Introduction
Sexual orientation, the sustained erotic attraction to members of one's own sex, the opposite sex, or both – homosexuality, heterosexuality, or bisexuality, respectively – is a primary component of human sexuality. Although Kinsey proposed that sexual orientation exists upon a continuum reflected by his scale (Kinsey et al. Reference Kinsey, Pomeroy and Martin1948), male sexual orientation tends to be bimodally distributed, with most men rating themselves as predominantly heterosexual (Kinsey scale 0–1) or homosexual (Kinsey scale 5–6) (Laumann et al. Reference Laumann, Gagnon, Michael and Michaels1994b ; Pillard & Bailey, Reference Pillard and Bailey1998; Bailey et al. Reference Bailey, Dunne and Martin2000). In contrast, women have a more continuous distribution across the non-heterosexual orientations, lower rates of homosexuality, higher rates of bisexuality, and less temporal stability (Laumann et al. Reference Laumann, Gagnon, Michael and Michaels1994b ; Bailey et al. Reference Bailey, Dunne and Martin2000; Diamond, Reference Diamond2008a , Reference Diamond b ). Although our focus is male sexual orientation, we note that female sexual orientation also appears moderately heritable (Bailey et al. Reference Bailey, Dunne and Martin2000; Alanko et al. Reference Alanko, Santtila, Harlaar, Witting, Varjonen, Jern, Johansson, von der Pahlen and Sandnabba2010; Langstrom et al. Reference Langstrom, Rahman, Carlstrom and Lichtenstein2010; Burri et al. Reference Burri, Cherkas, Spector and Rahman2011). Accurate prevalence estimates are currently difficult to obtain, but the following two conclusions are uncontroversial: homosexual orientation is much less common than heterosexual orientation among men, but it is not rare in respondents in Western industrialized nations. For example, in a large and careful survey of American males approximately 2% rated their sexual orientation as homosexual when defined as identity, 3–4% when defined psychologically (i.e. attraction and fantasy), and a few more did so when defined behaviorally (Laumann et al. Reference Laumann, Gagnon, Michael and Michaels1994a ). Partially independent familial aggregation of male and female homosexuality (Pillard & Weinrich, Reference Pillard and Weinrich1986; Bailey & Bell, Reference Bailey and Bell1993; Bailey & Benishay, Reference Bailey and Benishay1993; Bailey et al. Reference Bailey, Pillard, Neale and Agyei1993, Reference Bailey, Pillard, Dawood, Miller, Trivedi, Farrer and Murphy1999; Pattatucci & Hamer, Reference Pattatucci and Hamer1995; Schwartz et al. Reference Schwartz, Kim, Kolundzija, Rieger and Sanders2010) further suggests the utility of separate analyses by sex. Genetic epidemiological studies (family, twin, and segregation analyses; see reviews by Mustanski et al. Reference Mustanski, Chivers and Bailey2002; Schwartz et al. Reference Schwartz, Kim, Kolundzija, Rieger and Sanders2010) of males have shown (1) homosexuality to be more common in biological relatives of homosexual men compared to relatives of heterosexual men (or than expected from population surveys), with the sibling recurrence ratios falling between ~2 and ~4 (Pillard & Weinrich, Reference Pillard and Weinrich1986; Bailey & Pillard, Reference Bailey and Pillard1991; Bailey & Bell, Reference Bailey and Bell1993; Hamer et al. Reference Hamer, Hu, Magnuson, Hu and Pattatucci1993; Bailey et al. Reference Bailey, Pillard, Dawood, Miller, Trivedi, Farrer and Murphy1999; Schwartz et al. Reference Schwartz, Kim, Kolundzija, Rieger and Sanders2010); (2) genetic contributions suggested by higher concordances of sexual orientation for identical twins compared with same-sex fraternal twins (Bailey et al. Reference Bailey, Dunne and Martin2000; Kendler et al. Reference Kendler, Thornton, Gilman and Kessler2000; Kirk et al. Reference Kirk, Bailey, Dunne and Martin2000; Santtila et al. Reference Santtila, Sandnabba, Harlaar, Varjonen, Alanko and von der Pahlen2008; Alanko et al. Reference Alanko, Santtila, Harlaar, Witting, Varjonen, Jern, Johansson, von der Pahlen and Sandnabba2010; Langstrom et al. Reference Langstrom, Rahman, Carlstrom and Lichtenstein2010); (3) environmental contributions evidenced by monozygotic concordances far less than unity, with the best demonstrated specific environmental influence being the fraternal birth order effect (Blanchard & Bogaert, Reference Blanchard and Bogaert1996); and (4) non-Mendelian segregation patterns (Bailey et al. Reference Bailey, Pillard, Dawood, Miller, Trivedi, Farrer and Murphy1999; Rice et al. Reference Rice, Anderson, Risch and Ebers1999a ; Schwartz et al. Reference Schwartz, Kim, Kolundzija, Rieger and Sanders2010). Furthermore, genetic studies show some support for excess maternal transmission (Hamer et al. Reference Hamer, Hu, Magnuson, Hu and Pattatucci1993), whose presence would be consistent with X-linkage and which motivated focusing initial linkage approaches to the trait on chromosome X, although most studies have not found excess maternal transmission (Bailey et al. Reference Bailey, Pillard, Dawood, Miller, Trivedi, Farrer and Murphy1999; Rice et al. Reference Rice, Risch and Ebers1999b ; Schwartz et al. Reference Schwartz, Kim, Kolundzija, Rieger and Sanders2010). The aggregate evidence suggests multifactorial causation of male sexual orientation, both genetic and environmental (i.e. a complex genetics scenario).
