Evidence That the Satin Hair Mutant Gene Foxq1 Is among Multiple and Functionally Diverse Regulatory Targets for Hoxc13 during Hair Follicle Differentiation*

It is increasingly evident that the molecular mechanisms underlying hair follicle differentiation and cycling recapitulate principles of embryonic patterning and organ regeneration. Here we used Hoxc13-overexpressing transgenic mice (also known as GC13 mice), known to develop severe hair growth defects and alopecia, as a tool for defining pathways of hair follicle differentiation. Gene array analysis performed with RNA from postnatal skin revealed differential expression of distinct subsets of genes specific for cells of the three major hair shaft compartments (cuticle, cortex, and medulla) and their precursors. This finding correlates well with the structural defects observed in each of these compartments and implicates Hoxc13 in diverse pathways of hair follicle differentiation. The group of medulla-specific genes was particularly intriguing because this included the developmentally regulated transcription factor-encoding gene Foxq1 that is altered in the medulladefective satin mouse hair mutant. We provide evidence that Foxq1 is a downstream target for Hoxc13 based on DNA binding studies as well as co-transfection and chromatin immunoprecipitation assays. Expression of additional medulla-specific genes down-regulated upon overexpression of Hoxc13 requires functional Foxq1 as their expression is ablated in hair follicles of satin mice. Combined, these results demonstrate that Hoxc13 and Foxq1 control medulla differentiation through a common regulatory pathway. The apparent regulatory interactions between members of the mammalian Hox and Fox gene families shown here may establish a paradigm for “cross-talk” between these two conserved regulatory gene families in different developmental contexts including embryonic patterning as well as organ development and renewal.

It is increasingly evident that the molecular mechanisms underlying hair follicle differentiation and cycling recapitulate principles of embryonic patterning and organ regeneration. Here we used Hoxc13-overexpressing transgenic mice (also known as GC13 mice), known to develop severe hair growth defects and alopecia, as a tool for defining pathways of hair follicle differentiation. Gene array analysis performed with RNA from postnatal skin revealed differential expression of distinct subsets of genes specific for cells of the three major hair shaft compartments (cuticle, cortex, and medulla) and their precursors. This finding correlates well with the structural defects observed in each of these compartments and implicates Hoxc13 in diverse pathways of hair follicle differentiation. The group of medulla-specific genes was particularly intriguing because this included the developmentally regulated transcription factor-encoding gene Foxq1 that is altered in the medulladefective satin mouse hair mutant. We provide evidence that Foxq1 is a downstream target for Hoxc13 based on DNA binding studies as well as co-transfection and chromatin immunoprecipitation assays. Expression of additional medulla-specific genes down-regulated upon overexpression of Hoxc13 requires functional Foxq1 as their expression is ablated in hair follicles of satin mice. Combined, these results demonstrate that Hoxc13 and Foxq1 control medulla differentiation through a common regulatory pathway. The apparent regulatory interactions between members of the mammalian Hox and Fox gene families shown here may establish a paradigm for "cross-talk" between these two conserved regulatory gene families in different developmental contexts including embryonic patterning as well as organ development and renewal.
Recent progress in defining the molecular framework underlying normal and pathological development of skin and hair increasingly implicates the Hox family of transcriptional controller genes as key regulatory molecules (1). The hair follicle is of particular interest because both morphogenesis and cyclical renewal of this complex miniature organ employ many of the same genetic control mechanisms as required for embryonic patterning (2,3). Follicular morphogenesis is initiated during embryonic development and progresses in successive rostrocaudal waves across the epithelium (4), thus resulting in a fixed number of follicles at various stages of differentiation at birth (5). Furthermore, hair follicles and fibers display a great degree of regionally specified morphological heterogeneity defining hair follicle type (6), and there exists additional variation among follicles of the same type so that probably no two follicles of an individual are identical (7). In keeping with the role of the Hox gene system to define regional identities along the anteriorposterior axis during axial and paraxial patterning including limb development (8 -10), it has been proposed that this system might be employed to establish territorial specificity of the skin and its appendages (11). This concept is supported by the observation of distinct anterior-posterior expression domains for various Hox genes in developing skin as well as hair and feather follicles (11)(12)(13)(14)(15)(16). In parallel to this regionally restricted expression, certain members of the Hox family apparently are distinctly expressed in all follicles, as it applies to Hoxc13 (17) and probably several other members of this gene family (1). This universal follicular expression suggests essential functional roles during follicle morphogenesis and/or cycling. Support for this concept was provided by the lack of hair in Hoxc13 gene-targeted mice (17) and the delayed hair growth and subsequent loss of hair in Hoxc13-overexpressing transgenic mice (Tg(Hoxc13)61B1Awg, MGI: 3574566, also referred to as GC13 mice (18)), essentially affecting all body regions in both cases.
