Detailed Analysis of Allelic Variation in the ABCA4 Gene in Age-Related Maculopathy

Abstract

purpose. Age-related maculopathy (ARM) is one of the most common causes of blindness in older adults worldwide. Sequence variants in a gene coding for a retina-specific ATP-binding cassette (ABCA4) transporter protein, which is responsible for a phenotypically similar Mendelian form of retinal disease, were proposed to increase the risk of ARM. To examine the potential relationship of ABCA4 sequence variation and ARM risk in an independent data set, a clinically well-characterized population of 165 multiplex patients with ARM from 70 families, 33 unaffected relatives, and 59 unrelated control subjects with confirmed absence of ARM was screened for variants in any of the 50 exons and exon–intron boundaries of this gene.

methods. A combination of denaturing high-performance liquid chromatography (DHPLC) and bidirectional sequencing was used to detect ABCA4 sequence variants. The data set was analyzed with both case–control and family-based association analysis methods.

results. No evidence was found of significantly different allele frequencies of ABCA4 sequence variants in patients compared with control subjects, and no evidence for association or cosegregation with disease in family-based analyses.

conclusions. This study confirmed the very high degree of ABCA4 sequence polymorphism in the general population, which makes the detection of potential disease-associated alleles particularly challenging. While this study does not definitively exclude ABCA4 from contributing to a small or moderate fraction of ARM, it adds to the body of evidence suggesting that ABCA4 is not a major susceptibility gene for this disorder.

The ABCA4 protein (formerly ABCR; MIM 601691 belongs to the superfamily of ABC (ATP-binding cassette) transporter proteins and is found only in photoreceptor cells of the human retina. The protein is made up of 2273 amino acids and was previously characterized as the photoreceptor rim protein, ¹ because it localizes to the rims of the rod and cone outer segment disks. ² Its role is believed to be the transport of vitamin A derivatives across the photoreceptor disc membrane. ^{3 4} The ABCA4 gene coding for this protein is located on chromosome 1 (p13-p21). It comprises 50 exons, ranging in length from 33 to 266 bp, with an open reading frame (ORF) of 6819 bp and a total genomic length of approximately 140 kb. Gene expression occurs exclusively in retinal rod and foveal cone photoreceptors. ^{5 6} Variations in the group of ABC transporter proteins are the cause of various human diseases, such as cystic fibrosis ⁷ and progressive familial intrahepatic cholestasis. ⁸ Sequence variants in ABCA4, in particular, are responsible for autosomal recessive, juvenile-onset Stargardt disease (STGD1; MIM 248200), ⁵ cone-rod dystrophy, ⁹ and autosomal-recessive retinitis pigmentosa. ^{9 10} Even though ABCA4 has been identified as the causal gene for recessively inherited STGD, mutations in 30% to 40% of patients with STGD remain undetected with current screening technology. This, together with a large number of reported disease-associated mutations including compound heterozygous variants, illustrates the complexity and high degree of sequence variation of the ABCA4 gene and explains why routine genetic testing of patients with STGD is not yet feasible. ^{11 12}  

Some phenotypic overlap exists between the rare Mendelian phenotype of STGD and age-related maculopathy (ARM), a much more common late-onset disorder (also called age-related macular degeneration, AMD). Allikmets et al. ¹³ proposed that variations in the STGD gene may also be associated with an increased risk of ARM. Because ARM is the most common cause of blindness in older adults, with an estimated prevalence of 28% for early and 7% for severe forms in individuals over the age of 75 years, ^{14 15} any confirmed etiologic risk factors, genetic or otherwise, could be of great public health significance. After the initial report, the ABCA4 hypothesis was supported by one additional study ¹⁶ and not supported by others. ^{11 12 17 18 19 20} Therefore, the role of ABCA4 in ARM is still not unequivocally established and thus serves as a prime example for the difficulties often encountered in dissecting genetic risk factors for complex disorders. ²¹  

In this study, we present a comprehensive survey of ABCA4 sequence variation, to further test the proposed hypothesis. We have ascertained a clinically well-characterized population of 165 patients with AMD, 33 of their unaffected relatives (primarily siblings), and 59 unrelated control subjects, for a total of 257 samples. We screened the complete exonic sequence of the ABCA4 gene, as well as a range of flanking sequence, using a combination of denaturing high-performance liquid chromatography (DHPLC) and bidirectional sequencing. We quantified the degree of sequence polymorphism observed in our study population and conducted both case–control and family-based association analyses of ABCA4 variants and ARM risk. 