All previous genetic linkage studies of male sexual orientation were conducted on smaller sample sets. The two initial studies reported linkage to Xq28: first in 40 pairs of homosexual brothers (Hamer et al. Reference Hamer, Hu, Magnuson, Hu and Pattatucci1993) and then further supported in 33 additional such pairs (Hu et al. Reference Hu, Pattatucci, Patterson, Li, Fulker, Cherny, Kruglyak and Hamer1995). However, Xq28 linkage was not supported in three subsequent studies of 54 US pairs of homosexual brothers (Sanders et al. Reference Sanders, Cao, Zhang, Badner, Goldin, Guroff, Gershon and Gejman1998), 52 Canadian pairs of homosexual brothers (Rice et al. Reference Rice, Anderson, Risch and Ebers1999a ), or 73 additional pairs of homosexual brothers studied by the original research group (Mustanski et al. Reference Mustanski, Dupree, Nievergelt, Bocklandt, Schork and Hamer2005). Extended family samples [55 Canadian families (Ramagopalan et al. Reference Ramagopalan, Dyment, Handunnetthi, Rice and Ebers2010) and 146 US families with 155 independent pairs of homosexual brothers (Mustanski et al. Reference Mustanski, Dupree, Nievergelt, Bocklandt, Schork and Hamer2005)] that underwent genome-wide linkage scans (GWLS) found neither genome-wide significant linkage nor any further support for Xq28 linkage. The strongest findings from the larger of the reported GWLS were at chromosomes 7qtel (~7q35–q36), pericentromeric 8 (~8p21–p11), and 10qtel (~10q26) (Mustanski et al. Reference Mustanski, Dupree, Nievergelt, Bocklandt, Schork and Hamer2005). We chose to study male sexual orientation with an independent sample in a GWLS to seek more consistent findings by improvements in statistical power with a larger sample.
Method
Sample collection
We recruited families with two or more homosexual brothers in several primarily English-speaking counties, especially the United States, resulting in completed families from the United States (98.2%), Canada (1.6%), and the United Kingdom (0.2%). We excluded families known to have previously participated in other linkage studies of sexual orientation. Our primary source for identifying probands was through booths at community festivals, especially Gay Pride and related festivals. We also publicized our study through advertisements and articles in homophile media, liaisons with analogous groups (e.g. chapters of PFLAG; Parents, Families and Friends of Lesbians and Gays), and through an educationally oriented internet site (gaybros.com). We enrolled other interested members (available parents and any brothers, regardless of sexual orientation) in a family through the proband. The large majority (97.9%) of the studied families were of European ancestry, 1.6% were African American, and 0.5% were Asian; 95.1% were non-Hispanic and 4.9% were Hispanic. The average age of the studied brothers was 44.3 years [range 18.7–88.9, standard deviation (s.d.) = 10.7], and the sample was recruited from 2004 to 2008. Brothers completed a self-report questionnaire, and all subjects provided a DNA sample via blood (EDTA lavender top tubes, with venipuncture performed primarily through a vendor: LabCorp is the Laboratory Corporation of America, LabCorp, www.labcorp.com) or saliva (Oragene DNA Self-Collection kits, DNA Genotek, Canada). We obtained institutional review board approval from NorthShore University HealthSystem, and also convened and utilized a community advisory board. After complete description of the study to the subjects, all enrolled subjects provided written informed consent.
