TGFB1 as a Susceptibility Gene for High Myopia: A Replication Study With New Findings

Objective  To investigate the genetic association between transforming growth factor β1 (TGFB1) gene polymorphisms and high myopia in a Chinese population.

Methods  Six hundred adults were recruited for this case-control study, including 300 subjects with high myopia (−8.0 diopters or worse) and 300 control subjects (within ±1.0 diopters). Seven tag single-nucleotide polymorphisms (SNPs) and 1 coding SNP were genotyped. Their frequencies were compared between cases and controls by statistical tests.

Results  Four SNPs in the 5′ half of the gene showed significant differences in allele and genotype frequencies between cases and controls. The results remained significant after correction for multiple comparisons. The previously reported association of the coding SNP rs1800470 with high myopia was successfully replicated. The tag SNP rs4803455 in intron 2 was found to account for the positive results of the other 3 SNPs by stepwise logistic regression. The minor allele T of rs4803455 was protective against high myopia with an odds ratio of 0.67 (95% confidence interval, 0.53-0.86; P = .001).

Conclusion  TGFB1is a myopia susceptibility gene.

Clinical Relevance  TGFB1is the first myopia susceptibility gene successfully replicated. The functional significance of rs4803455 or the genuine causative SNPs in linkage disequilibrium with it remains to be determined.

Myopia is the most common disease of the human eye. High myopia, typically defined as −6 diopters (D) or worse, results in an increased risk of developing irreversible, sight-threatening diseases of the retina and choroid such as premature cataracts and glaucoma.¹Myopia development has become more prevalent in the younger generations, especially in Asian populations.^2-5Not surprisingly, the ocular manifestations of disease associated with high myopia are among the leading causes of registered blindness. Therefore, it is important to delineate the etiology of myopia. Myopia is a complex trait influenced by environmental and genetic factors and their interactions.^6,7One approach to understanding the molecular pathways underlying myopia is to identify the myopia susceptibility genes.⁷

In humans, myopia develops mainly because of excessive axial length rather than changes in corneal or lens power.⁸In animal models of myopia, active remodeling of the sclera has been shown to play a crucial role in ocular elongation.^9,10Scleral remodeling involves a reduced production of extracellular matrix resulting from reduced production of collagen and proteoglycans and from increased collagen degradation with concomitant increased activity of matrix metalloproteinase 2 (MMP2) and reduced activity of tissue inhibitors of MMP. Transforming growth factor β (TGF-β) is expressed in ocular tissues together with its receptors¹¹and is known to regulate the proliferation of fibroblasts and the production of collagen, MMP2, and tissue inhibitors of MMP.¹²As such, TGF-β is implicated in the regulation of scleral remodeling during myopia development.

The importance of TGF-β in myopia development is supported by many studies. In cell cultures, TGF-β and other growth factors were shown to stimulate the proliferation of chondrocytes and fibroblasts isolated from chick sclera.¹³Moreover, TGF-β levels were found to be reduced in chick retinal pigment epithelium and choroid in correlation with the axial elongation in myopia induced by form deprivation.¹⁴In the chick retina, the expression of the transcription factor ZENK was found to be enhanced by conditions suppressing ocular elongation (ie, plus defocus and termination of form deprivation) and suppressed by conditions promoting ocular elongation (ie, minus defocus and form deprivation).^15,16In human cultured fibroblasts, EGR1(early growth response gene type 1, the human homologue of ZENK) has been shown to activate TGFB1transcription via EGR1-binding sites of the TGFB1promoter.¹⁷Thus, TGFB1expression should be reduced as a result of less activation from the reduced ERG1(ZENK) level seen in ocular elongation—a situation compatible with results of an earlier study.¹⁴In the sclera of the tree shrew, the expression of all 3 TGF-β isoforms (TGFB1, TGFB2, and TGFB3) was found to be reduced in an isoform-specific and time-dependent manner in myopia induced by form deprivation.¹⁸More importantly, such in vitro changes in TGF-β concentrations also approximated the changes in vivo during myopia induction and were found to reduce collagen synthesis in scleral fibroblasts to an extent similar to that documented previously in induced myopia.¹⁸

Two recent studies investigated the genetic association between single-nucleotide polymorphisms (SNPs) of the TGFB1gene and high myopia but produced conflicting results.^19,20The present study serves to clarify this relationship with a case-control design and a much larger sample size. Located on chromosome 19q13.1, the human TGFB1gene (NCBI Entrez Gene 7040 and 190180) spans across 23 kilobases (kb) and has 7 exons (http://www.ncbi.nlm.nih.gov/).

