Genetic Variation in the Human Androgen Receptor Gene Is the Major Determinant of Common Early-Onset Androgenetic Alopecia

Abstract

Androgenetic alopecia (AGA), or male-pattern baldness, is the most common form of hair loss. Its pathogenesis is androgen dependent, and genetic predisposition is the major requirement for the phenotype. We demonstrate that genetic variability in the androgen receptor gene (AR) is the cardinal prerequisite for the development of early-onset AGA, with an etiological fraction of 0.46. The investigation of a large number of genetic variants covering the AR locus suggests that a polyglycine-encoding GGN repeat in exon 1 is a plausible candidate for conferring the functional effect. The X-chromosomal location of AR stresses the importance of the maternal line in the inheritance of AGA.

Androgenetic alopecia (AGA [MIM 109200]), or male-pattern baldness, is characterized by a defined pattern of hair loss from the scalp (Hamilton 1951). In whites, the proportion of affected males increases steadily with age, so that a male in his 50s has a 50% chance of having some degree of AGA (Hamilton 1951). Association of AGA with a variety of clinical phenotypes has been suggested, including coronary heart disease (Lotufo et al. 2000), benign prostatic hyperplasia (Hawk et al. 2000), prostate cancer (Oh et al. 1998), and disorders associated with insulin resistance (Matilainen et al. 2000). Androgen dependence is an important characteristic of AGA, and genetic disposition, which is assumed to be polygenic, plays the most substantial role in the development of AGA (Küster and Happle 1984; Ellis et al. 1998; Nyholt et al. 2003). Although the only factor known to influence onset age in patients with AGA is genetic predisposition, no systematic approach has hitherto been undertaken, to our knowledge, to identify the contributing genes.

As part of a genomewide linkage study of AGA, we investigated linkage to markers covering the X chromosome. The sample consisted of 95 families in which at least two brothers had early-onset AGA (391 genotyped individuals, including 201 affected men). We obtained evidence of linkage in chromosomal region Xq12-22 (nonparametric linkage [NPL] score of 2.70 [fig. 1]) (for X-chromosomal marker information, see table 1). This region contains the AR gene (MIM 313700), which is an obvious candidate for explaining the development of AGA, and an association with this region has been suggested elsewhere (Ellis et al. 2001), on the basis of results from the investigation of three polymorphic sites. In the present study, we have systematically explored the contribution of the AR gene to the development of early-onset AGA.

Figure 1.

Multipoint NPL analysis of chromosome X, calculated by the Allegro v1.2 software (Gudbjartsson et al. 2000). The X-axis is the chromosome location (top, cM; bottom, STR marker), and the Y-axis is the LOD score.

Table 1.

STR Markers Used for Linkage Analysis of Chromosome X

Marker	No. of Alleles	Location(cM)	Heterozygosity
DXS996	15	8.80	.9030
DXS7108	9	14.20	.6810
DXS9902	8	22.00	.6700
DXS7593	11	29.70	.8050
DXS1226	13	33.00	.8190
DXS1202	9	38.40	.7130
DXS9896	12	39.50	.7560
DXS9907	7	45.00	.4950
DXS1049	3	51.50	.5320
DXS1068	9	52.60	.6880
DXS6810	6	63.30	.6600
GATA144D04	9	71.30	.7340
DXS1003	9	72.40	.7140
DXS7132	6	83.30	.7170
DXS10070	8	83.59	.6740
DXS10071	3	83.62	.3180
DXS10072	4	84.11	.2110
DXS10073	5	84.35	.3330
AR_CAG	20	84.780	.9470
AR_GGN	11	84.790	.5210
DXS6800	8	93.20	.7230
DXS990	8	99.70	.7250
DXS6789	9	103.60	.7420
DXS6797	8	112.90	.7340
GATA172D05	7	116.20	.7870
DXS8081	5	120.50	.6040
DXS8067	10	130.40	.6880
DXS1212	6	130.40	.7310
GATA165B12	3	133.20	.6430
DXS1047	16	143.80	.8560
DXS8072	7	144.80	.7140
D3S2390	8	154.29	.8190
DXS8106	9	162.60	.7140
DXS7127	9	164.92	.6960
DXS998	6	173.40	.6170
DXS8069	4	173.40	.9290
DXS1073	10	184.30	.5640