Ethical standards
The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.
Measurement of sexual orientation
We assessed male sexual orientation as a primarily bimodal, psychological trait. Using data from standard Kinsey self-report questionnaires obtained on all participating brothers, we classified the sexual orientation of subjects as homosexual (homosexual identity and past year Kinsey 5–6 for feelings, i.e. sexual fantasies) or heterosexual (heterosexual identity and past year Kinsey 0–1 for feelings), or otherwise unknown (we collected no questionnaires on parents, and they were all classified as unknown phenotype, but nevertheless made their studied families more genetically informative). For the linkage analysis families, which contained a minimum of two homosexual brothers, no subjects were ultimately classified as unknown, intermediate, or bisexual. Note that the primary linkage analyses employed ‘affecteds-only analysis’ (i.e. focusing on the less common phenotypic variant), which in this case means that only the homosexual brothers’ phenotypes were considered certain (i.e. the heterosexual brothers were considered unknown by the analytical programs).
Genotyping
DNA samples were mailed (saliva) or couriered (blood via FedEx) to our laboratory. We isolated DNA following manufacturers’ recommendations from saliva (Oragene kits) or whole blood (QIAamp® DNA Blood Mini kit, Qiagen, USA). We measured DNA concentrations with PicoGreen, adjusting to 50 ng/μl. We performed initial identity, sex, and relationship screens using AmpFlSTR® Profiler Plus® [Applied Biosystems, USA; 10 simple tandem repeat polymorphism (STRP) markers] prior to shipping plates of DNA to the Vanderbilt Microarray Shared Resource (VMSR, now known as Vanderbilt Technologies for Advanced Genomics; USA). After receipt of samples provided as 15 μl aliquots in 96-well plates and performance of quality control (QC) procedures (primarily DNA concentration and A 260/A 280 ratios via Nanodrop Spectrophotometer, Thermo Scientific, USA), VMSR genotyped the sample on the Affymetrix 5.0 Genotyping Array [440 793 single nucleotide polymorphisms (SNPs)]. The initial linkage dataset consisted of 959 linkage and QC-related samples with BRLMM-P call rates ⩾98%. We conducted further QC iteratively at the sample and SNP level, incorporating inter- and intra-plate QC duplicates, following standard QC procedures for genome-wide association studies (GWAS) (Laurie et al. Reference Laurie, Doheny, Mirel, Pugh, Bierut, Bhangale, Boehm, Caporaso, Cornelis, Edenberg, Gabriel, Harris, Hu, Jacobs, Kraft, Landi, Lumley, Manolio, McHugh, Painter, Paschall, Rice, Rice, Zheng and Weir2010; Turner et al. Reference Turner, Armstrong, Bradford, Carlson, Crawford, Crenshaw, de Andrade, Doheny, Haines, Hayes, Jarvik, Jiang, Kullo, Li, Ling, Manolio, Matsumoto, McCarty, McDavid, Mirel, Paschall, Pugh, Rasmussen, Wilke, Zuvich and Ritchie2011), but further adapted for genome-wide linkage analyses such as the current study.
Sample QC
All 959 samples passed Affymetrix Power Tools QC (APTqc) and had call rates ⩾95%. A single HapMap sample (NA10857) was duplicated 13 times across the plates, and 12 internal duplicate pairs were placed within each plate (one pair per plate). Mean concordance between duplicate pairs was >99% for both inter-plate duplicates and intra plate duplicates. Checks for Mendelian errors, sex concordance, and relationship confirmation identified eight pairs of monozygotic (MZ) twins, four other unintentionally duplicated samples, and one additional error (possible cryptic adoptee within a brother pair), all of which were removed. Two of the eight MZ twin families had a third non-MZ homosexual brother, thus we removed the one of the MZ twins with the lower genotyping rate, but retained these families in the analysis. The other six MZ twin families (comprising seven individuals besides the proband) had no additional siblings and so were removed from the analysis. For the four non-MZ unintentionally duplicated samples, one family still had a genotyped homosexual sibling pair and thus the remainder of the family was retained. The other three families no longer had a sibling pair and thus were entirely dropped (i.e. the remaining sibling) from the linkage analyses. The remaining samples passed the heterozygosity filter (0.273–0.309 retained). We removed one member (the one with the lower genotyping rate) of each of 12 pairs of intra-plate QC duplicates, and for two of these QC duplicates removed the other pair member (because the family did not have a sibling pair for linkage analyses). Finally, after then removing the 13 HapMap inter-plate duplicates, 908 genotyped family members in 384 independent multiplex families remained for linkage analysis.
