Systematic Investigation of the Relationship between High Myopia and Polymorphisms of the MMP2, TIMP2, and TIMP3 Genes by a DNA Pooling Approach

Abstract

Purpose.: This study examined the relationship between high myopia and three myopia candidate genes—matrix metalloproteinase 2 (MMP2) and tissue inhibitor of metalloproteinase-2 and -3 (TIMP2 and TIMP3)—involved in scleral remodeling.

Methods.: Recruited for the study were unrelated adult Han Chinese who were high myopes (spherical equivalent, ≤ −6.0 D in both eyes; cases) and emmetropes (within ±1.0 D in both eyes; controls). Sample set 1 had 300 cases and 300 controls, and sample set 2 had 356 cases and 354 controls. Forty-nine tag single-nucleotide polymorphisms (SNPs) were selected from these candidate genes. The first stage was an initial screen of six case pools and six control pools constructed from sample set 1, each pool consisting of 50 distinct subjects of the same affection status. In the second stage, positive SNPs from the first stage were confirmed by genotyping individual samples forming the DNA pools. In the third stage, positive SNPs from stage 2 were replicated, with sample set 2 genotyped individually.

Results.: Of the 49 SNPs screened by DNA pooling, three passed the lenient threshold of P < 0.10 (nested ANOVA) and were followed up by individual genotyping. Of the three SNPs genotyped, two TIMP3 SNPs were found to be significantly associated with high myopia by single-marker or haplotype analysis. However, the initial positive results could not be replicated by sample set 2.

Conclusions.: MMP2, TIPM2, and TIMP3 genes were not associated with high myopia in this Chinese sample and hence are unlikely to play a major role in the genetic susceptibility to high myopia.

In myopia, the images of distant objects are focused in front of, rather than on, the retina under relaxed accommodation. Myopia is the commonest eye anomaly in the world and imposes a huge impact on the public health care system and the economy. ¹ In particular, subjects with high myopia, usually defined as ≤ −6.0 D, are more prone to ocular degenerative changes such as glaucoma and retinal detachment. Myopia is much more frequent in Asians (60%–80%) than in Caucasians (10%–25%), although its prevalence varies with time, the age of the subjects, and the ethnic origin of the population concerned. ² In Hong Kong, the prevalence is highest (70%) for age 19 to 39 years and then drops after age 40. ³  

Both environmental and genetic factors contribute to myopia, although the exact cause of myopia remains to be determined. ^{4 –6} Environmental factors such as lifestyle, schooling, near-work, and outdoor activities are known to contribute to differences in the prevalence of myopia. Estimates of heritability are high for refractive error and major ocular components, and shared genes between relative pairs could explain the strong correlation between refractive error and axial length. ^{7 –9}  

Myopia mainly results from elongated eyeballs caused by accelerated postnatal eye growth, rather than changes in corneal or lens power. ¹⁰ During myopia's development, the sclera undergoes active remodeling, which involves matrix metalloproteinases (MMPs) and tissue inhibitors of metalloproteinases (TIMPs)—the enzymes involved in the degradation of extracellular matrix. MMP2 is increased in the sclera of the eye with myopia induced by form deprivation in chicks when compared with the control eye, and the increased expression has been consistently shown for both the protein ^{11 –14} and the mRNA transcript. ^14,15 Increased scleral MMP2 expression in form-deprivation myopia has also been shown in tree shrews at both the protein ¹⁶ and the mRNA level, ^17,18 and in guinea pigs at the protein level. ¹⁹ An increased MMP2 transcript level has also been found in human scleral fibroblasts mechanically stretched in an in vitro system ²⁰ and in lens-induced myopia in the tree shrew, ²¹ but not in lens-induced myopia in the chick. ²² On the other hand, there has been less extensive study of TIMP expression in induced myopia. TIMP2 expression is found to be reduced in form deprivation myopia in chicks ¹⁴ and in lens-induced myopia in guinea pigs, ¹⁹ but at levels comparable to those of the control eye in lens-induced myopia in both tree shrews ²¹ and chicks. ²² Finally, the TIMP3 transcript level is found to be reduced in lens-induced myopia in the tree shrew. ²¹ These studies did not specifically examine any potential interaction among these three genes. 

We used a case–control study approach ²³ to examine the relationship between high myopia in a Han Chinese population and the tag single-nucleotide polymorphisms (SNPs) of three candidate genes. These three candidate genes were selected for this study because their involvement in scleral remodeling has been confirmed by extensive studies of animal myopia models, as has been summarized above. We performed the case–control study in three stages: (1) initial screening of DNA pools to identify putatively positive SNPs, (2) confirmation of positive SNPs by genotyping of individual DNA samples forming the original pools, and (3) replication of positive SNPs by an independent sample set (Fig. 1). The initial DNA-pooling step served to reduce the cost and time involved in individual genotyping. ^23,24 DNA pools were created by mixing equal amounts of DNA from many individuals sharing the same disease status. Thus, case pools were constructed from subjects with high myopia (cases) and control pools from emmetropic subjects (controls). Moreover, we adopted an optimal experimental design in the DNA-pooling step by creating small DNA pools, each constructed from 50 distinct individuals of the same disease status. ²⁵  

Figure 1.