Methods

Ascertainment of Study Subjects

The ARM index patients and their affected and unaffected relatives were ascertained in the southeastern United States as part of an ongoing genetic study that recruits both multiplex (two or more family members affected with ARM) and singleton (one family member affected with ARM) white families. In addition, an independent age- and ethnicity-matched control group without clinical signs of ARM was ascertained. Participating study sites were Duke University Medical Center (DUMC; Durham, NC), and Vanderbilt University Medical Center (VUMC; Nashville, TN). Only the multiplex families and the unrelated control subjects were included in our initial screening, with the goal of screening the singleton patients for particular variations of interest at a later date. The multiplex families were chosen as the initial screening set to allow for the examination of cosegregation of potential disease-associated variants with the disease phenotype, similar to the approach used by another study of ABCA4 in ARM. ²² The families were identified through a proband from the respective clinic populations or through a proband’s referral to the study site from local ophthalmologists. The protocol of the study conformed to the Declaration of Helsinki and was approved by the Institutional Review Boards of DUMC and VUMC. Informed consent was obtained from all subjects. For clinical diagnosis, stereoscopic fundus photographs were available for all study participants, and grading of ARM severity was performed by three of the authors (EAP, AA, MADLP) according to slightly modified criteria from the International ARM Study Group, ²³ as described previously. ¹⁸ Specifically, early ARM (grade 3) was defined as the presence of extensive (total extent ≥ area of a circle with 350 μm diameter) intermediate (≥63 to <125 μm), or any large (≥125 μm) soft drusen. Drusenoid retinal pigment epithelium (RPE) detachments without fluid were included as indicative of early ARM, whereas serous or hemorrhagic RPE detachments were considered a symptom of late (neovascular) ARM (grade 5). Grade 4 denoted atrophic ARM. Control subjects included 27 spouses of patients with ARM and 32 independent clinic-based control subjects, all of whom were at least 50 years of age and had undergone a complete ophthalmic examination. The majority of control subjects (52/59; 88.1%) were found to be without any ARM features (grade 1), and the others had only small (<63 μm) or nonextensive intermediate drusen (grade 2). The majority of patients (67.9%) and control subjects (64.4%) were women. Although the mean ages at examination were different in cases and control subjects (73.5 years in patients, 66.7 years in control subjects), all the control subjects were ascertained from the population considered to be at risk for ARM (≥50 years of age, with 95% of control subjects older than 55 years). Table 1 shows the frequency distribution of patients and control subjects for the five grades of ARM affection status, as well as their gender and age distribution. 

Laboratory Methods

The genomic sequence of ABCA4 was obtained by alignment of cDNA sequence (GenBank accession number XM_040469; http://www.ncbi.nlm.nih.gov/Genbank; provided in the public domain by the National Center for Biotechnology Information [NCBI], Bethesda, MD) to genomic sequence (from clone GI: 16165069) by using the BLAST program (http://www.ncbi.nlm.nih.gov/entrez/ provided in the public domain by NCBI, Bethesda, MD). We used primer sequences and polymerase chain reaction (PCR) conditions, as described previously. ²⁴ In some instances, nested primers were designed for sequencing. For the PCR fragment that contained exon 10, the reverse sequencing primer 5′-TGATCATGTTCATCTGTGTGC-3′ was designed to overcome technical problems encountered during sequencing with the original PCR primers. Primers for exons 1 to 49 were designed to amplify the exons and 10 to 162 bp of flanking sequence upstream and downstream. The primers for exon 50 were designed to extend amplification into an additional 463 bp of flanking sequence. Using DHPLC, we screened all exons in pools of up to five samples, which were designed to include all members of one or two families, if possible. DHPLC was conducted on a nucleic acid fragment analysis system (Model D7000 Transgenomic Wave with UV detector; Transgenomic Inc., Omaha, NE). PCR products were subjected to a heteroduplex reaction under the following conditions: initial denaturation at 95°C for 3 minutes; 60 cycles of denaturation at 95°C, minus 1.1°C per cycle. For each of the 50 exonic regions, 5 to 10 μL of heteroduplexed PCR template was injected and eluted from the column with a linear acetonitrile gradient at a constant flow of 0.9 mL/min. The optimum temperatures for heteroduplex and homoduplex DNA were determined using a temperature gradient on a subset of samples. Using the software allied with the system (WaveMaker Software ver. 4.0, and the DHPLC melt program; Transgenomic Inc.) the ideal temperatures were found to vary between 57°C and 65°C, depending on fragment sequence. Details on gradients, reagents, and optimal melting temperatures are available on request. Pool/exon combinations in which DHPLC indicated the presence of variable sequences were broken down into their component samples, each of which were then bidirectionally sequenced using an automated sequencer (model CEQ 2000 XL; Beckman Instruments, Fullerton, CA; or model 3100; Applied Biosystems, Foster City, CA). The wild-type sequence was derived from clone GI:16165069. 