Family structures
We confirmed full sibling relationships for the 351 families containing ⩾1 full sibling pair, and identified half-sibling relationships for the 35 families containing ⩾1 half-sibling pair. The latter group included 21 reported half-sibling families (already described in the questionnaires) and 14 previously undeclared half-sibling families (for which we corrected family structures). For all half-sibling pair families, we specified as paternal (11 families, which did not contribute linkage information for X chromosome analyses) if explicitly noted as sharing the father in the questionnaire, or otherwise as maternal (shared mother, 24 families). We also annotated one family for which the relative pair of homosexual brothers was explicitly described in the questionnaire (and consistent with the genotypes) as double first cousins (here, with the mothers being full sisters to each other, and the fathers being full brothers to each other). The 908 genotyped family members included 793 homosexual brothers (708/793 = 89% were Kinsey 6, the rest Kinsey 5), 33 heterosexual brothers (31/33 = 94% were Kinsey 0, the rest Kinsey 1), 49 mothers, and 33 fathers. Most families (361/384) had two homosexual brothers, but 21 families had three and two families had four, yielding 409 independent homosexual brother pairs (n – 1 method, see Table 1).
Table 1. SNP genotyped families and individuals analyzed for linkage
All counts reflect the 908 (793 + 33 + 82) SNP genotyped individuals analyzed for linkage. There were 11 paternal half-sibling pairs and 24 maternal half-sibling pairs in these families.
SNP QC
While not analyzed in the current study (except as a further indication of QC), we note that 34 duplicated samples genotyped with two related genotypic arrays at different times and sites [Affymetrix 5.0 at VMSR v. Affymetrix 6.0 at the Broad Institute (Shi et al. Reference Shi, Levinson, Duan, Sanders, Zheng, Pe'er, Dudbridge, Holmans, Whittemore, Mowry, Olincy, Amin, Cloninger, Silverman, Buccola, Byerley, Black, Crowe, Oksenberg, Mirel, Kendler, Freedman and Gejman2009)] revealed ⩾98.8% SNP concordance. After this concordance check, we removed 22 928 SNPs that were discrepant for any of the 34 platform-duplicated (Affymetrix 5.0 and 6.0) samples. We also removed 497 Affymetrix 5.0 SNPs not mapped to chromosomes 1–23 (i.e. those mapped to chromosomes 0, 24, or 25 per Affymetrix information). We then evaluated SNPs for genotyping call rates, deviations from Hardy–Weinberg equilibrium (HWE), and minor allele frequency (MAF) using PLINK (Purcell et al. Reference Purcell, Neale, Todd-Brown, Thomas, Ferreira, Bender, Maller, Sklar, de Bakker, Daly and Sham2007). We removed 78 112 SNPs (29 033 had call rate < 95%, an additional six SNPs showed significant deviations from HWE at p < 10−6, and an additional 49 073 had MAF < 0.01), leaving 362 681 SNPs for further analysis.
