Haplotypes at ATM Identify Coding-Sequence Variation and Indicate a Region of Extensive Linkage Disequilibrium

Abstract

Genetic variation in the human population may lead to functional variants of genes that contribute to risk for common chronic diseases such as cancer. In an effort to detect such possible predisposing variants, we constructed haplotypes for a candidate gene and tested their efficacy in association studies. We developed haplotypes consisting of 14 biallelic neutral-sequence variants that span 142 kb of the ATM locus. ATM is the gene responsible for the autosomal recessive disease ataxia-telangiectasia (AT). These ATM noncoding single-nucleotide polymorphisms (SNPs) were genotyped in nine CEPH families (89 individuals) and in 260 DNA samples from four different ethnic origins. Analysis of these data with an expectation-maximization algorithm revealed 22 haplotypes at this locus, with three major haplotypes having frequencies ⩾.10. Tests for recombination and linkage disequilibrium (LD) show reduced recombination and extensive LD at the ATM locus, in all four ethnic groups studied. The most striking example was found in the study population of European ancestry, in which no evidence for recombination could be discerned. The potential of ATM haplotypes for detection of genetic variants through association studies was tested by analysis of 84 individuals carrying one of three ATM coding SNPs. Each coding SNP was detected by association with an ATM haplotype. We demonstrate that association studies with haplotypes for candidate genes have significant potential for the detection of genetic backgrounds that contribute to disease.

Introduction

Qualifying and quantifying the genetic contribution to the etiology of common complex disease remains one of the great quests of modern medical genetics. The complexity of multifactorial diseases challenges the paradigms and tools of conventional genetic research. Traditional methods of genetic analysis do not have the statistical power or sensitivity for the task of teasing out a genetic contribution when it is subtle or when several genes may be working together (Risch and Merikangas 1996). Genomewide association studies, as well as population studies with candidate genes, have been touted as possible alternatives to linkage analysis (Risch and Merikangas 1996; Collins et al. 1997; Kruglyak 1999; Risch 2000). These approaches focus on finding either a causative variant or a genetic variant closely linked with the disease phenotype. Some studies utilizing single-nucleotide polymorphisms (SNPs) have succeeded in detecting the risk for disease, notably in the case of the apolipoprotein type E (apoE) gene and both coronary artery disease (Boerwinkle et al. 1996) and Alzheimer disease (Strittmatter and Roses 1995). These studies were able to directly assess the risk conferred by known apoE functional variants. In some other cases, however, the attempt to correlate single-locus alleles with phenotypes have produced mixed results (Josefsson et al. 1998; Kraft et al. 1998; Storey et al. 1998).

Haplotype association with disease by the linkage disequilibrium (LD) approach has been used successfully for the identification of genomic regions containing loci responsible for disease phenotypes (MacDonald et al. 1992; Yu et al. 1996). The same principle can be applied by use of haplotypes of biallelic markers to detect disease association. Using several SNPs distributed across 100–200 kb should result in statistical sensitivity that is greater than that in studies using fewer loci. Another strength of such an approach is the ability to use purely epidemiological populations for detection of chromosomal backgrounds lending risk for disease.

All of these approaches are, to one extent or another, dependent on LD. An understanding of LD relationships between markers will inform the efficacy and design of future LD-based strategies for detection of genetic contributions to common disease. Simulation studies have estimated the length of useful LD to be as low as 3 kb (Kruglyak 1999). Recent investigations support the notion that LD varies throughout the genome (Collins et al. 1999; Taillon-Miller et al. 2000) and that it can extend to considerable lengths, such as several hundred kilobases (Collins et al. 1999; Eaves et al. 2000; Moffatt et al. 2000; Taillon-Miller et al. 2000). Reports of such extreme differences indicate the need for further study of the extent and nature of LD.

Allelic variation leading to functional variants of genes may predispose to risk for seemingly sporadic cases of common disease (Lander 1996; Collins et al. 1997). Here we describe a strategy for exploring the possible effects of functional variants of genes involved in familial cancers. We use a resequencing approach to detect SNPs across a large (184 kb) genomic region containing the ATM gene. ATM is responsible for the autosomal recessive disease ataxia-telangiectasia (A-T) (Savitsky et al. 1995). A-T is characterized by cerebellar ataxia, oculocutaneous telangiectasia, immune deficiency, sensitivity to ionizing radiation, increased incidence of tumors, and chromosomal instability (Gatti et al. 1991). A-T heterozygotes may be at increased risk for development of cancers, most prominently—and controversially—breast cancer (Swift et al. 1987, 1991; Morrell et al. 1990; Stankovic et al. 1998; Gatti et al. 1999). With carrier frequencies estimated to be from 0.5% to >1% (Swift et al. 1986; Gatti et al. 1999), assessment of cancer risk for this population is a compelling endeavor. In addition, the ATM-gene product is centrally involved in cellular responses to DNA damage, including DNA double-strand break repair and signaling leading to cell-cycle arrest and apoptosis (reviewed in Rotman and Shiloh 1999). We genotyped 295 individuals from four ethnic groups, for 14 SNP markers that spanned 142 kb. An expectation-maximization algorithm estimated 22 ATM haplotypes from these data. Tests for recombination and LD revealed (a) no evidence for recombination in the white European American study population and (b) perfect disequilibrium extending the full length marked by these SNPs. We then conducted a model association study with these haplotypes and a population of samples that possessed one of three different coding SNPs (cSNPs) in the ATM gene. The results of this study provide strong support for the utility of complex SNP haplotypes as a means to detect polymorphisms in a population-based sample.