We tested 39 SNPs, two STRs (a polyglutamine-encoding CAG repeat and a polyglycine-encoding GGN repeat in exon 1 of AR), and one biallelic insertion polymorphism (XARx8insA), in a range of 4.4 Mb at the AR locus, for association with AGA (fig. 2). Our analysis included the three previously studied variants (Ellis et al. 2001): the CAG and GGN repeats and StuI RFLP (rs6152). TaqMan assays for genotyping of SNPs, listed in table 2, were designed by Applied Biosystems. PCRs were performed using 2.5 ng genomic DNA, AmpliTaq Gold DNA polymerase (Applied Biosystems), and an annealing temperature of 60°C for 45 cycles on Biometra T1 (Biometra) or Perkin Elmer GeneAmp 9700 (Applied Biosystems) thermocyclers. Fluorescence was measured with an ABI Prism 7900HT sequence detection system (Applied Biosystems). The fragment lengths of the CAG repeat, the GGN repeat, and XARx8insA were determined using the following fluorescence-labeled primers for PCR: CAG-F 5′-TCCAGAATCTGTTCCAGAGCGTGC-3′ and CAG-R 5′-GCTGTGAAGGTTGCTGTTCCTCAT-3′ (La Spada et al. 1991), GGN-F 5′-CCTGGCACACTCTCTTCACA-3′ and GGN-R 5′-GGATAGGGCACTCTGCTCAC-3′, and XARx8insA-F 5′-CACGGGAAGTTTAGAGAGCT-3′ and XARx8insA-R 5′-TCACCTTCTCGTCACTATTG-3′. Forty nanograms of genomic DNA was used in PCRs with AmpliTaq DNA polymerase (Applied Biosystems) in 38 cycles on a PTC-200 (MJ Research). The annealing temperatures were 62°C–60°C (touchdown PCR) for the CAG repeat, 58°C for the GGN repeat, and 63°C–55°C (touchdown PCR) for XARx8insA. MasterAmp PCR PreMix G (Epicentre) was used to amplify GGN fragments. Fragment lengths of the amplified products were analyzed on an ABI Prism 377 DNA sequencer (Applied Biosystems). Double-strand sequencing of genomic regions was performed with the ABI Prism BigDye Terminator Cycle Sequencing kit, version 2.0 (Applied Biosystems), and the ABI Prism 3730 DNA analyzer (Applied Biosystems).

Figure 2.

Gene and LD structure of the AR locus. A, Distribution of known genes (left) and typed SNPs and STRs (right) at the AR locus. The shown genomic region spans 4.4 Mb. Gene content information is based on Ensembl. B, LD in 198 individuals with AGA (upper right diagonal) and in 157 individuals without AGA (lower left diagonal) was measured with χ² and was visualized using the GOLD program (Abecasis and Cookson 2000).

Table 2.

Oligonucleotides for TaqMan Assays^[Note]