Linkage analysis
We conducted two-point and multipoint analysis for linkage using Merlin non-parametric linkage (Abecasis et al. Reference Abecasis, Cherny, Cookson and Cardon2002) using the Kong and Cox linear model (Kong & Cox, Reference Kong and Cox1997) and the S-pairs option. We chose S-pairs to assess independent sibling pairs rather than all possible sibling pairs (S-all), although the results for the latter were almost identical to the former as expected in a sample largely consisting of families genotyped for a single sibling pair. We calculated allele frequency estimates within the dataset (using founding parents when available, otherwise one brother per family), and used the deCODE genetic map (Kong et al. Reference Kong, Thorleifsson, Gudbjartsson, Masson, Sigurdsson, Jonasdottir, Walters, Gylfason, Kristinsson, Gudjonsson, Frigge, Helgason, Thorsteinsdottir and Stefansson2010). For multipoint analysis, we pruned SNPs for linkage disequilibrium (LD, r 2 > 0.16) using PLINK (Purcell et al. Reference Purcell, Neale, Todd-Brown, Thomas, Ferreira, Bender, Maller, Sklar, de Bakker, Daly and Sham2007) and MAF < 0.1, which left 44 766 autosomal SNPs and 621 chromosome X SNPs for multipoint analysis. As a further check of results, we examined the top results (two-point LOD > 2.2) and found no enrichment (compared to the remaining analyzed SNPs) of QC classes such as lower call rate, lower MAF, or lower p values for HWE deviations (online Supplementary Table S1). We estimated power for our GWLS using the formulas for non-parametric linkage analysis (Risch, Reference Risch1990) in 409 independent homosexual sibling pairs (online Supplementary Fig. S1). We found good power (>0.8) to detect suggestive (LOD > 2.2) linkage (Lander & Kruglyak, Reference Lander and Kruglyak1995) down to locus-specific sibling recurrence risk ratios of λ sibs~1.39 and to detect significant (LOD > 3.6) linkage (Lander & Kruglyak, Reference Lander and Kruglyak1995) down to λ sibs~1.52. Given that the overall sibling recurrence risk ratio is estimated to be ~2 to ~4, this suggests that we were able to detect moderate to major loci contributing to the trait.
Linkage models accounting for fraternal birth order effect
We incorporated the known environmental contribution of more older biological brothers from the same mother increasing the chance later born men would be homosexual, i.e. the fraternal birth order effect (Blanchard & Bogaert, Reference Blanchard and Bogaert1996), into our linkage analyses for the two strongest peaks. We did this by means of parametric models (dominant, recessive, and X-linked) using variable penetrance estimates to simulate increasing phenocopy rates (increasing importance of the number of older brothers relative to genetic contributions) as the number of older brothers increased. For the 826 genotyped brothers, we had the number of older brothers for 819 of them, with the number of older brothers ranging from 0 to 7 (online Supplementary Table S2). For the baseline penetrance (for no older brothers), we used 0.005 and increased this by 33% for each older brother a subject had, since it has been estimated that in demographically similar populations to our sample that each additional older brother increases the odds of male homosexuality by ~33% (Cantor et al. Reference Cantor, Blanchard, Paterson and Bogaert2002) (we assigned the baseline penetrance for the seven brothers with an unknown number of older brothers). Thus, the penetrances were: 0.005 for 0 older brother, and 0.0067, 0.0088, 0.012, 0.016, 0.021, 0.028, and 0.037 for 1–7 older brothers, respectively. Under the dominant model all carrier penetrances were set at 0.75 for all liability classes; under the recessive model heterozygote carriers matched the baseline (non-carrier) penetrances, with homozygote carriers having penetrance of 0.75 for all liability classes. All parents and heterosexual brothers were classified as phenotype unknown, and the population frequency of the homosexuality-associated allele was set to 0.02 in these parametric analyses (as in Hamer et al. Reference Hamer, Hu, Magnuson, Hu and Pattatucci1993). We compared the results from these variable penetrance models to those where all subjects’ penetrances were set to 0.005, i.e. as if there were no fraternal birth order effect. We used Merlin (Abecasis et al. Reference Abecasis, Cherny, Cookson and Cardon2002) to conduct multipoint linkage analyses with these parameters, and compared to the previous non-parametric linkage results on chromosomes 8 and X.
Supplementary chromosome X analysis dataset
For chromosome X, in addition to the dataset described above genotyped with SNPs, we had an independent set of 50 families (146 genotyped individuals, online Supplementary Table S3) that had been previously genotyped for 30 STRPs spanning chromosome X. This additional dataset included 56 independent homosexual brother pairs (same phenotyping as SNP-genotyped subjects, i.e. Kinsey 5–6), one of which was a maternal half-sibling pair. To integrate the STRP and SNP maps, we found the physical positions of the 30 STRPs on the hg19 build, and then translated these positions to the deCODE genetic map using the Affymetrix 5.0 annotation file. We ran multipoint linkage analysis with Merlin (minx) using the S-pairs option for the SNP-genotyped families and the STRP-genotyped families separately using the grid option of 1 cM (- -grid 0.01). We then summed LOD scores at each grid position across all SNP and STRP-genotyped families to achieve total LOD scores.