Subjects and Methods

Human Subjects

For SNP discovery, genomic DNA from five unrelated white European Americans was sequenced. This DNA was extracted from lymphoblast and fibroblast cell lines. For SNP genotyping, individuals from four ethnic groups were sampled: African American (n=71), Asian American (n=39), white European American (n=77), and Hispanic (n=73). All ethnic samples (self-described ethnicity) were part of a collection of 941 DNA purified samples from anonymous blood donors in community-based blood drives in southeastern and central Texas. Samples analyzed in the model association study were also from this DNA collection. Members of nine CEPH families were also analyzed. In all families, four grandparents, two parents, and four children were examined; since two of these families share a grandparent, 89 individuals were genotyped, and the number of segregating chromosomes is 70.

Samples from Great Apes

Six great-ape samples were genotyped in this study: two from common chimpanzees (Pan troglodyte), one from a bonobo (P. paniscus), two from western lowland gorillas (Gorilla gorilla), and one from an eastern lowland gorilla (G. g. graueri).

PCR and Sequencing Primers

Primers for DNA amplification and sequencing were designed by MacVector, version 6.0.1. The 184-kb genomic sequence of ATM was masked for repetitive sequence, by Repeat Masker. Thirty-six primer sets were designed to amplify regions containing little or no repeat sequence, distributed evenly throughout the sequence. Primers were selected that met strict criteria for melting temperature and that amplified regions containing very little or no repeat sequence. The same primers were used for PCR and sequencing reactions and are listed in Appendix A.

PCR Amplification of Genomic DNA

Genomic DNA from five unrelated individuals was amplified by means of 29 of the 36 primer sets mentioned above. The 50-μl reactions included DNA (200 ng), standard PCR buffer, dNTPs (0.1 mM each), Taq (0.5 μl; Perkin-Elmer), and primers (1 μM each). PCR was performed in a Perkin Elmer 9700 analyzer, with an initial denaturation at 95°C for 5 min, followed by 30 cycles of 95°C for 30 s, 60°C for 30 s, and 72°C for 30 s, and a final step at 72°C for 7 min. For all amplicons, 6 μl of PCR product was run on a 1.5% agarose gel.

DNA Sequencing

PCR products were purified and sequenced. Preparation of DNA for sequencing included incubation of ∼60 ng of PCR product with shrimp alkaline phosphatase (2 U; Amersham) and exonuclease I (10 U; Amersham) at 37°C for 15 min, followed by enzymatic inactivation at 80°C for 15 min. Sequencing of each PCR product was performed with the Thermo Sequenase [³³P]-radiolabeled terminator-cycle sequencing kit (Amersham Pharmacia), according to the manufacturer’s instructions. Sequencing reactions were performed in a Perkin Elmer 9700 analyzer, with an initial denaturation at 95°C for 1 min, followed by 35 cycles of 95°C for 30 s, 55°C for 30 s, and 72°C for 1 min. Samples were run on 6% polyacrylamide gels, fixed for 15 min in 5% acetic acid/20% methanol, and dried.

Multiplex PCR

Sequencing revealed 17 SNPs in 15 different regions of the gene. These 15 PCR amplicons were multiplexed into two PCR reactions. Multiplex group 8 amplifies eight fragments, and Multiplex group 7 amplifies seven fragments. The 50-μl reactions for group 7 included DNA (400 ng), standard PCR buffer (2×), dNTPs (0.2 mM each), and Taq (0.5 μl; Perkin-Elmer). The 50-μl reactions for group 8 included DNA (400 ng), standard PCR buffer (1.8×), dNTPs (0.2 mM each), and Taq (0.5 μl; Perkin-Elmer). Primers include some of those originally designed for sequencing and some of those newly designed to alter the size of the amplicons. Products were separated by ⩾20 bp, so that they could be resolved from one another on a 2.5% agarose gel. Multiplex PCRs were checked to have amplified all products, by running 6 μl of product on a 2.5% agarose gel. The concentrations and primer sequences used for PCR are listed in Appendix B.

Allele-Specific Oligonucleotide (ASO) Hybridizations

Genotypes for each SNP were determined in all sample populations, by ASO hybridizations. ASO hybridizations were performed as described by DeMarchi et al. (1994). We performed ASO hybridization for 14 SNPs for each individual typed. These 14 SNPs were chosen from the original 17 because they perform consistently well under standard ASO-hybridization conditions. Hybridizations were performed under conditions that allowed for annealing of only the probe that is an exact match for the substrate DNA. Genotypes for SNPs were read on at least two independent occasions. The sequences of the ASO-hybridization probes are listed in Appendix C.