SNP and Type of Oligonucleotide	Sequence of Oligonucleotide
rs1332720:
Primer-F	CCCATAGAGGGCAACCACTTG
Primer-R	GCACTGAAAAGGAAGTGAACAAAGG
Probe-1	CCAGAGTGATTTCTC
Probe-2	CCAGAGTAATTTCTC
rs1379146:
Primer-F	TGGAAAATATGGATGCAGGTAGCTT
Primer-R	AGCATGTCAGGCTCTTCTAAGAC
Probe-1	CCCTCAAGAGTTTTTGA
Probe-2	CCTCAAGAGATTTTGA
rs1158928:
Primer-F	CCTTTTTTACTGTATACATTAAGAACTTATTGTTGCT
Primer-R	GGTGGTGTGTAAATCACTTATAAATCTGGAA
Probe-1	AAAGGTTAAGAGACAAAAG
Probe-2	AAAAGGTTAAAAGACAAAAG
rs1385695:
Primer-F	ACACATCAAACCCACGCATAAAC
Primer-R	ACACTGAGGAAACTGTACAGCATTT
Probe-1	ACTCCATTTAGGCTATTT
Probe-2	CACTCCATTTAAGCTATTT
rs775362:
Primer-F	CTGCTTAATCCTATTCAAAGTCCTGGTT
Primer-R	AGCTCAGTCAGTTAAGCCACATTTA
Probe-1	ACTTTGCAGAATTCGTTGTG
Probe-2	CTTTGCAGAATTCATTGTG
rs1041668:
Primer-F	GCCTCCCTTTACCTACTAGACAACA
Primer-R	CAAATCACACTTAGGAGGTGCTACA
Probe-1	ATAAAATCAACTTCATGAAGTAG
Probe-2	AAAATCAACTTCATAAAGTAG
rs5964577:
Primer-F	CGTCTGGGAAATAATATTCAAAATCCAAAACA
Primer-R	AAGTGCACTGAGTGTTACAAAAAAGAAG
Probe-1	CACCAGTTGATACTTATTAG
Probe-2	CCAGTTGATACATATTAG
rs532649:
Primer-F	ACAACCAATATCTTGATCTATGTCACCG
Primer-R	TGTACTGGAGTCCCACAGATTCTAT
Probe-1	ATATCAAGTAGTATAAAGCTCA
Probe-2	ATCAAGTAGTATAGAGCTCA
rs989345:
Primer-F	CAGAGAAAGCACAAAATGGCTACAA
Primer-R	ACACTGAGTGAATATCTGAACACCTCTA
Probe-1	AGTCTGAGAATATACGCAAAG
Probe-2	CTGAGAATATGCGCAAAG
rs938059:
Primer-F	CTGGTGAACAATAAATTTGATATAAAAGCTATTTAATACATAGA
Primer-R	AACTCTTGAAGTGCCCTTCTTAATTCA
Probe-1	CATTTTCCATAGGTCCTCTAT
Probe-2	ATTTTCCATAGGTACTCTAT
rs925391:
Primer-F	GCTTGGGCAGGAAGGAGAAC
Primer-R	TGAGAAGGCCATATTCCAAAGCATT
Probe-1	TTCACATATAATCATGATATAAG
Probe-2	TTCACATATAATCATAATATAAG
rs5919287:
Primer-F	GGACTAGAAGAAGGGAAGACAATGG
Primer-R	GGTAAGTGGAAGGTACAATGCCAAA
Probe-1	TAGAGATGTTACCCCTGGTTT
Probe-2	TAGAGATGTTACCCTTGGTTT
rs2221799:
Primer-F	ACATGAAAGTTTGAAAGTCCAATCCTACT
Primer-R	GGATTTCAAGAGCAAAGTTTCAGACAA
Probe-1	TTCAAGATGAAGTATAAGTTAG
Probe-2	TTCAAGATGAAGTATAGGTTAG
rs10521339:
Primer-F	CATCTGTGCAGCTCTTTATGGATTG
Primer-R	TCAGGTCTTAGAAAGTTAGAACTAGAAGGG
Probe-1	AAAAGATTCTCATATAATACTAAC
Probe-2	AAAAGATTCTCATATTATACTAAC
rs2223841:
Primer-F	CTGGGCACTTTTAGCCTATTCAGAT
Primer-R	GGAATGAGACCCCTGTCAAGAA
Probe-1	CCAGAGTGCTAGTGGAGT
Probe-2	CAGAGTGCTGGTGGAGT
rs2207081:
Primer-F	GGATGAATTGTACTTACAAACTGTGACTAAC
Primer-R	GCAAATTGTTCAAAGATTAGTCACTTGGATA
Probe-1	TCTTTTAAAACAGAGCTTC
Probe-2	TTTTAAAGCAGAGCTTC
rs2497935:
Primer-F	GATCAAGCATGCCTGGTAAACTG
Primer-R	AGGCTCCTTGATCCTCTGTCAT
Probe-1	CTGCCAGAAGCTTAT
Probe-2	TGCCAGAGGCTTAT
rs962458:
Primer-F	ACCAAACCAATGACTTCTGGCTTTA
Primer-R	GTGCAGCAAGTATATGAATGAATAAACAAATCT
Probe-1	TGTGCGATGTTTTC
Probe-2	ATGTGCAATGTTTTC
rs6152:
Primer-F	AGCAGCAGCGGGAGAG
Primer-R	GTGCCCCCTAAGTAATTGTCCTT
Probe-1	CCCGAGGCTTCCCT
Probe-2	CCGAGGCCTCCCT
rs1337080:
Primer-F	TTTTTTAGGTAGAGACTCCAACATCATTACAG
Primer-R	CTCAACCCCACAGCATTTTATTTTCTAT
Probe-1	AACTATAAATTACATATGGAAAAG
Probe-2	TATAAATTACATGTGGAAAAG
XARx7_01:
Primer-F	GGTCAGAAAACTTGGTGCTTTGT
Primer-R	TGATGTGGGATGAAGACAGAATGG
Probe-1	CTCCTTCATGGGCATG
Probe-2	TCCTTCGTGGGCATG
rs2781516:
Primer-F	CGATGATGTTGGTTTCCTTGGAAGT
Primer-R	TCTGTTCTGACATGCAAATACTCTTTTCA
Probe-1	CACAGAGAAATAAATAACATTAA
Probe-2	CAGAGAAATAAATGACATTAA
rs2885913:
Primer-F	GCTCATCCAGGGAACAGATATAGGA
Primer-R	CAGTTTCTATCTCTGGAGGACTAGCA
Probe-1	TCAAAATGTGGAGTGACCA
Probe-2	CAAAATGTGGAATGACCA
rs2363785:
Primer-F	GTCTCTGGTGCCTCAGTGAAT
Primer-R	TCATAAACATCATGGCCTCAGCTT
Probe-1	CAATTATTCAGAGCAATTC
Probe-2	TTATTCCGAGCAATTC
rs1936313:
Primer-F	CCTTTAACCAAACTCTGGAGACACA
Primer-R	ACCCCATGAGAACTGCAATATGTC
Probe-1	CAAGCACCACTTTTACA
Probe-2	CAAGCACCATTTTTACA