In the progressive growth phase, anagen, the Hoxc13 expression pattern in the hair follicle bulb, originates in a conspicuous region of the matrix above the proliferative cell compartment (see Fig. 1 (18)). This expression domain includes a portion of the outer root sheath (ORS) 3 layer lining the upper dermal papilla (see Figs. 1 and 6A). This germinative layer is suggested to harbor stem cells that give rise to the diverse lineages differentiating into the individual hair follicle compartments (19), a proposition that has recently been supported by clonal labeling during hair follicle growth (20). This finding combined with the distinctly layered expression patterns of specific hair keratins in the matrix of human hair follicles suggest a high level of organization in the matrix where the compartment-specific lineages originate at distinct proximo-distal levels (19,20). According to this model, Hoxc13 expression in the upper internal ORS is consistent with a potential role in determining the fate of presumptive progenitors to differentiate along medulla and cortex-specific pathways (see Fig. 1). Furthermore, the continued expression of both mouse Hoxc13 and its human ortholog in differentiating cells of the medulla, cortex, cuticle of the hair shaft, and inner root sheath (IRS), and the companion layer overlaps well with the compartment-specific patterns of various hair keratin genes and suggests a role in their regulation (17)(18)(19). This idea is supported by in vitro DNA binding data showing sequencespecific interaction of human HOXC13 with cognate binding sites present in the promoter regions of human cortex-specific hair keratin genes hHa2 and hHa5 (21) as well as in the mouse Krtap16-5 promoter (22).
Currently it is not clear whether Hoxc13 acts primarily as a downstream effector of other regulatory molecules in the presumptive transcriptional control of these hair keratin genes or whether it might act also further upstream, i.e. whether it may influence the expression of other regulatory genes. The remarkable overlap of the Hoxc13 expression pattern with matrix-specific activity domains of several signaling molecules and transcription factors, including Wnt5a, Catnb, Bmp4, Fgf5, Tcf/Lef1, Foxn1, Foxq1, and Msx2 (for review, see Ref. 23) supports the idea of Hoxc13 being engaged in both mechanisms.
To address this question we determined the differential gene expression profile in postnatal skin of GC13 mice by DNA microarray analysis. 4 The results identified genes of different functional categories as potential downstream targets for Hoxc13 in distinct follicular compartments. Genes specifically expressed in the medulla formed the largest subgroup, including Foxq1 as the most conspicuous member of this group. Foxq1 sa is the mutated allele underlying the satin hair phenotype (24), a mouse mutation with a defect in the hair follicle medulla. Our data demonstrate that Foxq1 is a downstream target for Hoxc13 during medulla differentiation, thus demonstrating a functional link in mammals between two conserved families of regulatory genes essential for embryonic patterning and organ development.

MATERIALS AND METHODS
Transmission and Scanning Electron Microscopy (TEM and SEM)-Dorsal skin from two severely affected GC13 transgenic mice and from two age-and gender-matched FVB/NTac controls was removed after euthanasia by CO 2 asphyxiation using Institutional Animal Care and Use Committee-approved methods and cut into 1-mm strips in rostro-caudal orientation. Samples were fixed overnight in 2% glutaraldehyde, 2% paraformaldehyde in 0.1 M phosphate-buffered saline (PBS), pH 7.2, at 4°C and transferred into PBS for overnight shipment. Upon receipt tissues were post-fixed overnight in 1% osmium tetroxide, PBS and processed for embedding, sectioning and TEM analysis as described in detail (25). For SEM, parallel skin samples of about 1 cm 2 taken from the same animals were processed and analyzed essentially as described (25).
Gene Array Analysis-Total RNA (10 g) from post-partum day (PPD) 5 normal and GC13 skin of the trunk was converted into biotinylated, fragmented cRNA according to the Affymetrix protocol. Purification of biotinylated cRNA, hybridization in triplicate to MG-U74Av2 GeneChips (Affymetrix, San Diego, CA) containing probes for interrogating ϳ12,500 presumptive genes, and normalization of hybridization data were performed as previously described (26). Comparative expression analysis was conducted using the DNA Chip Analyzer tool (dChip (27,28)). Differential expression was scored by the following inclusion criteria; (i) genes had to be identified as "present" by GeneChip MAS 5.0 software in at least 2 of the 3 replicate arrays, (ii) p Յ 0.01 for the t test (unpaired, 2-tailed, assuming unequal variance), and (iii) changes in average intensity values of Ն2-fold. False discovery rate, estimated as the median number of genes discovered by iterative comparison of randomized sample groupings (29,30), approximated 0.00 for these criteria. Significant gene ontologies ( p Ͻ 0.005) were identified by dChip, which derives a probability score ( p value) based on hypergeometric distribution of ontology terms (28).
Real-time Quantitative (Q)-PCR-Total RNA from PPD 5 normal and GC13 skin was isolated by a CsCl step gradient, and polyadenylated RNA was extracted using the GenElute mRNA Miniprep Kit (Sigma). Complete cDNA was synthesized using a Marathon cDNA Amplification Kit (BD Biosciences Clontech, Palo Alto, CA). For standardization, cDNA samples were quantified using a PicoGreen dsDNA quantitation kit (Molecular Probes, Eugene, OR). Q-PCR was performed on an ABI PRISM 7000 sequence detection system using ABI Sybr Green PCR mixture and 20 M primers exactly as described by the manufacturer. Standard primers for ␤ 2 -microglobulin were used as controls and for normalization. Forward (F-) and reverse (R-) PCR primers specific for the genes tested (Table 1)  In Situ Hybridization-Freshly dissected scapular skin from euthanized (CO 2 asphyxiation) GC13 and FVB/NTac mice at PPD 5 was fixed in 4% paraformaldehyde, PBS at 4°C overnight and cryo-preserved by overnight incubation in 30% sucrose, PBS before embedding into Tissue-Tek OCT compound and storage at Ϫ80°C. Probes specific for genes tested (Table 1) were generated by PCR using total cDNA derived from FVB/ NTac mice at PPD 5 and the following forward (F-) and reverse PCR products were cloned into pCRII-TOPO vector (Invitrogen) for the generation of digoxigenin (Roche Applied Science)-labeled antisense and sense (control) RNA probes as described (22). Hybridization with 10-m cryosections and colorimetric signal detection using alkaline phosphatase-conjugated anti-digoxigenin antibodies (Roche Applied Science) was performed as previously described (22,31).