Estimation of Haplotypes and Frequencies

Haplotypes and their frequencies were estimated on the basis of unphased genotype data, by the computer program EMHAPFRE. Described in the work of Excoffier and Slatkin (1995), EMHAPFRE uses an expectation-maximization algorithm that determines the maximum-likelihood frequencies of multilocus haplotypes in diploid populations. Only individuals who were scored for all 14 SNPs were included in the data analysis.

Haplotype Assignment to Genotype Data

A short script written in Microsoft Excel Visual Basic and named “Assign” was used to assign genotypes to individual samples. The script was given, as input, the list of haplotypes produced by EMHAPFRE and the raw unphased genotype data. It produces a list of samples input, with a pair(s) of haplotypes that satisfies the genotype data assigned to each; in cases in which multiple pairs of haplotypes were listed, one pair is chosen, by use of a haplotype frequency–based method. A probability is calculated for each haplotype pair, by multiplication of the haplotypes’ frequencies in the control population. The haplotype pair with the highest probability is assigned to the individual.

Statistical Analysis for Recombination and LD

To test for recombination, we used the four-gamete test and the Hudson and Kaplan (1985) recombination statistic, R. For a given haplotype AB, mutation may result in either Ab or aB. Haplotype ab arises only in the case of either recombination or repeat mutation. The four-gamete test was executed on unphased genotype data, in a pairwise fashion, across all SNP loci. On the basis of the resulting matrix of the four-gamete test, R estimates the location and number of recombination events that have occurred in the sample.

Initial LD analysis was computed by performance of pairwise comparisons for all SNP loci. Fisher’s exact test was used to determine significance levels. SNPs having a minor-allele frequency of .05 were excluded from LD analyses. LD statistic D is a pairwise comparison of gametic frequencies such that D=p11p22–p12p21. D′, the relative disequilibrium, is D^′=D/|D|max, where |D|max=max(p1p2,q1q2) if D<0 and |D|max=min(q1p2,p1q2) if D>0. D′ ranges from 1 to −1, and this range is not influenced by allele frequency.

All recombination and LD statistics were generated by the software program DnaSP 3.00 (written by J. Rozas and R. Rozas, University of Barcelona).

Statistical Analysis for Association Study

Testing for significance in the model association study was done by use of contingency tables for independence. P values for significance of association at the haplotype level were determined by use of 2×2 tables and 3×3 tables for the genotype level. Significance values refer to a one-sided test.

Results

SNP Discovery

Our initial objective was to discover common neutral sequence variants spanning the length of the ATM gene. A gel-based resequencing strategy was employed to detect SNPs at the ATM locus. Genomic DNA of five unrelated individuals was amplified, by PCR, for [³³P]-radiolabeled sequencing. For detection of markers spanning the entire locus, PCR primers were designed for amplicons dispersed approximately evenly throughout the 184-kb genomic region containing the gene (fig. 1). Approximately 13.5 kb of the 184-kb total sequence was read in each individual. The nucleotide diversity, π, calculated for this sequence data was .00057. Seventeen SNPs were found, which span 142 kb and all of which are located in introns (table 1). This yielded an average of 1 SNP/794 nucleotides sequenced.

Figure 1.

Schematic of ATM. The 184 kb of the ATM locus is illustrated, with the 64 exons represented by black boxes. Twenty-nine ∼500-bp regions were amplified by PCR in five unrelated individuals. These regions were sequenced and found to contain 17 SNPs.

Table 1.

Seventeen ATM Noncoding SNPs Detected by Resequencing

SNPa	Location in Genomic Sequence with GenBank Accession Number U82828
Prior to 5′UTR t→ab	10182
IVS8−356t→c	34293
IVS19−1276a→g	57469
IVS21−77t→c	60136
IVS26+491c→gc	71049
IVS27−193c→tc	75083
IVS34+754g→a	85811
IVS46−257a→c	112721
IVS55+186c→t	121819
IVS57+3570t→c	127195
IVS58+997g→a	132032
IVS59+414g→tc	133986
IVS61−55t→c	142611
IVS62+60g→a	142789
IVS62+424g→a	143153
IVS62−973a→c	151964
IVS62−694c→a	152243

^aNomenclature is according to the guidelines recorded by the Ad Hoc Committee on Mutation Nomenclature (1996).

^bThis SNP is named in reference to the genomic sequence having GenBank accession number U82828 because of the highly variable nature of the 5′UTR.

^cNot used in genotyping or haplotype analysis.

Genotyping and Haplotype Development

To begin construction of haplotypes from these SNPs, we genotyped nine three-generation CEPH families (Dausset et al. 1990). By using three-generation families, we could determine haplotypes from genotype data, through inference. This allowed us both to determine the efficacy of the computer algorithm used to predict haplotypes (see below) and to optimize our genotyping assay. We performed ASO hybridization on nine CEPH families (89 individuals; 70 chromosomes), for 14 of the original 17 SNPs. These 14 SNPs were chosen from the original 17 because they performed consistently well under standard ASO-hybridization conditions.

