Common Polymorphisms in the SERPINI2 Gene Are Associated with Refractive Error in the 1958 British Birth Cohort

Abstract

Purpose.: To identify heritable genetic factors altering susceptibility to refractive error in the general population.

Methods.: This was a genetic association study of refractive error investigating genetic polymorphisms in regions previously reported through linkage. Two study panels were drawn from the British 1958 Birth Cohort, composed of 2211 persons 44 years of age at the time of visit. Two main outcomes were considered: refractive error as a continuous outcome (spherical equivalent) and myopia as a diagnosis (defined as spherical equivalent equal to or worse than−1.00 diopter). Genotyping was initially performed in 1188 subjects from the outer tertiles of the population distribution, using customized arrays of single nucleotide polymorphisms (SNPs) saturating regions of previously reported highly significant linkage. In a second stage, SNPs most significantly associated were validated in 1023 more persons. Findings were investigated further through human fetal expression studies.

Results.: Polymorphisms within the SERPINI2 gene were associated with refractive error in two different European subgroups from the 1958 British Birth Cohort (meta-analysis P = 7.4E-05 for rs9810473). Association was also significant for myopia (best association: OR = 0.80; 95% CI, 0.69–0.93; P = 0.003 for rs10936538). Expression profiling of SERPINI2 revealed that the gene is expressed in the retina and in other eye and CNS tissues.

Conclusions.: The novel association of SERPINI2 with refractive error and myopia is suggestive of a possible link between physiological pathways controlling eye growth and development and those controlling glucose metabolism. The findings indicate that SERPINI2 is a promising candidate for further investigations of the genetic susceptibility to myopia.

A major challenge for the 21st century is to find the genetic factors that influence susceptibility to common complex diseases and traits. Refractive error is an archetypal complex quantitative trait, with myopia at one end of the distribution and hypermetropia at the other. Refractive error is the commonest cause of visual impairment globally and has major socioeconomic costs for the affected person and for society as a whole; for example North Americans are reported to spend at least $12 billion annually on glasses, contact lenses, and refractive surgery. ¹  

Myopia affects at least 25% of white North Americans and Europeans ² and up to 60% of Asians but only 5% of Africans. ^3,4 The prevalence and severity of myopia have been increasing in recent years, especially in Southeast Asia. ^3,4 This suggests that, against a background of genetic susceptibility, changing environmental factors, possibly related to rapid urbanization, play a role in the development of refractive error. 

Family aggregation studies have shown a correlation between parental refractive error and that of their children, ⁵ and twin studies have reported high heritability of 60% to 90% in the general population. ⁶ However, until very recently, genetic research has focused on linkage analysis of cohorts of families, often with “syndromic” or “pathologic” myopia, whose similarities with common, population-wide myopia are uncertain. The first linkage study of refractive error based on dizygotic twins reported four linkage peaks on 11p13, 8p23, 3q26, and 4q12. ⁷ We investigated candidate genes in the top three linkage peaks from that study in a total of 2211 subjects from the 1958 British Birth Cohort and report findings from this candidate gene study and subsequent expression studies. 

Methods

Candidate Gene Association Study

The 1958 British Birth Cohort originally consisted of 17,000 persons who were born during one week in March 1958 and were subsequently followed up at intervals ⁸ with physical examination and collection of data on environmental, social, and lifestyle factors. In 2002 to 2003, when the participants were aged 44 to 45 years, a biomedical survey was undertaken of the 9377 individuals still actively participating in the study. Of all the subjects, 2485 (27%) were selected randomly and underwent noncycloplegic autorefraction with an autorefractor (Retinomax 2; Nikon, Tokyo, Japan) ⁹ (equipment costs precluded autorefraction of all subjects). After exclusion of 251 subjects for whom no biological material was available and 23 subjects with high anisometropia, this study was conducted in two stages. In 1188 subjects in a discovery stage, fine-mapping was performed of the regions of interest through custom genotyping. In a replication stage of another 1023 subjects, not overlapping with those participating in the first stage, associations were examined in subjects using genomewide data. Autorefractor measurements in this age group provided a reliable objective measure of refractive error at the point at which primary myopia became manifest but before myopia secondary to other ocular disorders occurred. ⁹ This avoided misclassification and ensured phenotypic stability. Blood was taken for the creation of immortalized cell lines to provide a continuous source of DNA. ⁸  