Statistical Methods

Characterization of Detected Variants.

For all detected ABCA4 sequence variants, tests of deviations from Hardy-Weinberg equilibrium (HWE) at each site and of linkage disequilibrium (LD) between pairs of sites were performed in samples of unrelated individuals. Empiric significance levels for the test of HWE were generated with a permutation approach (6000 permutations, GDA program; http://lewis.eeb.uconn.edu/lewishome/software.html; provided in the public domain by the University of Connecticut, Storrs, CT). ²⁵ Pair-wise LD for variants with minor allele frequency of 5% or more was estimated by the commonly used measures r ² (also called δ²) and D′ (GOLD program ²⁶ ). The maximum value of D′ = 1 indicates “complete LD,” meaning that two variants have not been separated by recombination during the history of the sample and that, at most, three of four possible two-locus haplotypes are observed. However, the relative magnitude of D′ values less than the maximum is more difficult to interpret and depends strongly on sample size. ²⁷ In contrast, the maximum value of r ² = 1 indicates “perfect LD,” meaning that two variants have not been separated by recombination during the history of the sample and have the same allele frequency. In this case, only two of four possible haplotypes are observed. The value of r ² is more directly related to the amount of information provided by one locus about the other, which makes intermediate values more easily interpretable. ²⁷ Because D′ has been used extensively in previous empiric studies, it is included here along with r ², which is preferred by population geneticists. 

DHPLC Sensitivity.

Instead of sequencing a subset of our samples for the entire ABCA4 gene to compare the results to those from the DHPLC screening, ¹¹ we chose a different approach to estimate at least an upper bound for the actual DHPLC sensitivity. For all exons in which DHPLC indicated the absence of variants in the sample pools, 10% of the samples were sequenced for that exon. In general, the sensitivity of DHPLC screening has been reported to be on the order of at least 90% to 95%. 

Case–Control Analysis.

For all detected variants, comparisons of allele frequencies between independent cases (n = 70) and control subjects (n = 59) were performed with the Fisher exact test. Where HWE in cases or control subjects was violated, the Armitage trend test, which is robust to deviations from HWE, ²⁸ was used instead to compare genotype frequencies. To include all the familial cases in the analysis of genotype frequencies, we used a modification of the Armitage trend test that appropriately accounts for correlations among related cases when comparing them with unrelated control subjects. ²⁹ For independent cases and control subjects, the total number of variants per person was compared with Wilcoxon’s rank sum test, both overall and by type of variant (missense, synonymous, intronic). 

Family-Based Analysis.

To assess further the evidence for allelic association of any variant with the ARM phenotype, the pedigree disequilibrium test (PDT) ^{30 31} was used. Because our family data set was primarily composed of discordant sibling pairs, with very few parent–offspring triads, the PDT is essentially a weighted version of the sibship disequilibrium test (SDT). ³² It compares the number of times an allele is transmitted to affected offspring with the number of times this allele is transmitted to their unaffected siblings. It is a valid test of association in both nuclear and extended pedigrees, even in the presence of linkage. 

In the presence of a disease-associated ABCA4 allele, we would expect to see evidence for increased identity-by-descent (IBD) sharing of this allele among sibling pairs affected with ARM, reflecting cosegregation of the variant allele and disease within a family. This was tested with the computer program SIBLINK (http://www.chg.duke.edu/software/siblink.html/ provided in the public domain by the Center for Human Genetics, Duke University Medical Center, Durham, NC). ³³  

Haplotype Analysis.

Theoretically, one could test all possible pairs and triplets of variant sites to examine potential joint effects of ABCA4 sequence polymorphisms, which may not be detected in tests of individual sites. However, this would cause a substantial multiple-testing problem. Therefore, we chose to estimate haplotype frequencies based on sites with minor allele frequency of at least 5% in control chromosomes, using the computer program MERLIN (http://www.sph.umich.edu/csg/abecasis/merlin/index.html/ provided in the public domain by the Center for Statistical Genetics, School of Public Health, University of Michigan, Ann Arbor, MI). ³⁴ It has been shown that sample size requirements increase dramatically when allele or haplotype frequencies decline below 5%, ³⁵ and we believed that our available sample size did not provide sufficient power to examine effects of less common haplotypes. The observed common haplotypes (≥5% frequency in either patients or control subjects) were ordered by their estimated frequencies, and the subset of those sites that distinguished between these haplotypes was examined for disease association. This is similar to the haplotype-tagging (htSNP) approach described by Johnson et al. ³⁵ We used the program haplo.score ³⁶ (http://www.mayo.edu.statgen/ provided in the public domain by the Mayo Foundation for Medical Education and Research, Rochester, MN) for case-control analysis, and TRANSMIT (http://www.cimr.cam.ac.uk/clayton/software/transmit.txt, provided in the public domain by the Cambridge Institute for Medical Research, University of Cambridge, UK) ³⁷ for family-based analysis of an association of ABCA4 haplotypes and ARM phenotype. 