We then used two different methods for deciphering the haplotypes derived from the genotype data, in a side-by-side comparison. First, haplotypes were inferred by hand. We began with homozygotes and predicted other haplotypes on the basis of transmission and by establishing the phase through the pedigrees. Seven haplotypes were identified in the sample of CEPH families. Subsequently, we subjected the same data set to an expectation-maximization algorithm, to estimate haplotypes and their frequencies. The computer program EMHAPFRE is a maximum-likelihood program developed to predict multilocus haplotypes from unphased genotype data (Excoffier and Slatkin 1995). It produces both a list of haplotypes and their estimated frequencies in the input sample population. The haplotype predictions from EMHAPFRE were in complete accordance with those that had been inferred manually, giving us confidence that this program was suitable for data of this nature.

Haplotype and Allele Frequencies

To determine frequencies of haplotypes and of individual SNPs in different ethnic populations, we performed ASO hybridization on anonymous African American (n=71), Asian American (n=39), white European American (n=77), and Hispanic (n=73) DNA samples collected in central and southeastern Texas. Genotype data were analyzed by the EMHAPFRE program. For the total population, 22 haplotypes and their frequencies were predicted by EMHAPFRE (table 2). Three predominant haplotypes were found at frequencies ⩾10%. An independent study that examined neutral sequence variants at the ATM locus also found three major haplotypes (Li et al. 1999).

Table 2.

ATM Haplotypes of 295 Humans from Five Ethnic Groups and of Three Species of Great Apes

		Frequency in Humansa
Haplotype	Sequence	Overall (n=295)	African American (n=71)	Asian American (n=39)	White European American (n=77)	Hispanic American (n=73)	CEPH (n=35)
1	ACTCTACTTCCCTC	.002				.007
2b	ACTCTACTTCTTTC	.313	.190	.500	.292	.315	.394
3	ACTCTCCTTCTTTC	.037			.065	.048	.061
4	ACTTCACTCCTCTC	.002	.007
5	ACTTTACTCTCCTCc	.002
6b	ACTTTACTTCCCTC	.066	.218	.013	.013	.027	.015
7	ACTTTACTTCTTTC	.019	.077
8	ATTCTACTTCTTTC	.012	.007	.051		.007
9	ATTCTCCTTCTTTC	.000		.013
10	ATTTCACTCCCCTC	.002	.007
11	ATTTCATCCTCCCC	.002		.013
12	TCTCTACTTCTTTC	.007				.021	.015
13	TCTTCACTCTCCTC	.010	.035			.007
14	TCTTCATCCTCCCC	.002				.007
15b	TTCTCACTCTCCTA	.090	.028	.013	.175	.041	.227
16	TTTCTATCCTCCCC	.005		.017		.007
17b	TTTTCACCCTCCTC	.100	.141	.068	.097	.110	.015
18	TTTTCACCCTCTTC	.002				.007
19	TTTTCACTCCTTTC	.002	.007
20	TTTTCACTCTCCTA	.002				.007
21b	TTTTCACTCTCCTC	.048	.162	.013	.006	.027
22b	TTTTCATCCTCCCC	.277	.113	.291	.351	.363	.273
	TTTCTACCCTCCTCc	…		.009
	ACTTTACCCTCCTCc	…	.007
Total		1.000	1.000	1.000	1.000	1.000	1.000
			Frequency in Great Apesd
			Chimpanzee (n=2)	Bonobo (n=1)	Gorilla (n=3)
1	TCTTTACTCTCCTC		.750	1.000	.000
2	TCTTTACTCTCTTC		.250	.000	.000
3	TATTTACTCTCCTC		.000	.000	1.000

^aSamples were genotyped by ASO hybridization, then haplotypes and their frequencies were estimated from unphased genotype data, by the EM algorithm EMHAPFRE.

^bHaplotype present in all four ethnic groups studied.

^cLow-frequency haplotypes in which some differences were seen in the combined data set and in individual ethnic populations.

^dSamples were genotyped by ASO hybridization and fluorescent sequencing.

The majority of SNPs identified in this study have a frequency, in all ethnic groups, of ⩾25% (fig. 2). Of the 14 SNPs, 3 (IVS19−1276a→g, IVS46−257a→c, and IVS62−694c→a) have a minor-allele frequency of <10% in most ethnic groups. SNP frequencies vary across ethnic groups. Three SNPs (IVS55+186c→t, IVS62+424g→a, and IVS62−973a→c) have a frequency of 11% in African Americans while being present at a frequency of >30% in all other ethnic groups. SNP IVS46−257a→c was not found in the samples from African Americans. Of the three low-frequency SNPs, two (IVS19−1276a→g and IVS62−694c→a) have a frequency of >18% in the white European American population and of <6% in the others. This is not surprising, given that the original five samples used for SNP detection were white European Americans.

Figure 2.

ATM SNP allele frequencies for 14 ATM SNPs in each of four ethnic groups. A total of 260 individuals (71 African American, 39 Asian American, 77 white European American, and 73 Hispanic) were genotyped by ASO hybridization.

