WNT and DKK Determine Hair Follicle Spacing Through a Reaction-Diffusion Mechanism

See allHide authors and affiliations

Science  01 Dec 2006:
Vol. 314, Issue 5804, pp. 1447-1450
DOI: 10.1126/science.1130088


Mathematical reaction-diffusion models have been suggested to describe formation of animal pigmentation patterns and distribution of epidermal appendages. However, the crucial signals and in vivo mechanisms are still elusive. Here we identify WNT and its inhibitor DKK as primary determinants of murine hair follicle spacing, using a combined experimental and computational modeling approach. Transgenic DKK overexpression reduces overall appendage density. Moderate suppression of endogenous WNT signaling forces follicles to form clusters during an otherwise normal morphogenetic program. These results confirm predictions of a WNT/DKK-specific mathematical model and provide in vivo corroboration of the reaction-diffusion mechanism for epidermal appendage formation.

The development of regularly arranged body parts has long fascinated experimental biologists and theoreticians alike. One area of long-standing debate has been the formation of epidermal appendages such as feathers and hairs. Theoretical models have provided seemingly simple solutions to complex developmental processes (1); in order to achieve regular patterns, the reaction-diffusion (RD) hypothesis of Alan Turing postulates a pair of activator and inhibitor with special characteristics (2) [supporting online material (SOM) text 1]. However, it remains largely unclear whether such predictions can be substantiated in molecular and mechanistic terms (3). Because canonical WNT signaling is essential for the induction of hair and feather follicles (4, 5) and forced stimulation of this pathway is sufficient to induce supernumerous appendages (6, 7), the pathway represents an appealing candidate for the primary signal that dictates follicle distribution. Here we set out to analyze its role in hair follicle arrangement by verifying predictions of a biologically adapted RD model.

The WNT pathway is active from the earliest stages of follicular development (5, 8) (Fig. 1A). Expression of the WNT inhibitor Dkk1 is directly controlled by secreted WNTs (9, 10). Further aspects of this pathway and the RD mechanism are discussed in SOM text 2 and fig. S1. In developing murine skin, mesenchymal Dkk1 expression is found adjacent to the early hair follicle bud (5) (Fig. 1B), whereas Dkk4, a further functional inhibitor of WNT signaling (11, 12), shows strong epithelial expression at discrete loci before hair placode formation (Fig. 1C). Weak expression in the early hair follicle bud indicates that Dkk4 expression marks the forming follicle (Fig. 1C). Five LEF/TCF consensus binding motifs are found within 700 base pairs (bp) upstream of the transcriptional start site of Dkk4, and regulation of the promoter by the canonical WNT signaling pathway was suggested by transfection studies (Fig. 1D). Hence, the available data support the role of WNT and DKK as primary determinants of hair follicle spacing patterns.

Fig. 1.

WNT signaling and expression of Dkk genes are associated with hair follicle formation. (A) WNT signaling in mesenchymal cells is associated with developing hair follicles (arrowhead). BATgal mice harboring a WNT-responsive lacZ gene were used as a reporter. (B) Mesenchymal Dkk1 expression adjacent to epithelial placodes and buds (arrowheads). (C) Strong epithelial Dkk4 expression at discrete loci prior to hair placode formation. Expression rapidly declines after follicle budding (arrow). (A to C) Scale bars, 100 μm. (D) Reporter gene expression [relative light units (RLUs) ± SEM] after endogenous (white) and stimulated (black) canonical WNT signaling. *P < 0.0001 (t test) for stimulated WNT signaling (black columns).