Electrophoretic Mobility Shift Assays (EMSAs)-Synthetic oligonucleotides (oligos) containing the Hoxc13 consensus binding sequence 5Ј-TT(A/T)ATNPurPur-3Ј (21) had GGG overhangs at their 5Ј ends to facilitate labeling with [ 32 P]dCTP using Klenow fragment. Sequences of sense strand oligos corresponding to the three most proximal putative binding sites upstream of Foxq1 (see Fig. 7) that started at nucleotide positions Ϫ2149 and Ϫ2188 on the non-coding strand and at position Ϫ2411 on the coding strand with respect to the Foxq1 translational start site (32) were 5Ј-(GGG)TAAGGA-CACCTTCATTAACAACAATGTCTTT-3Ј (oligo 2149), 5Ј-(GGG)AATGTATCTCCTCCATAAAACAGGCACGC-3Ј (oligo 2188), and 5Ј-(GGG)CAACAGTCAGCTTAATAGG-GACGAGGAATA-3Ј (oligo 2411); a mutated version of oligo 2411 (oligo 2411(M)) contained a change in the consensus binding sequence (5Ј-TTAATAGG-3Ј 3 5Ј-TGCCGAGG-3Ј). Annealed and labeled oligos were incubated with Hoxc13 protein synthesized using the TNT T7 coupled reticulocyte lysate system (Promega, Madison, WI). For this purpose, a full-length Hoxc13 cDNA was cloned into pcDNA3.1 vector (Invitrogen) and into a modified version of this vector (pcDNA3.1his) containing a Kozak (33) consensus translational start site and a polyhistidine (his) tag, thus resulting in expression vectors pHoxc13 and pHoxc13-his. Deletion of the homeodomain (hd) in pHoxc13-his by using EcoN1 and XhoI resulted in pHoxc13⌬hd-his. A TNT reaction with a luciferase expression plasmid was used as a negative control. Binding reactions and gel electrophoresis were performed exactly as previously described (34). For supershift assays, 1 l of HOXC13-specific antiserum (21) or anti-His antibody was added to the reaction; tyrosinase-related protein 1-specific antiserum (Santa Cruz Biotechnology, Santa Cruz, CA) was used as a negative control.
Transient Co-transfection Assays-For transient co-transfection assays using 3T3 mouse fibroblasts and the C2C12 mouse myoblast cell line, cells were treated and transfected as previously described (34). The reporter gene plasmid contained 9.1 kb of the Foxq1 promoter region in the pGL2-Basic Vector (Promega). Deletions of this promoter fragment were generated by digestion with MluI and Bst1107I and MluI and EagI, respectively (see Fig.  8A). Hoxc13 expression vectors used were pHoxc13, pHoxc13-his, and pHoxc13⌬hd-his as described above.
Each dish of 3T3 cells was transfected with 0.5 and 0.3 g of reporter and expression vector, respectively, in addition to 0.2 g of a ␤-galactosidase expression vector. C2C12 cells were transfected with double the amount of each plasmid. An enhanced green fluorescent protein expression plasmid was used to substitute for the expression plasmid as a negative control. Approximately 24 h after transfection, cells were harvested, luciferase and ␤-galactosidase assays were performed, and values were normalized as previously described (34). Each  (19). Accordingly, stem cells originating in the bulge region (not shown) migrate downward (black arrows) in the ORS (blue), which continues to line the dermal papilla (DP). There is evidence indicating that this germinative layer (green) harbors stem cells that give rise to the various lineages forming the follicular compartments (20). Based on the distinct hair keratin expression patterns, it has been proposed that upward and laterally moving cells derived from the upper internal ORS directly differentiate into precursors of the medulla and cortex (orange arrows), whereas cells derived from the lower internal ORS (gray arrows) undergo amplification as transient-amplifying cells in the keratin-free germinal matrix compartment (GMC) before differentiating into the cuticle and IRS unit. Note that Hoxc13 expression is found in the upper internal ORS and in all three hair shaft-forming compartments (cuticle, cortex, and medulla; shown in boldface) of the matrix. The cuticle of the hair shaft and the IRS cuticle are represented summarily as cuticle, and both are known to exhibit HOXC13 expression in human hair (21). In addition, Hoxc13 is expressed in the companion layer (not indicated here) of both mouse and human hair (17,21). transfection was performed in triplicate with at least three independent experiments.