Refractive error was quantified using the summary measure of spherical equivalent of refraction (SphE) in diopters (D). We used autorefraction readings to calculate SphE for each eye of every subject in the conventional way (SphE = S + C/2), where S is the sphere and C is the cylinder). Subjects with highly discordant SphE for both eyes (within the worst 5% of the sample) were excluded. ^10,11 Inclusion criteria for the first stage (linkage fine-mapping and SNP association discovery stage) were the mean spherical equivalent within either of the outer tertiles of the distribution for this trait (half the participants in each tertile). The lower tertile spanned the interval −12.875D to −1.3125D, whereas the upper tertile spanned the interval −0.3125D to +6.50D. As a categorical trait, myopia was determined as a mean spherical equivalent of −1D or less and nonmyopia as 0D or above. Subjects whose spherical equivalent was between 0 and −1D were not included in the qualitative analysis to remove uncertainties of myopia classification inherent in any arbitrary dichotomization of quantitative traits. ¹² All participants gave informed written consent to participate in genetic association studies, and the study and biomedical examination protocols were approved by the South East MultiCentre Research Ethics Committee and the Oversight Committee for the 1958 British Birth Cohort. The research adhered to the tenets of the Declaration of Helsinki. 

The discovery stage of the study was designed with two phases of genotyping. The first phase of genotyping consisted of fine-scale genotyping of 1536 single nucleotide polymorphisms (SNPs) covering 3 of the 4 linkage peaks on 11p13, 8p23, and 3q26 reported by Hammond et al. ⁷ on 590 persons of Caucasian ancestry from the 1958 Birth Cohort, with autorefractor measures randomly selected from the outer tertiles of the population distribution. These SNPs were selected so that they could provide information on genetic variations from 111 genes. The fourth peak reported in the original study ⁷ was the one with the weakest evidence for linkage and coincidentally was the widest one. Based on cost efficiency considerations, it was not analyzed as part of the present work. Linkage disequilibrium patterns from the HapMap Phase 2 CEU population were used to inform the choice of these SNPs. Selection of the tag SNPs was made using Tagger. ¹³ An r ² of at least 0.8 was required for genomic regions inside the genes, and an r ² of 0.7 or above was required for the genomic regions that were not known to code for any transcript. Genotyping was performed using a methylation assay platform (GoldenGate; Illumina, San Diego, CA). Approximately one-third of the strongest associations obtained (503 SNPs) from the first phase were taken forward and genotyped in an additional 598 subjects from the 1958 British Birth Cohort, also drawn from the outer tertiles of the refractive error distribution, in a second phase. As in the previous phase, genotyping was carried out using the same methylation assay platform (GoldenGate; Illumina). 

In the replication stage of the analysis, we initially obtained and used data from all (N = 1667) Caucasian participants in the 1958 Birth Cohort with autorefractor data who had undergone genotyping using genomewide SNP chip arrays in the course of other research. Subjects who had been included in the discovery stage of our study were excluded in the second stage using a relatedness analysis. Relatedness >80% calculated from loci genotyped at the first stage was interpreted as proof of identity, and the 644 samples concerned were removed from the replication panel. The subjects' Caucasian ancestry in the replication stage was confirmed by eigenvector analysis. ¹⁴ The 1023 subjects ultimately included in the replication stage had been genotyped using several platforms, including some of very high density (Affymetrix 5 and 6; Illumina Human1M-Duo and Omni 660W-Quad and Illumina HumanHap550 SNP chips). 

Statistical Analysis

After genotyping, SNPs were included in subsequent analytical stages if they were in Hardy-Weinberg equilibrium (P > 0.01), polymorphic (minor allele frequency at least) in the population sample, and above a threshold genotyping rate across the sample (>95%). 

