Abstract

Refractive error is a complex ocular trait governed by both genetic and environmental factors and possibly their interplay. Thus far, data on the interaction between genetic variants and environmental risk factors for refractive errors are largely lacking. By using findings from recent genome-wide association studies, we investigated whether the main environmental factor, education, modifies the effect of 40 single nucleotide polymorphisms on refractive error among 8461 adults from five studies including ethnic Chinese, Malay and Indian residents of Singapore. Three genetic loci SHISA6-DNAH9, GJD2 and ZMAT4-SFRP1 exhibited a strong association with myopic refractive error in individuals with higher secondary or university education (SHISA6-DNAH9: rs2969180 A allele, β = −0.33 D, P = 3.6 × 10–6; GJD2: rs524952 A allele, β = −0.31 D, P = 1.68 × 10−5; ZMAT4-SFRP1: rs2137277 A allele, β = −0.47 D, P = 1.68 × 10−4), whereas the association at these loci was non-significant or of borderline significance in those with lower secondary education or below (P for interaction: 3.82 × 10−3–4.78 × 10−4). The evidence for interaction was strengthened when combining the genetic effects of these three loci (P for interaction = 4.40 × 10−8), and significant interactions with education were also observed for axial length and myopia. Our study shows that low level of education may attenuate the effect of risk alleles on myopia. These findings further underline the role of gene–environment interactions in the pathophysiology of myopia.

INTRODUCTION

Myopia is the most common cause of visual impairment worldwide (1, 2). The global prevalence of myopia has been rising steadily over the past few decades, especially amongst urban East and Southeast Asian populations (3, 4). In Singapore, for example, the prevalence of myopia in young adults increased from an estimated 26% in the late 1970s to up to 80% by the middle 1990s (5, 6). This rapid rise of myopia prevalence in Asia may be attributed to changes in environmental factors, such as the increasing intensity of education, and possibly to gene and environment (G × E) interaction (7).

Recent genome-wide association studies (GWAS) have greatly advanced our understanding of the genetic architecture of myopia (8–14), with two large GWAS reporting >30, partially overlapping, genetic loci associated with refractive phenotypes (9, 10). Many of these recently identified genes are linked to known visually triggered signaling pathways in the development of human myopia, such as neuronal signaling in the retina and scleral extracellular matrix remodeling (i.e. GJD2, RASGRF1, BMP2 and BMP3), while others point to novel pathways involved in neuronal development and ion transport (i.e. KCNJ2, KCNMA1 and CD55).

The most solid evidence for environmental influences in adults studies on myopia is education, with a large number of studies reporting a strong correlation between the occurrence of myopia and a higher level of education (4, 5, 15–18). In Asian populations, persons aged 40 years or older with university education are nearly four times more likely to have myopia compared with those with primary and no formal education (18). Education level has largely been considered a surrogate of lifetime cumulative near work activities such as reading and writing, although other lifestyle factors including outdoor play patterns may be involved in myopia development (19–21). So far, most studies have investigated environmental and genetic factors for myopia separately. Whether there are interactions between genetic markers and environmental exposures on the risk of myopia remains to be elucidated.

In the present study, we examined whether education modifies the association between recently discovered single nucleotide polymorphisms (SNPs) and refractive error using data from five population-based cohorts in Singapore.

RESULTS

Table 1 shows the baseline characteristics of subjects included from five Singapore cohorts. After applying stringent quality-control procedures and retaining those with valid phenotype data, the present analyses were based on a total of 8461 participants aged 25 years and older, including Chinese from Singapore Chinese Eye Study (SCES) (n = 1710), SP2 (n = 1665) and Study on Strabismus, Amblyopia and Refractive Error in Singapore Children (STARS) (n = 741), as well as Malays from Singapore Malay Eye Study (SiMES) (n = 2257) and Indians from Singapore Indian Eye Study (SINDI) (n = 2088). Across all studies, individuals are more myopic in the higher education stratum (mean spherical equivalent (SE) = −2.3 D; SD = 3.0 D), compared with the lower education stratum (mean SE = −0.7 D; SD = 2.5 D). Chinese subjects, especially the young adults from STARS (mean SE = −2.8 D; SD = 2.85 D), tended to be more myopic compared with Malay (mean SE = −0.03 D; SD = 1.81 D) and Indian participants (mean SE = 0.04 D; SD = 2.07 D).

Table 1.

Characteristics of participants in five Singapore cohorts

Study All
 
Lower education group
 
Higher education group
 
 n Male (%) Agea (years) SEa (D) n Male (%) Agea (years) SEa (D) n Male (%) Agea (years) SEa (D) 
SCES 1710 51.6 57.5 (9.0) −0.72 (2.69) 1346 49.2 58.4 (9.1) −0.3 (2.2) 364 60.4 54.4 (7.7) −2.4 (3.1) 
SP2 1665 40.4 47.6 (10.9) −1.62 (2.87) 957 34.7 51.2 (10.0) −0.7 (2.4) 708 48.0 42.8 (9.9) −2.9 (3.0) 
STARS 741 52.4 38.5 (5.2) −2.80 (2.85) 328 50.9 39.5 (5.8) −1.8 (2.4) 413 53.5 37.6 (4.5) −3.6 (2.9) 
SiMES 2257 49.1 58.0 (10.8) −0.03 (1.81) 2105 47.8 58.6 (10.8) 0.02 (1.8) 152 67.1 49.1 (6.6) −0.81 (2.3) 
SINDI 2088 51.5 55.8 (8.8) 0.04 (2.07) 1618 47.1 56.6 (8.9) 0.23 (1.9) 470 66.6 53.2 (8.2) −0.63 (2.4) 
Total 8461 48.8 53.6 (11.3) −0.71 (2.54) 6354 46.1 56.0 (10.7) −0.18 (2.1) 2107 56.7 46.5 (10.3) −2.3 (3.0) 
Study All
 
Lower education group
 
Higher education group
 
 n Male (%) Agea (years) SEa (D) n Male (%) Agea (years) SEa (D) n Male (%) Agea (years) SEa (D) 
SCES 1710 51.6 57.5 (9.0) −0.72 (2.69) 1346 49.2 58.4 (9.1) −0.3 (2.2) 364 60.4 54.4 (7.7) −2.4 (3.1) 
SP2 1665 40.4 47.6 (10.9) −1.62 (2.87) 957 34.7 51.2 (10.0) −0.7 (2.4) 708 48.0 42.8 (9.9) −2.9 (3.0) 
STARS 741 52.4 38.5 (5.2) −2.80 (2.85) 328 50.9 39.5 (5.8) −1.8 (2.4) 413 53.5 37.6 (4.5) −3.6 (2.9) 
SiMES 2257 49.1 58.0 (10.8) −0.03 (1.81) 2105 47.8 58.6 (10.8) 0.02 (1.8) 152 67.1 49.1 (6.6) −0.81 (2.3) 
SINDI 2088 51.5 55.8 (8.8) 0.04 (2.07) 1618 47.1 56.6 (8.9) 0.23 (1.9) 470 66.6 53.2 (8.2) −0.63 (2.4) 
Total 8461 48.8 53.6 (11.3) −0.71 (2.54) 6354 46.1 56.0 (10.7) −0.18 (2.1) 2107 56.7 46.5 (10.3) −2.3 (3.0) 

SE, spherical equivalent; SCES, Singapore Chinese Eye Study; SP2, Singapore Prospective Study Program; STARS, Study on Strabismus, Amblyopia and Refractive Error in Singapore Children; SiMES, Singapore Malay Eye Study; and SINDI, Singapore Indian Eye Study.

aData presented are mean (s.e.).