Chromatin Immunoprecipitation (ChIP) Assays-ChIP assays were performed using methods from Ren and Dynlacht (35) with the following modifications. C2C12 mouse myoblast cells were plated at a density 3 ϫ 10 5 in 100-mm plates. After 48 h cells were transfected with a Hoxc13 expression vector where the Hoxc13 coding sequence was fused to a FLAG epitope at the amino terminus. After an additional 24 h cells were resuspended using trypsin/ EDTA (Sigma) and cross-linked using formaldehyde solution. Glycine (2.5 M) was added to stop the cross-linking reaction. Cells were lysed to extract chromatin and sonicated to generate DNA fragments of ϳ500 bp. Immunoprecipitation of FLAG-tagged protein-DNA complexes was achieved by incubating the lysate overnight at 4°C with EZview Red anti-FLAG M2 affinity gel (Sigma). After 2 washes with PBS the FLAG-tagged complexes were eluted with 2.5 M glycine, pH 3.5. Cross-links were reversed with 1% SDS, and protein was digested with proteinase K. DNA was purified by phenol/chloroform extraction. Precipitated Foxq1 DNA was detected by PCR using Foxq1-specific primer sets A and B (A-fwd, 5Ј-TCCTCCATCCTCCTCTCCTC, A-rev, 5Ј-ACTT-CATCCGTCACCACCTC; B-fwd, 5Ј-GGCCTGGAATCTCC-CTATTG; B-rev, 5Ј-AATTAGCCCAGTGCAAGGTG). Primer sets specific for Prx1 (5Ј-CCTGAGTTACCTGCACTCTG; 5Ј-AGGACTGAGGAGGATTCTTG) and Smad6 (5Ј-GCA-GCGTGGCGTACTGGGAGCA; 5Ј-GAG-TAACCCGGT-GGCACCTTG) genomic regions not containing Hoxc13 consensus binding sequences were used as controls. Annealing for all primers was at 60°C using standard conditions.

Structural Definition of Hair Defects in Hoxc13-overexpressing
Mice-Mouse Hoxc13 and its human ortholog are expressed in all three hair-forming compartments including hair shaft cuticle, cortex, and medulla as well as in the cuticle of the IRS and the companion layer ( Fig. 1 (17,18,21,22)). Accordingly, the follicular overexpression in GC13 transgenic mice may affect all of these compartments, and characterization of resulting hair defects has the potential to detect parallels to genetically defined hair mutants exhibiting similar defects. Mouse pelage hair fibers consist of four distinct types, guard, awl, auchenne, and zigzag hair, with the zigzag hair being the most common of truncal hair (6). With GC13 mutant mice, these hair types were difficult to distinguish when viewing skin and hair under SEM because most of the hair

TABLE 1 Gene hybridization and validation data for select skin and hair-specific genes
DNA microarray expression differentials of selected genes were validated by Q-PCR, and expression patterns in skin and hair are defined by ISH. FC, -fold change in expression value in GC13 mice relative to normal control mice. RT, reverse transcription. A space separates the group of "structural" genes from the transcription factor genes below (last three genes).

Gene symbol
Gene name Accession number

Foxq1 Is a Hoxc13 Regulatory Target in Hair
fibers were twisted and had an indistinct or poor quality cuticle (Fig. 2, A and B) compared with the straight appearance of hair from normal mice that exhibited a well defined and regular cuticular septation (Fig.  2, AЈ and BЈ). Furthermore, although normal mouse skin harbored a dense mass of hair fibers emerging sharply from the epidermis via the follicular osteum (Fig.  2AЈ), the mutant skin had few hair fibers emerging from ostea with a rosette-like mass of laminated cornified material surrounding the fiber (Fig. 2A). These changes are consistent with cutaneous orthokeratotic hyperkeratosis indicative of IRS defects. TEM analysis of the skin from two sets of mice revealed follicular hyperplasia compared with control mice (data not shown), which is consistent with earlier histopathologic results (18). This hyperplasia was accompanied by a marked increase in intermediate filaments within follicular keratinocytes. The IRS had large numbers of intermediate filaments that formed an intricate basket weavelike pattern near the hair fiber (Fig. 2C) not seen in the control (Fig. 2CЈ). Furthermore, we recently reported that the highly regular septation seen in the medulla of normal hair shafts is severely disrupted in the mutant (36). This abnormality turns out to be very interesting since there exist several mouse hair mutants with medulla defects of which the Foxq1-linked satin mutant (24) is highly relevant for this study because our array analysis identified Foxq1 as being down-regulated in skin of GC13 mice (Table 1). DNA Microarray Analysis Identifies Potential Regulatory Targets for Hoxc13 in Hair-As a step toward defining the molecular mechanisms underlying the distinct structural defects observed in hair of GC13 mice, we performed DNA microarray analysis of total skin from GC13 transgenic versus normal FVB/NTac control mice at PPD 5. Using criteria defined under "Materials and Methods," 226 genes were found as differentially expressed in mutant skin, 51 up-regulated and 175 down-regulated (see the supplemental list of differentially expressed genes). We selected a subset of differentially expressed genes that were either known or suspected to be active in most of the same hair follicle compartments where Hoxc13 is expressed (Fig. 1) for validation of their expression differentials by Q-PCR and for defining their expression patterns by in situ hybridization (ISH) analysis of normal and GC13 mouse skin at PPD 5 (Table 1). We identified Krt2-35 and Krtap5-4 as presumptive cuticle-specific genes and validated their down-regulation in GC13 skin by Q-PCR. The Krt2-35 expression pattern in the cuticle closely resembled the follicular pattern of its human ortholog HK6IRS2 (37,38) in the "neck" region of the bulb (Fig. 3A and AЈ). However, because of difficulty in resolving the cuticle of the IRS from that of the hair fiber (Fig. 3B), it is unclear whether Krt2-35 expression is exclu-sively restricted to the IRS cuticle as is the case for HK6IRS2 (37,38). By comparison, the keratin-associated protein (KAP)-encoding gene Krtap5-4, originally designated gUHS-SER-2 (39), exhibited a distinct cylindrical pattern in the cuticle restricted to a narrow region at the mid-level of the hair shaft that was shifted toward the bulb in GC13 mutant hair (Fig. 3C, CЈ and D). The down-regulation of both genes in GC13 hair might be a contributing factor in the development of the cuticular defects observed in these mice (Fig. 2).