Nucleotide Diversity.

Finally, we estimated the nucleotide diversity of ABCA4 in the two populations of independent patients and control subjects. Nucleotide diversity is a measure that combines the number of variable positions in a gene with the respective allele frequencies at these sites and can be compared with similar published estimates for this and other genes. It is defined as the mean number of nucleotide positions expected to be heterozygous in random population sampling of chromosome pairs ^{11 38} and is estimated as  

\[{{\sum}_{j\ {=}\ 1}^{m}}\ \left(1\ {-}\ {{\sum}_{i\ {=}\ 1}^{n}}\ p_{ij}^{2}\right)\]

where p _ij is the prevalence of the ith of n alleles at the jth of m polymorphic sites. 

Results

Characterization of Detected Variants

A total of 25 distinct ABCA4 nucleotide polymorphisms were detected in the 257 samples in this study. Fifteen occurred in the coding region of the gene (Table 2) , of which six were synonymous codon changes and nine were amino acid substitutions (missense variants). Six of the 15 variants occurred in at least 5% of control alleles, including three missense variants. Ten variable sites were detected in flanking intronic sequence (Table 3) , and seven of them occurred in at least 5% of control alleles. None of the intronic variants was an obvious splice site alteration. We did not detect any truncating variants. Most of the variants we detected have previously been reported in patients with ARM or in control subjects, but six of them were novel. We did not detect the variants G1961E (5882G→A, exon 42) and D2177N (6529G→A, exon 48) reported as disease-associated ¹⁶ in any of the 257 samples screened in this study. To verify the absence of these two variants, we resequenced exons 42 and 48 in all of our 165 ARM samples in both directions. In agreement with the DHPLC results, we did not detect G1961E and D2177N in this patient population. In our earlier work, ¹⁸ we had detected the D2177N variant in two of our sporadic patients with ARM. 

There was no evidence for deviation from HWE in the group of independent cases (n = 70) or control subjects (n = 59) for 22 of the 25 detected sequence variants. In cases, there was an excess of homozygous non–wild-type genotypes for 1240-14C→T (P = 0.003), compared with the number of heterozygous genotypes. In control subjects, we observed more heterozygous genotypes than expected for 1268A→G (P = 0.02), compared with the number of homozygous non–wild-type genotypes, and more homozygous than heterozygous non–wild-type genotypes for 1356+5delG (P = 0.01). All genotypes at these sites were reviewed in the laboratory and were confirmed with multiple sequencing techniques. When a Bonferroni correction was applied to maintain a global 0.05 significance level for 25 tests, the deviation from HWE was no longer significant for these three variants (P > 0.05/25 = 0.002). 

Linkage disequilibrium (LD) between pairs of variants with minor allele frequency of 5% or more is illustrated in Figures 1 and 2 . As expected, LD generally decayed with increasing distance. LD, as measured by r ² (Fig. 2) , was restricted to fairly small distances (r ² > 0.3 for distances of 1 to 3174 bp). LD, as measured by D′ (Fig. 1) , occasionally extended up to longer distances (D′ = 1 for distances up to 108 kb). Variants less than 5 kb apart showed a range of 0.39 to 1.0 for D′ and 0.006 to 1.0 for r ². 

DHPLC Sensitivity

The sequencing of 10% of the samples for those exons in which DHPLC did not detect the presence of sequence polymorphisms led to the detection of two additional variants: 3261G→A, a missense (Glu1087Lys) variant in exon 22, and 4848-16delAC in intron 34. 3261G→A was detected in two of six control subject chromosomes and in none of 36 patient chromosomes. Because the 4848-16delAC variant was observed in patients with ARM, it was sequenced in additional samples and detected at a similar frequency in patient (12/182; 0.066) and control (2/30; 0.067) chromosomes. With a total of 25 variants detected by DHPLC, our estimated upper bound of DHPLC sensitivity is 25/27 (92.6%), well within the range of general estimates. 

Case–Control and Family-Based Analysis

Tables 2 and 3 show allele frequencies of the 25 detected ABCA4 sequence variants in the subgroups of interest. For most of the variants, our observed allele frequencies were consistent with those reported previously. ^{11 12} Variants 1240-14C→T, 1268A→G, 1356+5delG, and 6006-16G→A were observed at a noticeably greater frequency in our data. Note that one previous study ¹¹ always observed changes at positions 5843 and 5844 in combination with each other (CA→TG), whereas we observed the 5844A→G change more frequently than the 5843C→T change. 