To test the hypothesis that education level could influence the genetic effect of specific variants on SE, we evaluated 40 candidate SNPs recently identified from three GWAS (see Supplementary Material, Table S1), where the current study sample was partially included in the two of three: GWAS on SE (9) and ocular axial length (22) by the Interactional Consortium for Refractive Error and Myopia (CREAM). In the current meta-analysis, 19 out of the 40 SNPs were associated with SE within one of the education strata. The risk alleles of these 19 SNPs were positively associated with a refraction shift towards myopia, consistent with the direction of the main association effect on SE in previous GWAS (Table 2; Supplementary Material, Table S1). Seven SNPs showed evidence of interaction at a PINT-value of <0.05, representing five loci: SHISA6-DNAH9, GJD2, ZMAT4-SFRP1, RBFOX1 and PRSS56. These SNPs that showed an interaction had larger main effects on myopic refractive error than those that did not display an interaction (average β = −0.12 versus −0.07, respectively; Supplementary Material, Fig. S1and Table S1).

Table 2.

SNPs with suggestive association with SE in either education group (P < 0.05) and SNP × education interactions in relation to SE from a meta-analysis of five Singapore studies

Ref Gene SNP Chr BP Allele Lower education group (n = 6354)
 
Higher education group (n = 2107)
 
SNP × education (n = 8461)
 
 Β s.e. P β s.e. P βINT s.e PINT PHET 
(9CD55 rs1652333 207470460 G/A −0.12 0.04 0.001 −0.04 0.07 0.549 0.08 0.08 0.333 0.167 
(11ZC3H11B rs994767 219775432 G/A −0.10 0.04 0.017 −0.08 0.08 0.336 0.02 0.09 0.804 0.100 
(9PRSS56 rs1656404 233379941 A/G −0.12 0.07 0.094 −0.32 0.16 0.050 −0.19 0.18 0.276 0.126 
(10 rs1550094 233385396 G/A −0.07 0.05 0.188 −0.30 0.11 0.0056 −0.23 0.12 0.049 0.397 
(9CHRNG rs1881492 233406998 T/G −0.14 0.06 0.028 0.13 0.13 0.335 0.27 0.15 0.067 0.474 
(9LOC100506035 rs9307551 80530671 A/C −0.08 0.04 0.023 −0.14 0.07 0.050 −0.05 0.08 0.501 0.144 
(9KCNQ5 rs7744813 73643289 A/C −0.11 0.04 0.013 −0.12 0.09 0.163 −0.01 0.10 0.883 0.168 
(9ZMAT4-SFRP1 rs7829127 40726394 A/G −0.07 0.06 0.251 −0.34 0.12 0.0029 −0.28 0.13 0.033 0.793 
(10 rs2137277 40734662 A/G −0.04 0.06 0.487 −0.47 0.12 1.46 × 10−4 −0.42 0.14 0.0021 0.500 
(10TOX rs72621438 60178580 C/G −0.12 0.04 0.002 −0.07 0.08 0.370 0.05 0.09 0.556 0.751 
(9 rs7837791 60179086 G/T −0.12 0.04 0.002 −0.04 0.08 0.551 0.07 0.08 0.382 0.631 
(9BICC1 rs7084402 10 60265404 G/A −0.08 0.04 0.031 −0.20 0.07 0.0047 −0.12 0.08 0.124 0.072 
(10LRRC4C rs1381566 11 40149607 G/T −0.04 0.05 0.428 −0.18 0.09 0.048 −0.14 0.10 0.162 0.512 
(9RDH5 rs3138144 12 56114769 G/C −0.19 0.06 0.001 −0.01 0.11 0.925 0.18 0.13 0.140 0.243 
(9PTPRR rs12229663 12 71249996 A/G −0.08 0.04 0.036 −0.12 0.08 0.116 −0.04 0.09 0.663 0.011 
(9, 10GJD2 rs524952 15 35005886 A/T −0.08 0.04 0.041 −0.31 0.07 1.68 × 10−5 −0.23 0.08 0.0038 0.524 
(9, 10RBFOX1 rs17648524 16 7459683 C/G −0.07 0.06 0.255 −0.37 0.12 0.0015 −0.29 0.13 0.026 0.868 
(9SHISA6-DNAH9 rs2908972 17 11407259 A/T −0.06 0.04 0.087 −0.30 0.07 3.45 × 10−5 −0.24 0.08 0.0035 0.236 
(10 rs2969180 17 11407901 A/G −0.05 0.04 0.165 −0.33 0.07 3.60 × 10−6 −0.28 0.08 4.78 × 10−4 0.377 
Ref Gene SNP Chr BP Allele Lower education group (n = 6354)
 
Higher education group (n = 2107)
 
SNP × education (n = 8461)
 