Among the differentially expressed cortex-specific genes, Krt2-18 was perhaps the most conspicuous potential target for Hoxc13, and its 2.5-fold down-regulation was validated by Q-PCR ( Table 1). The Krt2-18 expression pattern in cortical and precortical regions extensively overlaps with the Hoxc13 pattern, and its distinct proximal boundary in the follicular matrix is very similar to the Hoxc13 boundary in normal FVB/NTac mice (Fig. 4, A and  B). In apparent contradiction to the proposed inhibition of Krt2-18 by Hoxc13, the proximal expression boundaries of both Krt2-18 and Hoxc13 were lost in GC13 mice, as reflected by expression detected below their normal levels within the follicle (Fig. 4,  AЈ and BЈ).
The most heterogeneous group of potential Hoxc13 targets was the group of medulla-specific genes including the hair keratin gene Krt2-16 (40), the desmosomal cadherin gene Dsc2 (41), and the transcription factor gene Foxq1 (24, 32) ( Table 1) as well as the cysteine-rich secretory protein 1-encoding gene Crisp1, whose activity in hair as a novel site of expression (36,42) and its downregulation in the follicular medulla of GC13 mice was recently reported (36). Expression patterns of all these genes were defined by ISH (Fig. 5) and their down-regulation was validated by Q-PCR ( Table 1).
Effects of Foxq1 sa (Satin) Allele on Expression of Medullaspecific Genes and Hoxc13-The identification of Foxq1 as a potential downstream regulatory target for Hoxc13 is intriguing since defective Foxq1 alleles underlie the satin hair phenotype (24), which shares important characteristics with the hair defects seen in Hoxc13-overexpressing mice. Satin mice are  SEPTEMBER 29, 2006 • VOLUME 281 • NUMBER 39 named for their satiny sheen, which is similar to that of the GC13 hair coat and is in both cases due to the abnormal light refraction secondary to hair medulla defects (data not shown). Perhaps the most salient feature of the abnormally differentiated medulla in hair of Foxq1 sa /Foxq1 sa mice is the near absence of air spaces resulting in the loss of septation typical for this structure (6). This is accompanied by a lack of cortical ridges and abnormal keratinization (6). The Foxq1 sa allele carries an intragenic deletion resulting in a frameshift and a truncated protein product (24). To examine how expression of Hoxc13 and Foxq1 itself might be affected by these changes, we performed ISH on dorsal interscapular skin of Foxq1 sa /Foxq1 sa mice. Although Hoxc13 expression, normally seen in the lower medulla of FVB mice (Fig. 6A), was difficult to discern in the medulla of satin hair, expression clearly continued to be present in medulla precursor cells capping the dermal papilla (Fig.  6C). This suggests that activation of Hoxc13 in medulla precursor cells does not require functional Foxq1 to be expressed in the same region. Expression of the Foxq1 sa allele is restricted to the same area in Foxq1 sa / Foxq1 sa mice, thus showing that transcription of the defective allele is still initiated in a spatially appropriate manner (Fig. 6D). By contrast, expression of medulla-specific differentiation markers down-regulated in GC13 mice, including Krt2-16 and Dsc2, was no longer detectable in the hair of Foxq1 sa /Foxq1 sa mice (Fig. 6, F and H), indicating that their activity was either directly or indirectly dependent on functional Foxq1.

Foxq1 Is a Hoxc13 Regulatory Target in Hair
Evidence for Interaction of Hoxc13 with Foxq1 Promoter Region-The previous results are consistent with the concept of a regulatory relationship between Hoxc13 and Foxq1. In support of this idea, we found multiple copies of putative Hoxc13 binding sites in the Foxq1 promoter and 3Ј-flanking sequences (Fig. 7A). The consensus binding sequence (5Ј-TT(A/T)ATNPurPur-3Ј) has recently been defined for human HOXC13 (21). An overall amino acid sequence similarity of 98% between the mouse and human orthologs and 100% identity between the homeodomains of these two proteins suggests that this consensus sequence is relevant for mouse Hoxc13. We have recently shown that Hoxc13 does indeed interact specifically with presumptive cognate binding sites present in the Crisp1 promoter region (36), which was one of the medulla-specific genes down-regulated in hair of GC13 mice. In vitro DNA binding studies using EMSAs revealed sequence-specific interaction between Hoxc13 and double-stranded oligos corresponding to two of its three most proximal putative binding sites clustered in the Foxq1 promoter region (Fig. 7, A and B). Hoxc13-oligo complex formation was inhibited by HOXC13-specific antiserum (21), which reacts specifically with mouse Hoxc13 in both supershift and immunohistochemical assays (22,36), whereas anti-tyrosinase-related protein 1 antiserum (see "Materials and Methods") used as a control did not affect complex formation (Fig. 7B).