Unrelated southern Chinese subjects were recruited for this study. The entry criteria of spherical equivalent (SE) were −8.00 D or worse for both eyes for subjects with high myopia (cases), and within ±1.0 D for those with emmetropia (controls). This study was approved by the Human Subjects Ethics Subcommittee of The Hong Kong Polytechnic University and adhered to the tenets of the Declaration of Helsinki. Every subject gave written informed consent.

Each subject received a complete ocular examination, including visual acuity, refraction, slitlamp, and dilated fundus examinations, in the Optometry Clinic of The Hong Kong Polytechnic University. Objective refraction was measured using an open-field autorefractor (SRW-5000; Shin-Nippon Ophthalmic Instruments, Tokyo, Japan) after the subject was given 1 to 2 drops of 1% tropicamide per eye. Central corneal curvature was measured using autokeratometry (Canon RK-5 Auto Ref-keratometer; Canon, Inc, Tokyo). Axial length was measured using A-mode ultrasonography (Advent A/B System; Mentor, Santa Barbara, California) after 1 drop of 0.4% benoxinate hydrochloride was instilled in each eye to produce anesthesia. Subjects were excluded from the study if they showed obvious signs of ocular disease or other inherited disease associated with myopia. Venous blood was collected from the subjects after the ocular examination, and DNA was extracted from the leukocytes using a modified salt precipitation method.²¹

Tag SNPs were selected from a 35-kb region encompassing the TGFB1locus and 6 kb upstream and downstream of the gene, based on the HapMap data for Chinese subjects (release 21a; http://www.hapmap.org/index.html.en). Flanking sequences were included to capture SNPs in potential regulatory regions. We used the Tagger software²²for selection with the setting of the pairwise tagging algorithm r² > 0.8 and minor allele frequency of more than 0.1. Another SNP (rs1800470) was also included because it was recently found to be associated with high myopia.¹⁹Three different methods were used to genotype these SNPs,^23-25and the details are shown in the eTable. The choice of methods depended on the logistic arrangement for instrument use in our core laboratory and the cost of the assays.

The SNPs rs1800469, rs1800470, rs4803455, and rs12983047 were genotyped by means of restriction fragment length polymorphism under conditions recommended by the manufacturers (MBI Fermentas, Vilnius, Lithuania; or New England Biolabs, Beverly, Massachusetts) (eTable). Polymerase chain reaction (PCR) was performed in a 10-μL reaction mixture containing 20 ng of genomic DNA, 1.5mM or 2.5mM magnesium chloride, 0.1μM or 0.3μM each of the forward and reverse primers, 0.2mM each deoxynucleotide triphosphate and 1 × gold buffer (15mM TRIS hydrochloride and 50mM potassium chloride [pH, 8.0]), and 0.2 U of AmpliTaqGold DNA polymerase (Applied Biosystems, Foster City, California). Amplification was performed in 96-well plates with a PCR system (GeneAmp system 9700; Applied Biosystems), including 1 cycle of initial denaturation for 5 minutes at 95°C, 40 cycles of 30 seconds at 95°C, 30 seconds at the annealing temperature (eTable), and 30 seconds at 72°C, plus 1 cycle of final extension for 7 minutes at 72°C. Digested products were separated by electrophoresis in agarose or polyacrylamide gels as appropriate.

Two SNPs (rs11466345 and rs10417924) were genotyped by means of primer extension coupled with denaturing high-performance liquid chromatography as described previously.²³The PCR templates were generated using conditions mentioned in the preceding paragraph except for the annealing temperature and magnesium chloride and primer concentrations, which are as shown in the eTable. The thermal cycling condition for primer extension was modified as follows: an initial denaturation of 2 minutes at 95°C, followed by 50 cycles of 5 seconds at 95°C, 5 seconds at 43°C, and 5 seconds at 60°C. We used a DNA fragment analysis system (WAVE; Transgenomic, Omaha, Nebraska) to analyze the extension product.