Note.— Information about TaqMan “Assays on Demand” for SNPs rs12560201, rs1044165, rs708969, rs1485682, rs1385699, rs1204038, rs5919393, rs492933, rs1410127, rs1157321, rs7887862, rs1927232, rs5965536, and rs792952 was not provided by Applied Biosystems.

For case-control analysis, we compared allele frequencies in 198 males with early-onset AGA (including 95 unrelated and randomly chosen affected individuals from the linkage-analysis families), 188 control individuals, and 157 unaffected individuals. For family-based association analysis, we studied 179 families containing at least one affected male in the youngest generation. The family-based association sample included the 95 families from the linkage analysis and overlapped with the case sample, for 179 individuals. All affected males were aged <40 years (mean [± SD] 32.0±5.2) and had AGA that was representative of the most severely affected 10% of the distribution for the respective age class, on the basis of the classification of Hamilton (1951) (modified by Norwood [1975]). AGA classification, age, and ethnicity were the exclusive criteria used to select individuals for inclusion in the case sample. Unaffected males were aged >60 years (mean 67.9±6.2) and without AGA, representing the least-affected 20% of the distribution for this age class. Families and unrelated individuals both with and without AGA were recruited through various sources, including press reports and advertisements in magazines, newspapers, and placards. Control individuals were male blood donors from the blood transfusion center of the University Hospital Bonn, from whom information was available only on sex, age in years (mean 29.4 ± 8.6), and ethnicity. EDTA anticoagulated venous blood samples were collected from all individuals, and lymphocyte DNA was isolated by salting out with saturated NaCl solution (Miller et al. 1988). All participants were of German descent. The study was approved by the ethics committee of the University of Bonn, and informed consent was obtained from all participants.

Association analysis was conducted using a modification of the FAMHAP software (Becker and Knapp 2004a, 2004b) for X-chromosomal data. Case-control SNP single-marker analysis was performed using the χ² distribution of the 2×2 contingency table, and multiallelic markers were evaluated with the permutational version of the χ² test. For case-control haplotype analysis, P values were calculated from the χ² distribution with n−1 df of the respective likelihood-ratio test (n = number of different haplotypes). The family data were analyzed with the permutation-based association test for nuclear families (Zhao et al. 2000; Knapp and Becker 2003). For each marker (single-locus analysis) and each marker combination (haplotype analysis), we used 10¹⁰ permutation replicates.

Pairwise distances between haplotypes were calculated as allele mismatches. By resampling markers randomly with replacement, 100 bootstrapped data sets were generated as input for the program neighbor contained in the PHYLIP package.

A region of 1 Mb showed strong association with the lowest P value of 2.1×10^-12 for rs10521339 in the case-control analysis of affected and unaffected individuals (table 3). As expected, the association is stronger for comparisons between individuals with AGA and individuals without AGA than between individuals with AGA and an unselected control sample (table 3). We also performed a separate analysis of the 103 cases that were not included in the linkage analysis and found that the association was also present in this sample (data not shown). The association is further supported by family-based analysis, for which SNP rs938059 shows the lowest P value (4.03×10^-6) (table 3). AR is the only known gene in the strongly associated 1-Mb–spanning region (fig. 2 and table 3). The X-linked ectodysplasin-A2 receptor (XEDAR), which is located 900 kb 5′ of AR, is outside this region (fig. 2 and table 3). The significance of the association decreases within the 3′ part of the 180-kb–spanning AR gene, and oligophrenin 1 (OPHN1), which is located 320 kb 3′ of AR, is not within the block of strongest association (fig. 2 and table 3). rs6152 (P=6.66×10^-10 [table 3]) in exon 1 of AR corresponds to the StuI RFLP, for which an association has been described by Ellis et al. (2001).

Table 3.