Results
We detected suggestive two-point linkage (LOD ⩾ 2.2) (Lander & Kruglyak, Reference Lander and Kruglyak1995) for 352 SNPs (online Supplementary Table S1) analyzed in the 409 SNP-genotyped homosexual brother pairs (Table 1), with five SNPs exceeding the threshold for genome-wide significance (LOD ⩾ 3.6) (Lander & Kruglyak, Reference Lander and Kruglyak1995). One of the genome-wide significant SNPs was intergenic (rs13212974) with the nearest RefSeq gene being FRK (fyn-related kinase). The other four significant SNPs were located within introns of RefSeq genes: rs6990254 at CLVS1 (clavesin 1), rs2498600 at PTPRD (protein tyrosine phosphatase, receptor type, D), rs2221108 at GRM5 (glutamate receptor, metabotropic 5), and rs7964186 at DNAH10 (dynein, axonemal, heavy chain 10). However, we note that linkage signals are imprecise (especially for traits manifesting complex genetics) and thus larger regions containing additional genes are implicated. The two-point LOD maxima were 4.08 (rs6990254) at 8q12 for autosomes, and 2.99 (rs5925403) at Xq28 for chromosome X. Our genome-wide multipoint non-parametric linkage results are presented in Fig. 1, which shows the two strongest peaks (both with multipoint LOD > 2.5) to be at the pericentromeric region of chromosome 8 (~60–90 cM, based on multipoint drop-1 LOD support interval) and on Xqtel (~160–180 cM). These chromosomes are plotted in Fig. 2 a (chromosome 8) and Fig. 2 b (chromosome X), along with their two-point LOD results. We note that our strongest autosomal finding (based on two-point findings and supported by multipoint results) at pericentromeric chromosome 8 (8p11-q21, Figs 1 and 2 a) overlapped with the second strongest linkage peak in the next largest reported GWLS on 155 homosexual brother pairs which found a peak multipoint LOD = 1.96 at 8p12 (Mustanski et al. Reference Mustanski, Dupree, Nievergelt, Bocklandt, Schork and Hamer2005). We found little support (Fig. 1) for the 7q35–q36 (Mustanski et al. Reference Mustanski, Dupree, Nievergelt, Bocklandt, Schork and Hamer2005) or the 14q32 (Ramagopalan et al. Reference Ramagopalan, Dyment, Handunnetthi, Rice and Ebers2010) regions highlighted in previous smaller GWLS. However, our second strongest linkage region (Xq27–q28, Figs 1 and 2 b) overlaid the previously reported Xq28 linkage (Hamer et al. Reference Hamer, Hu, Magnuson, Hu and Pattatucci1993; Hu et al. Reference Hu, Pattatucci, Patterson, Li, Fulker, Cherny, Kruglyak and Hamer1995). After taking into account the fraternal birth-order effect by incorporating the number of older brothers into the linkage analysis using variable penetrance parametric models, the multipoint linkage peaks on chromosomes 8 and X remained largely unchanged, both compared to the non-parametric linkage results and compared to the parametric linkage results without such birth-order adjustment (online Supplementary Fig. S2). For the X-chromosome multipoint linkage analyses, we merged an independent set (i.e. different families) of 56 homosexual brother pairs (online Supplementary Table S3) previously genotyped for STRP markers (Sanders et al. Reference Sanders, Cao, Zhang, Badner, Goldin, Guroff, Gershon and Gejman1998) into the same genetic map (deCODE) to assess the effect on the Xq28 linkage peak. There was minimal influence from the smaller dataset on the multipoint curve except a small (~0.5 LOD) increase at the telomeric extreme of Xq28, slightly broadening the peak (online Supplementary Fig. S3).
Fig. 1. Multipoint LOD scores are plotted v. the chromosomal positions for the non-parametric linkage analysis. Adjacent chromosomes are separated by alternating black and gray lines. Maximum multipoint LOD scores are ~74 cM on chromosome 8q12 and ~175 cM on chromosome Xq28.
Fig. 2. Individual chromosomal plots for the strongest multipoint linkage peaks on chromosomes 8 and X. LOD scores are plotted v. the chromosomal positions for the non-parametric linkage analysis, the gray line indicating the multipoint values and the black dots indicating the two-point linkage results (with any two-point LOD < 1.0 undisplayed to enable visualization of the multipoint line). (a) Chromosome 8 has its multipoint linkage peak in the pericentromeric region containing several two-point LOD > 3.0, the highest being LOD = 4.08 at 8q12. (b) Chromosome X has its multipoint linkage peak in the pericentromeric region containing several two-point LOD > 2.5, the highest being LOD = 2.99 at Xq28.
