Hypospadias is a common congenital malformation in which the urethral meatus is on the ventral side of the penis ^{1, 2}. Familial patterns indicate that genetic factors substantially contribute to its etiology ³.

Early development of the external genitalia and urethra depends on the formation and outgrowth of the genital tubercle, the anlage for the penis. This early stage is largely androgen-independent and precedes sexual differentiation. Experimental studies support the contribution of many transcription factors and signaling molecules to this process, including Wilms tumor 1, sonic hedgehog, homeobox proteins, bone morphogenetic proteins, fibroblast growth factors, and Gli transcription factors ^4-9. Expression of the sex-determining region Y gene (SRY) and its direct target, SOX9, is critical to the initiation of the next phase, which is androgen-dependent and involves sexual differentiation and development of the penis and testes ¹⁰. Previous studies have investigated the association of hypospadias with genes that contribute to these events. A greater frequency of variants in cases than controls has been reported for BMP4, BMP7, HOXA4, HOXA6, and WT1 but not SRY or SOX9 among Chinese males ^{11, 12}, FGF8 and FGFR2 but not BMP7 or FGF10 among Swedish males ¹³, and one study did not observe differences among German males for HOXA13 or WTAP ¹⁴. Severity of the phenotypes varied. These studies have primarily involved sequencing, among few subjects. ¹⁵.

Our hypothesis was that variants in genes associated with genital tubercle and early urethral development are associated with human hypospadias. We examined close to 300 relatively common variants in a large population of California male infants in the following genes, given that they have been the subject of previous genetic association studies in humans: BMP4, BMP7, FGF8, FGF10, FGFR2, HOXA13, HOXD13, HOXA4, HOXB6, SRY and SOX9, WT1, and WTAP. We also examined SHH, GLI1, GLI2, and GLI3 (Table 1).

The study population included male infants born from 1990-2003 to mothers who were residents of eight California Central Valley counties and from 1990-1997 to mothers who were residents of Los Angeles, San Francisco, and Santa Clara counties, reflecting counties where case ascertainment was actively conducted by the California Birth Defects Monitoring Program (CBDMP) by reviewing medical records at hospitals and genetic centers ¹⁶.

Cases were classified as mild (meatus was limited to the coronal or glanular penis, British Pediatric Association [BPA] codes 752.605, 752.625), moderate (meatus on the penile shaft, BPA 752.606, 752.626), or severe (meatus at the peno-scrotal junction or perineal area, BPA 752.607, 752.627). Assignment of severity was finalized based on review by a medical geneticist (EJL or Dr. Cynthia Curry) ¹⁷. Cases for which the anatomical position was not sufficiently described (“not-otherwise-specified,” BPA codes 752.600, 752.620) were excluded. Cases classified as having a known single gene disorder or chromosomal abnormality were excluded. Mild cases without chordee (BPA752.605) were not ascertained by CBDMP except in 2004; thus, mild cases are under-represented and primarily those with chordee ¹⁷.

The underlying study population included 1,246,172 non-malformed live born male infants eligible for control selection. We randomly selected 931 with available bloodspots (the DNA source), in proportion to the underlying birth population for each year, to give an approximate 2:1 ratio of controls to cases from Central Valley counties and a 1:1 ratio from other counties. The ratio differed due to the presence of a secondary on-going study in the Central Valley.

Covariates were from birth certificates: maternal race-ethnicity, education, age, and parity; plurality; and infant birthweight and gestational age at delivery. In total, 667 (88% of eligible) cases and 931 (93% of eligible) controls were available for genotyping.

Genomic DNA was extracted from dried bloodspots using MasterPure™ Complete DNA and RNA Purification Kit (Epicentre Biotechnologies Madison, WI), and 10 ng genomic DNA was then used for whole genome amplification (Qiagen Repli-g® kit). TagSNPs that assay known common SNPs either directly or indirectly via linkage disequilibrium among measured and unmeasured SNPs were selected (http://gvs.gs.washington.edu/GVS/). The program provided tagSNPs that cover common variation at r²>0.80 across each candidate gene for a “cosmopolitan” population, including Hispanics. TagSNPs with minor allele frequencies (MAF) >10% were selected. For FGF10 and SRY we selected tagSNPs with MAF>1% because none had MAF>10%. For GLI2 and GLI3 we excluded the Yoruban (YRI) population from tagSNP selection to reduce the number of SNPs (both genes initially produced >100 tagSNPs). For FGFR2, we limited SNP selection to 2 non-intronic tagSNPs due to limited assay space. SNPs were genotyped using a custom multiplex Illumina GoldenGate assay.

We started with 324 SNPs. We excluded 19 for which the data indicated poor clustering of results and one with a call rate <90%. We also excluded 126 subjects (41 cases, 85 controls) with sample call rates <90%, leaving 626 cases and 846 controls for analyses. We further excluded 11 SNPs with p<0.01 for Hardy-Weinberg test of equilibrium among non-Hispanic white or Hispanic controls, leaving 293 for analysis. The single SNP we measured for SOX9 was excluded based on the Hardy-Weinberg test.

