FGF5 is a crucial regulator of hair length in humans

Identification of Causative Mutations in Fibroblast Growth Factor 5.

We performed genome-wide SNP genotyping on 14 members of FT1 to identify evidence for regions with linkage and homozygosity. Whereas SNP data for FT1 was uninformative for linkage, a maximum homozygosity score of 84 was obtained for a single region on chromosome 4q21.21 (Fig. 2 A and B), indicating a large continuous region of homozygous genotypes shared among affected individuals and not present in unaffected ones (7). Because exome sequencing has rapidly emerged as a powerful means to identify genes that cause rare Mendelian phenotypes, we simultaneously performed whole exome sequencing on two members of FT1, three members of FT2, and five unaffected ethnically matched controls. To identify variants that were consistent with our disease model, we used a filtering strategy in which we excluded (i) any variants heterozygous in one or more affected individuals; (ii) variants not present in all affected family members; (iii) variants present in unaffected ethnically matched controls; or (iv) variants annotated in public databases, including 1000 Genomes Project and the National Heart Lung and Blood Exome Variant Server containing exome data from 6,500 individuals. Using this filtering strategy, we identified a single gene, fibroblast growth factor 5 (FGF5), which resides on chromosome 4q21.21, directly within the region of homozygosity, and which harbored previously unidentified homozygous mutations in both our families (Fig. 2C).

In FT1, we identified a 1-base pair (bp) deletion at the donor splice site of intron 2 (c.459+1delG), whereas in FT2 we identified a 2-bp deletion within exon 1 (c.159_160delTA). Sanger sequencing was used to validate these mutations (Fig. S2), which demonstrated significant evidence for cosegregation in both families with a logarithm of odds (LOD) score of 5.73. We next sequenced FGF5 in unrelated probands from five additional FT families and identified an additional homozygous missense mutation in exon 3 (c.520T > C), in a single proband (FT3) from Pakistan (Fig. 2D). Using the rare variant burden test (8), we calculated that the number of rare damaging mutations in FGF5 identified in our cohort is greater than expected by chance (P = 4.5 × 10⁻¹⁰). None of the identified mutations were present in 50 ethnically matched controls, nor have they been reported in any public database. The fraction of familial trichomegaly cases caused by mutations in FGF5 is 42.5% (3/7) in this study.

Protein Consequences of FGF5 Mutations.

FGF5 is subject to alternative splicing, which generates two FGF5 transcripts, one producing a 268-amino-acid protein (FGF5), and a second, shorter 123-amino-acid protein, FGF5-short form (FGF5S). In FGF5S, exon 2 is skipped, and a frameshift due to the out-of-frame splicing of exon 1 to exon 3 results in translation of only 14 base pairs of exon 3 before a termination codon (Fig. S3). FGF5 contains a conserved receptor binding domain whose coding sequence spans all three exons, and interacts with its receptor, FGFR1 (9, 10). Moreover, FGF5S has been shown to act in an antagonistic manner by competitively binding FGFR1, preventing activation of FGFR1 by FGF5 (11). The mutation in FT2 is located within exon 2 and causes a frameshift affecting both FGF5 and FGF5S, whereas the mutations observed in FT1, and FT3 reside in exons 2 and 3, respectively, and therefore only affect full-length FGF5. In FT3 (Fig. 3A), the missense mutation (c.520T > C) converts a conserved tyrosine residue to histidine (p.Y174H). Examination of FGF-FGFR cocrystal structures (12, 13) shows that Y174 is positioned within a hydrophobic core and resides adjacent to several receptor-interaction surfaces where it likely stabilizes them in their native conformation (Fig. 3 B and C). We used the functional prediction programs Polyphen-2, SIFT, and Mutation Assessor (14–16) to determine the effect of the missense mutation in FT3. Polyphen-2 (14) predicted the change would be “probably damaging” with a score of “1”; SIFT (16) predicted the mutation would be “deleterious” with a score of “0”; whereas Mutation Assessor (15) indicated the functional impact of the mutation was “high,” within a disease-causing range, with an impact score of 3.535. Taken together, these predictions suggest that p.Y174H is likely to have a negative impact on the function of FGF5. We predict that substitution of the hydrophobic side chain of tyrosine with a hydrophilic histidine is likely to impact the function of FGF5 such that it would either fail to bind to and activate its receptor, FGFR1, or do so at significantly reduced levels.