 

A three-stage approach to testing genetic association based on an initial screen of DNA pools.

Methods

Subjects and DNA Samples

In the DNA-pooling–based initial study, 600 unrelated Southern Han Chinese subjects (sample set 1) were recruited: 300 high myopes with spherical equivalent (SE) ≤ −8.00 D in both eyes, and 300 emmetropic controls with SE within ±1.0 D in both eyes. Positive SNPs from the DNA-pooling–based initial screen were confirmed by individual genotyping of the original sample set 1 and, if confirmed, replicated by testing a second sample set (sample set 2). Sample set 2 consisted of 710 unrelated Han Chinese subjects with 356 cases and 354 controls. The same entry criteria were used for subject recruitment of both sample sets. 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. Signed, informed consent was obtained from all participants. All subjects were recruited from the Optometry Clinic of the Hong Kong Polytechnic University, and blood sample collection and DNA extraction were performed as described previously. ²⁶  

Construction of DNA Pools

For the DNA-pooling study, all DNA samples were accurately quantified (PicoGreen method; Invitrogen, Carlsbad, CA), according to the manufacturer's instructions, and diluted to a final concentration of 5 ± 0.3 ng/μL. Equal volumes of DNA solutions were mixed to create DNA pools. Six case pools and six control pools were constructed for sample set 1, each consisting of 50 distinct individuals of the same disease status. 

Selection of Tag SNPs

Three candidate genes were selected for study: MMP2, TIMP2, and TIMP3. Tagger implemented in Haploview (http://www.broadinstitute.org/haploview/Haploview, Board Institute, Cambridge MA) was used to select tag SNPs with the following settings: pairwise tagging algorithm, r ² ≥ 0.8 and minor allele frequency (MAF) > 0.1. The selection was based on the Han Chinese genotype data from the International HapMap Project database (release 23a, phase II; http://www.hapmap.org/) for these three loci and their flanking regions (3 kb upstream and 3 kb downstream of the genes). In total, 49 tag SNPs were selected from these three genes for analysis by the DNA-pooling strategy (Table 1). 

Table 1.

 

Summary of Tag SNPs in the MMP2, TIMP2, and TIMP3 Genes

Gene	GeneID	Chromosomal Location	Region Captured*	Tag SNPs, (n)	SNPs Captured at Mean r 2 = ?
MMP2	4313	16q13-q21	33.5 kb	17	43 (r 2 = 0.994)
TIMP2	7077	17q25	78.4 kb	17	52 (r 2 = 0.973)
TIMP3	7078	22q12.3	68.2 kb	15	37 (r 2 = 0.960)

*  The region captured includes the gene and its flanking region (3 kb upstream and 3 kb downstream).

Allele Frequency Estimation in DNA Pools

The same protocols were used for all 49 SNPs examined, unless stated otherwise. Touchdown PCR was performed in a total volume of 15 μL reaction mixture containing 25 ng of genomic DNA template, 0.1 μM of each primer (Supplementary Table S1) and 1.5 mM MgCl_2, 0.2 mM of each dNTP, and 0.2 U of DNA polymerase (HotStarTaq Plus; Qiagen, Hilden, Germany) in 1× PCR buffer provided by the manufacturer. There were a few exceptions: 0.3 μM of each primer was used for 3 SNPs (rs243845, rs11639960, and rs12600817) and 2.5 mM of MgCl₂ for 2 SNPs (rs11639960 and rs12600817). Amplification was performed in a PCR system (model 9700 GeneAmp; Applied Biosystems, Inc. [ABI], Foster City, CA). The touchdown thermocycling program included activation at 95°C for 5 minutes; six cycles of 95°C for 30 seconds, 64°C (initial annealing temperature) for 45 seconds, decreased by 1°C per cycle, and 72°C for 45 seconds; an additional 38 cycles of 95°C for 30 seconds, 58°C (final target annealing temperature) for 45 seconds, and 72°C for 45 seconds; and a final extension at 72°C for 7 minutes. There were a few exceptions: The initial and final annealing temperatures were 62°C and 56°C for four SNPs (rs11643630, rs243845, rs11639960, and rs12600817). 

PCR products were purified by using shrimp alkaline phosphatase and exonuclease I. Primer extension (PE) reactions were performed in a 25-μL reaction volume containing 10 μL purified PCR products, 1.5 μM of the specific primer (Supplementary Table S1), 50 μM of each appropriate ddNTP and/or dNTP (Supplementary Table S1), and 1 unit of DNA polymerase (Therminator; New England Biolabs, Beverly, MA) in a 1× reaction buffer supplied by the manufacturer. Thermocycling was performed with an initial denaturation step at 96°C for 1 minute, followed by 55 cycles of 96°C for 10 seconds, 43°C for 15 seconds, and 60°C for 1 minute. 