Association of AGA with Markers at the AR Locus^[Note]

	Frequency			TDTa			Affected/Control Analysis			Affected/Unaffected Analysis
SNP andAllele	Affected	Control	Unaffected	Transmitted	Nontransmitted	Pc	P	ORd	CI	P	ORd	CI	Positionb
rs12560201:						6.68 × 10−1	8.99 × 10−1	1.04	.55–1.82	9.38 × 10−1	.97	.5–1.87	63,335,397
G	.112	.108	.115	23	27
A	.888	.892	.885
rs1332720:						1.00	1.00	ND	ND	1.00	ND	ND	64,598,557
C	.995	1.000	1.000	0	1
T	.005	.000	.000
rs1044165:						1.00	9.50 × 10−1	.98	.55–1.89	1.58 × 10−1	1.53	.85–2.76	65,024,747
A	.124	.122	.178	22	23
G	.876	.878	.822
rs708969:						5.35 × 10−1	1.21 × 10−2	2.02	1.16–3.53	3.90 × 10−3	2.27	1.29–4	65,252,916
A	.875	.776	.755	36	29
T	.125	.224	.245
rs1379146:						3.25 × 10−1	2.68 × 10−4	2.61	1.55–4.45	4.72 × 10−4	2.60	1.51–4.5	65,389,382
T	.872	.723	.724	39	29
A	.128	.277	.276
rs1158928:						1.30 × 10−1	1.04 × 10−5	3.23	1.89–5.56	2.08 × 10−5	3.22	1.85–5.61	65,532,217
A	.883	.700	.701	42	27
G	.117	.300	.299
rs1485682:						1.30 × 10−1	1.97 × 10−5	3.13	1.82–5.35	3.08 × 10−5	3.16	1.81–5.48	65,598,612
C	.883	.707	.705	42	27
T	.117	.293	.295
rs1385699:						1.22 × 10−1	1.60 × 10−5	3.23	1.86–5.59	9.41 × 10−5	3.00	1.71–5.32	65,608,007
A	.887	.708	.722	44	28
G	.113	.292	.278
rs1385695:						2.97 × 10−3	1.03 × 10−5	4.68	2.24–9.78	2.40 × 10−8	6.57	3.17–13.63	65,687,533
A	.949	.799	.739	38	13
G	.051	.201	.261
rs775362:						4.10 × 10−3	2.09 × 10−5	4.39	2.13–9.21	3.73 × 10−8	6.40	3.11–13.39	65,762,194
G	.949	.809	.744	37	13
A	.051	.191	.256
rs1041668:						1.49 × 10−4	1.80 × 10−6	6.42	2.74–15.03	1.58 × 10−9	9.29	4.03–21.44	65,868,977
G	.036	.194	.258	9	39
A	.964	.806	.742
rs5964577:						2.50 × 10−3	4.24 × 10−6	4.99	2.32–10.68	1.20 × 10−8	7.02	3.29–15.08	65,954,512
A	.954	.806	.747	36	12
T	.046	.194	.253
rs532649:						5.96 × 10−6	5.42 × 10−7	6.38	2.75–14.71	2.33 × 10−10	9.62	4.17–22.13	66,050,269
A	.960	.807	.735	37	7
G	.036	.193	.265
rs989345:						4.65 × 10−6	1.60 × 10−6	6.24	2.74–14.66	6.13 × 10−10	9.56	4.2–22.29	66,084,055
A	.964	.811	.737	36	6
G	.036	.189	.263
rs938059:						4.03 × 10−6	1.10 × 10−6	6.55	2.82–15.25	9.20 × 10−10	9.63	4.18–22.17	66,119,748
C	.036	.195	.263	7	40
A	.964	.805	.737
rs925391:						4.76 × 10−6	5.97 × 10−7	7.28	3.03–18.03	1.98 × 10−10	11.15	4.65–27.42	66,123,458
G	.969	.811	.737	38	6
A	.031	.189	.263
rs5919287:						4.62 × 10−6	2.57 × 10−7	6.61	2.87–12.82	1.76 × 10−10	9.70	4.22–22.45	66,143,635
C	.964	.802	.734	39	7
T	.036	.198	.266
rs2221799:						4.62 × 10−6	3.37 × 10−7	6.53	2.84–15.18	1.39 × 10−9	9.21	4.04–21.47	66,159,637
A	.964	.804	.744	39	7
G	.036	.196	.256
rs10521339:						4.00 × 10−4	1.95 × 10−7	5.65	2.65–12.04	2.10 × 10−12	9.89	4.62–20.7	66,291,407
A	.954	.778	.677	11	41
T	.046	.212	.323
rs2223841:						9.40 × 10−6	1.20 × 10−6	5.59	2.62–11.94	1.72 × 10−11	9.48	4.48–20.05	66,353,192
A	.954	.789	.688	44	11
G	.046	.211	.312
rs2207081:						4.33 × 10−6	7.00 × 10−7	6.21	2.8–13.78	5.04 × 10−12	10.72	4.89–23.75	66,356,719
A	.959	.792	.688	44	10
G	.041	.208	.312
rs2497935:						1.79 × 10−5	3.10 × 10−6	5.00	2.41–10.36	3.17 × 10−11	8.69	4.23–17.87	66,447,287