 Β s.e. P β s.e. P βINT s.e PINT PHET 
(9CD55 rs1652333 207470460 G/A −0.12 0.04 0.001 −0.04 0.07 0.549 0.08 0.08 0.333 0.167 
(11ZC3H11B rs994767 219775432 G/A −0.10 0.04 0.017 −0.08 0.08 0.336 0.02 0.09 0.804 0.100 
(9PRSS56 rs1656404 233379941 A/G −0.12 0.07 0.094 −0.32 0.16 0.050 −0.19 0.18 0.276 0.126 
(10 rs1550094 233385396 G/A −0.07 0.05 0.188 −0.30 0.11 0.0056 −0.23 0.12 0.049 0.397 
(9CHRNG rs1881492 233406998 T/G −0.14 0.06 0.028 0.13 0.13 0.335 0.27 0.15 0.067 0.474 
(9LOC100506035 rs9307551 80530671 A/C −0.08 0.04 0.023 −0.14 0.07 0.050 −0.05 0.08 0.501 0.144 
(9KCNQ5 rs7744813 73643289 A/C −0.11 0.04 0.013 −0.12 0.09 0.163 −0.01 0.10 0.883 0.168 
(9ZMAT4-SFRP1 rs7829127 40726394 A/G −0.07 0.06 0.251 −0.34 0.12 0.0029 −0.28 0.13 0.033 0.793 
(10 rs2137277 40734662 A/G −0.04 0.06 0.487 −0.47 0.12 1.46 × 10−4 −0.42 0.14 0.0021 0.500 
(10TOX rs72621438 60178580 C/G −0.12 0.04 0.002 −0.07 0.08 0.370 0.05 0.09 0.556 0.751 
(9 rs7837791 60179086 G/T −0.12 0.04 0.002 −0.04 0.08 0.551 0.07 0.08 0.382 0.631 
(9BICC1 rs7084402 10 60265404 G/A −0.08 0.04 0.031 −0.20 0.07 0.0047 −0.12 0.08 0.124 0.072 
(10LRRC4C rs1381566 11 40149607 G/T −0.04 0.05 0.428 −0.18 0.09 0.048 −0.14 0.10 0.162 0.512 
(9RDH5 rs3138144 12 56114769 G/C −0.19 0.06 0.001 −0.01 0.11 0.925 0.18 0.13 0.140 0.243 
(9PTPRR rs12229663 12 71249996 A/G −0.08 0.04 0.036 −0.12 0.08 0.116 −0.04 0.09 0.663 0.011 
(9, 10GJD2 rs524952 15 35005886 A/T −0.08 0.04 0.041 −0.31 0.07 1.68 × 10−5 −0.23 0.08 0.0038 0.524 
(9, 10RBFOX1 rs17648524 16 7459683 C/G −0.07 0.06 0.255 −0.37 0.12 0.0015 −0.29 0.13 0.026 0.868 
(9SHISA6-DNAH9 rs2908972 17 11407259 A/T −0.06 0.04 0.087 −0.30 0.07 3.45 × 10−5 −0.24 0.08 0.0035 0.236 
(10 rs2969180 17 11407901 A/G −0.05 0.04 0.165 −0.33 0.07 3.60 × 10−6 −0.28 0.08 4.78 × 10−4 0.377 

P, P-value of SNP effect on SE in each education stratum; PINT, P-value for interaction between SNP and education on spherical equivalent. β, beta regression coefficient of the SNP effect in each education stratum; βINT, interaction beta coefficient, representing the differences in spherical equivalent per risk allele at each SNP comparing individuals in the higher education group to the lower education group; s.e., standard error. PHET, P-value for heterogeneity I2 of SNP effect between five study cohorts. Allele is listed as risk allele/other allele reported from prior GWAS. Genome NCBI build 37; Three genetic loci (in bold) with PINT < 0.01.

The associations at SNPs rs2969180 in SHISA6-DNAH9, rs524952 in GJD2 and rs2137277 in ZMAT4-SFRP on SE were highly significant in the higher education stratum (rs2969180 A allele, β = −0.33 ± 0.07 D, P = 3.60 × 10−6; rs524952 A allele, β = −0.31 ± 0.07 D, P = 1.68 × 10−5; rs2137277 A allele, β = −0.47 ± 0.12 D, P = 1.46 × 10−4; Table 2), while the signals were greatly attenuated in the lower education stratum (all P ≥ 0.041). The strongest predictor in the main model was level of education, as higher educational level was associated with a significant shift towards myopia (β = −1.27, P ≤ 6.00 × 10−6 for all models; Supplementary Material, Table S2). SNPs at three loci also showed significant main effects on SE (SNP main effect: rs2969180 A allele, β = −0.11 ± 0.03 D, P = 8.47 × 10−4; rs524952 A allele, β = −0.12 ± 0.03 D, P = 1.53 × 10−4; rs2137277 A allele, β = −0.13 ± 0.06 D, P = 1.94 × 10−2). Among these genetic variants, SNP rs2969180 in SHISA6-DNAH9 exhibited the most significant interaction with education (βINT for interaction = −0.28 ± 0.08 D, PINT for interaction: 4.78 × 10−4). SNPs rs524952 in GJD2 and rs2137227 in ZMAT4-SFRP showed nominal significance (βINT = −0.23 ± 0.08 D, PINT = 0.0038; βINT = −0.42 ± 0.14 D, PINT = 0.0021; respectively). The magnitude of the effect on myopic refraction was at least 2.8-fold greater per risk allele in the higher education group compared with that in the lower education group. For these three loci, there was no evidence of heterogeneity in the interaction effects across studies (Q test: P-value for heterogeneity ≥0.377, I2 ≤ 5%; Fig. 1).

Figure 1.

Forest plot of the effects of the interaction between genetic variants at SHISA6-DNAH9, GJD2 and ZAMT4-SFRP1 and education on SE in a meta-analysis of five Singapore studies. The interaction beta represents the difference in diopters of SE per risk allele of rs2969180, rs524952 or rs2137227 comparing individuals within the higher education group to the lower education group. Between-study heterogeneity was quantified by the I2 value, and low heterogeneity was defined as an I2 value of 0–25%. SCES, Singapore Chinese Eye Study; SP2, Singapore Prospective Study Program; STARS, Study on Strabismus, Amblyopia and Refractive Error in Singapore Children; SiMES, Singapore Malay Eye Study; and SINDI, Singapore Indian Eye Study.

Figure 1.

Forest plot of the effects of the interaction between genetic variants at SHISA6-DNAH9, GJD2 and ZAMT4-SFRP1 and education on SE in a meta-analysis of five Singapore studies. The interaction beta represents the difference in diopters of SE per risk allele of rs2969180, rs524952 or rs2137227 comparing individuals within the higher education group to the lower education group. Between-study heterogeneity was quantified by the I2 value, and low heterogeneity was defined as an I2 value of 0–25%. SCES, Singapore Chinese Eye Study; SP2, Singapore Prospective Study Program; STARS, Study on Strabismus, Amblyopia and Refractive Error in Singapore Children; SiMES, Singapore Malay Eye Study; and SINDI, Singapore Indian Eye Study.

Next, we assessed whether education modifies the combined genetic effects of carrying increasing number of risk alleles at these three loci on refractive error. The strong interaction with education on SE in the same direction was observed for combined genetic effects (PINT = 4.40 × 10−8, Fig. 2A; Supplementary Material, Table S3). Each additional copy of risk alleles of either SHISA6-DNAH9, GJD2 or ZMAT4-SFRP1 was associated with a −0.35 D shift towards myopia in the higher education group (P = 2.01 × 10−13), while the corresponding figure was a −0.06 D myopic shift in the lower education group (P = 0.014). Accounting for the interaction term of the combined genetic effects with education led to an average of a 1.5-fold increase in the explained variance in SE (Chinese: 0.7% versus 0.46%; Malays: 0.1% versus 0.06% and Indians: 1.0% versus 0.6%) compared with a model with the main effects alone. Consistent with the analysis of SE, a similar effect was observed on axial length, which is one of the most important ocular biometric determinant of refraction (PINT = 6.00 × 10−5; Fig. 2B). Furthermore, we found that subjects with increasing number of risk alleles at these three loci were more likely to have myopia within the higher education stratum compared with the lower education stratum (PINT = 0.012 and 0.0013 for mild or moderate myopia, respectively; Fig. 2C and D). For individuals carrying one additional copy of risk allele at three loci, the odds ratios (ORs) for mild myopia were 1.24 (95% CI: 1.12–1.38, P = 6.23 × 10−5), for moderate myopia 1.35 (95% CI: 1.20–1.52, P = 1.05 × 10−6) and 1.40 for high myopia (95% CI: 1.19–1.65, P = 4.96 × 10−5; Supplementary Material, Table S3) in the higher education group, whereas significantly reduced ORs for mild, moderate or high myopia were observed (OR = 1.06, 95% CI: 1.00–1.12, P = 0.034; OR = 1.06, 95% CI: 0.98–1.15, P = 0.15; OR = 1.14, 95% CI: 0.99–1.32, P = 0.79, respectively) in the lower education group.