We considered two phenotypic outcomes: the spherical equivalent as transformed using rank-based inverse normal transformation to overcome nonnormality of phenotype distribution (Fig. 1) and myopia as a categorical disease status using the criterion of a spherical equivalent of −1.00 D or more extreme for myopia, with all subjects with 0 D or higher classified as nonmyopes, as in previously published studies. ^3,15 Association was tested using linear and logistic regression models; PLINK ¹⁴ and Stata 10 (Stata Corp, College Station, TX) were used for the analysis. Correction for multiple testing was made either through the Bonferroni method or through phenotype-swapping 1 million permutations to control for multiple testing where linkage disequilibrium was so high that the information made available from multiple SNP loci became highly redundant. 

Figure 1.

 

Full distribution of spherical equivalent refractive error in the 1958 British Birth Cohort. Vertical dashed lines: values −1.3125 D and 0.3125 D, used to define our tertiles.

All genotypes obtained at the various stages of the work were converted with their positive strand alleles as reference. Information on variation at loci that were not directly genotyped in our assays at both stages were imputed using the program MACH ¹⁶ using the HapMap Phase 2 CEU population as a reference. Four hundred Markov Chain iterations accounting for the 300 most common haplotypes overlapping SERPINI2 were used to provide maximal quality of imputation (error rate equaled 2e-04). Metal was used to calculate fixed-effect inverse variance weighted meta-analysis. 

Expression in Fetal Tissues.

Human embryonic and fetal material was provided by a licensed tissue bank, Human Developmental Biology Resource (HDBR), UCL Institute of Child Health, with National Research Ethics approval. 

Cloning.

We designed genomic amplicons to cover the SERPINI2 coding regions (see Table 4). PCRs were performed in 96-well plates. Fetal total cDNA (50 ng, supplied by HDBR) was amplified using the Invitrogen Platinum® Taq DNA polymerase PCR Kit. PCR products were purified using a PCR purification kit (QIAquick; Qiagen, Valencia, CA) and were ligated (pGEM T-Easy; Promega, Madison, WI). 

Production of DIG-Labeled Probes.

Antisense and sense probes were prepared by linearizing plasmids for SERPINI2 A, B, and C with SpeI and SacII, respectively. Digoxigenin-UTP was incorporated into riboprobes during in vitro transcription using the DIG RNA labeling mix (Roche, Indianapolis, IN) according to the manufacturer's instructions. Antisense and sense probes were generated using T7 and SP6 polymerase, respectively. Probe concentration was measured (Nanodrop; Thermo Scientific, Waltham, MA). 

Tissues Tested.

Sections from seven human embryos and four fetuses were tested. Samples ranged from 47 to 63 days of gestation. 

In Situ Hybridization.

In situ hybridization was carried out as described by Wilkinson. ¹⁷ Human embryonic and fetal tissue was dissected and fixed in 4% paraformaldehyde (PFA) in phosphate-buffered saline (PBS) overnight at 4°C. After fixation, tissues were dehydrated and embedded in paraffin wax. Sagittal sections were cut at 8 μm using a standard microtome and attached to microscope slides (Superfrost Plus; VWR, Leicestershire, UK). Before hybridization, tissue sections were dewaxed, hydrated, fixed in 4% PFA/PBS, and rinsed twice with PBS. Proteins were removed by incubation with proteinase K (20 mg/mL) in PBS. After washing with PBS, the sections were refixed in the same PFA solution and treated with 0.1 M triethanolamine containing 0.25% acetic anhydride. Slides were dehydrated through an alcohol series and air dried. 

Hybridization solution contained riboprobe (300 ng DIG-labeled RNA probe), RNAguard (1 μL/mL), and tRNA (0.5 mg/mL) in hybridization buffer (50% formamide, 0.3 M NaCl, 20 mM Tris-HCl, pH 7.5, 5 mM EDTA, pH 8.0, 10% dextran sulfate, and 1× Denhardt's solution). A 100-μL aliquot of hybridization probe was added to each slide, which was incubated in a sealed chamber moistened with 50% formamide/1× standard saline citrate (SSC) overnight at 65°C. Stringency washes were performed in the following order: 2× SSC (twice at 65°C), 50% formamide/2× SSC (twice at 65°C), 2× SSC (twice at 65°C), 0.2× SSC (65°C), and 0.2× SSC (65°C cooled to room temperature). Slides were then incubated for 1 hour in 150 mM NaCl and 100 mM Tris-HCl, pH 7.5, containing 10% fetal calf serum (FCS). For antibody detection, slides were incubated in anti-digoxigenin antibody conjugated with alkaline phosphatase (anti-Dig antibody diluted 1:1000, containing 2% FCS) overnight at 4°C. Expression patterns were visualized using the NBT/BCIP system, and sections were mounted in aqueous mounting medium (VectaMount; Vector Laboratories, Burlingame, CA) and analyzed using an imaging system (Axioplan 2; Zeiss, Thornwood, NY). 