Denaturing high-performance liquid chromatography (DHPLC) was performed (WAVE Nucleic Acid Fragment Analysis System; Transgenomic, Inc., Omaha, NE). PE products were analyzed with a 6% linear gradient change of the working elution buffer over a 3-minute period and a different starting concentration of buffer B, dependent on the SNP concerned (Supplementary Table S1). ²⁷ Relative allele frequencies in DNA pools were estimated based on the intensity of primer-extended products by DHPLC. For each DNA pool, the analysis included a single PCR followed by a single PE reaction and a single DHPLC analysis. Each DNA pool was analyzed in triplicate (Fig. 2). In other words, there were 36 sets of readings for six case pools and six control pools (sample set 1) for each SNP. 

Figure 2.

 

Nested design of the DNA-pooling study. There are two subject groups (case group G₂ and control group G₁), six DNA pools per group (P₂₁ to P₂₆ for the case group, and P₁₁ to P₁₆ for the control group), and three technical replicates (R_{ij 1} to R_{ij 3}) for each DNA pool. Note that there is no link between any pools of the case group and any pools of the control group. Therefore, the level of the case group is not cross-classified with that of the control group, but is nested with the respective group (i.e., the pools are nested within the group). Each DNA pool was constructed by mixing equal amounts of DNA from 50 distinct individuals of the same subject group.

Individual Genotyping

The positive findings (three SNPs) in the DNA-pooling–based initial study were confirmed by individual genotyping of the same sample set (set 1) (MassArray iPLEX chemistry; Sequenom, San Diego, CA; Supplementary Table S2) according to the manufacturer's instructions. These three SNPs were grouped together with SNPs of other on-going studies for genotyping, using a method (MassArray iPLEX; Sequenom) employed by a local service provider (http://genome.hku.hk/portal/). The confirmed positive results were tested by a follow-up replication study on sample set 2. 

For sample set 2, two SNPs of the TIMP3 gene (rs135029 and rs137485) were genotyped by unlabeled probe melting analysis. ²⁸ This method uses asymmetric PCR to generate single-stranded (ss)DNA product and an unlabeled probe that is 3′-blocked by a phosphate group to prevent probe extension. After PCR, the unlabeled probe and a saturating dsDNA dye are added to ssDNA target for high-resolution melting analysis. An asymmetric PCR reaction was performed in a 10-μL reaction mixture containing 10 ng of genomic DNA, 3.5 mM MgCl₂, 0.1 μM forward primer (excess), 0.01 μM reverse primer (limiting) (Supplementary Table S2), 0.2 mM of each dNTP, 1× PCR buffer, and 0.2 U DNA polymerase (HotStarTaq Plus; Qiagen). Amplification was performed in 96-well plates (model 9700 GeneAmp; ABI), including 1 cycle of initial denaturation for 5 minutes at 95°C, 50 cycles of 30 seconds at 95°C, 50 seconds at 55°C for annealing, and 25 seconds at 72°C, plus 1 cycle of final extension for 5 minutes at 72°C. After PCR, a 10-μL reaction mixture containing 8.4 μL PCR product, 0.5 μM unlabeled probe (International DNA Technologies, Coralville, IA), and 2.5 mM green fluorescent nucleic acid stain (SYTO 9; Invitrogen) were prepared in 96-well white plates. Melting was performed in a thermocycler (LightCycler 480; Roche Diagnostics, Mannheim, Germany) with heating of samples to 95°C for 1 minute and then cooling to 50°C for 1 minute. The melting data were then collected between 50°C and 95°C with a heating rate of 0.11°C/s at five acquisitions per degree centigrade, by using the melting-curves analysis mode. Samples were again cooled to 40°C for 10 seconds, and the melting curves were analyzed (LightCycler 480 Software, ver. 1.5; Roche). 

Statistical Analysis

High myopia was examined as a dichotomous trait. Subjects were classified as affected (cases) or unaffected (controls). In PE, unequal representation of the two alleles of a SNP can result from differential incorporation of ddNTPs and was corrected with a factor, known as the k correction factor, ²⁹ estimated based on the average of three independent replicate readings from a heterozygous sample. Relative allele frequencies of a given SNP were estimated from the heights (i.e., intensities) of the two peaks representing the two extension products in the DHPLC elution profile with correction by the k correction factor. ²⁹ The relative allele frequencies of a SNP were compared between the pools of the case group and the pools of the control group by nested analysis of variance (nested ANOVA; for an explanation, see Supplementary Materials). ³⁰ SNPs with P ≤0.10 were followed up by genotyping individual samples forming the DNA pools (sample set 1). A lenient threshold of P ≤ 0.10 was used to avoid excluding any potentially significant SNPs (all analyses: STATA, ver. 8.2; StataCorp, College Station, TX). 