If WNTs and WNT inhibitor(s) represent the two components required by the RD hypothesis, it should be possible to derive, from a WNT/DKK-specific RD model (SOM text 3), predictions about the outcome of experimental alterations of activating and inhibitory functions. Our computational modeling showed that moderate overexpression of activator during either the initial or a subsequent inductive wave increases follicular density (Fig. 2, A and B). By contrast, strong overexpression of activator completely disrupts the patterning process. Moderate overexpression of inhibitor during the initial inductive wave increases the interfollicular spacing (Fig. 2A). In line with the prediction of defective pattern formation after further enhancement of inhibitor expression, hair and feather follicle induction is indeed blocked in the presence of strong Dkk1 expression (4, 5); however, the role of WNTs in appendage formation and patterning is still contentious. During a subsequent inductive wave, increased inhibitor expression blocks the development of new follicles in the interfollicular space (Fig. 2B). In addition, excess inhibitor gives rise to ringlike zones of high activator levels around preexisting appendages (Fig. 2B). If levels of activator above a threshold determine where appendages form, this suggests the possibility of follicle cluster formation. The ringlike zones of high activator levels are converted to discrete spots (Fig. 2B) if preexisting follicles become insensitive to activator and inhibitor (SOM text 4). Thus, moderate overexpression of inhibitor by the appendages predicts a quantitative as well as a qualitative read-out: The number of follicles is reduced, and their distribution is changed.

Fig. 2.

A WNT/DKK-specific model of a reaction-diffusion system predicts changes in epidermal appendage distribution after transgenic interference with endogenous signaling. Diagrams for excess activator or inhibitor production correspond to moderate overexpression that was restricted to the forming appendages. (A) Modeling of the first inductive wave. The calculated distribution of activator is shown for a 100 × 100 area (arbitrary units). (B) Modeling of a subsequent inductive wave. The first wave is shown in blue, the second in red. The calculated distribution of activator is depicted for a 200 × 200 area. The inset reflects the difference in activator distribution between second and first inductive wave for a first wave follicle. An activator and inhibitor insensitivity of first wave follicles might reflect the in vivo situation most closely.

We used transgenic expression of inhibitor during hair follicle formation to experimentally test these theoretical predictions (Fig. 3 and SOM text 5 and fig. S2). In mouse, temporally well defined successive waves of induction during embryogenesis and early postnatal life give rise to three major hair follicle types (designated guard, awl, and zigzag) (13) (Fig. 3A). In compliance with the model, new follicles are initiated in between previously induced appendages (Fig. 3B). We analyzed four Foxn1::Dkk2 mouse lines in detail, and their phenotype was correlated to the level of transgene expression (Fig. 3, C and D). At high levels of ectopic Dkk2 mRNA, mice appeared almost hairless; at lower levels, the hair coat is present but structurally abnormal (Fig. 3D). Suppression of hair follicle formation at high levels of Dkk2 expression is consistent with previous results for Dkk1 (5) and conforms to the prediction of our computational modeling. However, guard hair follicle induction still took place even in severely affected mice (Fig. 3, D and E) because of the delayed onset of Foxn1 promoter activation relative to guard hair follicle induction (Fig. 3A and fig. S2).

Fig. 3.

Suppression of WNT signaling increases interfollicular spacing in Foxn1::Dkk2 mice. (A) Schematic of the timing of Foxn1 promoter activity and the induction of the three major hair follicle types. (B) E18.5 back skin reveals the existence of successive inductive waves that can be distinguished by the developmental stage of follicles (1, guard; 2 and 3, awl; 4, zigzag). (C) Levels of Dkk2 expression. TG, Foxn1::Dkk2 transgenic mice. (D) Appearance of wild-type and differently affected Foxn1::Dkk2 mice. (E) Strongly affected Foxn1::Dkk2 mice only develop large guard hair follicles. (F) Densities of guard hair follicles (black) and of the full complement of follicles (white) are reduced after WNT signaling is suppressed. (G) The density of guard hair follicles is reduced at E14.5, that is, immediately after the first inductive wave, after WNT signaling is suppressed. Emerging follicles are characterized by localized WNT signaling, visualized by whole-mount lacZ staining of skin from BATgal mice of the indicated genotypes (top views). (B, C, E, and G) Scale bars, 100 μm.