Following our previous work, we genotyped 106 ancestry informative marker (AIM) SNPs to discriminate Native American, African, and European ancestry ^18-21, and we used Structure 2.1 to estimate individual ancestry estimates (IAE) ^{22, 23}. Four SNPs were excluded that had call rates <90%. Structure was run using the admixture model with unlinked markers, with 50,000 burn-in iterations and 50,000 further iterations. Structure provided variables reflecting the proportions of Native American, African and European ancestry for each subject. Given that the three proportions sum to one, analyses included only two (Native American and African).

We used logistic regression to compare homozygous and heterozygous variant genotypes with homozygous wildtype (the more frequent allele among controls was designated as wildtype). We considered the presence of population stratification by examining models restricted to self-identified non-Hispanic white and Hispanic subjects that contained product terms to estimate interaction. For SNPs for which the overall p-value for the product term was <0.10 (n=21), we focused on stratified results. We conducted analyses of all cases grouped together as well as separately by severity.

For the 9 genes for which there were >5 SNPs (BMP7, FGF8, FGF10, HOXB6, GLI1, GLI2, GLI3, WT1, WTAP), we examined haplotypes. We used Haploview 4.2 to determine the LD structure and to define haplotype blocks and their frequencies based on all subjects’ genotypes ²⁴. The most common haplotype was the reference. Maximum likelihood estimates of odds ratios (OR) and their corresponding 95% confidence intervals (CI) were calculated from logistic regression models to estimate relative risks.

We also evaluated genetic risk scores created by combining high-risk SNPs. For each individual we counted the number of genes in which they carried an associated variant (p<0.01). For variants with ORs<1, the reference genotype (homozygous wildtype) was scored as the risk genotype. We calculated scores overall and separately by severity (applying the p-value criterion within each group, such that a somewhat different set of variants was scored within each group). We then estimated ORs and 95% CIs associated with the risk scores, using logistic regression.

All ORs were adjusted for the two ancestral proportion variables and for maternal residence in the Central Valley due to the differing case-control ratio based on this variable inherent to the study design. In addition, non-stratified results were adjusted for maternal race-ethnicity (Hispanic, non-Hispanic white, or other). Two cases and two controls had missing race-ethnicity, such that SNP-based analyses included 844 controls and a maximum 624 cases (81 mild, 319 moderate, 209 severe, 15 undetermined).

The study was approved by the California Health and Human Services Agency Committee for the Protection of Human Subjects and Institutional Review Boards at Stanford University and Children’s Hospital Oakland.

Comparing case versus control mothers, the former were more likely non-Hispanic white, more highly educated, older, and nulliparous (Table 2). Cases were more likely to be low birthweight and delivered before 37 weeks of gestation.

There were 28 SNPs for which any comparison had p<0.01 (Table 3). (Results for other SNPs are available in a supplementary table). The homozygous variant genotypes for four SNPs in BMP7 were associated with at least 2-fold increased risk of hypospadias, regardless of severity. Five SNPs for FGF10 were associated with 3-4-fold increased risks, regardless of severity; for four of these, results were restricted to whites. For GLI1, six SNPs were associated: three with reduced risk and three with increased risk, primarily among moderate cases. For GLI2, two SNPs were associated with 2-fold increased risks for moderate phenotypes, but only among whites, and one SNP was associated with 2-fold increased risk, but only for moderate hypospadias (severe phenotypes were also associated with increased risk for two of the SNPs, albeit p>0.01). For GLI3, one SNP was associated with increased risk of mild, one with moderate, and one with severe hypospadias. For SHH, one SNP was associated with increased risk of moderate hypospadias. For WT1, six SNPs were associated with approximately 2-fold increased risk, primarily for severe hypospadias.

The similarity in results across multiple SNPs for certain genes reflected high linkage disequilibrium in some instances but not others. For example, for BMP7 the pair-wise R-squared values for three of the four SNPs ranged from 0.83 to 0.95 (the fourth SNP, rs607007, was rare). For the other genes, the ranges were as follows: 0.43-0.97 for FGF10; 0.54-0.91 for GLI1; 0.28-0.65 for GLI2; 0.03-0.09 for GLI3; and 0.48-0.86 for WT1, with the exception of rs12293750, for which there was only one homozygous variant subject (data not shown).

Haplotype analyses, which included all cases as one group, produced two results with p<0.01. For a block in GLI3 that included rs2237421, rs2237420, rs1527499 and rs3801165, the OR for the ACTG haplotype versus ACGG was 1.5 (95% CI 1.1, 2.0), p=0.0079. For a block in WTAP that included rs963800, rs2758313, rs2842972, rs4709364, rs3822849 and rs1440, the OR for the CCCCTG versus CCTCTG haplotype was 3.3 (95% CI 1.5, 7.3), p=0.002.