Genotype data of individual samples (sample set 1 or 2) were analyzed by the PLINK package (ver. 1.07; http://pngu.mgh.harvard.edu/∼purcell/plink/index.shtml, developed by Shaun Purcell, Massachusetts General Hospital, Boston, MA). ³¹ Hardy-Weinberg equilibrium (HWE) testing was performed with an exact test for controls and cases separately. Single-marker association analysis was performed with the χ² test or exact test as appropriate. Haplotype analysis was also performed with logistic regression based on the Wald test. Multiple comparisons were corrected by permutation tests. Permutation tests are based on resampling theory and are widely accepted as the gold standard for correction of multiple comparisons. ³² In each permutation, the genotype data structure and the number of cases and controls were kept unchanged while the phenotype status of the subjects was randomly swapped (permutated). The statistic was calculated with each permutation, and an empiric P value was generated based on 10,000 permutations. Permutation of the phenotype status among study subjects is valid under the assumption of the null hypothesis. To control experiment-wise (instead of marker-wise) type I error rates, each permutation involved all individual SNPs and all haplotypes for a given sample set genotyped individually (sample set 1, sample set 2 or combined sets, each separately), and this process was repeated 10,000 times. As such, the generated empiric P values controlled the experiment-wise (or more correctly, family-wise) type I error rates for a given sample set. 

Results

Analysis of Ocular Data

This study had two sample sets collected from Han Chinese in Hong Kong. Sample set 1 consisted of 300 high myopes (cases) and 300 emmetropes (controls). The characteristics of these subjects have been reported previously. ²⁶ The cases (n = 356) and controls (n = 354) of sample set 2 were recruited using the same entry criteria as for sample set 1; their characteristics are summarized in Table 2 with the ocular data being shown for the right eye only, as has been done previously. ²⁶ Subjects of sample set 2 were on average older than those of sample set 1: 34.0 years for cases and 33.1 years for controls of set 2 (Table 2) and 27.7 years for cases and 24.9 years for controls of set 1. ²⁶ However, cases and controls of set 2 had very similar refractive error and axial length as their counterparts in set 1. The mean SE for the right eye was −10.30 D for cases and 0.08 D for controls of set 2, and −10.53 D for cases and 0.03 D for controls of set 1. The mean axial length of the right eye was 27.64 mm for cases and 23.73 mm for controls of set 2, and 27.76 mm for cases and 23.85 mm for controls of set 1. 

Table 2.

 

Characteristics of Study Subjects of Sample Set 2

Characteristic	Cases (n = 356)	Controls (n = 354)
Age, mean (SD), y	34.0 (9.1)	33.1 (9.5)
Females, n (%)	236 (66.3)	209 (59.0)
Spherical equivalent, mean (SD), D	−10.30 (2.46)	0.08 (0.53)
Axial length, mean (SD), mm	27.64 (1.18)	23.73 (0.82)
Corneal power, mean (SD), D	44.38 (3.71)	43.97 (1.51)
Anterior chamber depth, mean (SD), mm	3.34 (0.38)	3.20 (0.41)
Lens thickness, mean (SD), mm	4.30 (0.50)	4.33 (0.57)

 Data are for right eyes only.

Analysis of Pooled DNA Results

Results of pooled DNA analyses are summarized in Table 3. The k correction factor ranged from 0.65 to 1.45, with a mean of 1.07. The estimated frequencies of the first eluted allele ranged from 0.0719 to 0.8848 for case pools and from 0.0939 to 0.8574 for control pools. The difference (case pools − control pools) in estimated allele frequencies ranged from −0.0597 to 0.0401. Of the 49 SNPs tested by the DNA-pooling approach, only 3 showed significant differences in allele frequencies between case pools and control pools: rs2003241 (difference, 0.0329; nested ANOVA P = 0.0119), rs135029 (difference, −0.0597; P = 0.0010) and rs137485 (difference, 0.0351; P = 0.0727). These three SNPs were then genotyped for individual samples forming the DNA pools (sample set 1) for confirmation. The remaining 46 SNPs did not show significant differences in allele frequencies between case pools and control pools, and hence were not tested any further (Fig. 1). 

Table 3.

 