Localization of FGF5 in the Hair Follicle.

FGF5 acts as a stimulus for catagen entry by binding to FGFR1 located within the dermal papilla, a mesenchymal signaling center located at the base of the hair follicle (17). However, the origin of FGF5 has not been clearly established within mouse skin. One hypothesis is that FGF5 originating within the hair follicle outer root sheath acts to induce catagen (18), or alternatively, the source may reside in the follicle macroenvironment, potentially from FGF5-expressing macrophages that surround the hair follicle (19). To localize the isoforms of FGF5 in normal human hair follicles, we used PCR to amplify fragments of FGF5, and FGF5S in tissue from an unaffected individual, and found that both were expressed within occipital scalp skin in addition to plucked hair fibers (Fig. S4A). Immunohistochemistry on scalp skin also confirmed the presence of FGF5 within the upper outer root sheath cells of human hair follicles, and interestingly, in small round cells surrounding the follicle (Fig. S4B). In mouse skin, global changes in the macroenvironment have been shown to modulate hair cycling, which occurs in a synchronized wave (20). In contrast, human hair growth occurs in a mosaic pattern, not in a synchronized wave, and fluctuations in the macroenvironment have yet to been characterized. The presence of FGF5 in adjacent perifollicular cells invites further exploration of the role of these cells in the human hair cycle.

We took advantage of the localization of FGF5 within the outer root sheath to interrogate whether FGF5 protein was present within FT hairs. When a hair fiber is plucked from the skin it contains one to two layers of outer root sheath cells that remain attached to the hair. We performed whole mount immunofluorescence using patient and unaffected control plucked forearm hair fibers, with antibodies specific for both FGF5 and FGF5S. In FT1, we found that FGF5 was completely absent, and very low levels of detectable FGF5S were present, whereas in FT2 neither FGF5 nor FGF5S were present in the tissue surrounding plucked hair fibers, consistent with the predicted mechanism of action of the mutations (Fig. S5).

Phenotypical Effect of Mutations in FGF5.

To our knowledge, this study is the first description of FGF5 gene mutations that are causative for trichomegaly in humans. Previously, mutations in FGF5 were found to be causative for the angora mouse phenotype (21), which is characterized by an abnormally long body hair coat (22). Within our families, hair growth was most striking on the eyelashes. However, to determine the impact of FGF5 mutations on body hairs (Fig. S6), we obtained and measured plucked forearm hair fibers from adult male members of families FT1 and FT2. Not only were hairs significantly longer than those collected from controls, but they also showed an increased variance in length, which suggests that there is a dysregulation of a physiological process (Fig. 4A). However, hairs from affected family members were not any thicker than control hair fibers. We later categorized plucked hair fiber root tips into either anagen or telogen based on their morphology and found a notable shift in the anagen:telogen ratio toward anagen, in patient hairs compared with ethnically matched controls (Fig. 4B). In FT1, 65% of fibers were in anagen, and in FT2, 59% were in this cycle stage. This was compared with only 17% of control hair fibers being in anagen (83% in telogen), indicating that in humans, mutations in FGF5 do not only affect eyelash hair follicles but also affect hair growth at other sites that predominantly reside in telogen, such as the forearm. Whereas mutations in FGF5 result in longer hair fiber length, we cannot exclude the possibility that affected hair fibers are also growing faster than shorter hairs from control individuals. Longer eyelashes have previously been shown to have both a longer anagen duration, and a faster rate of growth than short eyelashes, with both of these features contributing to the observed length (23).

Overexpression of FGF5 in Human Hair Follicles Induces Catagen.

In mouse skin, exogenous introduction of FGF5 by s.c. injection of recombinant proteins has previously been shown to induce premature entry into catagen, whereas FGF5S introduction can antagonize this effect (24), presumably by competitively binding the FGF5 receptor, FGFR1 (11). To assess the effect of FGF5 on the human hair cycle, we used the well-established organ culture model (25) for hair fiber elongation and grew microdissected human scalp hair follicles in the presence of recombinant FGF5 (Fig. 4C). Strikingly, the human follicles entered catagen prematurely after 2–4 d in the presence of recombinant FGF5, whereas untreated hairs entered catagen after 5–7 d (Fig. 4D). Their growth was significantly (P = 0.03) reduced after incubation with recombinant FGF5 protein, allowing us to conclude that, whereas a lack of FGF5 results in increased hair length as seen in our patients, overexposure to FGF5 inhibits hair growth in an organ culture model by initiating catagen.

