Introduction

All-trans-retinol dehydrogenase (RDH8) is the enzyme that catalyses the reduction of all-trans-retinal to all-trans-retinol in the presence of NADPH (Rattner et al. 2000). It is a member of the short chain dehydrogenase/reductase family and is located in the outer segments of photoreceptors (hence also known as photoreceptor retinol dehydrogenase). It is very important in the visual cycle and begins the rhodopsin regeneration pathway by reducing all-trans-retinal, the product of bleached and hydrolysed rhodopsin (Rando 2001). This is a rate-limiting step in the visual cycle (Saari et al. 1998). The human RDH8 gene consists of six exons spanning about 9 kilobasepairs (kb), and maps to chromosome 19p13 (Rattner et al. 2000).

Myopia (or short-sightedness) is a complex eye trait with both genetic and environmental risk factors contributing to its genesis (Saw et al. 1996). It is much more common in Orientals than in Caucasians (Saw et al.1996; Wong et al. 2000). In Hong Kong, the prevalence peaks at age 19–39 at 70% (Goh and Lam 1994). In parallel with our existing efforts to identify myopia susceptibility genes, we identify single nucleotide polymorphisms (SNPs) in myopia candidate genes, which can then be tested by association studies. Association studies are powerful in detecting genes with small effects in complex traits (Lander and Schork 1994; Johnson and Todd 2000). Retinoic acid is an important signalling molecule in the eye and has been shown to be a potential mediator between refractive error and compensatory eye growth (Seko et al. 1998; Mertz and Wallman 2000). In other words, genes encoding enzymes, binding proteins, or receptors involved in the metabolism of retinoic acid are potential candidate genes. Here we report the identification of SNPs for the RDH8 gene and its pattern of linkage disequilibrium (LD) in a Chinese population.

This study employed denaturing high-performance liquid chromatography (DHPLC) to screen SNPs in DNA pools. In DHPLC, products amplified by polymerase chain reaction (PCR) are fractionated in a special column under partially denaturing conditions by ion-pair reverse-phase liquid chromatography (Xiao and Oefner 2001). Heteroduplexes formed from different alleles have melting properties different from perfectly matched homoduplexes and can be differentially eluted from the latter. Potential SNPs were then confirmed and characterised by direct DNA sequencing. DNA samples were pooled before PCR amplification to increase the throughput efficiency of SNP detection by DHPLC (Wolford et al. 2000).

Materials and methods

Subjects

Blood samples were obtained with written consent from 20 university students and used in the screening stage of the study. Anonymous blood samples were also collected from 150 unrelated blood donors from the Hong Kong Red Cross Blood Transfusion Service and used to establish SNP allele frequencies for the local Chinese population. DNA was extracted with a modified salting-out method (Miller et al. 1988). The study was approved by the University’s Human Subject Ethics Subcommittee.

PCR amplification

Twenty-four primer pairs (Table 1) were designed with the software Oligo (version 6.57; Molecular Biology Insights, Cascade, USA) to amplify the six RDH8 exons and their immediate flanking regions, and non-coding sequences about 2.5 kb upstream of the start codon and 2.0 kb downstream of the stop codon. NT_011295 was used as the reference sequence. The size of the PCR fragments ranged between 220 and 439 basepairs (bp). In the screening stage, four DNA pools were constructed, with each consisting of equal amounts from five individuals. If at least one heterozygous sample was present in a pool, then the relative frequency of the less frequent allele in the pool would be at least 10% and the DHPLC elution profile would show multiple peaks (Wolford et al. 2000). For any pool showing multiple peaks, its constituent samples were amplified and analysed individually by DHPLC to identify homozygous and heterozygous samples. In the genotyping stage, 150 samples were individually amplified before DHPLC analysis.

Table 1 Polymerase chain reaction (PCR) fragments, primer sequences and annealing temperatures (Ta), and conditions of denaturing high-performance liquid chromatography (DHPLC) analysis
Full size table

The 50-μl reaction volume contained 1× Gold Buffer (15 mM Tris-HCl, 50 mM KCl, pH 8.0), 0.2 mM of each dNTP, 0.3 μM of each primer, 1U AmpliTaq Gold DNA polymerase (Applied Biosystems, Foster City, USA) and 100 ng genomic DNA (pooled or individual). The MgCl2 concentration was 1.5 mM for the amplification of all fragments except four, namely, Rdh831, Rdh833, Rdh836 and Rdh837, for which it was 2.5 mM. Touch-down PCR was used to avoid excessive optimisation of the reaction conditions. PCR amplification consisted of initial denaturation for 5 min at 95°C, eight touch-down cycles, 35 main cycles and final extension for 7 min at 72°C. Both touch-down and main cycles consisted of 30 s at 95°C, 30 s at the annealing temperature and 45 s at 72°C. The annealing temperature of the main cycles for each primer pair is shown in Table 1, and the initial annealing temperature for the touch-down cycles was 7°C above this with 1°C reduction for each successive touch-down cycle.