To begin to describe the haplotype phylogeny at the ATM locus, we wanted to determine what haplotypes were present in each ethnic population. The genotype data were analyzed, by EMHAPFRE, as four separate data sets segregated by ethnic group. However, this analysis led to small discrepancies from what was predicted from the complete data set. In each case, changes were found in the lowest-frequency haplotypes (table 2). The efficacy of EMHAPFRE is known to decay as data sets decrease in size (Excoffier and Slatkin 1995). Thus, a second approach to ascription of haplotypes and their frequencies to each ethnic group was taken. To this end, a simple script was written in Microsoft Excel Visual Basic. This script, named “Assign,” takes a list of haplotypes and a data set of unresolved genotypes and then assigns to each individual sample one or more pairs of haplotypes that can resolve its genotype data; Assign lists every pair of haplotypes that can resolve an individual’s genotype data. We input each ethnic group’s data set individually with the 22 haplotypes. In this way we were able to determine which of the haplotypes suggested by EMHAPFRE were necessary for resolution of our genotype data, thus further refining the results. The genotype of every sample in this study could be accounted for by at least one pair of the 22 haplotypes predicted by EMHAPFRE from the complete data set. Six of the 22 haplotypes exist in all ethnic populations, and 11 of them are unique to a single population and hereafter are referred to as “private” haplotypes (table 2); each of these 11 haplotypes has a frequency of <1%.

We analyzed primate DNA in order to approximate an ancestral ATM haplotype. Three haplotypes were found in 12 chromosomes (table 2). Two common chimpanzees, one bonobo, and three gorillas were genotyped by ASO hybridization and fluorescent sequencing; in cases in which ASO hybridization gave ambiguous results, fluorescent sequencing was used to confirm the genotype. None of the ape haplotypes was found among the 22 human haplotypes. One ape haplotype differs from a human haplotype by a single-base variant. This human haplotype is one of the least common (frequency .007) and occurs only in our African American study group. Only one of the human SNPs showed variation in the apes; the remainder were monomorphic. One common chimpanzee was heterozygous for IVS62+424g→a. The gorillas shared all but one allele with the chimpanzees. At IVS8−356t→c, gorillas are homozygous for a third allele (A), which is not found in either humans or chimpanzees.

Intragenic Recombination and LD

The small number of haplotypes seen in our study population suggests the possibility that recombination is reduced at the ATM locus. This is further evidenced by the results of the four-gamete test (fig. 3) (Hudson and Kaplan 1985). For a given haplotype AB, mutation may result in either Ab or aB. Haplotype ab arises only in the case of either recombination or repeat mutation. For the purpose of this analysis, we will consider repeat mutation to be rare and will use the four-gamete test as a measure of recombination. The four-gamete test was executed on unphased genotype data, in a pairwise fashion across SNP loci. This was done for each ethnic group separately. Interestingly, the four-gamete test found no site pairs with four gametes in the samples from white European Americans, implying a complete lack of recombination in that population. Low recombination was indicated for the other groups, as shown in figure 3.

Figure 3.

Four-gamete test for recombination in ATM. White boxes denote site pairs having four gametic types, which implies that recombination has occurred between these two sites. Also shown is the Hudson and Kaplan recombination statistic R, which is an estimate of the number and sites of recombination events needed to explain the results of the four-gamete matrix. A white box containing an “×” denotes a potential site of recombination. The asterisk (*) denotes an SNP that is not polymorphic in the sample population.

Another test for recombination is that of Hudson and Kaplan (1985). Based on the resulting matrix of the four-gamete test, the Hudson and Kaplan parameter R is an estimate of the minimum number of recombination events in the history of the sample. For the white European American population, this estimate is 0 (fig. 3). For the other ethnic groups, R ranges from 4, in Hispanics, to 2, in Asian Americans. The predicted sites of recombination are similar among ethnic groups. African Americans and Hispanics share two possible recombination sites in the 5′ end of the gene, and a third, in the 3′ end, could also be in the same location. The Asian American population shares one of the 5′ end sites and has, in the middle of the gene, another potential site of recombination, which is also present in Hispanics.

Further support for the hypothesis that there is minimal recombination at the ATM locus is provided by the results of Fisher’s exact test (Weir 1996). We computed all possible pairwise comparisons between sites, to determine the degree of nonrandom association between sites. The majority of site pairs across all data sets show significance (P<.001), indicating that there is extensive disequilibrium at this locus (fig. 4). It has been demonstrated that alleles with frequencies ⩽.05 do not have the power for detection of disequilibrium (Lewontin 1995; Goddard et al. 2000). In this analysis, we included only SNPs having an allele frequency >.05. The Hispanic and Asian American populations were in complete disequilibrium. In the white European American population, the pattern of equilibrium followed the SNP with the lowest-frequency (.06) allele.

Figure 4.

Fisher's exact test for LD in ATM. White boxes denote site pairs that do not have a significant value by Fisher’s exact test, indicating linkage equilibrium. Gray columns and rows denote SNPs that have a minor-allele frequency ⩽.05. The asterisk (*) denotes an SNP that is not polymorphic in the sample population.

Disequilibrium was next measured by use of the statistic D′ , in a pairwise fashion across the 14 SNP loci (fig. 5). D^′=D/|D|max, where D=p11-p1p2 and |D|max=max(p1p2,q1q2) if D<0 and |D|max=min(q1p2,p1q2) if D>0. D′ ranges from 1 to −1, and this range is not influenced by allele frequency. A score of either 1 or −1 is considered to represent perfect disequilibrium. Interestingly, the results of this test are virtually superimposable on the results of the four-gamete test. The majority of site pairs are in perfect disequilibrium. The white European American population is in perfect disequilibrium across all sites. For the other groups, the sites with |D^′|<1 are exactly the same sites that have four gametes. We conclude that the ATM locus exhibits reduced recombination and extensive disequilibrium in all four ethnic groups, with the white European American population being the most extreme case.