Patient Blood Collection.

Informed consent, for blood/hair collection and taking and using photographs, was obtained from all subjects and approval for this study was provided by the Institutional Review Board of Columbia University. Peripheral blood samples were collected from family members, as well as unrelated healthy control individuals of Pakistani origin (50 individuals). Genomic DNA was isolated from these samples using the Puregene DNA isolation kit (Gentra System). We collected DNA from 21 members of FT1 and 9 members of FT2. We also collected DNA from probands of five other families who presented with trichomegaly.

Genome-Wide Linkage and Homozygosity Analysis.

At the beginning of our study, we only had DNA from 14 of the 21 members of FT1. To identify regions in the genome with evidence for linkage and/or homozygosity, genome-wide SNP-based genotyping was performed on these original 14 members using the Affymetrix Human Mapping 10K 2.0 Array according to the manufacturer’s protocol. Quality control and data analysis was initially performed with Genespring GT (Agilent software). SNPs with greater than 5% missing data or SNPs that violated a Mendelian inheritance pattern were removed from the dataset before analysis. To reduce type I error introduced by linkage disequilibrium between markers, haplotypes were inferred from genotype data and analyzed for linkage under the assumption of a fully penetrant disease gene with a frequency of 0.001 transmitted by a dominant mode of inheritance. Homozygosity Mapper was used to identify homozygous regions shared among affected individuals (7). In Homozygosity Mapper, the homozygosity score is calculated as a function of marker genotype, sample size, and length of homozygous block. The score is relative and analysis dependent. In a single analysis, a homozygosity score is computed for each marker. A “maximum score” is also yielded in this process, which marks the region(s) of the genome with the highest level of homozygosity in a particular analysis (7).

Whole Exome Capture Followed by Massive Parallel Sequencing.

Two members of FT1, three members of FT2, and five ethnically matched controls were subject to whole exome sequencing. Agilent SureSelectXT Human All Exon v.2 44 Mb kit (Agilent) was used for exome capture, which were then analyzed by massively parallel sequencing on an Illumina HiSEq 2000. Briefly, for each sample, fragment libraries were constructed using 3 μg of genomic DNA as starting material, following the Agilent standard library preparation protocol for Illumina TruSeq. A total of 500 ng of captured library was hybridized with the baits for 48 h, then washed and eluted using the Agilent protocol. Real-time PCR was used to quantify the libraries, which were subsequently sequenced with standard protocols for the Illumina HiSEq 2000 instrument. Standard Illumina TruSeq multiplexing (barcoding) protocol was used to sequence multiple samples in one lane.

Bioinformatics Analysis.

Paired-end reads (read size: 101 bp) were mapped to the human reference genome National Center for Biotechnology Information (NCBI) build 37 using BWA (40) (version 0.5.9), allowing up to five mismatched, inserted or deleted bases (indels). Alignment was then refined using GATK (41) (version 1.0.6020) by performing local multiple sequence alignment around inferred putative indels and known ones from the 1000 Genomes Project, and subsequently recalibrating base quality scores. We then called single nucleotide variants (SNVs) and indels jointly from all of the samples in one batch of experiments using GATK (Unified Genotyper). In addition to the default variant filters in GATK, we applied additional filters to ensure low false discovery rate. These filters included: (i) number of reads that have zero-mapping quality is more than 10% of all of the reads mapped to the locus, or (ii) genotype quality score smaller than 30, or (iii) quality over depth less than 5.0, or (iv) strand bias score larger than −0.1. In total, this method allows us to detect about 17,000–20,000 coding SNVs, with an average transition to transversion ratio of 3.3 for known coding variants and 2.9 for novel coding variants, which indicates low false discovery rate (41). Called variants were annotated with Annovar and filtered under a recessive model for potentially damaging coding and splice site variants (3).

Sequence Validation.