Pooled DNA Analysis of Tag SNPs in the MMP2, TIMP2 and TIMP3 Genes

SNP*	Alleles† (1st/2nd)	k Correction Factor Peak Height Ratio (1st/2nd Allele)	Estimated Frequencies of First Allele			nANOVA P
			Case Pools	Control Pools	Diff (Case − Control)
MMP2
rs11643630	T/G	1.01	0.4573	0.4534	0.0039	0.9029
rs1477017	G/A	1.12	0.3064	0.2958	0.0106	0.5585
rs865094	G/A	0.96	0.3268	0.3335	−0.0067	0.7647
rs11076101	C/T	1.01	0.7637	0.7855	−0.0218	0.3412
rs17301608	C/T	1.06	0.6557	0.6801	−0.0244	0.2008
rs11646643	G/A	1.02	0.1817	0.1702	0.0115	0.6305
rs2241146	G/A	1.01	0.7749	0.7736	0.0013	0.9427
rs9928731	C/T	1.11	0.5414	0.5015	0.0399	0.1435
rs12599775	C/G	1.25	0.1659	0.1819	−0.0160	0.5207
rs243847	C/T	1.09	0.4084	0.4196	−0.0112	0.7028
rs243845	G/A	1.05	0.7187	0.6907	0.0280	0.3367
rs243843	G/A	1.04	0.4285	0.4466	−0.0181	0.4689
rs183112	G/A	1.32	0.6643	0.6718	−0.0075	0.7514
rs1992116	G/A	1.02	0.6671	0.6945	−0.0274	0.2274
rs11639960	G/A	1.07	0.2801	0.2763	0.0038	0.8187
rs243835	C/T	1.05	0.4218	0.3817	0.0401	0.2328
rs1861320	G/T	1.23	0.7969	0.8195	−0.0226	0.5003
TIMP2
rs4789932	C/T	1.07	0.3519	0.3551	−0.0032	0.9259
rs8080623	C/T	1.15	0.2968	0.2975	−0.0007	0.9770
rs8179091	C/T	0.96	0.4432	0.4465	−0.0033	0.9115
rs7212662	C/A	1.17	0.2912	0.2972	−0.0060	0.8225
rs8066695	G/A	1.03	0.3740	0.3907	−0.0167	0.5461
rs12600817	C/T	0.65	0.5475	0.5774	−0.0299	0.4249
rs4789860	G/A	1.09	0.2553	0.2584	−0.0031	0.9410
rs2889529	C/T	1.23	0.3255	0.3202	0.0053	0.8840
rs2376999	C/T	1.01	0.3446	0.3235	0.0211	0.3401
rs2003241	G/A	0.97	0.2458	0.2129	0.0329	0.0119
rs7502935	C/T	1.45	0.6973	0.6914	0.0059	0.8301
rs6501258	A/T	1.20	0.5172	0.5019	0.0153	0.5409
rs6501256	G/A	1.01	0.2361	0.2166	0.0195	0.4293
rs11868442	G/A	1.36	0.6678	0.6783	−0.0105	0.7285
rs2277698	G/A	1.21	0.7445	0.7360	0.0085	0.7413
rs9905930	G/T	1.01	0.7625	0.7660	−0.0035	0.8442
rs16971783	T/A	1.00	0.1017	0.1042	−0.0025	0.8263
TIMP3
rs1962223	G/C	1.08	0.6029	0.6245	−0.0216	0.4599
rs9619311	G/A	0.99	0.1290	0.1520	−0.0230	0.2400
rs242089	C/T	1.04	0.5167	0.5126	0.0041	0.8976
rs80272	G/A	1.01	0.1050	0.1306	−0.0256	0.2691
rs8140818	G/A	1.01	0.0719	0.0939	−0.0220	0.2107
rs242076	C/T	0.92	0.5287	0.5623	−0.0336	0.4067
rs715572	C/T	1.04	0.6400	0.6452	−0.0052	0.8863
rs242072	G/A	1.01	0.4711	0.4663	0.0048	0.8683
rs135029	C/T	1.03	0.7898	0.8495	−0.0597	0.0010
rs241890	C/A	1.18	0.5981	0.5901	0.0080	0.7674
rs1427385	G/A	1.03	0.5344	0.5391	−0.0047	0.8962
rs9609643	C/T	1.12	0.8848	0.8574	0.0274	0.1511
rs9862	G/A	0.95	0.5242	0.5439	−0.0197	0.5089
rs11547635	A/G	0.98	0.3424	0.3270	0.0154	0.5151
rs137485	A/T	0.85	0.2440	0.2089	0.0351	0.0727

*  SNPs are arranged in the order of 5′→3′ along the respective gene.

†  The first allele has a shorter elution time, whereas the second allele has a longer one. The alleles are named with reference to the sense strand of the respective gene.

Confirmation of Pooled DNA Results by Individual Genotyping

The genotypes of the three follow-up SNPs were in HWE (P > 0.05, exact test) for sample set 1. The only exception was rs135029 in the case group (P = 0.0250). Deviation from HWE in cases can signify marker–disease association. ³³ Single-marker analysis showed that rs135029 of the TIMP3 gene was associated with high myopia (P _asym = 0.0069, allelic test), whereas the other two SNPs (rs2003241 and rs137485) showed no significant differences between cases and controls (sample set 1, Table 4). In addition, haplotypes consisting of rs135029 and rs137485 (both in the TIMP3 gene) were also associated with high myopia (P _asym = 0.0178, omnibus test; sample set 1, Table 5). These results remained significant after correction of multiple comparisons across single markers and haplotypes by permutation tests: P _emp = 0.0162 (allelic test, Table 4) and P _emp = 0.0496 (omnibus test, Table 5). Therefore, both rs135029 and rs137485 were further tested in a replication study using sample set 2. SNP rs2003241 was not tested any further (Fig. 1). Note that the asymptotic P value is indicated as P _aysm and the empiric P value as P _emp (also, see the footnotes to Tables 4, 5). 

Table 4.