Two SNPs (rs2241716 and rs12981053) were genotyped by means of allele-specific PCR (eTable) performed in real-time PCR systems. Each genotype of rs2241716 was determined according to the difference of threshold cycle numbers for a 2-tube allele-specific PCR performed in the presence of nucleic acid stain (SYBR Green I; Molecular Probes, Eugene, Oregon) in a real-time PCR system (7500 PCR system; Applied Biosystems), as described previously but with slight modifications.²⁴The genotype of rs12981053 was based on the melting temperatures of the amplicons amplified in single-tube allele-specific PCR with tailed primers in the presence of the nucleic acid stain in a real-time PCR system (LightCycler 480 system; Roche Applied Science, Basel, Switzerland) according to a published protocol with minor modifications.²⁵

Ocular data were analyzed using a commercially available statistical software package (STATA, version 8.2; StataCorp, College Station, Texas). Genotypes were tested for Hardy-Weinberg equilibrium (HWE) for controls and cases separately by exact tests as executed in Haploview²⁶(version 4.0; http://www.broad.mit.edu/mpg/haploview/). Genetic association analysis was performed with the GenAssoc package²⁷(http://www-gene.cimr.cam.ac.uk/clayton/software/) executed within the statistical software. Controls and cases were compared for allele and genotype frequencies by means of χ²tests. Without knowledge of possible genetic models in operation, only the additive model was also tested using the Cochran-Armitage trend test, which is more conservative but does not rely on an assumption of HWE.²⁸Odds ratios (ORs) and the corresponding 95% confidence intervals (CIs) were calculated with reference to the more frequent (major) allele or the more frequent homozygotes. Multiple comparisons were corrected by false discovery rate.²⁹After adjustment of 24 comparisons (8 SNPs each compared for both allele and genotype frequencies and also for additive model) and with a level of .05 for the false discovery rate, the cutoff Pvalue for significant association was .0208. For the 4 SNPs found to be significant after multiple testing correction, stepwise logistic regression was used to determine whether any one of these 4 SNPs could account for the positive effects of the other SNPs.²⁷

Haplotype blocks were constructed using Haploview with an algorithm known as the solid spine of linkage disequilibrium (LD).²⁶Haplotype frequencies were estimated by Haploview using an accelerated expectation-maximization algorithm³⁰and then compared between controls and cases. Multiple testing was corrected by generating empirical Pvalues based on 10 000 permutations by Haploview.

This study recruited 300 control subjects and 300 high myopia cases. For controls, the mean age was 24.9 (range, 17-46) years; mean SE was 0.03 (range, −1.00 to 0.88) D; and mean axial length was 23.85 (range, 21.24-27.71) mm (Table 1). For cases, the mean age was 27.7 (range, 15-48) years; mean SE was −10.53 (range, −24.00 to −8.00) D; and mean axial length was 27.76 (range, 24.62-31.29) mm. The ocular data given herein are for right eyes. The correlation between right and left eyes was the best for SE (0.97) and axial length (0.96) when all 600 subjects were included in the analysis, but less striking when controls and cases were analyzed separately. As expected, the partial correlation of SE with other ocular components was the best for axial length (Table 1). The proportion of female subjects was lower among the controls than the cases (56.3% vs 71.7%; P < .001).

Seven tag SNPs were selected from the HapMap database for Chinese subjects, including 1 (rs1800469) in the 5′ flanking region of the gene, 3 (rs2241716, rs4803455, and rs11466345) in introns, and 3 (rs12983047, rs10417924, and rs12981053) in the 3′ flanking region (Figure). The coding SNP rs1800470 (C/T) involves an amino acid change (proline to leucine) at codon 10 and was added because of a previous positive finding.¹⁹This coding SNP, however, was not documented in the HapMap database. For easy reference, the 8 SNPs were also designated as S1 to S8 from the direction of 5′ to 3′ (Figure).

The genotypes were all in HWE for the controls and cases except for rs4803455 (S4) for the case group (P = .04). The genotypes of S4 had been confirmed by a second method (2-tube allele-specific PCR method).²⁴Departure from HWE for genotypes in cases can be a sign of marker-disease association.³¹

Although there were fewer female subjects in the control group than in the case group (Table 1), the comparison of allele and genotype frequencies did not show any significant differences between female and male subjects in either group for all 8 SNPs. This justified the direct comparison of genetic data between controls and cases without stratification by sex. Four tag SNPs did not differ significantly between the controls and cases for allele frequencies or genotype frequencies (direct comparison or additive model): S5 to S8 (Table 2).

The other 4 SNPs (S1 to S4) all showed significant differences between controls and cases at the nominal cutoff value of P < .05 (Table 2). After correction for multiple testing by means of the false discovery rate, all of these differences remained significant except for 2 direct comparisons of genotype frequencies (one for S1, and the other for S2). With the less frequent allele as the disease allele, the ORs were all less than 1.00, indicating that the less frequent alleles were protective against high myopia.