Exome sequencing validation was performed for FT1 and FT2 in individual exons. All three exons were also sequenced in affected probands from an additional five index cases with familial trichomegaly. Individual exons were amplified using Platinum PCR SuperMix (Invitrogen) and the following primers: exon 1 forward (5′-CTACAGAGCCCAGAATCAGC-3′) and reverse (5′-CCTACAAAACCCGTC CTAGG-3′), exon 2 forward (5′-CACAATGATATAAATAAGAGCCTG-3′) and reverse (5′-AAATTATTACTTCTGGAAGCGCAC-3′), and exon 3 forward (5′-GATCGCCACAACTTAAATTTCC-3′) and reverse (5′-TTTCTCACAAGGCCAAGAGG-3′). Amplified PCR products were purified, then sequenced using the forward primer in an ABI Prism 310 Automated Sequencer, using the ABI Prism Big Dye Terminator Cycle Sequencing Ready Reaction kit (Invitrogen).

Statistical Analyses of Causal Variants.

Two-point linkage analysis was performed for the disease variants with Superlink Online v1.5 (42). Data from all family members whose DNA was available for genotyping were included in the analysis. A fully penetrant disease allele with a population frequency of 0.001 and a recessive mode of inheritance was assumed based on studies in animal models. Variant burden for FGF5 in our cohort of FT families was calculated with Fisher’s exact test, comparing the distribution of rare FGF5 variants in seven index cases (two index cases from FT1 and FT2 and five additional index cases) to those reported in the National Heart Lung and Blood Exome Sequencing Project (8).

Functional Predictions of Nonsynonymous Mutations.

The crystal structure of FGF5 was based on a close protein homolog, FGF20. PyMol was used to generate a ribbon diagram of the protein, with the location of the substituted amino acid highlighted. We used three functional mutation prediction programs to determine the effect of p.Y174H on the protein. Polyphen-2 (14) generates a qualitative prediction based on sequence, phylogenetic, and structural information, with output scores close to 1 confidently predicting a damaging mutation. The program SIFT (16) predicts if mutations are deleterious or tolerable based on both sequence homology and the physical properties of amino acids. A score close to 0 is predicted to be highly deleterious. Lastly, Mutation Assessor (15) creates a functional impact score based on amino acid conservation, with scores above 1.9 indicating mutations that are predicted to have a medium or high impact on the protein function.

Plucked Hair Fiber Analysis.

Hair fibers were plucked from the forearm of one affected adult male in FT1, one affected adult male in FT2, and two ethnically matched adult male controls. Individual hairs were straightened to extend, and measured in millimeters from the root tip through to the tapered end under a dissecting microscope. Broken hairs were discarded, leaving a lower number of total hair fibers that were used (48 from FT1, 43 from FT2, and 20 and 29, respectively, from the controls) in the final analysis. We calculated variance between samples using an F test and then assessed if there were significant differences in hair length between patient and control hairs using an unpaired two-tailed t test.

Fibers were also categorized into either anagen, or telogen groups based on the morphology of the root tip end. Anagen fibers have an elongated end, and often appear curled, or distorted at their tip. Comparatively, telogen fibers have a club shaped root tip. Percentages of fibers in either anagen or telogen were then calculated for affected males and controls.

Hair Follicle Organ Culture.

Hair follicles were obtained as discarded tissue after hair transplantation procedures, in accordance with Helsinki guidelines after approval from the Columbia University Institutional Review Board. Follicles were microdissected from intact scalp skin, carefully cleaned of excess tissue, and transected just below the site of attachment of the sebaceous gland to the follicle. Each follicle was cultured in 1 well of a 24-well plate containing Williams Medium E (Sigma-Aldrich) supplemented with 2 mM glutamine, 1× antibiotics-antimycotics, 10 ng/mL hydrocortisone, and 10 mg/mL insulin. Recombinant long form FGF5 (R&D Systems) was reconstituted in sterile PBS containing 1 μg/mL heparin and 0.1% BSA for a stock solution of 10 µg/mL. For treatment of hair follicles, recombinant FGF5 was used, diluted 1/1,000, for a working concentration of 10 µg/μL. Control follicles were cultured in medium with the addition of heparin and BSA only. Follicles from two different control individuals were used in this analysis, for a total of 8 control follicles, and 10 follicles treated with FGF5. Media was changed every 3 d, but follicles were photographed each day to measure fiber growth rates. Fiber lengths were measured in millimeters, as the total height of the follicle with the new fiber, minus the starting height, and plotted as a growth rate graph using Microsoft Excel. Error bars were based on SE, whereas an unpaired two-tailed t test was performed on follicles at day 8, to compare differences between follicles in the treatment (FGF5) group versus control hair follicles.