Figure 2.

Combined effects of SHISA6-DNAH9, GJD2 and ZMAT4-SFRP1 stratified by educational level on (A) spherical equivalent (SE), (B) axial length, (C) risk of moderate myopia (SE ≤−3.0 D) and (D) mild myopia (SE ≤−1.0 D). The combined genetic effect of SHISA6-DNAH9 rs2969180, GJD2 rs524952 and ZMAT4-SFRP1 rs2137277 was assessed on myopia-related traits, including a variable for the number (or the dosage) of risk alleles to quantify the risk per additional allele for all the three SNPs. The beta coefficients (A and B) or ORs (C and D) with 95% CI were depicted, from a meta-analysis using data from five Singapore cohorts. The sample sizes for each outcome are as follows: SE (n = 8461), axial length (n = 6590), moderate myopia (n = 1472 cases/4530 controls) and mild myopia (n = 2926 cases/4530 controls).

Figure 2.

Combined effects of SHISA6-DNAH9, GJD2 and ZMAT4-SFRP1 stratified by educational level on (A) spherical equivalent (SE), (B) axial length, (C) risk of moderate myopia (SE ≤−3.0 D) and (D) mild myopia (SE ≤−1.0 D). The combined genetic effect of SHISA6-DNAH9 rs2969180, GJD2 rs524952 and ZMAT4-SFRP1 rs2137277 was assessed on myopia-related traits, including a variable for the number (or the dosage) of risk alleles to quantify the risk per additional allele for all the three SNPs. The beta coefficients (A and B) or ORs (C and D) with 95% CI were depicted, from a meta-analysis using data from five Singapore cohorts. The sample sizes for each outcome are as follows: SE (n = 8461), axial length (n = 6590), moderate myopia (n = 1472 cases/4530 controls) and mild myopia (n = 2926 cases/4530 controls).

DISCUSSION

In our study, the most promising interactions between SNPs and education on refractive error were detected at three genetic loci (SHISA6-DNAH9, GJD2, ZMAT4-SFRP1). Similar interaction patterns were also observed for axial length and myopia.

Among these three genetic loci, GJD2 was identified as the first robust myopia-susceptibility gene. It was first discovered in a GWAS of European participants at the genetic marker rs634990 (23), followed by replication in several studies (24–26). Both the CREAM and 23andMe GWAS recently independently pinpointed SNP rs524952 (9, 10), in strong LD with previously identified SNP rs634990 as being involved in the variation of refractive error. The GJD2 gene encodes the connexin protein and GJD2 knockout mice show defects in retinal photoreceptors, implying an important role in the retinal signal transmission (23).

SHISA6-DNAH9 and ZMAT4-SFRP1 are relatively novel loci reported in two recent large GWAS (9, 10). SHISA6 (shisa, homolog 6) has no known link to vision, but the top SNP rs2969180 also lies 94 kb upstream of DNAH9, the nearest gene to SHISA6. The DNAH9 gene encodes an axonemal dynein heavy chain 9, a component to intraflagellar transport (IFT) responsible for cilia motility of the photoreceptor (27). IFT in the connecting cilium is critical for the survival of photoreceptors cells, and consequently defects of IFT result in retinal degeneration, such as retinitis pigmentosa (28, 29). We further examined interactions between education and genetic variants in the gene DNAH9 and found a cluster of SNPs with significant signals (SNP rs4792159, P for interaction = 6.37 × 10−5; Supplementary Material, Fig. S2). The ZMAT4 gene belongs to the zinc finger family, consistent with recent reports of an involvement of several zinc finger proteins in myopia development (11, 30); a neighboring gene SFRP1 (secreted frizzled-related protein 1) is implicated in the growth of photoreceptors cells (31) and the Wnt signaling (32). These genes in or near these loci are linked to signal transduction processes in the retina and hence are likely to be involved in the development of refractive errors.

Education level is thought to reflect the accumulated effect of near work tasks during the first three decades of life (17, 33). Near work, such as reading and writing, has long been considered an important environmental risk factor for the development of myopia (20, 34). Other putative environmental risk factors (e.g. time spent outdoors) might mitigate this process as well (35), whereas the relationship between education level and outdoor activity warrants investigation. One study in Singapore found that intensive near work activity, measured by the increased number of books read per week, had a greater effect on the risk of having myopia in children of myopic parents compared with those with no myopic parents (36). When viewing near objects, eye generates extra optical power to focus the target on the retina to maintain a clear visual image. Across animal models, negative-lens-induced defocus as in near work tasks is known to promote the excessive eye elongation, resulting from the loss of protoglycans and sclera thinning through the turnover of extracellular matrix materials (37). Five genetic loci that showed interactions with education (P for interaction < 0.05; Table 2) can be placed within the biological context of the visually evoked signaling cascade that begins in the retina and mediates sclera remodeling (9, 10). Three genes, DNAH9, SFRP1 and GJD2, play a pivotal role in the retina by maintaining photoreceptor function (27), determining photoreceptor cell degeneration (38) or enabling molecular transduction between these neuronal cells (23). An altered morphology of the photoreceptors, such as photoreceptor elongation (39) and reduced photoreceptor cells density associated with retinal stretching and retinal imaging, has been reported in myopic eyes (40). A recent study also showed that a mutation in opsin gene is responsible for both photosensitive cone dysfunction and vision defect in families with Bornholm Eye Disease and myopia (41). These findings show that genes modulating visual imaging and neurotransmission in the process of retinal cell development play an important role in the ocular growth.

The myopia risk-increasing alleles of rs524952 in GJD2, rs2969180 in SHISA6-DNAH9 and rs2137227 in ZMAT6-SFRP1 are very common, with 60% of our subjects carrying at least three risk alleles at these loci. However, a strong association between these risk alleles and myopia was observed only in individuals with a high level of education (25% of the total sample). This conveys an important message that the influence of genes on the susceptibility to myopia could be highly dependent on which environmental factors are present. Moreover, the risk alleles of the majority of susceptibility loci (13 of 19 SNPs; Table 2) in our study had less or no influence on myopic shift in the lower education category compared with the higher education category. Thus, for certain risk allele carriers, the hereditary predisposition to myopia could be latent or suppressed, if less exposed to the myopiagenic environment associated with education level. Conversely, it is possible that increased exposure to a protective environment, such as outdoor activity, may differentially affect individuals who are more genetically predisposed to develop myopia compared with those at lower risk.