Results

Of the original 1536 SNPs, 1472 (95.83%) were genotyped successfully in 590 subjects in phase 1 of the study. Of all the SNPs included in the analysis, the strongest association with spherical equivalent was observed for a genomic segment spanning 36 kb on chromosome 3 (P = 0.000022 for rs9810473 and other SNPs in perfect LD with it; data not shown.), which overlaid the serine protease inhibitor clade I member 2 (SERPINI2) genomic coding region. This association was stronger than what would have been expected from simple multiple testing (P = 0.03 after Bonferroni multiple testing adjustment). SNPs in which the strongest association was observed, including 20 SNPs overlapping the SERPINI2 gene and flanking adjacent regions, were genotyped in an additional 598 subjects from the British Birth Cohort (Table 1 and Supplementary Table S1). Imputation from genotypes of all subjects participating in phases 1 and 2 of the discovery stage was performed as described to improve the accuracy of the localization of the association signal. 

Table 1.

 

Best Association Results for the SNPs Located within the SERPINI2 Genomic Sequence or Its Immediate Vicinity from the Discovery Stage for the Spherical Equivalent

Chromosome	SNP	Position on Build 36	N	Regression Slope	SE	P
3	rs6784768	168639049	1188	−0.1019	0.029	0.00043
3	rs9864094	168642525	1188	−0.1019	0.029	0.00043
3	rs9844202	168642528	1188	−0.1019	0.029	0.00043
3	rs9810473	168642674	1188	−0.1019	0.029	0.00043
3	rs9848929	168643017	1188	−0.1019	0.029	0.00043
3	rs9884112	168645652	1188	−0.1019	0.029	0.00043
3	rs2272138	168646881	1188	−0.1019	0.029	0.00043
3	rs13085856	168647355	1188	−0.1019	0.029	0.00043
3	rs6444326	168641564	1188	−0.1019	0.029	0.00043
3	rs35216909	168642415	1188	−0.1019	0.029	0.00043

 Effect sizes and SEs are standardized. Results shown are uncorrected.

A summary of the probabilities for association with refractive error for polymorphisms, directly genotyped or imputed, within the SERPINI2 genomic sequence is shown in Table 2 and Supplementary Table S2). Analysis of the 1188 subjects genotyped revealed strong association for a number of SNPs adjacent to one another and located within the 3′ end of the SERPINI2 gene transcript (best association P = 0.00043 for rs9810473 [standard linear regression coefficient β = −0.10] and a number of other SNPs in proximity with it). This SNP and others remained significant after correcting for multiple testing using 1 million phenotype-swapping permutations (P = 0.0006). 

Table 2.

 

Summary of the Association with Refractive Error for Selected SNPs for the SERPINI2 Loci, in Each of the Phases of This Study, Directionality of Effect, and Meta-Analysis Association

SNP	Bp	Allele 1	Allele 2	Stage 1 P	Stage 2 P	Direction of Effect	Standardized Effect	Meta P
rs6784768	168639049	T	C	0.0004	0.02	—	−0.1112	0.00007
rs9810473	168642674	A	G	0.0004	0.02	++	0.1112	0.00007
rs9864094	168642525	T	C	0.0004	0.02	++	0.111	0.00008
rs9844202	168642528	T	C	0.0004	0.02	—	−0.1108	0.00009
rs9848929	168643017	A	G	0.0004	0.02	—	−0.1108	0.00009
rs9884112	168645652	T	G	0.0004	0.02	++	0.1108	0.00009
rs2272138	168646881	A	G	0.0004	0.02	++	0.1108	0.00009
rs13085856	168647355	T	C	0.0004	0.02	—	−0.1108	0.00009
rs9848915	168642985	A	G	0.0006	0.02	—	−0.1091	0.0001
rs10936539	168674811	C	G	0.004	0.03	—	−0.1019	0.0004
rs13074812	168638486	C	G	0.06	0.05	—	−0.0923	0.001
rs9860065	168645314	A	G	0.09	0.02	—	−0.0658	0.02

 Stage 1 and stage 2 refer to the discover and replication stages, respectively. Values shown are uncorrected.