PCR.

Primers were designed against the FGF5 sequence deposited in the NCBI database (NM_004464). A forward primer (5′-CGCTATGTCTTCCTCTTCTGC-3′) was designed within the first exon of FGF5 and a reverse primer, spanning exons 1 and 2 (5′-CAAAACACTTAACATATTGGCTTCG-3′) to amplify FGF5, or spanning exons 1 and 3 (5′-TGAACTTGGCTTAACATATTGGC-3′) to amplify FGF5S. Forward (5′-CTCGGAGGATGATGATGATG-3′) and reverse (5′-CCTCCAATTCTGTGGTCAGG-3′) primers for the FGF5 receptor, FGFR1, were also designed. Finally, forward (5′-CACAGCCCAAGATAGTTAAGTG-3′) and reverse (5′-GCATAAAGTGTAAGTGTATAAGCATA-3′) beta-2-microglobulin primers were used as housekeeping controls. PCR was performed on cDNA from whole scalp skin, plucked hair fibers, microdissected dermal papillae, whole blood peripheral blood mononuclear cells, and T cells isolated from blood. cDNA was amplified using Platinum PCR Supermix for 30 cycles, with an annealing temperature of 55 °C, and a 30-s extension at 72 °C. Products were separated on 2% (wt/vol) agarose gels.

Whole Mount Immunofluorescence.

To perform whole mount immunofluorescence, plucked forearm hair fibers from affected males in FT1 and FT2, or from ethnically matched controls, were adhered to microscope slides using crazy glue, leaving the root tips exposed on the lower half of the slide. Hairs were rehydrated for 15 min in PBS pH 7.4 before fixation for 10 min in 4% paraformaldehyde. Hairs were then washed in 3 × 10 min washes of PBS and placed in blocking solution [2% (wt/vol) fish skin gelatin, 1% (vol/vol) donkey serum, 0.5% Triton X-100 in PBS] for 1 h. Primary antibodies were applied overnight at 4 °C in a humidified chamber, both diluted 1/100. We used two FGF5 antibodies; a polyclonal antibody D01P (Abnova), raised against amino acids 1–123, predicted to recognize both FGF5 and FGF5S, and a polyclonal antibody (Abbiotec) that recognizes a 27-amino-acid epitope at the C terminus, and therefore recognizes only FGF5, not FGF5S. After primary antibody incubation, hairs were washed 3 × 10 min in PBS, and a donkey anti-rabbit 488-conjugated secondary antibody (Molecular Probes) at 1:800 was applied for 1 h at room temperature. Antibodies were washed off for 3 × 10 min in PBS, and coverslips applied using Vectashield containing DAPI (Vector Labs). Hairs were visualized on a LCM Excitor confocal microscope (Zeiss) and both patient and control hairs were photographed under the same conditions.

Immunohistochemistry.

For immunohistochemistry to localize FGF5, we used 8-µM paraffin sections of normal occipital scalp tissue adhered to positively charged slides. Slides were deparaffinized with xylene then rehydrated through decreasing gradients of ethanol. Antigen retrieval was performed by placing slides in boiling sodium citrate buffer (pH 6) for 5 min, then allowed to cool to room temperature. Slides were washed in Tris-buffered saline pH 8.4 (3 × 3 min), then incubated with blocking solution [2% (vol/vol) goat serum, 1% (wt/vol) BSA, 0.1% fish skin gelatin, 0.1% Triton X-100, and 0.05% Tween-20 in TBS] for 1 h. A primary antibody against FGF5 (Abbiotec) was diluted 1/100 and incubated on slides overnight at 4 °C in a humidified chamber. After washing off primary antibody with 3 × 3 min PBS washes, a secondary antibody, biotin-XX goat anti-rabbit (Molecular Probes) at 1/1000 was applied for 45 min. This was washed off with TBS, and a tertiary, streptavidin-alkaline phosphatase (Zymed) at 1/300 was applied for 30 min at room temperature. Slides were washed and alkaline phosphatase was visualized using Sigma Fast Red (Sigma) for ∼10 min. Slides were washed and counterstained using Hematoxylin QS (Vector Labs) before mounting coverslips using glycergel.