The genotype and allele frequency comparisons did not indicate any disease-associated variants of the ABCA4 gene, in the analysis either of unrelated cases and control subjects or of familial cases and unrelated control subjects (Tables 2 and 3) . This was consistent with the family-based analyses: the PDT did not detect evidence for excess transmission of a particular allele to affected offspring, when compared with transmissions to unaffected offspring (n = 57 discordant sibling pairs). SIBLINK did not reveal any evidence for increased IBD sharing among affected sibling pairs (n = 115) within the multiplex families. 

The mean number of variable ABCA4 sites, overall and by type of variant, in patients with ARM and control subjects is shown in Table 4 . No significant differences between patients and control subjects were observed. 

Haplotype Analysis

Table 5 shows the six haplotypes with at least 5% estimated frequency in either patients with ARM or control subjects. Only polymorphic sites with at least 5% minor allele frequency were included in this analysis. These six haplotypes account for only 58.0% (patients with ARM) and 56.0% (control subjects) of observed haplotypes, indicating that there is a large number of rare haplotypes with frequencies of less than 5%. Table 5 also shows that variation between the six most common haplotypes was due to only the first four polymorphisms (sites 302+26, 1240-14, 1268, and 1356+5), with the remaining sites consistently including the wild-type allele. Previous reports indicated that the common haplotypes of many genes account for at least 80% of the total number of observed haplotypes and that these haplotypes can generally be “tagged” with two to five single-nucleotide polymorphisms (htSNPs) capturing the haplotype variability. ³⁵ However, given the substantially larger size of the ABCA4 gene, compared with the genes examined by Johnson et al., ³⁵ our finding was not altogether surprising. Results of tests of an association between disease and haplotypes of the four htSNP sites (302+26, 1240-14, 1268, and 1356+5) with the programs haplo.score ³⁶ and TRANSMIT ³⁷ were nonsignificant. 

Nucleotide Diversity

As suggested by previous studies, the ABCA4 gene is characterized by a very high degree of sequence polymorphism. In this study, its nucleotide diversity, including all the polymorphic sites, was estimated as 4.24 in patients with ARM and 4.28 in control subjects. Nucleotide diversity in the ABCA4 coding region was estimated as 1.99 in patients with ARM and 1.96 in control subjects. This is even higher, but certainly comparable to a previous estimate of 1.28 in individuals without ARM. ¹¹ Taking into account the size of the ABCA4 gene (6819 bp; ORF), this translates to a density of 2.9 × 10⁻⁴ sites/nucleotide, much greater than the corresponding values for other ophthalmic genes investigated in a similar fashion, such as the VMD2 gene in individuals without Best disease (0.14 sites, 0.8 × 10⁻⁴ sites/nucleotide) ³⁹ and the EFEMP1 gene in individuals without Malattia Leventinese or Doyne honeycomb retinal dystrophy (0.003 sites, 0.03 × 10⁻⁴ sites/nucleotide). ⁴⁰  

Discussion

The genetic dissection of complex traits has turned out to be substantially more challenging than anticipated, given the successful determination of the genetic basis of many Mendelian traits. Replication of reported associations has been difficult for many complex phenotypes, and ARM is no exception. None of the obvious candidate genes—that is, those responsible for other forms of retinal degeneration—have been convincingly demonstrated to play a role in ARM etiology. The initial suggestion of an ABCA4 association with ARM risk ¹³ prompted several subsequent studies by different research groups with heterogeneous findings. Our study is consistent with those previous studies ^{11 12 17 18 19 20} that did not detect an increased ARM risk conferred by ABCA4 sequence variants. Our control individuals were clinically examined for the presence of ARM and were old enough to manifest signs of the disease (all ≥50 years; 95% were ≥55 years). We did not see statistically significant increased frequencies of any sequence variants in patients with ARM compared with control subjects with clean maculae. Our conclusions were identical when we restricted control individuals to those with complete absence of macular drusen (grade 1 only) or to those of at least 55 years of age. We also did not observe any evidence for cosegregation of ABCA4 variants with the ARM phenotype in multiplex families. We confirmed the previously reported high nucleotide diversity of the ABCA4 gene, with an average of more than four variable sites in both patients with ARM and the control population, suggesting that it is unusual for an individual in the general population to be homozygous for the consensus ABCA4 sequence. ¹¹ Consistent with this observation, we also inferred a relatively small number of common (≥5% frequency) and a correspondingly large number of rare ABCA4 haplotypes in the general population. Strong levels of LD (D′ > 0.8) seemed to extend only rarely over distances beyond 5 kb. 