We extracted SERPINI2 genotypes within the associated interval (168522964–168806403 on genomic build 36) from 1023 additional subjects participating in the 1958 British Birth Cohort in the replication stage. We fitted the genotypes and their refractive error measurements and SERPINI2 genotypes in linear regression models and noticed that a number of SNPs identified in association with the phenotype in the discovery stage were also nominally associated, with the same directionality of association in the replication stage (Supplementary Table S3). Evidence for association remained strong for SNP rs9810473 (P = 0.02). In the second stage, the strongest effects were observed for rs6784768; each copy of the rarer allele T in the replication stage lowered refraction by an average 0.26 SE units (corresponding to 0.52 D) versus 0.1 SD units in the first stage (corresponding to 0.27 D). 

Combining both results confirmed the association at that locus (P = 7.4 × 10⁻⁵), as shown in Table 2 and Figure 2. The overall meta-analysis standardized effect was −0.11 (approximately −0.22 D change for each additional copy of susceptibility allele) and SE. SE = 0.03 for these SNPs. 

Figure 2.

 

Association in and around the SERPINI2 gene. The blue diamond represents the SNP with the best association (shown in a logarithmic scale on the y-axis). The colors of the other diamonds represent the linkage disequilibrium with the lead SNP (red for a pairwise r ² 0.8–1, orange for r ² 0.6–08, and yellow for r ² 0.4–0.6 between the SNP they represent and the lead SNP). The blue line represents the recombination rate as calculated in the HapMap CEU subjects. SERPINI2 is superimposed at the bottom of the figure.

We further looked at associations between SERPINI2 and clinical myopia. We compared allele frequency distributions of subjects with myopia (<−1 D) with those of nonmyopic subjects (≥0 D) and found good evidence for associations between SERPINI2 SNPs and myopia, with consistency in the effect for all loci studied (Supplementary Tables S4, S5). The main association originated from a group of tightly linked markers, as shown in Figure 2. As an example, rs10936538 is associated with myopia on both stages. The minor T allele on that locus significantly decreased the odds of having myopia (meta-analysis odds ratio [OR] = 0.80; 95% confidence interval [CI], 0.69–0.93; uncorrected P = 0.004) as shown in Figure 3. The SNP rs6784768, previously associated with refractive error, was equally associated with myopia (meta-analysis OR = 1.44; 95% CI, 1.10–1.89 for allele T). Association with myopia was overall significant for a number of other SNPs (Table 3). 

Figure 3.

 

ORs (horizontal axis) for SNP rs10936538 in the two panels used in this study and the meta-analysis ORs (vertical axis).

Table 3.

 

Association with Myopia for Selected SNPs in Each of the Stages of this Study and Meta-Analysis of their Effects

SNP	Allele 1	Allele 2	Discovery Stage				Replication Stage				Meta P	Heterogeneity of Effect
			OR	Lower Bound	Upper Bound	P	OR	Lower Bound	Upper Bound	P
rs9848915	A	G	1.63	1.16	2.30	0.005	1.33	0.80	2.22	0.28	0.004	0.52
rs9884112	G	T	1.64	1.16	2.33	0.005	1.33	0.80	2.22	0.28	0.004	0.50
rs9864094	C	T	1.64	1.16	2.33	0.005	1.33	0.80	2.22	0.28	0.004	0.50
rs2272138	G	A	1.64	1.16	2.33	0.005	1.33	0.80	2.22	0.28	0.004	0.50
rs9844202	T	C	1.64	1.16	2.33	0.005	1.33	0.80	2.22	0.28	0.004	0.50
rs13085856	T	C	1.64	1.16	2.33	0.005	1.33	0.80	2.22	0.28	0.004	0.50
rs9848929	A	G	1.64	1.16	2.33	0.005	1.33	0.80	2.22	0.28	0.004	0.50
rs11926880	C	A	0.84	0.70	1.00	0.06	0.73	0.56	0.95	0.02	0.004	0.39
rs10936538	T	C	0.83	0.69	0.99	0.04	0.75	0.58	0.98	0.04	0.004	0.54
rs6784768	G	A	1.64	1.16	2.33	0.005	1.20	0.79	1.83	0.40	0.007	0.26

 All the values are uncorrected.