Discussion
We have detected genome-wide significant linkage to pericentromeric chromosome 8 with multipoint support, and replicated linkage to chromosome Xq28. In context with the previous linkage scans, it seems likely that genes contributing to variation in male sexual orientation reside in these regions. As usual with linkage peaks for complex traits, there are a number of genes of potential relevance under each broad peak, such as transcription factors, microRNAs, and various brain-expressed genes including some with neurodevelopment, neuroendocrine, and/or neurotransmission roles (e.g. CHRNB3, CHRNA6, SNTG1, NPBWR1, OPRK1, RGS20, PENK, CRH, TRPA1, GDAP1, SLITRK2, CNGA2, GABRE, GABRA3, GABRQ, PLXNB3, L1CAM, GDI1, PLXNA3), and at Xq27–28 the testes-specific members of the SPANX (sperm protein associated with the nucleus on chromosome X) family (these ones being ampliconic genes independently acquired since human-mouse divergence; Mueller et al. Reference Mueller, Skaletsky, Brown, Zaghlul, Rock, Graves, Auger, Warren, Wilson and Page2013). We comment further on selected examples, acknowledging the speculative nature of genotype-behavior correlations across species (e.g. human and mice) and even between more closely related species (e.g. within rodents) (Baud et al. Reference Baud, Hermsen, Guryev, Stridh, Graham, McBride, Foroud, Calderari, Diez, Ockinger, Beyeen, Gillett, Abdelmagid, Guerreiro-Cacais, Jagodic, Tuncel, Norin, Beattie, Huynh, Miller, Koller, Alam, Falak, Osborne-Pellegrin, Martinez-Membrives, Canete, Blazquez, Vicens-Costa, Mont-Cardona, Diaz-Moran, Tobena, Hummel, Zelenika, Saar, Patone, Bauerfeind, Bihoreau, Heinig, Lee, Rintisch, Schulz, Wheeler, Worley, Muzny, Gibbs, Lathrop, Lansu, Toonen, Ruzius, de Bruijn, Hauser, Adams, Keane, Atanur, Aitman, Flicek, Malinauskas, Jones, Ekman, Lopez-Aumatell, Dominiczak, Johannesson, Holmdahl, Olsson, Gauguier, Hubner, Fernandez-Teruel, Cuppen, Mott and Flint2013). Arginine vasopressin is a hormone that mediates social and affiliative behaviors (see review by Ebstein et al. Reference Ebstein, Knafo, Mankuta, Chew and Lai2012), and one of its receptors (AVPR2, arginine vasopressin receptor 2) is located on Xq28, although AVPR2 is primarily expressed in the kidney. However, neuropeptides B/W receptor 1 (NPBWR1) on 8q11.23 is expressed in limbic regions including the hippocampus, has been shown to modulate social interactions in mice (Nagata-Kuroiwa et al. Reference Nagata-Kuroiwa, Furutani, Hara, Hondo, Ishii, Abe, Mieda, Tsujino, Motoike, Yanagawa, Kuwaki, Yamamoto, Yanagisawa and Sakurai2011), and a common functional variant (Tyr135Phe, rs33977775, CEU MAF~10%) has been reported as associated with altered emotional responses to facial expressions in humans, especially valence (with heterozygotes perceiving expressions in more pleasant terms) (Watanabe et al. Reference Watanabe, Wada, Irukayama-Tomobe, Ogata, Tsujino, Suzuki, Furutani, Sakurai and Yamamoto2012). Another brain-expressed gene, CNGA2 (cyclic nucleotide gated channel alpha 2) on Xq28, has been shown in mice to be critical for regulation of odor-evoked socio-sexual behaviors, including major histocompatibility complex (MHC)-related odors (Mandiyan et al. Reference Mandiyan, Coats and Shah2005; Spehr et al. Reference Spehr, Kelliher, Li, Boehm, Leinders-Zufall and Zufall2006); and although more uncertain, MHC-related odors may have relevance in social communication and mate selection in humans (Wedekind et al. Reference Wedekind, Seebeck, Bettens and Paepke1995; Havlicek & Roberts, Reference Havlicek and Roberts2009; Milinski et al. Reference Milinski, Croy, Hummel and Boehm2013). However, linkage can only indicate a region and not particular genes, a task better suited to other approaches such as resequencing approaches, GWAS, and meta-analyses thereof.