Foxq1 Is a Hoxc13 Regulatory Target in Hair
To determine whether these Hoxc13-oligo interactions are dependent on the DNA binding function of the Hoxc13 hd, we performed a second set of EMSAs using poly histidine-tagged Hoxc13 (Hoxc13-his) compared with His-tagged Hoxc13 in which the hd was deleted (Hoxc13⌬hd-his). The results show a lack of specific complex formation between Hoxc13⌬hd-his and oligos 2411 and 2188 (Fig. 7C, lanes 9 and 19) compared with Hoxc13-his (Fig. 7C). Combined, these data suggest that Hoxc13 specifically interacts with at least two of the cognate binding sites found in the Foxq1 promoter region in vitro.
Further evidence for presumptive regulatory interactions between Hoxc13 and the Foxq1 promoter region was obtained by transient co-transfection assays in NIH3T3 cells and in C2C12 myofibroblasts. A Foxq1-luciferase reporter gene construct (Foxq1-luc) was utilized that included the two bona fide Hoxc13 binding sites and 12 additional putative Hoxc13 sites within 9 kb of Foxq1 upstream sequences (Fig. 8A). Surprisingly, in contrast to the repression of Foxq1 gene expression by endogenous Hoxc13, co-transfection of this reporter with a Hoxc13 expression vector resulted in ϳ10and 5-fold up-regulation of reporter gene activity in 3T3 and C2C12 cells, respectively, compared with control assays without Hoxc13 expression vector (Fig. 8B). Similar results were observed with the vector expressing His-tagged Hoxc13 (Fig. 8C). To determine how truncation of the promoter region affects this activation, we sequentially removed segments of 4 and 8.5 kb in Foxq1⌬Bluc and Foxq1⌬E-luc, respectively (Fig. 8A). Remarkably, Foxq1⌬B-luc showed a 3-fold up-regulation compared with Foxq1-luc, thus suggesting the presence of repressor functions in the deleted region. On the other hand, Foxq1⌬E, in which most of the promoter region was deleted, resulted in complete loss of activity. Furthermore, although co-transfection of Foxq1-luc with Hoxc13⌬hd-his versus Hoxc13-his expectedly resulted in a clear reduction of luciferase activity (Fig. 8C), this reduction was only about 50%, thus indicating that part of the Hoxc13-mediated activation of Foxq1-luc was hd-independent. hd-independent transcriptional activation has been reported previously, a particularly relevant case being the regulation of the Bmp4 promoter through Hoxa13 in which one of the three separate activation domains is functional in a hd-independent manner, whereas the other two require the hd (44).
To provide evidence for in vivo occupancy of the Foxq1 promoter region by Hoxc13, we performed ChIP assays with chromatin extracted from C2C12 cells that were transfected with a FLAG epitope-tagged Hoxc13 expression vector. Testing of precipitated chromatin by PCR using two primer sets, Foxq1(A) and Foxq1(B), specific for the proximal Foxq1 promoter region that included the presumptive Hoxc13 binding sites tested by EMSA, resulted in the amplification of Foxq1-specific DNA fragments in both cases (Fig. 9). Control reactions performed with precipitated chromatin and primers specific for exon 1 of Smad6 and the 5Ј-untranslated region of Prx1 that do not harbor Hoxc13 binding site consensus sequences remained negative (Fig. 9).
In summary, although we currently lack sufficient information to explain the opposite effect of Hoxc13 on in vivo Foxq1 gene expression compared with the results of transient transfection assays, taken together the data derived from EMSAs, co-transfection, and ChIP assays do suggest a role for Hoxc13 in the transcriptional regulation of Foxq1.