 

Allelic Association Tests of TIMP2 and TIMP3 SNPs Genotyped Individually

Gene, SNP	Alleles*		Genotype Counts (11/12/22)		Minor Allele Freq.		OR (95% CI)†	Allelic Test‡
	1	2	Cases	Controls	Cases	Controls		P asym	P emp
Sample Set 1
TIMP2, rs2003241	A	G	189/93/10	201/84/12	0.1935	0.1818	1.08 (0.81–1.45)	0.6078	0.9374
TIMP3, rs135029	C	T	188/106/5	223/69/6	0.1940	0.1359	1.53 (1.12–2.09)	0.0069	0.0162
TIMP3, rs137485	T	A	209/84/4	228/62/7	0.1549	0.1279	1.25 (0.90–1.73)	0.1828	0.4466
Sample Set 2
TIMP3, rs135029	C	T	260/87/8	261/83/9	0.1451	0.1431	1.02 (0.76–1.37)	0.9142	0.9966
TIMP3, rs137485	T	A	259/87/6	271/66/9	0.1406	0.1214	1.18 (0.87–1.62)	0.2870	0.5231
Combined (Sets 1 and 2) §
TIMP3, rs135029	C	T	448/193/13	484/152/15	0.1674	0.1398	1.26	0.0344	0.0693
TIMP3, rs137485	T	A	468/171/10	499/128/16	0.1471	0.1244	1.24	0.0917	0.1268

*  The major allele is designated as 1 and the minor allele as 2. Sample set 1 has 300 cases and 300 controls, and sample set 2 has 356 cases and 354 controls. Note that the total genotype counts may not add up to these expected numbers, because a few samples failed to be genotyped in a random fashion.

†  The OR is calculated for the minor allele (allele 2) with the major allele (allele 1) as the reference. The 95% CIs are indicated within brackets. Since rs135029 is of interest, its genotypic ORs and their 95% CI are given here with CC as the reference genotype: 1.82 (1.27– 2.61; set 1), 1.06 (0.75–1.50; set 2), 1.37 (1.08–1.77; combined set, not adjusted for age here) for genotype CT, and 0.99 (0.30–3.29; set 1), 0.90 (0.34–2.37; set 2), and 0.94 (0.44–2.00; combined set, not adjusted for age here) for genotype TT, simply serves to avoid cluttering in the table.

‡  Allele frequencies are compared by χ² test to give the asymptotic P value (P _asym). Multiple comparisons are corrected by 10,000 permutations across single marker allelic tests (this table) and omnibus tests of haplotypes (Table 5) for a given sample set (set 1, set 2 or combined, each separately). The empirical P value is indicated as P _emp.

§  Single-marker (this table) and haplotype (Table 5) analyses are performed for the combined sample set (sets 1 and 2) with adjustment for age as a covariate to account for the effect of age. This adjustment has to be made by logistic regression; the relevant PLINK command does not give the 95% CI for the OR.

Table 5.

 

Haplotype Analysis of 2 TIMP3 SNPs rs135029 and rs137485*

Haplotype	Haplotype Freq.		OR	P asym	P emp
	Cases	Controls
Sample Set 1
Omnibus test	—	—	—	0.0178	0.0496
TA	0.1431	0.1102	1.38	0.0785
CA	0.0118	0.0186	0.62	0.3350
TT	0.0522	0.0254	2.11	0.0201
CT	0.7929	0.8458	0.69	0.0172
Sample Set 2
Omnibus test	—	—	—	0.2380	0.4293
TA	0.1028	0.1033	1.00	0.9780
CA	0.0368	0.0185	2.00	0.0463
TT	0.0425	0.0387	1.10	0.7260
CT	0.8179	0.8395	0.86	0.2890
Combined (Sets 1 and 2) †
Omnibus test	—	—	—	0.0690	0.1381
TA	0.1219	0.1070	1.20	0.1560
CA	0.0247	0.0180	1.35	0.2680
TT	0.0463	0.0320	1.43	0.0829
CT	0.8071	0.8430	0.77	0.0118

*   P _asym is obtained with the Wald test based on logistic regression. Multiple comparisons are corrected by 10,000 permutations across single-marker allelic tests (Table 3) and omnibus tests of haplotypes (this table) for a given sample set (set 1, set 2, or combined, each separately). The empirical P value is indicated as P _emp. Note that PLINK does not generate CIs for ORs in haplotype analysis.

†  Single-marker (Table 4) and haplotype (this table) analyses are performed for the combined sample set (Sets 1 and 2) with adjustment for age as a covariate to account for the effect of age. This has to be done using logistic regression, the relevant PLINK command does not give the 95% CI for the OR.

Replication Study Based on Sample Set 2

The genotypes of both rs135029 and rs137485 were in HWE (P > 0.05, exact test). Single-marker and haplotype analyses did not show any significant differences in allele or haplotype frequencies between cases and controls (sample set 2, Tables 4, 5). We combined the sample sets (656 cases and 654 controls in total) and reanalyzed the data with adjustment for age as a covariate, because the mean age differed very significantly between sample sets 1 and 2 (difference, 7.46 years, P < 10⁻⁴, by t-test). The results remained the same, without significant differences in allele or haplotype frequencies between cases and controls (combined, Tables 4, 5). In other words, the initial positive results in sample set 1 could not be replicated independently by sample set 2. 