Our findings may also begin to provide a genetic basis for the considerable evolutionary paradox that homosexual men are less motivated than heterosexual men to have procreative sex and yet exist as a stable non-trivial minority of the population (Wilson, Reference Wilson1975; Bell & Weinberg, Reference Bell and Weinberg1978). Linkage to Xq28 is especially relevant to the X-linked sexually antagonistic selection hypothesis that women with genetic variant/s predisposing to homosexuality in men have a reproductive advantage compared with other women, i.e. that fertility costs of variants that increase the likelihood of a man's homosexuality are balanced by increased fecundity when expressed in a woman (Miller, Reference Miller2000; Gavrilets & Rice, Reference Gavrilets and Rice2006; Camperio Ciani et al. Reference Camperio Ciani, Cermelli and Zanzotto2008; Rahman et al. Reference Rahman, Collins, Morrison, Orrells, Cadinouche, Greenfield and Begum2008). Modeling of the sexually antagonistic hypothesis (Gavrilets & Rice, Reference Gavrilets and Rice2006) predicts such loci to be strongly over-represented on chromosome X with 1–2 loci there, and allows for strong asymmetries in effect size (i.e. large decrease in reproductive fitness in males can be balanced by smaller increase in females). The putative effects of sexually antagonistic genes are necessarily even more speculative and uncertain than the existence of such genes. Speculation has included the possibility that such genes increase sexual attraction to men in both homosexual men and their heterosexual female relatives, leading to increased reproduction in the females (Camperio Ciani et al. Reference Camperio Ciani, Cermelli and Zanzotto2008). An empirical study focusing on female maternal relatives of homosexual men found several personality and self-reported health differences (Camperio Ciani et al. Reference Camperio Ciani, Fontanesi, Iemmola, Giannella, Ferron and Lombardi2012); it is unclear if any of the differences were related to increased fecundity. A different possibility, overdominance (male heterozygote advantage) (MacIntyre & Estep, Reference MacIntyre and Estep1993), could conceivably help explain the paradox of autosomal genes for male homosexuality, including our findings of linkage to the pericentromeric region of chromosome 8.
Limitations of this research include statistical power considerations (despite the sample size) and challenges inherent to linkage mapping of traits with complex genetics. However, the larger size of the currently studied sample should provide a more stable estimate of the degree of linkage, such as at chromosome Xq28, compared to earlier smaller samples, in part by reducing the influence of stochastic variation. We note that while not studied here, female sexual orientation merits its own scientific study. Future investigations of sexual orientation using GWAS may shed further light on the development of human sexual orientation. Regarding any scenario that research in this area will result in a prenatal genetic test for homosexuality, the small magnitude of effects suggested herein are inconsistent with a test that those motivated to influence their children's sexual orientation would find useful. Furthermore, we agree with Murphy (Reference Murphy1997, Reference Murphy2012) that fear of this unlikely scenario should not prevent further research. Indeed, factual information about sexual orientation can help prevent distorted and hostile views (reviewed in Murphy, Reference Murphy1997). Finding genetic linkage contributes to the overall societal debate by extending support for genetic influence on variation in male sexual orientation from the epidemiological into the molecular realm. While our study results provide further evidence for early (prenatal) biological influences on variation in male sexual orientation, we also emphasize that genetic contributions are far from determinant but instead represent a part of the trait's multifactorial causation, both genetic and environmental.
Supplementary material
For supplementary material accompanying this paper visit http://dx.doi.org/10.1017/S0033291714002451.
Acknowledgements
This work was supported by NICHD: the Eunice Kennedy Shriver National Institute of Child Health and Human Development (A.R.S., grant no. R01HD041563) for the SNP-genotyped sample, and by intramural NIH funds for the STRP-genotyped sample. We thank the men and their families for their participation, and Juliet J. Guroff (deceased) for her contributions to collecting the family sample studied by STRPs (simple tandem repeat polymorphisms), along with other individuals who assisted in the study of that sample at the intramural NIH (Qiuhe Cao, Jing Zhang, and Lynn R. Goldin). We thank Timothy F. Murphy for his work on the community advisory board and study website, and Besiana Liti for technical assistance.
Declaration of Interest
None.