Our data set cannot definitively exclude ABCA4 from influencing ARM risk. However, the fact that we did not detect a single copy of the only variants that have so far been reported as potentially disease associated, G1961E (5882G→A) and D2177N (6529G→A), in any of our 165 patients with ARM, suggests that their contribution to this phenotype is small at best. The probability of detecting either of these two variants, given an estimated carrier frequency of 3.4% in ARM cases, ¹⁶ is 99.7% for a sample size of 165 individuals, and 91.1% for 70 independent patients. The 95% binomial confidence interval for our observed proportion of 0 in a sample of 165 is (0.0,0.02), which does not include the previously estimated value of 0.034. In addition to the other negative ABCA4 screening studies, several published studies that performed linkage analysis with highly polymorphic markers distributed evenly across the entire human genome have so far not detected any evidence for an ARM susceptibility locus in the 1p region harboring ABCA4, even when the analyzed data sets were of substantial size (Swaroop A, Yashar BM, Ghiasvand N, Yoshida S, Zareparsi S, Stringham H, Richards J, Boehnke M, Jacobson S, Abecasis G, ARVO Abstract 832, 2002; Iyengyar SK, Schick J, Reading K, Milliard C, Brey N, Liptak R, Klein R, Klein B, Elston R, ARVO Abstract 1844, 2002). ^{41 42}  

The population-attributable risk of these two ABCA4 sequence variants has been estimated at 8% to 10%. ¹⁶ However, with an assumed relative risk of 5 and a carrier frequency of 0.95% in control subjects, ¹⁶ this value would be only 3.8%. This illustrates that, with risk factor frequencies as low as these, the estimated attributable risk is quite sensitive to very small changes in the assumed parameters. Although large sample sizes allow for the estimation of relative risks for exposures of small prevalence, we believe that a strategy of screening such large data sets for ABCA4 variants would have to weigh the relatively high costs of this approach against the benefits. We have estimated the case–control sample size necessary for 80% power to detect an ABCA4 effect on ARM risk. The frequency of potentially disease-associated variants has been consistently estimated to be no higher than 1% in a nonpatient population. ^{11 12 16 20} Because of uncertainty about the true effect size, we have assumed three plausible relative risk values for an ABCA4 variant: 5, estimated for G1961E; ¹⁶ 3, estimated for D2177N; ¹⁶ and 1.5, the lower confidence limit for G1961E. ¹⁶ Power calculations using commercial software (nQuery-Advisor; Statistical Solutions, Saugus, MA) and DSTPLAN (http://odin.mdacc.tmc.edu/anonftp; provided in the public domain by the Department of Bioinformatics M. D. Anderson Cancer Center, University of Texas, Houston, TX) indicate that approximately 600 (300 case–control pairs), 1,600, and 16,000 samples, respectively, would have to be tested to achieve 80% power, at the 5% significance level, for detecting these relative risks. The most definitive test would be sequencing of the entire gene, which, with an approximate cost of $10 per exon for a single forward and reverse reaction, would generate a total cost of $300,000, $800,000, and $8,000,000, respectively. A much more feasible study might employ commercial assays (e.g., TaqMan-based Assay-by-Design or Assay-by-Demand; Applied Biosystems, Inc.) to genotype only the two previously implicated variants. With an approximate cost of $1 per sample (reagents and labor) for each variant, the expense would amount to $1,200, $3,200, and $32,000, respectively. 

However, this study would hinge entirely on the assumption that at least one of the two variants is indeed the causal one, or in very high LD with it. An alternative to sequencing of the entire gene is the design of a DNA chip, which would test all the published ABCA4 variants. If a per-sample cost of less than $500 ($10 × 50) could be achieved, the total cost would be less than that calculated for whole-gene sequencing, although still substantial. Given these prohibitive costs and the necessary sample sizes, our own results and those of several other studies, we argue that the currently available technology does not allow for a definitive resolution of the ABCA4 controversy in ARM. Because costs are virtually certain to decrease over time, this will probably be feasible within the next decade or so, if large collaborative studies can be arranged to obtain a patient and control population of sufficient size. However, even if significantly different allele frequencies between patients with ARM and genetically matched control subjects were found, the lack of consistent familial cosegregation of ABCA4 variants and disease phenotype described thus far ^{22 43} makes the benefit of screening relatives of patients with ARM for such rare variants questionable. 