Figure 5.

D′, measured as D^′=D/|D|max in a pairwise fashion across 14 SNP loci. A score of either 1 or −1 is considered perfect disequilibrium. Black boxes denote site pairs with perfect disequilibrium; white boxes denote site pairs with |D^′|<1. The asterisk (*) pair denotes an SNP that is not polymorphic in the sample population.

Association Study

Ultimately, we aim to use these ATM haplotypes for association studies in populations with cancer. To evaluate the potential that these haplotypes have for identification of a particular mutation or polymorphism, we performed a model association study. We tested the ability of these haplotypes to detect, by association, three different cSNPs in the ATM gene. These cSNPs were found by sequencing the reverse transcriptase–PCR products from ATM mRNA isolated from peripheral blood lymphocytes from cancer patients. cSNP1 is located in exon 4 and results in Ser49Cys. Positioned in exon 38, cSNP2 results in Asp1853Asn; and cSNP3, which results in Pro1054Arg, is located in exon 23. A population of 941 individuals was screened for these three cSNPs, by ASO hybridization. The resulting frequencies of the cSNPs in this population are shown in table 3.

Table 3.

Frequencies for Three ATM cSNPs in the Control Population

	Frequency in 941 Individuals
	White European Americans	African Americans	Asian Americans	Hispanic Americans
cSNP1	.005	.000	.001	.001
cSNP2	.066	.001	.002	.017
cSNP3	.015	.001	.000	.005

In the model association study, samples from white European Americans in the 941-individual collection that were found to possess one of the three cSNPs were considered to be the “case” population. The “control” population consisted of samples from white European Americans from the same collection that were randomly chosen and negative for the cSNPs; because of the low frequency of these cSNPs in other ethnic groups, only samples from white European Americans were used in this association study. All case and control samples were genotyped for the 14 ATM neutral sequence variants, via ASO hybridization. To assign haplotypes to individual samples, we used Assign and the initial 22 ATM haplotypes.

Each cSNP showed a significant association with a different, specific ATM haplotype (table 4). cSNP1 showed an association with haplotype 2, cSNP2 with haplotype 15, and cSNP3 with haplotype 17. Haplotype 2 was present at a frequency of .29 in the control population (no. of chromosomes [c] = 152) and at a frequency of .64 in the cSNP1 population (c=14); haplotype 15 was present at a frequency of .07 in the control population (c=112) and at a frequency of .57 in the cSNP2 population (c=56); and haplotype 17 was present at a frequency of .08 in the control population (c=146) and at a frequency of .52 in the cSNP3 population (c=54). These are 2-fold, 8-fold, and 6.5-fold increases in the frequencies of haplotypes 2, 15, and 17, respectively; and the P values for these associations are .0166, .0000, and .0000, respectively. Genotype correlations were also present, with P values of .0478, .0000, and .0000, respectively (table 4).

Table 4.

Association of ATM Haplotypes and ATM cSNPs in Individual “Case” and Control Populations^[Note]

		Frequency ina		P for
	Haplotype	Control Population	“Case” Population	Genotype Associationb	Haplotype Associationc
cSNP1	2	.29 (c=152)	.64 (c=14)	.0478	.0166
cSNP2	15	.07 (c=112)	.57 (c=56)	.0000	.0000
cSNP3	17	.08 (c=146)	.52 (c=54)	.0000	.0000

Note.— Samples carrying one of three ATM cSNPs were genotyped, by ASO hybridization, for the 14 ATM noncoding SNPs.

^aEach cSNP was found to occur on separate ATM haplotypes. c = total number of white European American chromosomes genotyped.

^bBy 2×2 contingency table.

^cBy 3×3 contingency table.

One of the great challenges in studying the genetics of a complex disease such as cancer is its multifactorial etiology. As presented in table 4, the data from our simulated association study model a scenario in which all “cases” are caused by a single mutation. To more accurately simulate an association study with a complex disease, we reanalyzed our data. We considered the three groups of samples carrying the variant cSNPs as one “case” population. In this analysis, two of the three haplotypes that originally had shown an association demonstrated a significant increase in frequency (table 5). No increase in frequency was apparent for haplotype 2, which had previously shown a twofold increase in the cSNP1 population; haplotype 15 showed a fourfold increase (P=.0002); and haplotype 17 showed a threefold increase (P=.0002). Thus, we successfully demonstrated the ability of these ATM haplotypes to discern members of our case population who carry a particular SNP. The results of these studies indicate a significant potential for the use of haplotypes extending over a large genomic region, to detect disease associations through a case-control–study design in a general population.

Table 5.

Association of ATM Haplotypes and ATM cSNPs in Combined Case Population^[Note]

		P for
Haplotype	Frequency in Combined Case Population (c=124)	Genotype Association	Haplotype Association
2	.19	.9644	.7606
15	.27	.0000	.0002
17	.24	.0000	.0002

Note.— See footnotes to table 4.