The key determinants of statistical power in the G × E interaction studies are the sample size, the magnitude of the interaction, the allele frequency, the strength of the association for the main effect and the significance threshold. The phenotypic variation accounted for by the interaction term at the individual locus is modest in our data, i.e. average R2 of 0.16% for the most significant SNP rs2969180. Using the same minor allele frequency and magnitude of the interaction term and the main effect of SNP rs2969180 reported in this study, we calculated a post hoc statistical power of 79% to detect the reported interaction effect size at an alpha-level of 1.25 × 10−3 under additive inheritance model (QUANTO (42)). To detect half of the original interaction effect size of this locus, a study of ∼30 000 participants is required to have desired statistical power of 80% at the same level of the statistical rigor. Notably, the top SNPs that showed an interaction have the effect sizes larger than those that did not display an interaction (Supplementary Material, Fig. S1). Our study might not be sufficiently powered to examine G × E interactions for those SNPs with smaller effect sizes. A large-scale study will be beneficial in mapping the details of the interplay between genetics and environmental factors in this scenario.

Finally, several limitations of our study warrant some attention. First, our selection of SNPs relied on a subset of previously identified SNPs achieving genome-wide significance and several SNPs were further excluded after quality-control procedures. Thus, a number of potentially interesting loci may have been missed due to such stringent criteria for SNP inclusion. Second, the transferability of the GWAS index SNPs is influenced by many factors across different populations. There could be other genetic loci interacting with education that failed to be detected in this study, possibly due to the linkage disequilibrium variation and allelic heterogeneity between our samples and previous GWAS cohorts. Although our study contains the largest Asian sample collection available, a much larger sample size will be required to examine the source of such interpopulation heterogeneity (43). In addition, we dichotomized education levels into two categories, higher education (high school, polytechnic or university) versus lower education (lower secondary education or below); this is somewhat arbitrary and may not adequately reflect the true underlying risk factors for myopia. The duration of education (i.e. years of formal education) would likely be a better proxy variable, but was not available in the current data.

In summary, we found that education level influenced the association between recently discovered three genetic loci (SHISA6-DNAH9, GJD2 and ZMAT4-SFRP1) and refractive error. The genetic effects of these loci on myopic refraction were significantly larger within subjects who had a higher level of education compared with a lower education level. Our findings highlight the important role of G × E interactions in the development of myopia.

MATERIALS AND METHODS

Study population

Study participants of Chinese, Malay and Indian origin were drawn from five Singapore cohorts: (i) the SCES, a population-based, cross-sectional survey on eye diseases in Chinese aged 40–80 years (n = 2226) with random sampling drawn from the residents in the Southwestern part of Singapore (18, 44); (ii) the SiMES, a population-based, random sampling survey of eye diseases among elderly Malays residing in the Southwestern part of Singapore (n = 3280) (45, 46); (iii) the SINDI, a population-based survey of eye diseases in a sample of elderly Indians in the Southwestern part of Singapore (n = 3400) (44); (iv) the Singapore Prospective Study Program (SP2), surveying a sample of individuals aged 25 years or older from the entire Singapore population (n = 2867) (47) and (v) the STARS, a population-based survey of Chinese families with children residing in the Southwestern and Western region of Singapore (n = 870; Chinese parents aged 22–58) (48). Individuals <25 years old or with anisometropia >5 D were excluded, as well as those who had undergone cataract surgery, laser refractive procedures or other intra-ocular procedures that could alter refraction.

All studies adhered to the tenets of the Declaration of Helsinki. Ethics approvals were obtained from the Institutional Review Boards of the Singapore Eye Research Institute, Singapore General Hospital, National University of Singapore and National Healthcare Group, Singapore. All participants provided written, informed consent before recruitment into the studies.

Measurements of refractive error, axial length and covariates

All studies used a similar protocol for ocular phenotype measurements. Non-cycloplegic refraction was measured by autorefractor (Canon RK-5, Tokyo, Japan), and ocular axial length was measured using optical laser interferometry (IOLMaster V3.01, Carl Zeiss; Meditec AG Jena, Germany) (44, 45). SE was calculated as the sphere power plus half of the cylinder power for each eye. The mean values of the right and left eyes were used as a quantitative outcome. When data from only one eye were available, the SE of this eye was used. Moderate myopia and mild myopia were defined as SE ≤−3.0 and SE ≤−1.0 D, respectively. Non-myopia controls were defined as SE >−0.5 D.

Age, gender, height and level of education were obtained from all participants who underwent an ophthalmologic examination. For education, subjects chose the most appropriate level of education he/she had completed from the following options: (i) no formal education, (ii) completed primary education, (iii) completed lower secondary (iv) completed high school or polytechnic or (v) completed university. Level of education was categorized into a dichotomous variable: the lower education group including those who had completed lower secondary education or less (i)–(iii) and the higher education group comprising of high schoolers, diploma holders (iv) or university graduates (v).

Genotyping and imputation

We included 40 index SNPs known to be associated with refractive phenotypes (P-value < 5.0 × 108), identified from a recent GWAS meta-analysis of refractive error from the international CREAM (9), a large GWAS of age of onset of myopia (10), and from a GWAS meta-analysis of axial length (22). These 40 SNPs represent 31 unique genetic loci, where 11 loci were associated with refractive error, 8 loci were associated with age of myopia onset, 10 loci were associated with both phenotypes above and 1 locus was associated with axial length.

For SCES, SiMES, SINDI and STARS, the DNA samples were genotyped using the Illumina Human610 Quad Beadchips (Illumina Inc., San Diego, USA). SP2 samples were genotyped using Illumina Human 1Mduov3 and 610 Quad Beadchips. The detailed quality-control procedures of genotyping are provided elsewhere (11, 49). In brief, each study applied stringent quality-control filters to remove SNPs with: (i) per-SNP call rate <95%; (ii) minor allele frequency (MAF) <1% and (iii) Hardy–Weinberg equilibrium (HWE) P-value < 10−6, and individuals with the following conditions: (i) per-sample call rate <95%, (ii) excessive heterozygosity, (iii) cryptic relatedness, (iv) gender discrepancies and (v) deviation in population structure. Population structure in each study was ascertained using principal component analysis (PCA) with the EIGENSTRAT program (50) (Supplementary Material, Figs. S3–S7). Due to the presence of population structure, we adjusted for the top five principal components in the association analyses for SiMES and SINDI datasets. From the PCA plots, we did not observe any evidence that eigenvalues (PC scores) along the top five eigenvectors were more likely to be clustered in any education stratum in each study. No obvious correlation between the top five eigenvectors and SE was noted in our data (data not shown). Imputation was performed using the genotyped data passed the quality-control filtering, together with the 1000 genomes phase 1 cosmopolitan panel haplotypes (March 2012 release). The Markov Chain Haplotyping software was used in the imputation procedure (Minimac software, http://genome.sph.umich.edu/wiki/Minimac, last accessed date on January, 2013). Two SNPs (rs994767, rs1960445) not in 1000 genomes-based imputed data were extracted from the data imputed with reference to HapMap release 22 CHB + JAP population data using IMPUTE version 2(9). These 40 SNPs in our analysis had MAFs >0.01, good imputation quality (imputation quality score >0.44; average: 0.90) if imputed, and were in HWE (P-value > 0.10) (Supplementary Material, Table S1).