The number of follicles per square millimeter is reduced by about 30% in adult Foxn1:: Dkk2 (medium) mice (Fig. 3F). Regarding the first inductive wave that gives rise to guard hair follicles, a similar reduction is evident (Fig. 3F). In Foxn1::Dkk2 (strong) mice, which only develop guard hair follicles, the density is further reduced (Fig. 3F). The emergence of hair follicle buds is accompanied by localized WNT signaling in mesenchymal cells of the prospective dermal papilla (Fig. 1A). As expected, the density of these cell clusters is reduced in Foxn1::Dkk2 transgenic mice (Fig. 3G). However, we could not detect any difference between mice with high and medium mRNA levels. Thus, the significant decrease of guard hair follicle density in Foxn1::Dkk2 (strong) as compared with Foxn1::Dkk2 (medium) mice suggests the existence of more than one inductive wave for guard hair follicles; apparently, all but the first wave are blocked by strong transgene expression. For the first inductive wave, the results confirm the predictions of our model with respect to the quantitative effects of inhibitor overexpression.

As suggested by our simulations, WNT and DKK proteins may control the natural increase of follicle density during subsequent inductive waves (Fig. 2B and SOM text 6 and figs. S3 and S4). In order to investigate inhibitor effects on these waves, we next examined transgenic mice with an abnormal hair coat in more detail. These mice are characterized by a misdistribution of hair shafts. Whereas emerging hair shafts are almost evenly distributed over the skin surface in wild-type mice, bundles of hair shafts are separated by large areas of interfollicular epidermis in Foxn1::Dkk2 mice (Fig. 4A). Histological sections confirmed that usually more than three follicles are tightly clustered in transgenic mice (Fig. 4B); all follicles within a cluster gave rise to bona fide hair shafts. By contrast, follicles are clearly separated from each other in wild-type skin (Fig. 4B). Strong follicle clustering was also observed in Foxn1::Dkk1 mice (fig. S5), which corroborated its independence of the inhibitor's identity. A ringlike pattern of WNT signal-receiving cells around preexisting follicles confirms the model'sprediction for activator distribution and indicates an initiation of cluster formation at about embryonic day 17.5 (E17.5) (Fig. 4C). Unequivocal morphological indications of hair follicle cluster formation were observed at E18.5 and postnatal day 1 (P1) (Fig. 4D and fig. S6). Thus, hair follicle clusters in Foxn1::Dkk2 mice form during a normal inductive program by misdistribution of the normal complement of epidermal appendages (Fig. 4D and SOM text 7 and fig. S6).

Fig. 4.

A normal sequence of inductive waves gives rise to hair follicle clusters after moderate suppression of WNT signaling. (A) Hair distribution on the back of wild-type and Foxn1::Dkk2 mice. (B) Hair follicle distribution in the back skin of 10-day-old wild-type and Foxn1::Dkk2 animals. (C) Aberrant distribution of WNT signal–receiving cells in Foxn1::Dkk2 mice at E17.5. Canonical WNT signaling is visualized by whole-mount lacZ staining of back skin from BATgal mice of the indicated genotypes (top views). (D) New follicles (arrowheads) emerge at E18.5 and P1 in Foxn1::Dkk2 mice. All scale bars, 100 μm.

We note that further signaling pathways are involved in interfollicular patterning. Studies on feather development demonstrated that fibroblast growth factors (FGFs) promote follicle formation, whereas BMP and EGF signaling confers interfollicular fate (14, 15). Because these pathways appear to be downstream of WNT signaling and exert feedback control (5, 14, 1619), we propose that they mainly mediate and modulate WNT signals, thereby contributing to the stabilization and refinement of the patterning process. Indeed, ablation of the BMP receptor IA appears to have no major impact on hair follicle induction and distribution (20). However, suppression of BMP signaling that may reduce Dkk1 expression (19) causes an increase in hair follicle density (21), consistent with our simulations (fig. S3C). Given that LEF1 plays an important role in WNT signaling, which controls ectodysplasin signals (5, 22), our results may also explain the misdistribution of follicles in EdaA1 and Lef1 transgenic mice (23, 24). However, amoresophisticated systemsbiology approach will be needed in the future to include the full complexity and dynamics of the WNT signaling pathway (25) in a model of interfollicular patterning. In conclusion, our combined experimental and computer modeling approach presents compelling evidence for WNT signaling and a reaction-diffusion mechanism as key determinants of hair follicle spacing patterns.

Supporting Online Material

Materials and Methods

SOM Text

Figs. S1 to S6


References and Notes

View Abstract

Stay Connected to Science

Navigate This Article