DHPLC analysis

In the screening stage, PCR products were denatured for 5 min at 95°C and then cooled to 25°C at a rate of 1°C/min to allow for heteroduplex formation. Denatured products were analysed with the WAVE DNA Fragment Analysis System (Transgenomic) using a DNASep column kept at a certain temperature in an oven. Next, 5 μl aliquots of denatured products were automatically injected into the column and eluted with a linear acetonitrile (ACN) gradient in 0.1 M triethylammonium acetate (TEAA) buffer, pH 7.4, at a constant flow rate of 0.9 ml/min. The gradient was composed of a range of proportions (percent) of buffer B increased at a rate of 2%/min and created by mixing buffers A and B, where buffer A was 0.1 M TEAA with 0.025% ACN and buffer B 25% ACN in 0.1 M TEAA. The gradient was calculated automatically by the software WAVEMAKER (version 4.1, Transgenomic) on the basis of the DNA sequence and hence was specific to each PCR product (Table 1). The column was cleaned with 75% ACN after each analytical run. The eluate was monitored by an ultraviolet light detector at 260 nm. The WAVEMAKER software also predicted the melting domains of input DNA fragment sequence. Column temperatures for screening of each fragment were selected on the basis of its melting domains whereby the domains were 75–90% in helical conformation at the given temperatures. Consequently, the PCR fragments were each analysed at three to seven column temperatures (Table 1).

For each identified and characterised SNP, 150 Chinese samples were genotyped to establish the allele frequencies. The samples were first analysed individually without mixing with a reference sample and using the same DHPLC conditions as above, but only at a single column temperature at which the multiple-peak elution profile was most distinctive (Table 1). Known homozygous and heterozygous samples were run in parallel as controls. Samples showing a single peak (homozygotes) were then mixed with a reference homozygous sample in equal volumes and re-analysed again using the same conditions. With this second analysis, the sample had the same genotype as the control homozygote if a single peak was observed, and was homozygous for the other allele if multiple peaks were seen. Similarly, samples showing multiple peaks (heterozygotes) at the first analysis were also mixed with a reference heterozygous sample and re-analysed. Identical multiple peaks as in the first analysis confirmed that the sample had the same heterozygous genotype as the control heterozygote. A distorted multiple-peak elution profile indicated the presence of additional SNP in the fragment and would be further investigated.

DNA Sequencing

The DNA sequences of representative samples showing distinct elution patterns were determined by direct sequencing. After amplification as described above, the PCR products were purified with shrimp alkaline phosphatase (Amersham Biosciences, Piscataway, USA) and exonuclease I (New England Biolabs, Beverly, USA). Cycle sequencing was performed on the purified products using the Big Dye Terminator Cycle Sequencing Ready Reaction Kit (Version 2.0 or 1.1, Applied Biosystems). Forward and/or reverse primers were used as sequencing primers as appropriate. The sequencing products were purified by ethanol precipitation and analysed in ABI PRISM 310 Genetic Analyzer (Applied Biosystems) according to the manufacturer’s instructions.

Statistical analysis

Allele frequencies were calculated using the simple gene counting method. Testing for Hardy-Weinberg equilibrium was performed using chi-square test. LD pattern and haplotype frequencies were also established for common SNPs with minor allele frequency >0.05 in the Chinese population under study. Allelic association and hence LD between a pair of SNPs were determined with the expectation-maximisation algorithm (Excoffier and Slatkin 1995) using the ASSOCIATE programme (Ott 1985), as described by Yip et al. (2003). Briefly, the output of ASSOCIATE allowed the determination of the Lewontin’s LD parameter (D) and its maximum value (Dmax), and hence the calculation of the standardised Lewontin’s LD parameter (D’=D/Dmax). ASSOCIATE also gave the p value for the statistical significance of allelic association and hence LD. In order to obtain an overall significance level (α’) of 0.05, Bonferroni procedure was applied to correct for multiple testing (Weir 1996). There were 21 pairwise LD values. Thus, the significance level (α) for each individual testing was obtained from α=1 – (1 – α’)1/21 and found to be 0.0024. In addition, another commonly used LD measure (r2) was also calculated; r2=D2/[PA PB (1 – PA) (1 – PB)] where the PA and PB are the frequencies of allele A and allele B at two different loci. Finally, frequencies of haplotypes across all seven common SNP sites were estimated with the EH programme (Terwillger and Ott 1994).