Discussion

We adopted an efficient, three-stage approach to investigating the relationship between high myopia and the tag SNPs of three candidate genes (MMP2, TIMP2, and TIMP3). There are many experimental studies using animal myopia models that suggest the involvement of these genes in myopia development. In the initial stage, 49 tag SNPs were screened with a DNA-pooling approach, and 3 SNPs passed the lenient threshold of P ≤ 0.10 and were followed up. In the second stage, these three putatively positive SNPs were genotyped for individual samples forming the original DNA pools. In the third stage, two SNPs from stage 2 were genotyped for individual samples from a second sample set. However, the initial positive results could not be substantiated in the replication study. It is interesting to note that rs135029 of TIMP3 gave an OR of 1.26 for the combined sample set (P _asym = 0.0344; Table 4), but did not survive after correction for multiple comparisons (P _emp = 0.0693; Table 4). In view of this borderline significance, we explored the potential functional role of this SNP in the literature and using a web-based tool (FuncPred; http://manticore.niehs.nih.gov/snpfunc.htm, National Institute of Environmental Health Sciences, National Institutes of Health, Bethesda, MD) for prediction of SNP functions, but without success. In other words, MMP2, TIMP2, and TIMP3 were not associated with high myopia in the Han Chinese population under study and are thus unlikely to play a major role in the genetic susceptibility to high myopia. 

A recent Japanese study examined two functional promoter SNPs of the MMP2 gene in a case-control study involving 725 high myopes (SE ≤ −6.0 D) and 546 population-based controls and found no association of these two SNPs with high myopia. ³⁴ These two promoter SNPs were rs243865 and rs2285053 (named C −1306T and C −735T, respectively, in the report) and were not examined in the present study. The SNP rs243865 had an MAF < 0.10 in Han Chinese and hence did not satisfy the criteria of selecting tag SNPs in our study, whereas the other SNP rs2285053 was not documented in the HapMap database. 

A U.S.-based group recently examined 146 tag SNPs from 14 MMP and 4 TIMP genes in 55 Amish families (358 individuals; mean SE −1.61 D) and 63 Ashkenazi families (535 individuals; mean SE −3.56 D). ³⁵ The tag SNPs were selected from the HapMap Caucasian (CEU) database with the criteria of MAF ≥ 0.15 and r ² ≥ 0.7. In particular, 6 tag SNPs from MMP2, 11 from TIMP2, and 12 from TIMP3 were included, which are expectedly fewer than those examined in our study (Table 1), because of their less restrictive criteria of SNP selection. Two SNPs were found to be significantly associated with ocular refraction by quantitative trait analysis, using family-based association testing in the Amish families, but not the Ashkenazi families. Both sets of families were sampled from largely endogamous, rapidly expanding, but isolated populations in the United States. The prevalence of refractive errors is high in Jewish populations, ³⁶ but relatively low in the Old Order Amish. ³⁷ The behavioral and environmental factors are more conducive to myopia development in the Jewish populations than in the Amish populations, and probably explain the discrepancy in the genetic association results, as suggested by the authors. ³⁵ They also anticipated that the positive results could not be replicated in South Asian Chinese and Japanese populations with a high prevalence of environmentally induced myopia. ³⁵ Indeed, in our study we could not could not replicate the findings. One of the positive SNPs in the Amish population was rs9928731 (P = 0.00026) within the MMP2 gene. ³⁵ This SNP was also screened by the DNA-pooling approach in the present study: the estimated frequency of the C allele was 0.5414 in case pools and 0.5015 in control pools, which were not statistically significant (difference 0.0399; nested ANOVA P = 0.1435; Table 3). The frequency of the C allele in controls is similar to that in Han Chinese documented in the HapMap database (0.5015 vs. 0.4560). It is worth noting that the phenotype definition was different for these two studies: quantitative measures of refractive errors in the American study, but dichotomous trait of high myopia (affected versus unaffected) in our study. 

All three association studies (Japanese, U.S., and ours) focused on common polymorphisms in the genes under study, and hence assumed the hypothesis of common disease–common variants. ³⁸ Strong linkage disequilibrium (LD) between common tag SNPs and common casual variants is critical to the success of this indirect LD mapping approach. Sequence variations must have similar allele frequencies to be highly correlated and in strong LD. However, rare casual variants may also contribute to myopia development—the other side of the story being the hypothesis of common disease rare variants. ³⁸ This indirect approach is of low power in detecting association with rare variants because of the weak LD between common tag SNPs and rare casual variants. Therefore, direct mapping must be performed to detect association with rare causal variants, which must first be identified. Rare variants can be identified for direct association studies by sequencing of good candidate genes or even the whole genome for a very large number of samples. ³⁹  