A	.949	.790	.684	45	12
G	.041	.208	.316
rs962458:						2.89 × 10−3	2.87 × 10−4	7.32	2.12–25.3	2.56 × 10−6	11.10	3.27–37.74	66,528,985
G	.015	.102	.146	2	17
A	.985	.898	.854
CAGe:						2.71 × 10−1 f							66,548,181
18	.042	.085	.110	13	12		NS	ND	ND	NS	ND	ND
19	.110	.106	.110	30	39		NS	ND	ND	NS	ND	ND
20	.194	.112	.103	50	19		2.66 × 10−2	1.91	1.07–3.41	2.20 × 10−2	2.10	1.10–3.99
(continued)
21	.147	.191	.205	32	31		NS	ND	ND	NS	ND	ND
22	.136	.122	.096	32	28		NS	ND	ND	NS	ND	ND
23	.126	.090	.068	26	30		NS	ND	ND	NS	ND	ND
24	.084	.080	.096	19	36		NS	ND	ND	NS	ND	ND
25	.047	.064	.075	12	14		NS	ND	ND	NS	ND	ND
rs6152:						5.88 × 10−5	3.80 × 10−6	5.26	2.45–11.3	6.66 × 10−10	8.21	3.86–17.45	66,548,648
T	.046	.201	.282	11	40
C	.954	.799	.718
GGNe:						7.79 × 10−5 f							66,549,360
23	.651	.495	.421	76	43		2.06 × 10−3	1.91	1.26–.2.88	2.07 × 10−5	2.57	1.656–3.98
24	.182	.367	.461	21	68		5.39 × 10−5	.38	.24–.62	2.62 × 10−8	.26	.161–.42
rs1204038:						9.79 × 10−5	1.90 × 10−6	5.45	2.55–11.67	2.83 × 10−9	7.72	3.62–16.46	66,571,246
A	.046	.208	.271	11	39
G	.954	.792	.729
rs5919393:						1.58 × 10−5	6.00 × 10−7	5.82	2.73–12.41	1.69 × 10−8	7.88	3.7–16.77	66,608,378
C	.046	.218	.274	11	38
T	.954	.782	.726
rs1337080:						5.73 × 10−3	6.71 × 10−4	6.73	1.92–23.14	4.75 × 10−6	10.78	3.12–36.19	66,661,940
A	.985	.907	.859	15	2
G	.015	.093	.141
XARx7_001g:						1.00	6.79 × 10−2	3.95	.81–19.25	1.45 × 10−1	3.21	.61–16.76	66,725,646
A	.010	.039	.032	3	3
G	.990	.961	.968
XARx8insAh:						3.29 × 10−3	5.09 × 10−5	4.11	1.99–8.47	6.62 × 10−8	5.94	2.94–12.02	66,727,141
delA	.944	.804	.739	34	13
A	.056	.196	.261
rs2781516:						3.28 × 10−2	1.00 × 10−2	2.02	1.18–3.48	2.86 × 10−4	2.69	1.56–4.64	66,885,052
A	.128	.228	.282	21	38
G	.872	.772	.718
rs2885913:						4.46 × 10−2	1.49 × 10−4	2.31	1.49–3.57	4.29 × 10−5	2.53	1.62–3.97	66,898,824
G	.255	.441	.465	40	64
A	.745	.559	.535
rs2363785:						1.73 × 10−2	1.82 × 10−4	2.28	1.47–3.53	2.04 × 10−4	2.34	1.49–3.68	66,954,319
G	.745	.562	.555	65	37
T	.255	.438	.445
rs1936313:						2.48 × 10−2	7.91 × 10−5	2.37	1.54–3.64	9.15 × 10−4	2.15	1.36–3.37	66,983,669
T	.744	.551	.575	65	39
C	.256	.449	.425
rs492933:						2.44 × 10−2	1.15 × 10−2	1.76	1.13–2.72	2.39 × 10−2	1.69	1.07–2.65	67,046,865
C	.736	.614	.623	67	40
T	.264	.386	.377
rs1410127:						2.83 × 10−2	8.02 × 10−3	1.79	1.16–2.76	2.13 × 10−2	1.71	1.08–2.67	67,063,402
C	.735	.608	.619	66	40
T	.265	.392	.381
rs1157321:						7.83 × 10−4	1.59 × 10−5	2.89	1.75–4.7	2.73 × 10−5	2.93	1.75–4.87	67,104,762
G	.848	.659	.656	64	27
T	.152	.341	.344
rs7887862:						1.09 × 10−1	3.31 × 10−1	1.34	.74–2.41	1.28 × 10−1	1.65	.86–3.2	67,148,136
C	.152	.118	.098	36	21
T	.848	.882	.902
rs1927232:						6.77 × 10−1	1.77 × 10−1	1.36	.87–2.11	8.23 × 10−1	1.06	.66–1.69	67,165,912
C	.719	.653	.708	49	44
T	.281	.347	.292
rs5965536:						1.56 × 10−1	3.57 × 10−1	1.33	.73–2.41	9.50 × 10−1	1.02	.57–1.84	67,300,796
C	.154	.121	.151	35	24
G	.846	.879	.849
rs792952:						1.39 × 10−1	8.45 × 10−1	1.05	.65–1.69	8.56 × 10−1	1.05	.64–1.71	67,766,144
C	.249	.240	.240	53	36
G	.751	.760	.760

