The communication between BMP signaling and Wnt signaling might be cell context-dependent. In the bulge area, BMP signaling inhibits the Wnt/β-catenin signal partially through the PTEN-Akt pathway and may also be partially through Smad-mediated signaling, thus controlling activation/arresting of stem cells (Fig. 7). Because expression of BMP is constant, dynamic expression of Noggin plays a role in coordination with Wnt signaling to fully activate β-catenin by temporarily overriding BMP signaling, leading to initiation of a new HF growth cycle . Although we favor this hypothesis, the regulation of β-catenin and its relationship to Wnt/Akt signaling has been explored in a number of contexts [33, 68–70, , ]. Recent evidence suggests that Akt may directly phosphorylate β-catenin, resulting in stabilization and activation of β-catenin by 14-3-3ζ . Thus, both activation of Wnt signaling and inhibition of BMP signaling are required for HF anagen-phase induction. Consistent with this hypothesis, coordination between Noggin and Wnt to activate stem cells has also been described previously in both the HF and intestinal system [29, 52].
During hair shaft cell differentiation, however, BMP signaling plays a synergistic role with Wnt signaling in favoring cell differentiation. This is supported by the coexistence of β-catenin and p-Smad1/5/8 in the cell nuclei located in the precortex region (Fig. 1K). Wnt/β-catenin signaling is required for hair shaft differentiation through regulation of hair shaft-specific gene expression . Recently, several groups have shown that target deletion of Bmpr1a will lead to loss of hair due to disruption in HF differentiation [22, , –25]. Blocking BMP signaling will affect Lef1 expression, which is indispensable for hair shaft cell differentiation . Consistent with these findings, in our Bmpr1a mutant mice, we observed not only expansion of EP-SCs associated with activation of the p-PTEN-pAkt-β-catenin pathway, but also impaired hair shaft cell differentiation evidenced by the lack of AE13 (hair shaft cell marker) and CK10 (mature keratinocyte marker) expression (data not shown). These observations indicate that antagonistic (in EP-SCs) and synergistic (in precortex progenitor cells) interactions between the BMP and Wnt pathways play essential roles in control of stem cell activation and regulation of proper HF differentiation.
The authors indicate no potential conflicts of interest.
We appreciate Dr. O. Tawfik for pathological consultation. We thank Dr. P. Dijke for providing anti-BMPR1A anti-serum, Dr. T.T. Sun for AE13 and AE15 antibodies, Drs. A. McMahon and B. Hogan for providing NogginLacZ and BMP4LacZ mice, and Dr. C.G. Lobe for Z/EG reporter mice. We thank Dr. R. Krumlauf for scientific discussion and Drs. O. Pourquié and R. Kopan for critically reviewing the manuscript. We are grateful to D. di Natale, D. Stenger, and J. Haug for assistance on manuscript editing. We thank J. Ross for plasmid preparation, D. Stark for imaging assistance, and H. Marshall and K. Porter for technology assistance. This work is supported by the Stowers Institute for Medical Research. J.Z. is currently affiliated with the Department of Pathology, Loyola University Medical Center, Maywood, IL.