DISCUSSION
Functional Complexity of Hoxc13 in Hair-Despite the great heterogeneity of the tissue used in the present study, i.e. total skin from the trunk of 5-day-old Hoxc13-overexpressing mice, which consists of a wide variety of cell types at different stages of differentiation, we succeeded in identifying a considerable number of presumptive Hoxc13 target genes. During the validation process involving Q-PCR, co-localization of expression with Hoxc13 in hair and in silico analysis of promoter regions for the presence of putative Hoxc13 binding sites, many of these emerged as potential direct targets for Hoxc13 regulation in differentiating hair follicles. Consistent with previous data published by this group and others (18,21,22,36), the data pre-  Hoxc13, Foxq1, Krt2-16, and Dsc2 expression patterns in hair follicles of FVB/NTac and SB/LeJ-Foxq1 sa /Foxq1 sa mice at PPD 5. A and B, normal Hoxc13 (A) and Foxq1 (B) patterns (purplish-blue staining); in both cases strong expression is seen in the pre-medulla regions (red arrows) and the upper differentiated medulla (M; yellow arrows). C and D, Hoxc13 and Foxq1 expression patterns in hair of Foxq1 sa /Foxq1 sa (satin) mice. The loss of properly differentiated medulla structures in the mid to upper regions is accompanied by a loss of expression of the Foxq1 sa allele (D) but apparently not of Hoxc13 expression (C) in the central regions of these hairs; both genes continue to be expressed in the pre-medulla region (red arrows) capping the dermal papilla (DP); the pigmented satin hair has brownish-colored melanin granules (gray arrows) absent in albino FVB hair. E and G, normal Krtap16 (E) and Dsc2 (G) expression as seen in the medulla of hair from FVB mice (yellow arrows) is not detectable in hair of satin mice (F and H). Ctx, cortex; Cu, cuticle; M, medulla. SEPTEMBER 29, 2006 • VOLUME 281 • NUMBER 39 sented here suggest that Hoxc13 regulates the expression of various sets of "structural" genes (i.e. genes encoding hair keratins, KAPs, and desmosomal cadherins) that are required for proper terminal differentiation of distinct hair follicle compartments.

Foxq1 Is a Hoxc13 Regulatory Target in Hair
Two of the four KAP genes found to be down-regulated, Krtap8-2 and Krtap6-3 (see the supplemental list of differentially expressed genes), were also isolated by us in a previous screen for differentially expressed genes in postnatal GC13 skin by using a different method, i.e. cDNA suppression subtraction hybridization (18). Both KAP genes were identified as members of a novel KAP gene domain on MMU16 in a region of conserved linkage to HSA21q22.11 (18,22), and most of the cDNAs isolated by suppression subtraction hybridization in the previous study corresponded to this domain (18,22), a bias that was not observed in the DNA microarray data set presented here. The up-regulation of Krt2-1 in the epidermis (Table 1) was also noted in the previous screen (18), and although this finding is consistent with the epidermal hyperproliferation and ichthyosiform condition observed in GC13 mice, a potential mechanistic link to the overexpression of Hoxc13 in hair follicles is cur-rently obscured since epidermal Hoxc13 expression was not detectable by ISH in both normal and GC13 mice (data not shown).
In addition to implicating Hoxc13 in the regulation of structural genes, we have validated Hoxc13-induced expression changes for two upstream regulatory transcription factors (Foxq1 and Foxn1, Table 1) and have obtained differential expression data for several other regulatory factors including Tcf3, Sox18, and Smad7 (supplemental list of differentially expressed genes). These regulatory genes are distinctly expressed in the hair follicle matrix (Refs. 24 and 45-49; for review, see Ref. 23), and many, including Hoxc13, are conspicuously expressed in the internal ORS layer surrounding the dermal papilla, also known as germinative layer (Fig. 1), which recently has been demonstrated to harbor restricted self-renewing stem cells (20). These cells are organized in distinct sectors arranged symmetrically on either side of the dermal papilla, and their position along the proximal-distal axis determines the specific compartment lineage of transient amplifying or differentiating cells originating from these precursors (20,19). According to this model and assuming that the presumptive regulatory tar- gets for Hoxc13 in this critical region can be confirmed as genuine targets as we have done in the case of Foxq1, Hoxc13 is likely to have an important function in the organization of the matrix and its lineage trajectories, at least as far as this affects the organization of the upper lineages contributing to the hair shaft compartments. The disorganized expression of Krt2-18 and of Hoxc13 itself in the follicular matrix of GC13 mice is consistent with this view, as this might indicate mixing of differentiating with proliferating cells in the proximal bulb region (Fig. 4).
The identification of Foxq1 as differentially expressed in GC13 skin was fortuitous and will assist future work in this area since there is a mutant allele, satin, which causes a hair phenotype (24). There are at least four other differentially expressed genes in GC13 skin whose mutated alleles are linked to known mouse hair mutants with distinct hair shaft defects. These include Foxn1 (nude (46)), the X-linked Nsdhl gene (bare patches (51)), Tgf-␣ (waved 1 (52)), and P (pink-eyed (53)) (see the supplemental list of differentially expressed genes). Although of these four genes we have thus far only validated the expression differential for Foxn1 by Q-PCR (Table 1), it will be of interest to determine whether Foxn1 itself and any of the remaininggenesarepartofaHoxc13dependent regulatory network.
Hoxc13, Transcriptional Activator or Repressor?-The changes in Foxq1 expression in skin of GC13 versus normal mice as determined by DNA microarray, Q-PCR, and ISH (Table 1; Fig. 5) suggest that Foxq1 is a downstream target for Hoxc13 in hair follicles. The regulation of Foxq1 by Hoxc13 in a transfection system (Fig. 8) in conjunction with data suggesting Hoxc13 binding to Foxq1 promoter sequences both in vitro (Fig. 7) and in vivo (Fig. 9) strongly support the concept that Hoxc13 directly regulates Foxq1 during hair follicle differentiation.