A stepwise logistic regression procedure was used to detect which of these significant SNPs contributed the main effects (Table 3). Given the main effect of S4, none of the other 3 SNPs remained significant in the forward procedure. This was consistent with the finding that, given the main effects of S2 or S1, only S4 still contributed a significant effect (P = .04 and P = .03, respectively). Given S3, none of the other 3 SNPs contributed significant effects anymore. However, the main effect of S3 did not come out given the effect of S2 or S1. On the other hand, the backward procedure eliminated S3 from the alternative model first, then S2, and finally S1. Thus, the stepwise regression procedure demonstrated that S4 (rs4803455) contributed a significant main effect to high myopia and could explain the positive association results of the other 3 SNPs (Table 3). This finding is also consistent with its strongest effect size (greatest difference of the OR from 1.00) and the corresponding smallest Pvalues (Table 2).

Two haplotype blocks were always constructed for the 8 SNPs studied, whether the controls and cases were examined separately or combined (Figure). Block 1 always consisted of the 4 SNPs (S1 to S4) that are located at the 5′ half of the gene and were found to be associated with high myopia in the analysis described in the preceding paragraphs. Block 2 always had S6 to S8, and S5 was included in block 2 only for the control group. Haplotype frequencies were also compared between controls and cases for the 2 blocks separately (Table 4). Only the following 2 haplotypes in block 1 were found to show significant differences: TCGG (P = .003) and CTAT (P = .002). These positive results remained significant after correction for multiple comparisons by permutation (or by false discovery rate).

Lin et al¹⁹conducted a case-control study between rs1800470 (S2) and individuals with high myopia. The choice of this coding SNP for investigation was a good one because this SNP had been correlated with many other complex traits and diseases, as noted by the authors. The study involved 201 case subjects with high myopia (SE, less than −6.0 D) and 86 control subjects (SE, greater than −0.5 D), all older than 16 years and ethnic Chinese in Taiwan. They found more allele C in the case subjects with high myopia than in the controls (P < .001; OR, 1.83; 95% CI, 1.27-2.63). If allele C was used as the reference, as in the present study, the OR was 0.55 (95% CI, 0.38-0.79).

Another recent study recruited 660 Japanese subjects, including 330 case subjects with high myopia (SE, less than −9.25 D) and 330 control subjects (SE, greater than −2.0 D).²⁰Six of the selected 10 SNPs were nonpolymorphic, and the remaining 4 SNPs were each genotyped in fewer than 100 cases with high myopia and fewer than 100 controls. Three of these genotyped SNPs were in perfect LD with each other and thus were effectively represented by one of them in the information content. Therefore, this study effectively genotyped only 2 SNPs for less than 200 samples. No significant association was detected. One of the SNPs genotyped was rs1800469 (S1).

The present study involved 300 case subjects with high myopia (SE, less than −8.0 D) and 300 controls (SE, within ±1.0 D), all ethnic Chinese subjects in Hong Kong. It captured the SNP genetic information in a 35-kb region encompassing the TGFB1locus by genotyping only 7 tag SNPs selected from the HapMap database for Chinese. An additional SNP, rs1800470 (S2), was also genotyped for the reasons explained in the “Results” section. This SNP was found in strong LD (r² > 0.8) with rs1800469 (S1) in the Chinese population (Figure).

One direct approach to identifying potentially functional sequence variants is to sequence the whole candidate gene for all case and control samples. A more cost-effective alternative is to sequence a small number of samples to reveal common sequence variants in the exons in the initial phase. Five nonsynonymous SNPs in TGFB1exons are currently documented in the dbSNP database (http://www.ncbi.nlm.nih.gov/SNP/index.html), but most of them have very low minor allele frequencies except rs1800470 (codon 10 SNP), which was tested in this study. A very large sample size is required to detect genetic association, if any, when the minor allele frequencies are low.⁷Therefore, genetic association studies generally aim to identify common variants responsible for complex diseases. Within a small genomic region, the SNPs are usually in strong LD with each other, and hence there are usually only a few haplotypes across many SNP sites.³²These observations make possible an SNP selection strategy in which a subset of representative SNPs (tag SNPs) are selected for genotyping to capture the few common haplotypes of a given region. Thus, there is no need to type other redundant SNPs. This is one of the main reasons for establishing the International HapMap Project, and selection of tag SNPs from HapMap data for association studies is also widely adopted.^32,33Any identified positive SNP can be causative for the complex disease under study or more commonly a surrogate marker in LD with a genuine causative SNP.