Results for the risk score analysis reflect the increase in risk due to having risk-associated SNPs in multiple genes (Table 4). As expected, a higher number of risk genes corresponded with higher ORs. For example, among severe cases a score of two or three was associated with at least a 3-fold increased risk, whereas a score of one was associated with a 1.6-fold increased risk.

SNPs in several genes that contribute to genital tubercle and early urethral development were associated with hypospadias risk, including BMP7, FGF10, GLI transcription factors, SHH and WT1. Results did not suggest an association with BMP4, FGF8, FGFR2, SRY or WTAP. This study included a more in-depth investigation of variants in these genes than previous studies, in a large, racially-ethnically diverse study population.

Experimental studies indicate that bone morphogenetic proteins and fibroblast growth factors contribute to genital tubercle development ^{4, 6}. Two studies have examined their contribution to hypospadias in humans. Sequence variations in BMP4 were reported among three cases and in BMP7 among six cases, out of 90, and none among 190 controls ¹¹. Among 60 Swedish boys with familial, isolated hypospadias, four had variants in FGF8 and seven had variants in FGFR2, which were not observed among 96 controls ¹³. Several FGFR2 relatively frequent polymorphisms were observed but not different among cases versus controls. No sequence variations were reported in FGF10 or BMP7 ¹³. In the current study, SNPs in BMP4, FGF8, and FGFR2 were not associated with hypospadias, whereas, several SNPs in BMP7 and FGF10 were associated with at least 2-fold increased risks of hypospadias, regardless of severity. Findings for FGF10 were predominately among non-Hispanic whites.

SHH contributes to epithelial-mesenchymal interactions and patterning in the genital tubercle ²⁵, and Shh knockout mice have genital tubercle agenesis ⁹. Gli transcription factors are encoded by three genes with overlapping function (GLI1-3), which are regulated by SHH ²⁵. GLI genes are associated with limb and craniofacial development, which involve developmental processes related to tissue patterning that may apply to urethral development ²⁵. Gli2 mutant mice have defective urethral formation ^{9, 25}. To our knowledge, the current study represents the first investigation of GLI1-3 or SHH and hypospadias in humans. Our study provided some evidence of an association for selected variants in these genes, but results were inconsistent across phenotypes and race-ethnicity.

WT1 is a zinc-finger transcription factor that contributes to normal development of the genitourinary system ^{26, 27}. WTAP contributes to WT1 function ²⁸. Sequence variations in WT1 were reported in three of 90 Chinese cases and zero of 276 controls ¹², and zero of 35 Swedish cases ²⁹. Variants in WT1 were reported among six of 80 severe hypospadias cases with cryptorchidism; all six eventually developed Wilms tumor or nephropathy ³⁰. No variants were observed among 70 cases without cryptorchidism. Utsch et al. sequenced the exons of WTAP ¹⁴. They observed two variants and stated that their frequency was not different between cases and controls. In the current study, a WTAP haplotype and several SNPs in WT1 were associated with hypospadias.

Experimental evidence also suggests that homeobox genes contribute to early genital tubercle development ^{4, 5}. Hoxa13 knockout mice have hypospadias ⁴, and Hoxa13 mutants have altered androgen receptor expression ⁴. In humans, mutations in HOXA13 cause hand-foot-genital syndrome, which includes hypospadias ⁵. However, no sequence variants in HOXA13 were observed among 37 cases with hypospadias ¹⁴. Variants in HOXA4 were observed among three subjects and in HOXA6 among two subjects, among 90 total, and none among controls ¹¹. The current study did not provide support for an association of common variants in several homeobox genes with hypospadias.

Our study is strengthened by its size, population-based controls, ancestry informative markers, and inclusion of multiple genes. We chose to highlight results that met a relatively modest criterion for statistical significance (p<0.01) rather than conducting formal correction for multiple testing, given that our study focused on candidate genes. Our approach minimizes Type II errors (false negatives). However, we acknowledge that the trade-off is an increased possibility of false positive results. As such, we emphasize the need for replication of our findings in additional study populations. Our approach of investigating tagSNPs seemed appropriate, given that it captures the majority of genetic variation and is cost-efficient, and that minimal examination of the studied genes in humans preceded the current study. However, most tagSNPs are intronic and have no known functional consequences. Two exonic SNPs are included in Table 3 but are non-synonymous variants (rs2228226 in GLI1 and rs16754 in WT1). Thus, observed associations are likely driven by linkage disequilibrium with other less common unmeasured variants and merit further genetic inquiry. Elucidation of underlying causal variants could be useful for genetic counseling purposes or for directing mechanistic studies. Also of note, we were able to explore results by phenotypic severity, but sample size for some phenotype-specific results were limited, especially for mild cases. Under-ascertainment of mild cases or misclassification of severity are potential limitations but unlikely to be responsible for our results.

This study found substantial evidence for an association of hypospadias with genes involved in genital tubercle development.