Note.— ND = not determined; NS = not significant.

^aFamily transmission/disequilibrium test (Knapp and Becker 2003).

^bUCSC Genome Bioinformatics Plus Strand Build 35.

^cPermutation-based value for association.

^dOdds ratio.

^eOnly alleles with frequencies ⩾0.05 in controls.

^fGlobal P value.

^gGCTTTGTCTAATGCTCCTTC[G/A]TGGGCATGCTTCCCCTCCCC (intron 6 of AR).

^hACAAGCAAACAAAAAAAAAA[A/delA]GCAAAAACAAAACAAAAAAT (exon 8 of AR).

The long range of the associated SNPs implies the presence of a large haplotype block, visible in fig. 2B. In principle, this is not unexpected in a location close to the centromere, where recombination events occur at a relatively low frequency (Nagaraja et al. 1997). To test whether the size of the haplotype block stands out even in comparison with X-chromosomal loci with similar low recombination frequencies, we analyzed average pairwise linkage disequilibrium (LD) (measured by |D′|) between SNPs retrieved from the HapMap Project database in 1-cM–sized windows. Average pairwise LD was found to be inversely correlated with recombination rate (Spearman’s ρ=-0.598; P<.001). An average LD higher than that at the AR locus was displayed by only the six windows covering the centromere; these windows showed distinctively smaller recombination rates (fig. 3A). This result might suggest that the predominant AR haplotypes are evolutionarily more recent, perhaps indicating positive selective pressure acting at this locus (Bamshad and Wooding 2003). Since androgens mediate a wide range of developmental and physiological responses through the androgen receptor (AR) and are especially important in the male reproductive system (Lee and Chang 2003), it is conceivable that variability in AR can have an impact on selection. In accordance with this hypothesis, the haplotype with the highest frequency (0.45 [fig. 3B]) (which also confers risk of AGA) in the German population seems to be evolutionarily recent, as indicated by the low sequence identity with the ancestral haplotype (fig. 3B).

Figure 3.

The AR locus: LD relative to the X-chromosome and haplotype structure. A, Average pairwise LD (measured by |D′|) between SNPs retrieved from the HapMap Project database, plotted over X-chromosomal recombination rates in 1-cM–sized windows. An average LD higher than that at the AR locus was displayed by only the six windows covering the centromere and showed distinctively smaller recombination rates. B, Neighbor-joining tree of frequent haplotype sequences (>5% in the samples from the individuals with AGA, controls, and individuals without AGA) within the most strongly associated haplotype block (genetic markers rs1385695–XARx8insA). The CAG repeat, XARx7_01, and XARx8insA were omitted from tree construction, because the corresponding Pan troglodytes allele could not be retrieved from the chimpanzee genomic sequence. The AGA haplotype is shown in red, and the chimpanzee haplotype is shown in orange. The AGA haplotype shows low sequence identity to the ancestral (chimpanzee) haplotype. Haplotype frequencies of the respective samples are indicated in parentheses (affected/control/unaffected [in %]). GGN-23 is indicated as “23”; GGN-24 is indicated as “24.”