Results

Twenty-four PCR fragments (Table 1) were amplified to cover the 5’ flanking region, six exons and their immediate flanking intron sequences, and the 3’ flanking region of the RDH8 locus, which amounted to a total genomic distance of about 7 kb. DHPLC analysis found 15 SNPs in nine of these fragments. Most fragments harboured one or two SNPs, but one fragment (Rdh858) contained three SNPs (Tables 1 and 2). It is interesting to note that five of these 15 SNPs were identified not during the screening stage but during the genotyping stage when 150 samples were genotyped by DHPLC (Table 2). This was not unexpected in view of the much larger sample size for genotyping purpose. On the other hand, two PCR fragments showing potential heteroduplex peaks on initial DHPLC analysis of the DNA pools were finally confirmed not to contain any SNP on repeated DHPLC analysis and direct DNA sequencing of individual samples of the positive pools. This gave a false positive rate of 18% for DHPLC analysis for screening sequence variations.

Table 2 Details of the single nucleotide polymorphisms (SNPs) identified in the all-trans-retinol dehydrogenase (RDH8) gene in Hong Kong Chinese population (n=150)
Full size table

Characteristics of identified SNPs

Table 2 shows the details of the SNPs identified in this study. Of the 15 identified, 12 were transitions (seven A/G and five C/T transitions), two transversions, and one was a two-base insertion/deletion. Only three were located in coding regions, while the rest were in non-coding regions with nine found upstream of the start codon and three downstream of the stop codon. One coding SNP was a synonymous transition (8117C>T or RDH8E61) in exon 6. The other two coding SNPs were adjacent to each other in the same codon (202) located in exon 5 (7826T>C and 7827G>A, or RDH8E5a and RDH8E5b, respectively). Among the 150 samples genotyped, only three of the four possible haplotypes were found for these two adjacent SNPs: ATG (methionine), ACG (threonine) and ACA (threonine). The ACA haplotype was found only in four individuals heterozygous for both SNPs, who, in other words, carried the two-site genotype ATG/ACA. This was determined by single-strand conformation polymorphism analysis (data not shown).

Ten of the 15 SNPs were novel and not previously reported. Seven SNPs (two novel and five previously documented in NCBI with reference SNP ID numbers) were common with the minor allele frequency >0.05. The observed genotype frequencies were in Hardy-Weinberg proportions for all SNPs.

Three common SNPs (RDH851, RDH8E61 and RDH836) were of particular interest, because they might have been missed if DHPLC analysis had been performed only at the single column temperatures recommended by the software WAVEMAKER (Tables 1 and 2). They showed distinct heteroduplex peaks at temperatures 2, 1 and 5°C, respectively, above those recommended. These findings testify the usefulness of our approach to selecting multiple column temperatures on the basis of the melting domains, whereby the domains are 75–90% in helical conformation at the given temperatures.

Analysis of LD and haplotypes

LD pattern and haplotype frequencies were established for the seven common SNPs. Table 3 shows the pairwise LD measures (both |D’| and r2) and the corresponding statistical significance for allelic association. Strong to complete LD was observed across four SNPs at the 3’ end of the RDH8 locus, namely, RDH8E5a, RDH8E61, RHD8E62b and RHD836. The pairwise LD measures were >0.75 for |D’| and >0.42 for r2, with all p values for allelic association <0.0001. While this LD block was obvious and certain, LD among other SNP pairs was either questionable (with inconsistent LD measures) or absent. In particular, the |D’| values showed significant (p<0.0024) LD between RDH851 on one hand and three other SNPs (RDH855b, RDH8E5a and RDH836) on the other hand. However, the r2 values were all <0.09 for these three SNP pairs. Similar situations occurred for three other SNP pairs (RDH855b-RDH8E61, RDH851-RDH8E12b, and RDH8E12b-RDH836) that had |D’|≥0.39, but r2<0.05 with p values significant at the 0.05 level but not at the Bonferroni-adjusted level (0.0024).