Discussion

In this study, we have presented a strategy for uncovering the genetic contribution to complex disease. Specifically, we have demonstrated the utility of a complex SNP-based–haplotype approach to association studies and have detected significant LD at the ATM locus, extending ∼142 kb. The results of this study provide proof of principle for the use of SNP-haplotype data in the detection of genetic factors contributing to complex disease.

We sequenced 13.5 kb of the ATM gene in five unrelated individuals and detected 17 SNPs in noncoding regions. We then utilized these neutral sequence variants spanning 142 kb of the ATM gene to construct haplotypes for this genomic locus. The expectation-maximization algorithm EMHAPFRE (Excoffier and Slatkin 1995) was used to predict haplotypes from genotype data on 295 individuals from four ethnic groups. Twenty-two haplotypes and their frequencies were predicted by EMHAPFRE, for the total population. Three of these 22 haplotypes have a frequency of ⩾10%. This concurs with the findings of Li et al. (1999), who also used neutral sequence variants to detect three major haplotypes at the ATM locus. Six of the 22 haplotypes exist in all four ethnic populations in our study and are also the most commonly occurring haplotypes. There are 11 private haplotypes, each of which has a frequency of <1%.

We verified the reliability of the haplotype-prediction algorithm by using several tests. First, we genotyped individuals from nine three-generation CEPH families (n=87). This allowed us to determine haplotypes by inspection of allele segregation. The CEPH genotype data were also analyzed by EMHAPFRE, and the resulting haplotypes agreed completely with those inferred on the basis of transmission data. Next, we used Assign, a script written in Microsoft Excel Visual Basic, to assign pairs of haplotypes to individual genotypes. Given the 22 haplotypes predicted by EMHAPFRE, Assign successfully resolved the genotype data for all 295 individuals in this study. The results of EMHAPFRE were tested against another haplotype-prediction program, one that does not use the expectation-maximization algorithm and that does not assume that Hardy-Weinberg is in effect. This program, termed “Data Mining,” uses the resulting matrix of the four-gamete test to inform the process of haplotype prediction so that recombination may influence outcome (N. Wang, R. Chakraborty, M. Kimmel, and L. Jin, personal communication). There were minor differences in the results of this comparison. For the population of white European Americans, the outcome of each program was identical. This is not surprising, since the four-gamete test reveals no evidence for recombination in this population. The results of these trials confirm that EMHAPFRE was successful in estimating the correct haplotypes necessary to sufficiently resolve our data set. We feel confident that the size and diversity of our data set has allowed us to describe in relative depth the haplotype architecture of ATM. Consequently, we have chosen to use, as the foundation for further studies, the 22 haplotypes predicted from the complete data set.

With a minimal amount of sequencing (13.5 kb in five individuals), we were able to detect highly informative neutral sequence variants spanning a large genomic region. In sequencing 10 chromosomes form white European Americans, we found SNPs that have a common occurrence in four different ethnic groups. In all ethnic groups, the majority (11 of 14) of SNPs identified in this study have a minor-allele frequency of ⩾25%. SNPs with frequencies in the range of .2–.5 have the highest information content for association and LD studies (Kruglyak 1997). Although most SNPs had a high minor-allele frequency in all ethnic groups, allele frequencies varied across ethnic groups. This is in accordance with several other studies that have found population differences in SNP-allele frequencies (Lai et al. 1998; Nickerson et al. 1998; Cargill et al. 1999; Halushka et al. 1999; Goddard et al. 2000). Variations in allele frequencies are most pronounced in the African American population. Four SNPs (IVS21−77t→c, IVS55+186c→t, IVS62+424g→a, and IVS62−973a→c) have a minor-allele frequency that is reduced by 40%–75% in African Americans, compared with that in other ethnic groups. A fifth SNP, IVS46−257a→c, was not found in the African American samples. These differences illustrate that there is population structure in SNP-allele frequencies that is an important factor to consider when SNP-based association and LD studies are designed.

Comparison of genotype data from six great apes was instructive for approximating ancestral haplotypes and SNP alleles. Genotyping revealed three haplotypes in this population, none of which is identical to the human ATM haplotypes. Of the 14 SNPs, 2 showed variation in the ape population. One common chimpanzee was heterozygous for IVS62+424g→a, and all three gorillas were homozygous for a third allele (A) at IVS8−356t→c. The extent of homozygosity in this sample indicates that most of the SNPs found varying in the human population have arisen since the divergence of the human lineage from the last common ancestor shared with the chimpanzee. This agrees with the assertion by Hacia et al. (1999)—that is, that most current neutral human polymorphisms are not shared with the chimpanzee (Hacia et al. 1999). It may also imply that these SNPs are not hypermutable sites, since more variation might be expected in the 12 primate chromosomes analyzed. Although these SNPs are common in man, they are not due to hypermutability; rather, they are old enough to be found throughout diverse ethnic groups.