Statistical analysis

For each study, a linear regression model at each genotyped or imputed SNP was conducted to determine the association with the mean SE and the interaction with educational attainment. We assumed an additive genetic model where the number of risk allele carried is an ordinal variable (0, 1, 2). For the imputed SNP, we used risk allele dosage, a continuous variable ranging from 0 to 2. Primary analysis included age, age squared, sex and education as covariates as well as SNP and a SNP × education interaction term. Education was treated as a dichotomous variable (0 = lower education; 1 = higher education) and mean-centered age as a continuous variable. Age squared was also adjusted to address the quadratic pattern of SE with age. The beta estimates of SNP effects and corresponding standard errors (s.e.) by the level of education were derived from the primary model. Main effects of SNPs and education levels were estimated from the same primary model of linear regression, except no interaction terms included. The top five principal components of genetic ancestry were included to account for the potential effects of population substructure in SiMES and SINDI (49). The linear regression analyses of the interaction models and parameters estimates were conducted with R version 2.15.2 software (http://www.r-project.org/, last accessed date on September, 2013).

A fixed-effects meta-analysis on the regression coefficients of SNP × education interaction terms as well as SNP effects was conducted using conventional inverse-variance weighting (51). The meta-analysis was performed using METAL software (www.sph.umich.edu/csg/abecasis/metal, last accessed date on May, 2013). A Cochran's Q test or I2 index was used to assess heterogeneity of the beta coefficients across studies. In the meta-analysis, we tested a total of 40 index SNPs that had genome-wide significant evidence of association with refractive errors and related traits in prior GWAS. We used a conservative P-value threshold of 1.25 × 10−3 as a stringent correction for the multiple testing of 40 SNPs at 31 unique loci to achieve a type 1 error rate of 0.05.

To evaluate the combined genetic effect of the multiple genetic loci on refractive phenotypes, we summed the number (or the dosage) of risk alleles of the SNPs carried by each individual. Relations with quantitative traits of SE and axial length were evaluated by a linear regression model, whereas relations with myopia were evaluated by a logistic regression model. Height was also adjusted for the outcome of axial length.

SUPPLEMENTARY MATERIAL

Supplementary Material is available at HMG online.

FUNDING

This study was supported by the National Medical Research Council, Singapore (NMRC 0796/2003, NMRC 1176/2008, STaR/0003/2008; CG/SERI/2010), Biomedical Research Council, Singapore (06/1/21/19/466, 09/1/35/19/616 and 08/1/35/19/550). R.W. acknowledges funding from the US National Eye Institute grant #1K08EY022943.

Acknowledgements

We acknowledge the Genome Institute of Singapore for the genotyping for all study populations. We thank all the participants who volunteered to take part in the studies, and the research teams from the Singapore Eye Research Institute and the National University of Singapore who collected the DNA samples and phenotyped the subjects.

Conflict of Interest statement. None declared.

REFERENCES

1
Saw
S.M.
Katz
J.
Schein
O.D.
Chew
S.J.
Chan
T.K.
Epidemiology of myopia
Epidemiol. Rev.
 , 
1996
, vol. 
18
 (pg. 
175
-
187
)
2
Neelam
K.
Cheung
C.M.
Ohno-Matsui
K.
Lai
T.Y.
Wong
T.Y.
Choroidal neovascularization in pathological myopia
Prog. Retin. Eye Res.
 , 
2012
, vol. 
31
 (pg. 
495
-
525
)
3
Lin
L.L.
Shih
Y.F.
Hsiao
C.K.
Chen
C.J.
Prevalence of myopia in Taiwanese schoolchildren: 1983 to 2000
Ann. Acad. Med. Singapore
 , 
2004
, vol. 
33
 (pg. 
27
-
33
)
4
Wong
T.Y.
Foster
P.J.
Hee
J.
Ng
T.P.
Tielsch
J.M.
Chew
S.J.
Johnson
G.J.
Seah
S.K.
Prevalence and risk factors for refractive errors in adult Chinese in Singapore
Invest. Ophthalmol. Vis. Sci.
 , 
2000
, vol. 
41
 (pg. 
2486
-
2494
)
5
Wu
H.M.
Seet
B.
Yap
E.P.
Saw
S.M.
Lim
T.H.
Chia
K.S.
Does education explain ethnic differences in myopia prevalence? A population-based study of young adult males in Singapore
Optom. Vis. Sci.
 , 
2001
, vol. 
78
 (pg. 
234
-
239
)
6
Au Eong
K.G.
Tay
T.H.
Lim
M.K.
Race, culture and Myopia in 110,236 young Singaporean males
Singapore Med. J.
 , 
1993
, vol. 
34
 (pg. 
29
-
32
)
7
Saw
S.M.
Chua
W.H.
Wu
H.M.
Yap
E.
Chia
K.S.
Stone
R.A.
Myopia: gene-environment interaction
Ann. Acad. Med. Singapore
 , 
2000
, vol. 
29
 (pg. 
290
-
297
)
8
Nakanishi
H.
Yamada
R.
Gotoh
N.
Hayashi
H.
Yamashiro
K.
Shimada
N.
Ohno-Matsui
K.
Mochizuki
M.
Saito
M.
Iida
T.
, et al.  . 
A genome-wide association analysis identified a novel susceptible locus for pathological myopia at 11q24.1
PLoS Genet.
 , 
2009
, vol. 
5
 pg. 
e1000660
 
9
Verhoeven
V.J.
Hysi
P.G.
Wojciechowski
R.
Fan
Q.
Guggenheim
J.A.
Hohn
R.
Macgregor
S.
Hewitt
A.W.
Nag
A.
Cheng
C.Y.
, et al.  . 
Genome-wide meta-analyses of multiancestry cohorts identify multiple new susceptibility loci for refractive error and myopia
Nat. Genet.
 , 
2013
, vol. 
45
 (pg. 
314
-
318
)
10
Kiefer
A.K.
Tung
J.Y.
Do
C.B.
Hinds
D.A.
Mountain
J.L.
Francke
U.
Eriksson
N.
Genome-wide analysis points to roles for extracellular matrix remodeling, the visual cycle, and neuronal development in myopia
PLoS Genet.
 , 
2013
, vol. 
9
 pg. 
e1003299
 
11
Fan
Q.
Barathi
V.A.
Cheng
C.Y.
Zhou
X.
Meguro
A.
Nakata
I.
Khor
C.C.
Goh
L.K.
Li
Y.J.
Lim
W.
, et al.  . 
Genetic variants on chromosome 1q41 influence ocular axial length and high myopia
PLoS Genet.
 , 
2012
, vol. 
8
 pg. 
e1002753
 