Table 3 Pairwise linkage disequilibrium (LD) measures for common single nucleotide polymorphisms (SNPs) of the all-trans-retinol dehydrogenase (RDH8) locus and the corresponding statistical significance for allelic associationa
Full size table

Table 4 shows the seven-site haplotypes together with the corresponding frequencies estimated by the EH programme. Although there are theoretically a total of 128 possible haplotypes for seven bi-allelic SNPs, six major seven-site haplotypes gave a total cumulative frequency of ~0.75. The cumulative frequency added up to ~0.95 with ten additional seven-site haplotypes.

Table 4 Frequencies of haplotypes across seven common single nucleotide polymorphisms (SNPs) of the All-trans-retinol dehydrogenase (RDH8) locusa
Full size table

Discussion

We are interested in mapping susceptibility genes predisposing humans to complex eye diseases including myopia. One major mapping approach is association studies based on either unrelated cases and controls from the same population (population based) or families with at least one affected sib (family based). Both types of association studies require identification and characterisation of suitable SNPs within and around potential candidate genes followed by testing with appropriate sample sets. Discovery and characterisation of SNPs in candidate genes form part of our concerted mapping effort based on association studies. This study reports the SNPs in the RDH8 locus and their LD pattern. The gene product is one of the proteins involved in the metabolism of retinoic acid, which may play a role as a mediator between refractive error and compensatory eye growth (Seko et al. 1998; Mertz and Wallman 2000).

SNP screening based on DNA pooling and DHPLC

DHPLC analysis at a partially denaturing temperature separates homoduplex and heteroduplex DNA fragments very efficiently if the injected sample contains more than one allele. Conventionally, a test sample is mixed with a reference sample of wild-type homozygous genotype before injection to allow detection of sequence variations (SNPs or pathological mutations). If the test sample is heterozygous, then the variant allele frequency in the mixed sample is 25% (one variant allele in a pool of four alleles). It was shown previously that DHPLC analysis could detect one variant allele (5%) in a pool of 20 alleles (Wolford et al. 2000). Consequently, we increased the screening throughput five-fold by using DNA pools composed of samples from five individuals instead of mixing every sample with a reference. In the DNA pool, every constituent sample forms a reference for all other samples in the pool. If at least one of the five constituent samples is heterozygous for an SNP, then the variant allele frequency is at least 10% (one variant allele in a pool of ten alleles). This can easily be detected by DHPLC. In addition, our pooling strategy still had a wide safety margin. DNA pooling has proved very efficient in SNP discovery when coupled with DHPLC analysis.

The same study also reported that fluorescent cycle sequencing could detect variants in the same pools if the frequency of the less frequent allele was at least 10% (Wolford et al. 2000). However, we found this approach very unreliable, and heterozygous peaks very often could not be detected in the sequencing traces for the DNA pools showing heteroduplex signals in DHPLC. As a result, we adopted the approach that individual samples of the positive pools were re-amplified and re-analysed by DHPLC before individual representative samples were sequenced to characterise the SNPs.

The present study highlighted the inadequacy of the WAVEMAKER software in predicting the column temperature to be used for DHPLC analysis of a particular PCR fragment (Table 1). There are recommendations to conduct the analysis also at temperatures 2°C below and above the predicted temperature (Colosimo et al. 2002). Even with this recommendation, the SNP RDH836 (or 10254G>A) would still be missed, since it was detected at a temperature 5°C above that recommended by WAVEMAKER. This justifies the selection of multiple temperatures at which the melting domains are 75–90% in helical conformation.

Interestingly, DHPLC failed to detect at all selected temperatures an SNP found by direct sequencing in one particular sample. This SNP was located in a GC-rich domain and was analysed at a temperature best for this domain. At this particular column temperature, all other domains melted completely and the elution peak appeared very early in the chromatogram. On the other hand, cycle sequencing failed to confirm convincingly in heterozygous samples an SNP detected by DHPLC. This particular SNP could only be confirmed by cycle sequencing of the two types of homozygous samples.

SNP genotyping by DHPLC

In genotyping SNPs for population samples with DHPLC analysis, all samples were first analysed without mixing with a reference sample. This allowed the detection of homozygotes (of either type) and heterozygotes. A second analysis required mixing with a reference sample (either a homozygous or a heterozygous sample, depending on the initial results) allowed accurate genotyping of the SNPs concerned and simultaneous detection of any additional novel SNPs present in the same PCR fragment. A similar, but not identical, DHPLC strategy was also recently reported for genotyping purpose (Schaeffeler et al. 2001). In our study, the second analysis additionally required the heterozygous sample to be mixed with a reference heterozygous sample and then re-analysed. This extra step would allow additional SNPs within the same fragment to be identified if this additional SNP did not reveal itself in the first analysis. This approach has proved very successful in our hands.