Serine protease inhibitor clade I member 2 (SERPINI2) is a member of the plasminogen activator inhibitor-1 family, a subset of the serpin superfamily, and these proteins act as tissue-specific plasminogen-activator (tPA) inhibitors. ¹⁸ The limited literature on gene function indicates that SERPINI2 has a complex role in the control of growth, with a knockout mouse model having apoptosis of the acinar cells. ¹⁹ SERPINI2 expression has been reported to be inversely correlated with tumor growth in breast, pancreatic, and other types of cancer. ^{20 –22} However, there was no published report of SERPINI2 expression in ocular tissues at the time of our study. For this reason we investigated the presence of SERPINI2 in human eyes. Using in situ hybridization (Table 4), we sought to assess the expression of SERPINI2 transcript expression in the human fetal and embryonic eye and brain. 

Table 4.

 

Primers Used for SERPINI2 Cloning

Exons	Primer Sequence (5′→3′)	Size of PCR Product (bp)
2–3	Forward: TGG ACA CAA TCT TCT TGT GGA	438
	Reverse: CAC AAG CCT TTG CAT CTT GA
5–7	Forward: TGG CTC TCT GAG ATG CAA GA	179
	Reverse: TGC GTC ACT TGG GAA ACA TA
2–3	Forward: GAG CCA AAG GAA AAG CAC AG	245
	Reverse: TGC ATC TTG AAA ATC CAC CA

Antisense probes to SERPINI2A and SERPINI2C showed similar patterns of expression in the eye and brain. SERPINI2B antisense probe did not yield any reproducible results. 

We found SERPINI2 to be expressed in the anterior epithelium of the lens and within the lens (Fig. 4A). The anterior surface of the lens is covered by an epithelial cell layer that extends to the equatorial zone. These cells proliferate and at the equatorial zone migrate posteriorly, elongate, and differentiate to produce the lens fibers. SERPINI2 may therefore have a role in the production of lens fibers. SERPINI2 expression was also seen in discrete cells within the neural retina and mesenchyme around the eye. Additional markers would be needed to identify which cell types express SERPINI2 in the neural retina. 

Figure 4.

 

RNA in situ hybridization of SERPINI2 in fetal and embryonic eye and brain. At 54 days of gestation (Carnegie stage 22), SERPINI2 is expressed in the anterior lens epithelium (LE), within the lens (L), in discrete cells of the neural retina (NR), and in mesenchyme (ME) around the eye (A). Expression is also seen in the cortical plate (CP), hippocampus (HP), mesencephalon (M), and thalamus (TH) (B). At 63 days of gestation, SERPINI2 is expressed in the seam of the fusing eyelids (FE) and within the lens (LE) and neural retina (NR) (C).

SERPINI2 was expressed in the cortical plate, hippocampus, mesencephalon, and thalamus (Fig. 4B). The mesencephalon includes the superior colliculi, an area known to have a visual function. SERPINI2 was expressed in the fetal lens and neural retina at 63 days of gestation (Fig. 4). SERPINI2 was also shown to be expressed at the site of fusion of the eyelids. 

Discussion

Based on a joint analysis of two subpopulations drawn from the 1958 British Birth Cohort, we report that polymorphisms in a novel candidate gene, SERPINI2, are associated with refractive error. According to our results, SERPINI2 polymorphisms increase susceptibility to refractive error; odds of having myopia were also significant in the presence of specific alleles (OR = 0.80; 95% CI, 0.69–0.93 for allele T of rs10936538). 