Sequencing of the transcribed region and of 3.4 kb of conserved sequences in the 5′ region and intron 1 of AR in 12 individuals revealed only two additional variants of the associated haplotype (XARx7_01 and XARx8insA [table 3]). The additional variants were noncoding and did not show a stronger association than other tested markers. Of the two repeat polymorphisms in the coding region, the CAG repeat was not associated with AGA (affected/unaffected global P value of .1), whereas the GGN repeat was highly associated (affected/unaffected global P value of .0001). The study by Ellis et al. (2001) also showed a larger effect for the GGN repeat than the CAG repeat. However, the pooling of alleles in their study renders an exact comparison of results difficult. The GGN allele of 23 repeats showed a difference of affected versus unaffected allele frequencies of 0.23, which was in the range of the strongest-associated SNPs (rs1385695–XARx8insA [table 3]) but with a lower frequency in controls. This suggests that GGN-23 either is closer to the AGA mutation or is itself the AGA-susceptibility allele. Previously obtained functional data, in which shorter repeat alleles of the GGN repeat were associated with higher protein levels and thereby higher AR activity (Ding et al. 2005), support the possibility of a causal role for the repeat, and this would be compatible with current understanding of the involvement of androgens in AGA. Several studies have also suggested an effect of CAG repeat lengths on AR transactivating activity (Mhatre et al. 1993; Chamberlain et al. 1994; Kazemi-Esfarjani et al. 1995; Choong et al. 1996; Nakajima et al. 1996; Beilin et al. 2000; Ding et al. 2004). Since both repeats modulate AR activity and since we observe an association only between the GGN repeat and AGA, it may be that cells of the hair follicle lack cofactors that interact with the CAG-encoded domain.

Interestingly, shorter alleles for the GGN repeat have also been associated with prostate cancer (Hsing et al. 2000; Chang et al. 2002), whereas longer alleles have been associated with endometrial cancer (Sasaki et al. 2005), which would be in accordance with the differing effects of androgens on the endometrium and the prostate (androgens exert an inhibitory effect on endometrial cell proliferation, whereas they have a mitogenic effect in the prostate). However, the association findings with prostate cancer remain controversial, and no effect was shown in a large meta-analysis (Zeegers et al. 2004). The association with endometrial cancer has yet to be confirmed.

It remains possible that an as-yet-undetected variant in either a regulatory region affecting the expression level or an intronic variant affecting the splicing pattern of AR might be responsible for AGA susceptibility. The latter seems unlikely since we did not detect alternatively spliced transcripts of AR in human hair follicles of seven individuals representing different haplotypes. Previous studies have identified AR regulatory elements up to position −737 of the AR transcription start site (Faber et al. 1991, 1993; Supakar et al. 1993), as well as exonic enhancers in exon 1 (Faber et al. 1993) and exons 4 and 5 (Grad et al. 2001). Our sequencing analysis of 12 individuals with the associated haplotype revealed no variability in these regulatory elements of AR. However, there may be additional regions with regulatory effect that have not yet been fully characterized (Lower et al. 2004). Haplotypes carrying the GGN-24 allele show clearly higher frequencies in individuals without AGA than in those with AGA (fig. 3B). Since this effect is strikingly weaker in the ACAAAAAGCATTTAAG-24-ATA haplotype than in the other GGN-24–carrying haplotypes (fig. 3B), it is likely that further functionally relevant variability exists that modifies the protective effect of GGN-24–bearing haplotypes.

It is interesting to note that genetic variation in AR, which is located on the X chromosome, cannot explain the resemblance of fathers and sons with respect to the development of AGA (Küster and Happle 1984; Ellis et al. 1998), since sons always inherit the X chromosome from their mothers. The fact that family studies of AGA have typically stressed the resemblance of fathers and sons is understandable, given the differences in patterns of hair loss between males and females. Our genetic data, however, stress the relative importance of the maternal line in the inheritance of AGA, since we estimate an etiological fraction of 0.46 that can be attributed to having ⩽23 GGN repeats within AR. This suggests that the average phenotypic resemblance should be greater between affected males and their maternal grandfathers than between affected males and their fathers. It is likely that the remaining etiological fraction is due to genetic variation at autosomal loci, which could explain the similarity of the AGA pattern of fathers and sons. Some autosomal candidate genes have been investigated in the past, including the insulin gene (Ellis et al. 1999), the 5α-reductase genes (Ellis et al. 1998), and the hairless gene (Hillmer et al. 2001, 2002), but none of these has been associated with AGA. A systematic linkage-based approach should enable the identification of additional loci.