12
Li
Z.
Qu
J.
Xu
X.
Zhou
X.
Zou
H.
Wang
N.
Li
T.
Hu
X.
Zhao
Q.
Chen
P.
, et al.  . 
A genome-wide association study reveals association between common variants in an intergenic region of 4q25 and high-grade myopia in the Chinese Han population
Hum. Mol. Genet.
 , 
2011
, vol. 
20
 (pg. 
2861
-
2868
)
13
Shi
Y.
Qu
J.
Zhang
D.
Zhao
P.
Zhang
Q.
Tam
P.O.
Sun
L.
Zuo
X.
Zhou
X.
Xiao
X.
, et al.  . 
Genetic variants at 13q12.12 are associated with high myopia in the Han Chinese population
Am. J. Hum. Genet.
 , 
2011
, vol. 
88
 (pg. 
805
-
813
)
14
Li
Y.J.
Goh
L.
Khor
C.C.
Fan
Q.
Yu
M.
Han
S.
Sim
X.
Ong
R.T.
Wong
T.Y.
Vithana
E.N.
, et al.  . 
Genome-wide association studies reveal genetic variants in CTNND2 for high myopia in Singapore Chinese
Ophthalmology
 , 
2011
, vol. 
118
 (pg. 
368
-
375
)
15
Gwiazda
J.
Deng
L.
Dias
L.
Marsh-Tootle
W.
Association of education and occupation with myopia in COMET parents
Optom. Vis. Sci.
 , 
2011
, vol. 
88
 (pg. 
1045
-
1053
)
16
Wong
T.Y.
Foster
P.J.
Johnson
G.J.
Seah
S.K.
Education, socioeconomic status, and ocular dimensions in Chinese adults: the Tanjong Pagar Survey
Br. J. Ophthalmol
 , 
2002
, vol. 
86
 (pg. 
963
-
968
)
17
Lee
Y.Y.
Lo
C.T.
Sheu
S.J.
Lin
J.L.
What factors are associated with myopia in young adults? A survey study in Taiwan Military Conscripts
Invest. Ophthalmol. Vis. Sci.
 , 
2013
, vol. 
54
 (pg. 
1026
-
1033
)
18
Pan
C.W.
Zheng
Y.F.
Anuar
A.R.
Chew
M.
Gazzard
G.
Aung
T.
Cheng
C.Y.
Wong
T.Y.
Saw
S.M.
Prevalence of refractive errors in a multiethnic Asian population: the Singapore epidemiology of eye disease study
Invest. Ophthalmol. Vis. Sci.
 , 
2013
, vol. 
54
 (pg. 
2590
-
2598
)
19
Morgan
I.G.
Ohno-Matsui
K.
Saw
S.M.
Myopia
Lancet
 , 
2012
, vol. 
379
 (pg. 
1739
-
1748
)
20
Ip
J.M.
Saw
S.M.
Rose
K.A.
Morgan
I.G.
Kifley
A.
Wang
J.J.
Mitchell
P.
Role of near work in myopia: findings in a sample of Australian school children
Invest. Ophthalmol. Vis. Sci.
 , 
2008
, vol. 
49
 (pg. 
2903
-
2910
)
21
Wojciechowski
R.
Nature and nurture: the complex genetics of myopia and refractive error
Clin. Genet.
 , 
2011
, vol. 
79
 (pg. 
301
-
320
)
22
Cheng
C-Y.
Schache
M.
Ikram
M.K.
Young
T.L.
Guggenheim
J.A.
Vitart
V.
MacGregor
S.
Verhoeven
V.J.M.
Barathi
V.A.
Liao
J.
, et al.  . 
Nice loci for ocular axial length identified through genome-wide association studies, including shared loc with refractive error
Am. J. Hum. Genet.
 , 
2013
, vol. 
93
 (pg. 
264
-
277
)
23
Solouki
A.M.
Verhoeven
V.J.
van Duijn
C.M.
Verkerk
A.J.
Ikram
M.K.
Hysi
P.G.
Despriet
D.D.
van Koolwijk
L.M.
Ho
L.
Ramdas
W.D.
, et al.  . 
A genome-wide association study identifies a susceptibility locus for refractive errors and myopia at 15q14
Nat. Genet.
 , 
2010
, vol. 
42
 (pg. 
897
-
901
)
24
Hayashi
H.
Yamashiro
K.
Nakanishi
H.
Nakata
I.
Kurashige
Y.
Tsujikawa
A.
Moriyama
M.
Ohno-Matsui
K.
Mochizuki
M.
Ozaki
M.
, et al.  . 
Association of 15q14 and 15q25 with high myopia in Japanese
Invest. Ophthalmol. Vis. Sci.
 , 
2011
, vol. 
52
 (pg. 
4853
-
4858
)
25
Jiao
X.
Wang
P.
Li
S.
Li
A.
Guo
X.
Zhang
Q.
Hejtmancik
J.F.
Association of markers at chromosome 15q14 in Chinese patients with moderate to high myopia
Mol. Vis.
 , 
2012
, vol. 
18
 (pg. 
2633
-
2646
)
26
Verhoeven
V.J.
Hysi
P.G.
Saw
S.M.
Vitart
V.
Mirshahi
A.
Guggenheim
J.A.
Cotch
M.F.
Yamashiro
K.
Baird
P.N.
Mackey
D.A.
, et al.  . 
Large scale international replication and meta-analysis study confirms association of the 15q14 locus with myopia. The CREAM consortium
Hum. Genet.
 , 
2012
, vol. 
131
 (pg. 
1467
-
1480
)
27
Supp
D.M.
Witte
D.P.
Potter
S.S.
Brueckner
M.
Mutation of an axonemal dynein affects left-right asymmetry in inversus viscerum mice
Nature
 , 
1997
, vol. 
389
 (pg. 
963
-
966
)
28
Hong
D.H.
Pawlyk
B.S.
Shang
J.
Sandberg
M.A.
Berson
E.L.
Li
T.
A retinitis pigmentosa GTPase regulator (RPGR)-deficient mouse model for X-linked retinitis pigmentosa (RP3)
Proc. Natl Acad. Sci. USA
 , 
2000
, vol. 
97
 (pg. 
3649
-
3654
)
29
Badano
J.L.
Mitsuma
N.
Beales
P.L.
Katsanis
N.
The ciliopathies: an emerging class of human genetic disorders
Annu. Rev. Genomics Hum. Genet.
 , 
2006
, vol. 
7
 (pg. 
125
-
148
)
30
Shi
Y.
Li
Y.
Zhang
D.
Zhang
H.
Lu
F.
Liu
X.
He
F.
Gong
B.
Cai
L.
Li
R.
, et al.  . 
Exome sequencing identifies ZNF644 mutations in high myopia
PLoS Genet.
 , 
2011
, vol. 
7
 pg. 
e1002084
 