Our case subjects had extreme refractive errors (mean SE −10.53 D for set 1 ²⁶ ; and −10.30 D for set 2; Table 2), which would enhance the homogeneity of the myopia phenotype, enrich the genetic components of the contributing factors, and hence increase the power of our study (albeit in a subtle manner). The three candidate genes were chosen for study because they have been shown to be involved in sclera remodeling in myopia development in many studies. ^{11 –22} Our negative finding may imply that these genes do not carry common sequence variants that are capable of influencing their function and/or regulation in the relevant ocular tissue. However, the contribution of behavioral and environmental effects on high myopia should not be overlooked. Our DNA-pooling–based initial screen adopted a lenient threshold of P ≤ 0.10 in order to avoid missing potential SNPs. For rs135029 of TIMP3, the power of the third stage study (sample set 2) is 73% under an allelic model and 78% under a genotypic model. The power is calculated based on the following assumptions with the online Genetic Power Calculator (http://pngu.mgh.harvard.edu/∼purcell/gpc/ developed by Shaun Purcell, Massachusetts General Hospital, Boston, MA): OR and allele frequencies obtained for rs135029 for sample set 1 (Table 4), a disease prevalence of 0.05 for high myopia in our local Chinese population, ⁴⁰ and a significance level set at α = 0.025 because two SNPs were examined in the third stage. One disadvantage of DNA-pooling strategy is that it makes haplotype analysis very difficult, if not impossible. ²⁴ Algorithms are available for estimating haplotype frequencies in small DNA pools constructed from a few (<10) individuals. In other words, our current pooling protocol might miss some potential SNPs for follow-up in the second-stage analysis if high myopia is associated with certain haplotypes, but not individual SNPs. This is one of the reasons that a lenient threshold of P ≤ 0.10 was used for selecting SNPs for follow-up study by individual genotyping in the second stage. 

Association testing of DNA pools has been shown to be an effective initial screen of SNPs and candidate genes for subsequent detailed follow-up study. ^24,41 The major advantages are a tremendous reduction in DNA usage and in the amount of genotyping work when compared with individual genotyping. For example, our study required for each SNP 36 PCRs and subsequent analyses for six case pools and six control pools (Fig. 2), plus three separate PCRs for heterozygotes to determine the k correction factor. The amount of genotyping was only about one fifteenth that necessary for genotyping of 600 individual samples. It has also been shown that the use of small DNA pools of approximately 50 individuals is superior to use of fewer, larger DNA pools in candidate gene studies. ²⁵ In addition to these advantages, the small DNA pools allows the use of a standard statistical method (nested ANOVA) for data analysis, without the need of directly estimating the variance components of the error sources, while it properly handles variations arising from sampling of subjects and technical errors, which are due to unequal amounts of DNA being mixed together, errors in PCR and primer extension, and errors in DHPLC analysis. 

The present study used DHPLC analysis to estimate the relative allele frequencies of DNA pools. DHPLC analysis is in fact a rate-limiting step because samples have to be injected and analyzed sequentially. The throughput can be greatly increased if quantitative genotyping is conducted with a mass spectrometer ⁴² (e.g., using the MassArray iPLEX method; Sequenom). However, the local service provider would only entertain a request for classic genotyping, based iPLEX method, not quantitative genotyping. The DNA-pooling strategy may become less attractive as the unit cost of genotyping decreases substantially with the availability of high-throughput genotyping platforms such as whole-genome genotyping arrays. Nevertheless, the total cost of whole-genome genotyping for a large number of samples is still prohibitive for many research groups. In fact, genome-wide association studies can be within the reach of even small- to medium-sized research groups if the DNA-pooling strategy is applied. ⁴³ Interestingly, errors due to array variations are much greater than those due to pool construction, and hence it is recommended to have multiple arrays per DNA pool for a few pools rather than multiple DNA pools with less arrays per pool. ⁴⁴  

In conclusion, we used a DNA-pooling strategy to screen 49 tag SNPs from three candidate genes (MMP2, TIMP2, and TIMP3). Three tag SNPs passed the threshold (P ≤ 0.10) and were tested by individual genotyping of samples forming the DNA pools. Two SNPs from the TIMP2 gene were found associated with high myopia by single-marker analysis or haplotype analysis. However, the initial positive results could not be replicated by an independent second sample set. Overall, these three candidate genes are unlikely to play a major role in the genetic susceptibility to high myopia in the Chinese population. 

Supplementary Materials

Text s01, DOC - Text s01, DOC 

Text s02, DOC - Text s02, DOC 

Text s03, DOC - Text s03, DOC 

Footnotes

 Supported by a Grant PolyU 5411/06M, account code B-Q04A from the Research Grant Council of Hong Hong and Grants J-BB7P, 87MS, and 87LV from The Hong Kong Polytechnic University.

Footnotes

 Disclosure: K.H. Leung, None; W.C. Yiu, None; M.K.H. Yap, None; P.W. Ng, None; W.Y. Fung, None; P.C. Sham, None; S.P. Yip, None

The authors thank all participants in the Myopia Genetics Study.