We agree with the proponents of the ARM-ABCA4 hypothesis that ABCA4 is an excellent candidate gene for this phenotype, based on our current knowledge of its function and the fact that it is responsible for several types of retinal disorders demonstrating at least some degree of overlap with ARM features. It has been proposed that there is a gradient of severity of retinal disease inversely related to residual ABCA4 activity, based on functional analyses of ABCA4 mutations in a family with both STGD and retinitis pigmentosa phenotypes. ⁴⁴ With this hypothesis in mind, it is tempting to speculate that mild mutations may lead to ARM, but this has so far not been confirmed convincingly in human studies. As for experimental data on ABCA4, in a recent study, ⁴⁵ investigators specifically examined mice heterozygous for a null mutation in the mouse abcr gene and found that the resultant phenotype was similar to that seen in abcr knock-out mice, but manifested itself at a slower, age-related rate, suggesting a similarity to the human ARM phenotype. However, it is not clear that the two sequence variants proposed to be disease-associated in humans (G1961E, D2177N) are equivalent to a null mutation, even though they were shown to have some effect on the function of the ABCA4 protein. ⁴⁶ An animal model can be an invaluable tool for assessing the potential significance of sequence variation in a candidate gene; however, caution is necessary in inferring functional consequences in the context of complex human disease from data generated on mice. 

A recent editorial ²¹ pointed out some additional sources for an erroneous classification of ABCA4 variants as disease-associated: Because of the nature of the biological processes that give rise to the existence of more than one DNA sequence version, allele frequencies tend to vary between different populations. For this reason, the careful matching of patient and control populations is especially crucial in genetic studies in general and in multicenter genetic studies in particular. In the large ABCA4 consortium study, ¹⁶ matching by age was certainly incomplete, with an average age of 39.2 years in a set of general population control subjects (n = 653; average age of 73.4 in another set of age-matched control subjects, n = 605), compared with 74.8 years in patients (n = 1218). Because only the two sequence variants previously implicated in ARM were screened, it is unclear whether the genetic matching of patients and control subjects was sufficiently complete. That said, several recent empiric studies have suggested that the impact of population stratification in reasonably well-designed genetic–epidemiologic studies may not be as large as initially suspected. ^{47 48} Nevertheless, the report of a much higher prevalence of the G1961E allele in healthy individuals of Somali ancestry ²⁰ was illuminating, and, in the absence of high prevalences of either ARM or STGD in Somalia, certainly raised doubts about the potential disease association of this allele. Finally, the importance of a detailed clinical assessment of the phenotype under study cannot be overemphasized. In the case of ABCA4, care should be taken to avoid misclassifying even a small number of individuals with end-stage STGD as having atrophic ARM, which may be sufficient to imply a role of the STGD gene in the more common ARM phenotype. 

In summary, our study adds to the body of evidence arguing against a substantial role of ABCA4 in ARM. We cannot exclude the possibility that ABCA4 variants may only do harm in the presence of unknown modifiers of a genetic or environmental nature or may affect only subgroups of patients. Future large-scale studies providing sufficiently detailed data to define clinical subgroups and including information about environmental risk factors for ARM are needed to explore this hypothesis. 

 Supported by Grant EY12118 from the National Eye Institute (MAP-V) and by Grant AG11268 from the National Institute on Aging.

 Submitted for publication September 18, 2002; revised February 5, 2003; accepted February 14, 2003.

 Disclosure: S. Schmidt, None; E.A. Postel, None; A. Agarwal, None; I.C. Allen Jr, None; S.N. Walters, None; M.A. De La Paz, None; W.K. Scott, None; J.L. Haines, None; M.A. Pericak-Vance, None; J.R. Gilbert, None

 The publication costs of this article were defrayed in part by page charge payment. This article must therefore be marked “advertisement” in accordance with 18 U.S.C. §1734 solely to indicate this fact.

 

Table 1.

 

Study Population Screened for ABCA4 Sequence Variants

	ARM Grade (n, %)*									Mean Age (SD) [Range]	Female (n, %)	Total
		1	2	3	4	5
ARM patients	—	—	37 (22.4)	33 (20.0)	95 (57.6)	73.5 (9.7)	112 (67.9)	165
						[41–93]
Control subjects	52 (88.1)	7 (11.9)	—	—	—	66.7 (8.5)	38 (64.4)	59
						[50–85]
Unaffected relatives	25 (75.8)	8 (24.2)	—	—	—	66.1 (11.2)	18 (54.6)	33
						[39–84]
								257

Table 2.