The Serpin gene family contains numerous members, and SERPINI2 belongs to the serine protease inhibitor (serpin) family. In a mouse model of pancreatic insufficiency, deletion of the SERPINI2 gene of causes apoptosis of the acinar cells. ¹⁹ Many of the symptoms in this model of pancreatic insufficiency can be attributed to malnutrition, with the resultant effects on growth and immune function. ¹⁹  

Our finding of an association of Serpini2 with refractive error and myopia raises intriguing questions about pathogenesis. SERPINI2 is involved in acinar cells of the pancreas and may be important to control glucose metabolism. The latter has been hypothesized to be linked to the development of myopia. ^23,24 Experimental findings in animal models indicate that insulin and its metabolic pathways are implicated in retinal signaling, which results in excessive eye growth and myopia. ^23,25 Signaling between the neural retina and lens is known to be necessary for eye growth and development. ²⁶ Expression of SERPINI2 in the anterior epithelium of the lens, within the lens, and in discrete cells within the retina is compatible with a role in eye development. Thus it is possible that SERPINI2 is implicated in refractive error through more than one mechanism. Further functional characterization is needed of the properties of the SERPINI2 protein and the physiologic pathways in which it participates. 

Polymorphisms located on chromosome 3 that we report here to be in association with refractive error and myopia are unlikely to be alone in predisposing to these traits. The presence of other SNPs that were initially found in association in the first genotyping group but were not followed up in the second (such as those in chromosomes 8 and 11), as well as variants that were not associated at all in the present work either because they lacked sufficient power or because they lay outside the linkage regions of interest to us, suggest that SERPINI2 polymorphisms could potentially be just a few of what is likely to be a complex constellation of genetic factors behind refractive error and myopia. 

There are some issues to be considered regarding the results of our study. First, although the effect size and direction of effect were similar in discovery and replication data sets, the replication data set achieved lesser statistical significance than the discovery cohort. The most likely explanation is that the distribution of the refractive error was different in the two panels, though they were drawn from the same cohort. Although the opposing tertiles were selected for the discovery stage, subjects used in the replication stage were drawn from the remainder of the same cohort, which meant the replication sample was not enriched for extreme cases of refractive error. Richness in extremes of phenotypes is reflected in the power of association analyses. This was particularly true when the effect on myopia as a dichotomous trait was studied. Second, by comparing myopic with all nonmyopic subjects, we are likely to be reporting effect sizes higher than would be obtained comparing myopic with emmetropic subjects. This comparison was made necessary by the design adapted in this study, which was aimed at ensuring the highest power for refractive error association detection. Third, our panel consisted of Caucasian subjects, and it is possible that they were not fully generalizable to populations of different racial or ethnic composition. Fourth, we cannot exclude the possibility that variants other than those in SERPINI2, either in short or long distances from this gene, account for at least part of the linkage signal and variance of spherical equivalent observed in the original linkage study. Many complex genetic traits are controlled by multiple loci of small effect. ²⁷ Although the power to detect effects such as the ones reported here, at a significance threshold of α = 0.001 would be 0.90 for N = 1188 (our discovery sample), for smaller effect sizes the power would have been considerably smaller. 

Refractive error is increasingly relevant to public health in both developed and developing countries. The present lack of knowledge about environmental risk factors makes it difficult to develop effective strategies for prevention and treatment. Identifying genetic variants could lead to better understanding of the pathogenesis of refractive error. In this study, we have observed the association of variants in SERPINI2 with refractive error. Investigation of the function of this gene—only very recently recognized to be expressed in the human eye—in relation to ocular development could lead to a better understanding of the pathophysiology of this complex trait and thus ultimately to therapeutic or preventive strategies. 

Supplementary Materials

Table st1, DOC - Table st1, DOC 

Footnotes

 Supported by the Medical Research Council Component Project Grant G0301069 and Health of the Public Project Grant G0000934. PGH is funded by the EU MyEuropia Marie Curie Research Training Network.

Footnotes

 Disclosure: P.G. Hysi, None; C.L. Simpson, None; Y.K.Y. Fok, None; D. Gerrelli, None; A.R. Webster, None; S.S. Bhattacharya, None; C.J. Hammond, None; P.C. Sham, None; J.S. Rahi, None

Human embryonic and fetal material was provided by the MRC/Wellcome-funded Human Developmental Biology Resource, UCL Institute of Child Health.