The statistical power of a genetic association has always been a concern, particularly when the results are negative.⁷One solution is to use meta-analysis to combine the data from different studies, if possible. Meta-analysis was performed using the MIX program³⁴(http://www.mix-for-meta-analysis.info) to combine the data for SNPs that were genotyped in more than one study. For rs1800469 (S1), a significant association with high myopia was still demonstrated on meta-analysis of the Japanese data²¹and our data (OR, 0.78 for allele C with reference to the major allele T; 95% CI, 0.64-0.96; z = 2.39; P = .02 for the fixed-effects model using Mantel-Haenszel weights). For rs1800470 (S2), significant association with high myopia was even more convincing on meta-analysis of the Taiwanese data²⁰and our data (OR, 0.66 for allele T with reference to the major allele C; 95% CI, 0.55-0.81; z = 4.15; P < .001 for the fixed-effects model using Mantel-Haenszel weights). Therefore, the present study successfully replicated the Taiwanese study, and added further strength to the significance role of TGFB1as a susceptibility gene for high myopia.

More important is the novel discovery that rs4803455 (S4) could account for the positive results of the other 3 SNPs (Table 3). This is consistent with the following results of haplotype analysis: the 2 haplotypes (TCGG and CTAT) associated with high myopia showed less marked differences in frequencies between controls and cases, as revealed by their χ²-based Pvalues (Table 4) being greater than those for S4 alone (Table 2). This SNP is located in intron 2 of the TGFB1locus, and, to our knowledge, has not yet been reported to have any functional role in regulating the gene expression of TGFB1. We performed a search for potential transcription factors that may be affected by this SNP with the SNPInspector software (http://www.genomatix.de/). Two transcription factors (TCF11 and BRN-2) are potentially associated with the major allele G but not the minor allele T, whereas 3 transcription factors (LHX3, ATBF1, and DLX3) are associated with allele T but not allele G. Whether these transcription factors regulate the TGFB1expression, particularly in relation to myopia development, remains to be determined experimentally.

Another possibility is that rs4803455 (S4) was not the causative SNP directly associated with myopia development but was in LD with another genuine causative SNP in the flanking regions. It is intriguing that S4 does not capture or represent any other SNP currently documented in the HapMap database for Chinese in the 35-kb region examined in this study or even a larger region of 100 kb if tag SNPs are selected on the basis of r² > 0.8 and a minor allele frequency of more than 0.1. However, many neighboring SNPs are not genotyped in the HapMap Project. Thus, such neighboring SNPs are worth exploring in this regard. Meanwhile, the low LD between the haplotype blocks (Figure) indicates the existence of a recombination hot spot in the middle of the TGFB1locus—a finding consistent with the HapMap data. Because S4 is upstream of this recombination hot spot, potentially causative SNPs should also be searched upstream of this hot spot, particularly those in the 5′ regulatory region of the gene.

Although this study focused on identifying a susceptibility gene for myopia, it must be stressed that myopia is a complex trait that is caused by environmental and genetic factors.^6,7Environmental factors such as excessive close work may interact with genetic risk factors to produce an abnormal eye growth leading to myopia. The importance of environment and its interaction with genes has been reviewed recently.³⁵

In conclusion, 4 SNPs in the 5′ half of the TGFB1locus were found to be associated with high myopia in Chinese subjects, and the positive finding in a recent report was successfully replicated. Moreover, rs4803455, a tag SNP in intron 2, could account for the positive results of the other 3 SNPs. The study supports the role of TGFB1as a susceptibility gene for high myopia and provides the foundation for further studies to explore the functional effect of rs4803455 or to identify the genuine causative variants in LD with this SNP.

Correspondence:Shea Ping Yip, PhD, MPhil, FIBMS, Department of Health Technology and Informatics, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong (shea.ping.yip@polyu.edu.hk).

Submitted for Publication:May 6, 2008; final revision received July 31, 2008; accepted September 5, 2008.

Author Contributions:As the principal investigator, Dr Yip had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Financial Disclosure:None reported.

Funding/Support:The study was supported by grants G-YX1B, 87MS, and A362 from The Hong Kong Polytechnic University (PolyU). The following instruments used in this study were purchased using PolyU Big Equipment grants or departmental funds: the Wave DNA fragment analysis system (G.52.27.9027), the 7500 real-time PCR system (G.52.27.D032), and the LightCycler 480 real-time PCR system (1.55.27.DD02).