SNP genotyping based on direct DHPLC analysis requires two injections per sample. But this approach does offer many advantages. There is no need to re-design primers, re-optimise PCR, and use other new reagents (like restriction enzymes). Sample injection is automated and throughput is high. Reproducibility is very high. Furthermore, this approach allows identification of new SNPs within the same fragment (Table 2), because DHPLC is used as both a screening and a diagnostic method in this setting.

Analysis of LD and haplotypes

D’ and r2 are two most commonly used LD measures (Ardlie et al. 2002; Weiss and Clark 2002). In some ways, r2 is complementary to D’ but is increasingly recommended as the measure of choice for quantifying and comparing LD in the context of mapping genes by association studies. In addition, r2 is less affected by sample size and allele frequency. Both measures were estimated for seven common SNPs identified in the RDH8 locus in this study (Table 3). As expected, the r2 values were lower than the |D’| values. The |D’| measure reached its maximum value of 1 for some SNP pairs, but the r2 measure was at most 0.76 (for the SNP pair RDH8E5a-RDH836) in this study. The case of |D’|=1 (complete LD) occurs if, and only if, two SNPs have not been separated by recombination (Ardlie et al. 2002). In this case, at most, three out of the four possible two-site haplotypes are observed in the sample. On the other hand, the case of r2=1 (perfect LD) occurs if, and only if, two SNPs have not been separated by recombination and, in addition, have the same allele frequency (Ardlie et al. 2002). In this case, exactly two out of the four possible two-site haplotypes are observed in the sample. From the practical viewpoint of mapping genes using association studies, an r2 value of at least 1/3 is the recommended minimum, because this requires an increase of sample size by less than three-fold in order to compensate for the weak LD between an SNP marker and a susceptibility locus.

Both LD measures strongly supports the notion that the four common SNPs (RDH8E5a, RDH8E61, RHD8E62b and RHD836) at the 3’ end of the RDH8 locus forms a LD block with |D’|>0.75 and r2>0.42 (Table 3). This conclusion is also supported by haplotype analysis for these four SNPs, which spans a genomic distance of about 2.5 kb. For the sake of clarity and easy discussion, we denote the major allele of each SNP as 1 and the minor allele as 2. As estimated by EH on the basis of the observed frequencies of genotypes in the Chinese population under study, the observed four-site haplotype frequency across these four SNPs (in the order listed above) was 0.3164 for 1-1-1-1, 0.3826 for 2-2-2-2 and 0.1325 for 2-1-2-2, with a total observed frequency of 0.8315 (Table 4). The expected haplotype frequencies could be obtained by multiplying the corresponding allele frequencies and were found to be 0.1197, 0.0287 and 0.0424, respectively, with an expected total frequency of only 0.1908. The large differences between the observed and the expected haplotype frequencies testify the strong LD among these four SNPs, which can be regarded as forming a haplotype block.

That three common SNPs (RDH855b, RDH851 and RDH8E12b) at the 5’ end of the locus does not exhibit significant useful LD is also supported by a similar analysis of the haplotype frequencies. As estimated by EH, the observed three-site haplotype frequency was 0.5198 for 1-1-1, 0.2534 for 1-2-1 and 0.1430 for 2-1-1, with a total observed haplotype frequency of 0.9162 (Table 4). The expected haplotype frequencies were 0.5808, 0.2113 and 0.1134, respectively, with an expected total frequency of 0.9055; all were not much different from the observed frequencies. In other words, these three SNPs were in linkage equilibrium, which is not expected on the basis of short genomic distance (~1.6 kb) spanning by these SNPs.

In other words, LD pattern established by pairwise LD measures is supported by multiple-site haplotype analysis. Our results indicate that two sets of markers are required for association studies involving the RDH8 gene. One SNP is required in the 3’ haplotype block. Of the four SNPs in this haplotype block, RDH8E5a is best suited for this purpose because it involves a non-synonymous change (Met/Thr) and might have functional consequences. Since strong and useful LD does not exist among the common SNPs at the 5’ region of the gene, two to three SNPs are needed.

Currently, international concerted efforts are being made to produce a haplotype map for the human genome, which serves to facilitate LD mapping for complex human traits or diseases through association studies. This study reports the SNPs within and around the RDH8 gene and the corresponding LD pattern. This serves to provide the groundwork for association studies mapping genes involved in complex eye diseases.