 

Polymorphisms and Rare Sequence Variants in Exons of the ABCA4 Gene

Exon	Nucleotide Change	Effect	Allele Frequency*					P , †	P , §	Reference, ∥
						Independent ARM (n = 140)	All ARM (n = 330)				Control Subjects (n = 118)
6	589G→C	Asp197Asn	0.000	0.000	0.009	0.46	0.12	—
6	635G→A	Arg212His	0.030	0.026	0.000	0.13	0.11	W, R
10	1268A→G	His423Arg	0.394	0.371	0.427	0.62, ‡	0.34	W, R
10	1269C→T	His423His(syn)	0.033	0.039	0.031	1.0	0.74	W
18	2701A→G	Thr901Ala	0.000	0.003	0.000	NA	0.58	W, R
23	3495C→T	Asn1165Asn(syn)	0.000	0.003	0.000	NA	0.75	—
30	4469G→A	Cys1490Tyr	0.007	0.003	0.000	1.0	0.59	W
37	5206T→C	Ser1736Pro	0.009	0.008	0.000	1.0	0.44	W
40	5603T→A	Asn1868Ile	0.100	0.102	0.054	0.29	0.18	W
40	5682G→C	Leu1894Leu(syn)	0.293	0.272	0.298	1.0	0.64	W
41	5814A→G	Leu1938Leu(syn)	0.160	0.169	0.218	0.33	0.38	W
42	5843C→T	Pro1948Leu	0.052	0.038	0.054	1.0	0.50	W
42	5844A→G	Pro1948Pro(syn)	0.199	0.192	0.205	1.0	0.77	W
44	6069C→T	Ile2023Ile(syn)	0.040	0.050	0.044	1.0	0.82	W
44	6079C→T	Leu2027Phe	0.000	0.000	0.009	0.48	0.13	W

Table 3.

 

Polymorphisms and Rare Sequence Variants in Introns of the ABCA4 Gene

Exon	Nucleotide Change	Allele Frequency					P	P	Reference
				Independent ARM (n = 140)	All ARM (n = 330)	Control Subjects (n = 118)
3	302+26A→G	0.379	0.392	0.411	0.69	0.77	W
10	1240−14C→T	0.298	0.267	0.250	0.52	0.80	W
10	1356+5delG	0.302	0.277	0.372	0.47	0.20	W
17	2653+32T→A	0.000	0.003	0.000	NA	0.62	—
17	2653+33G→A	0.000	0.003	0.000	NA	0.62	—
30	4353−4T→C	0.000	0.003	0.000	NA	0.59	—
42	5836−43C→A	0.169	0.160	0.196	0.62	0.42	W
42	5836−11G→A	0.169	0.160	0.188	0.74	0.54	W, R
44	6006−16G→A	0.176	0.164	0.167	0.86	0.95	W, R
50	6819+372A→G	0.045	0.062	0.049	1.0	0.72	—

Figure 1.

 

Pairwise linkage disequilibrium (LD) for ABCA4 sequence variants with minor allele frequency of 5% or more in patients with ARM (n = 70), measured by D′ (program GOLD ²⁶ ). The plot for control subjects (n = 59) is qualitatively very similar (data not shown).

Figure 2.

 

Pair-wise linkage disequilibrium (LD) for ABCA4 sequence variants with minor allele frequency of 5% or more in patients with ARM (n = 70), measured by r ² (program GOLD ²⁶ ). The plot for control subjects (n = 59) is qualitatively very similar (data not shown).

Table 4.

 

Mean Number of ABCA4 Sequence Variants in Independent Patients with ARM and Control Subjects

Type of Variant	ARM (n = 70)	Control (n = 59)	P *
Any	3.9 (2.5)	4.4 (2.7)	0.34
Missense	0.8 (0.7)	0.8 (0.6)	0.46
Synonymous	1.0 (1.1)	1.2 (1.1)	0.31
Intronic	2.1 (1.5)	2.3 (1.8)	0.64

Table 5.

 

Common ABCA4 Haplotypes in Patients with ARM and Control Subjects, Based on Sites with Minor Allele Frequency of 5% or More

Exon																									Frequency
3 302+26 A/G	10 1240-14 C/T	10 1268 A/G	10 1356+5 7G/6G	40 5603 T/A	40 5682 G/C	41 5814 A/G	42 5836-43 C/A	42 5836-11 G/A	42 5843 C/T	42 5844 A/G	44 6006-16 G/A	50 6823+372 A/G	Cases (%)	Controls (%)
A	C	G	7G	T	G	A	C	G	C	A	G	A	22.8	29.9
A	C	A	6G	T	G	A	C	G	C	A	G	A	7.9	7.3
G	T	A	7G	T	G	A	C	G	C	A	G	A	7.9	5.6
G	C	A	7G	T	G	A	C	G	C	A	G	A	5.4	2.5
A	T	A	6G	T	G	A	C	G	C	A	G	A	7.7	5.7
G	C	A	6G	T	G	A	C	G	C	A	G	A	6.3	5.0
Sum													58.0	56.0

The authors thank the patients with ARM, their families, and the volunteers who served as control subjects for participating in this research and the personnel at the Duke Center for Human Genetics for assistance in data management and preparation of this manuscript.