The confounding aspect of Hoxc13 regulation of Foxq1 is that in co-transfection assays Foxq1 was activated, whereas down-regulation in GC13 versus normal mice suggests that Hoxc13 acts as a repressor of Foxq1 in vivo in hair follicles. These opposing effects are likely reconcilable with each other when considering that plasmid-based reporter gene activation in cultured cells is unlikely to reflect an authentic response of the corresponding gene in its native cellular and nuclear environment. The absence of appropriate chromatin structure, lack of potentially critical control elements not included in the reporter construct, and importantly, cell type-specific differences in co-factor supply are only some of many issues that have to be considered when interpreting these results. Even if the same co-factors were present in diverse cell types, the regulation of targets still may be altered by signaling cascades. For example, Hox-Pbx cofactor complexes may be switched from transcriptional repressor to activator in response to protein kinase A signaling or cell aggregation (54).
Numerous studies in recent years revealed a continuously broadening spectrum of regulatory mechanisms affecting Hoxtarget gene interaction including sequence-specific binding of Hox proteins as monomers (55) and in association with one or several cofactors (56). The most thoroughly characterized cofactors known to selectively and cooperatively increase binding affinity and specificity of Hox proteins belong to the class of FIGURE 8. Subregions of Foxq1 promoter are differentially responsive to Hoxc13. A, maps of Foxq1-luciferase (luc) reporter gene constructs Foxq1-luc, Foxq1⌬B-luc, and Foxq1⌬E-luc containing ϳ9, 5, and 0.5 kb, respectively, of Foxq1 promoter region. B, co-transfection assays with Foxq1-luc and Hoxc13 expression vector in NIH3T3 cells and C2C12 myofibroblasts indicate an ϳ5-10-fold up-regulation of reporter gene activity compared with control assays without Hoxc13 expression vector. C, co-transfection of Foxq1⌬B-luc with Hoxc13-his expression vector reveals an ϳ3-fold enhancement of expression compared with Foxq1-luc, whereas removal of most of the Foxq1 promoter region in Foxq1⌬E-luc, including all of the presumptive Hoxc13 binding sites, results in complete loss of luciferase activity. Interestingly, co-transfection of Foxq1-luc with Hoxc13⌬hd-his, in which the homeodomain was deleted, resulted in only a 50% reduction of expression compared with co-transfection with Hoxc13-his. TALE (three-amino acid loop extension) homeodomain proteins and include members of the Pbx, Meis, and Prep groups (56,57). Potential cofactor requirements for Hoxc13 are currently unknown. The in vitro cofactor-independent binding of human and mouse Hoxc13 to HOXC13 consensus binding sequences present in the promoter regions of several hair keratin and KAP genes (21,22) as well as Crisp1 (36) and Foxq1 (Fig. 7, A and B) does not preclude Hoxc13-cofactor interaction in vivo. Examination of the human hHa5 and hHa2 hair keratin gene promoter regions known to drive reporter gene expression in PtK2 cells upon co-transfection with a HOXC13 expression vector in a homeodomain-dependent manner (21) identified perfect matches to both Meis/Prep heterodimer and Pbx consensus binding sequences (56) located upstream of multiple monomeric HOXC13 sites (21). However, although expression of Pbx1-4, Meis1a, Meis1b, and Meis2 as well as Prep1 and Prep2 was indicated in the bulb of human anagen hair follicles by reverse transcriptase PCR, immunofluorescence studies localized the corresponding proteins primarily in the IRS and mid-to lower cortex, and none of them showed nuclear colocalization with HOXC13 (58). In light of these results, the significance of multiple putative Pbx1 as well as composite Pbx1/Meis1 and Meis1/Prep1 sites found to be interdigitated with the Hoxc13 binding sites in the Foxq1 genomic 5Ј-and 3Ј-flanking regions (data not shown) is currently uncertain.
The circumstance that none of the putative Hoxc13 binding sites (Fig. 7A) was found to form a potential heterodimer site with any of the TALE cofactors might lend support to the concept of a monomeric role for Hoxc13. Although there is generally little information about the regulation of Hox target genes in a cofactor-independent manner, data showing repression of the Drosophila spalt gene through multiple Ubx monomeric binding sites have led to the proposition that regulation of some Hox target genes evolved via the accumulation of multiple monomeric binding sites (59). Because Hox proteins generally bind with low affinity to monomeric sites, the cumulative effects of interactions with multiple low affinity binding sites may range from relatively high levels of expression to complete repression of a target gene depending on Hox protein concentration. The quantitative changes in Foxq1 expression observed in GC13 versus normal mice (Table 1; Fig. 5, E and F) are consistent with this model. This type of modulated response offers greater adaptability and, thus, greater system stability than a simple on-off switch type of regulation.
Recent evidence that Hox-mediated transcriptional responses depend on compartment-specific contextual information (60,43) provides a further argument for reconciling the opposing effects of Hoxc13 on Foxq1 regulation in in vivo versus cultured cells. According to this model, the contextual co-factors do not control the binding properties of Hox⅐TALE protein complexes but modulate the transcriptional output depending on whether cognate binding sites required for their recruitment are present in cis-regulatory regions of target genes (43). Although this has not yet been shown in mammals, it is likely that cell type-specific co-factors will be of universal relevance in determining the quality of Hox-controlled transcriptional responses.