31
Rodriguez
J.
Esteve
P.
Weinl
C.
Ruiz
J.M.
Fermin
Y.
Trousse
F.
Dwivedy
A.
Holt
C.
Bovolenta
P.
SFRP1 regulates the growth of retinal ganglion cell axons through the Fz2 receptor
Nat. Neurosci.
 , 
2005
, vol. 
8
 (pg. 
1301
-
1309
)
32
Elzi
D.J.
Song
M.
Hakala
K.
Weintraub
S.T.
Shiio
Y.
Wnt antagonist SFRP1 functions as a secreted mediator of senescence
Mol. Cell. Biol.
 , 
2012
, vol. 
32
 (pg. 
4388
-
4399
)
33
Wong
L.
Coggon
D.
Cruddas
M.
Hwang
C.H.
Education, reading, and familial tendency as risk factors for myopia in Hong Kong fishermen
J. Epidemiol. Community Health
 , 
1993
, vol. 
47
 (pg. 
50
-
53
)
34
Saw
S.M.
Chua
W.H.
Hong
C.Y.
Wu
H.M.
Chan
W.Y.
Chia
K.S.
Stone
R.A.
Tan
D.
Nearwork in early-onset myopia
Invest. Ophthalmol. Vis. Sci.
 , 
2002
, vol. 
43
 (pg. 
332
-
339
)
35
Goss
D.A.
Clinical accommodation and heterophoria findings preceding juvenile onset of myopia
Optom. Vis. Sci.
 , 
1991
, vol. 
68
 (pg. 
110
-
116
)
36
Saw
S.M.
Hong
C.Y.
Chia
K.S.
Stone
R.A.
Tan
D.
Nearwork and myopia in young children
Lancet
 , 
2001
, vol. 
357
 pg. 
390
 
37
Rada
J.A.
Matthews
A.L.
Brenza
H.
Regional proteoglycan synthesis in the sclera of experimentally myopic chicks
Exp. Eye Res.
 , 
1994
, vol. 
59
 (pg. 
747
-
760
)
38
Jones
S.E.
Jomary
C.
Grist
J.
Stewart
H.J.
Neal
M.J.
Modulated expression of secreted frizzled-related proteins in human retinal degeneration
Neuroreport
 , 
2000
, vol. 
11
 (pg. 
3963
-
3967
)
39
Liang
H.
Crewther
D.P.
Crewther
S.G.
Barila
A.M.
A role for photoreceptor outer segments in the induction of deprivation myopia
Vision Res.
 , 
1995
, vol. 
35
 (pg. 
1217
-
1225
)
40
Chui
T.Y.
Song
H.
Burns
S.A.
Individual variations in human cone photoreceptor packing density: variations with refractive error
Invest. Ophthalmol .Vis. Sci.
 , 
2008
, vol. 
49
 (pg. 
4679
-
4687
)
41
McClements
M.
Davies
W.I.
Michaelides
M.
Young
T.
Neitz
M.
MacLaren
R.E.
Moore
A.T.
Hunt
D.M.
Variations in opsin coding sequences cause x-linked cone dysfunction syndrome with myopia and dichromacy
Invest. Ophthalmol. Vis. Sci.
 , 
2013
, vol. 
54
 (pg. 
1361
-
1369
)
42
Gauderman
W.J.
Sample size requirements for matched case-control studies of gene-environment interaction
Stat. Med.
 , 
2002
, vol. 
21
 (pg. 
35
-
50
)
43
Luan
J.A.
Wong
M.Y.
Day
N.E.
Wareham
N.J.
Sample size determination for studies of gene-environment interaction
Int. J. Epidemiol.
 , 
2001
, vol. 
30
 (pg. 
1035
-
1040
)
44
Lavanya
R.
Jeganathan
V.S.
Zheng
Y.
Raju
P.
Cheung
N.
Tai
E.S.
Wang
J.J.
Lamoureux
E.
Mitchell
P.
Young
T.L.
, et al.  . 
Methodology of the Singapore Indian Chinese Cohort (SICC) eye study: quantifying ethnic variations in the epidemiology of eye diseases in Asians
Ophthalmic. Epidemiol.
 , 
2009
, vol. 
16
 (pg. 
325
-
336
)
45
Foong
A.W.
Saw
S.M.
Loo
J.L.
Shen
S.
Loon
S.C.
Rosman
M.
Aung
T.
Tan
D.T.
Tai
E.S.
Wong
T.Y.
Rationale and methodology for a population-based study of eye diseases in Malay people: the Singapore Malay eye study (SiMES)
Ophthalmic Epidemiol.
 , 
2007
, vol. 
14
 (pg. 
25
-
35
)
46
Saw
S.M.
Chan
Y.H.
Wong
W.L.
Shankar
A.
Sandar
M.
Aung
T.
Tan
D.T.
Mitchell
P.
Wong
T.Y.
Prevalence and risk factors for refractive errors in the Singapore Malay Eye Survey
Ophthalmology
 , 
2008
, vol. 
115
 (pg. 
1713
-
1719
)
47
Tan
C.E.
Emmanuel
S.C.
Tan
B.Y.
Jacob
E.
Prevalence of diabetes and ethnic differences in cardiovascular risk factors. The 1992 Singapore National Health Survey
Diabetes Care
 , 
1999
, vol. 
22
 (pg. 
241
-
247
)
48
Dirani
M.
Chan
Y.H.
Gazzard
G.
Hornbeak
D.M.
Leo
S.W.
Selvaraj
P.
Zhou
B.
Young
T.L.
Mitchell
P.
Varma
R.
, et al.  . 
Prevalence of refractive error in Singaporean Chinese children: the strabismus, amblyopia, and refractive error in young Singaporean Children (STARS) study
Invest. Ophthalmol. Vis. Sci.
 , 
2010
, vol. 
51
 (pg. 
1348
-
1355
)
49
Fan
Q.
Zhou
X.
Khor
C.C.
Cheng
C.Y.
Goh
L.K.
Sim
X.
Tay
W.T.
Li
Y.J.
Ong
R.T.
Suo
C.
, et al.  . 
Genome-wide meta-analysis of five Asian cohorts identifies PDGFRA as a susceptibility locus for corneal astigmatism
PLoS Genet.
 , 
2011
, vol. 
7
 pg. 
e1002402
 
50
Price
A.L.
Patterson
N.J.
Plenge
R.M.
Weinblatt
M.E.
Shadick
N.A.
Reich
D.
Principal components analysis corrects for stratification in genome-wide association studies
Nat. Genet. 38
 , 
2006
(pg. 
904
-
909
)
51
Kanoni
S.
Nettleton
J.A.
Hivert
M.F.
Ye
Z.
van Rooij
F.J.
Shungin
D.
Sonestedt
E.
Ngwa
J.S.
Wojczynski
M.K.
Lemaitre
R.N.
, et al.  . 
Total zinc intake may modify the glucose-raising effect of a zinc transporter (SLC30A8) variant: a 14-cohort meta-analysis
Diabetes
 , 
2011
, vol. 
60
 (pg. 
2407
-
2416
)

Supplementary data