The results of this study show a remarkable lack of recombination at the ATM locus. This effect is most profound in the white European American population, in which no evidence for recombination is detected by the four-gamete test and in which D′ shows perfect disequilibrium across all SNPs. Low recombination is implicated for the African American, Asian American, and Hispanic groups as well. The possibility of low recombination was suspected on the basis of the seemingly small number of haplotypes found at this locus. Twenty-two haplotypes with 14 loci is not considerably greater than the n+1 (i.e., 15) that would be expected if there is no recombination. Another study, performed in parallel with this one, used the same approach as that described here and serves as a direct comparison: D. Trikka, Z. Fang, A. Renwick, S. Jones, R. Chakraborty, M. Kimmel, and D. L. Nelson (unpublished data) used neutral sequence variants dispersed across the BLM, WRN, and RECQL loci, to derive haplotypes for these regions; their study used the same sample population, with fewer SNPs (8, 13, and 11 respectively) for haplotype construction, and found considerably larger numbers of haplotypes (50, 56, and 47, respectively) at each locus. The key difference between these loci and ATM is the amount of recombination and LD reported. Trikka et al. found more evidence for recombination and linkage equilibrium when the four-gamete test and Fisher’s exact test were used. For ATM, the four-gamete test revealed few site pairs with four gametes. The Hudson-Kaplan recombination statistic R ranged from 0, in white European Americans, to 4, in Hispanics. Analysis by both Fisher’s exact test and D′ indicated extensive LD for ATM, in all ethnic groups studied. Figure 5 shows extensive disequilibrium, with >72% of site pairs having perfect disequilibrium in all ethnic groups.

Using a model association study, we have successfully demonstrated the ability of ATM haplotypes to identify chromosomes carrying specific coding polymorphisms. The three cSNPs that we used as candidates for detection had varying frequencies in our control population of white European Americans (cSNP1, .005; cSNP2, .066; and cSNP3, .015). When each of the three cSNP populations was analyzed individually, each cSNP showed a significant association with a different ATM haplotype. cSNP1 showed an association with haplotype 2, cSNP2 with haplotype 15, and cSNP3 with haplotype 17 (P=.0166, .0000, and .0000, respectively); the increase in haplotype frequency in cases versus controls was 2-fold, 8-fold, and 6.5-fold, respectively. To model the potential for multiallelic etiology of a complex disease, we combined the three populations of samples carrying the cSNPs into one “case” population. In this analysis, two haplotypes demonstrated a readily detectable increase in frequency: haplotype 15 showed a fourfold increase, and haplotype 17 showed a threefold increase in frequency; no frequency increase was apparent for haplotype 2, which had previously shown a twofold increase in the cSNP1 population.

The association that becomes undetectable (i.e., haplotype 2 with cSNP1) involves the haplotype occurring most commonly (frequency .29) in the general population. Haplotype 15 shows the greatest increase in frequency and is the least common of the three haplotypes, with a control frequency of .05. This leads us to an important point for future association studies. Haplotypes with lower frequencies in control populations may be more effective for detection of associations. However, it is important to note that haplotype 17, which is the third most frequent haplotype (frequency .10), nevertheless showed a 2.6-fold increase in frequency in the combined cSNP population. An additional factor contributing to detection in this study is frequency of the mutation. In the case of cSNP1 and haplotype 2, in which the association becomes undetectable, the most frequent haplotype was associated with the least common SNP (cSNP1, .006). The difference in frequency between cSNP1 and the other cSNPs is a factor of 10. Both the haplotype frequency and the cSNP frequency contribute to detection. This underscores the idea that several factors, including frequency of haplotype, frequency of mutation, and age of mutation, contribute to limits of detectability.

This model association study demonstrates proof of principle for the use of complex SNP haplotypes covering candidate genes, in the detection of genetic factors contributing to complex disease. We have successfully demonstrated the ability of these ATM haplotypes to discern members of our “case” population that carry a particular coding SNP. The results of these studies indicate that haplotypes extending over a large genomic region have a significant potential for detection of disease associations.

There is much interest in the use of SNPs in genomewide association studies and other LD-based strategies. Our approach and analyses bear on those strategies, in several regards. First, LD estimates from simulation studies have been as low as 3 kb of meaningful LD (Kruglyak 1999). This calculation suggests that a very-high-density map with as many as 0.5–3 million SNPs would be necessary for effective association studies (Kruglyak 1999). Our results and those of other studies (Collins et al. 1999; Eaves et al. 2000; Moffatt et al. 2000; Taillon-Miller et al. 2000) indicate, to the contrary, that significant LD can be found extending as far as several hundred kilobases. This should reduce the number of SNPs necessary for genomewide linkage studies. Comparison of LD at ATM versus the results of the LPL study (Clark et al. 1998) in which LD patterns were complex over just 9.7 kb supports the idea that LD varies widely throughout the genome, indicating that some regions will require SNPs that are more densely spaced. Second, higher-frequency (.2–.5) SNPs are more robust, whereas rare SNPs may be less useful and, in some analyses, may confound results. More than half of the SNPs used to construct haplotypes in the LPL study had a relative allele frequency of <.2. This resulted in 67 of 71 individuals having a unique haplotype. By using fewer markers (14) with higher frequency (.20), we were able to effectively elucidate the haplotype architecture and the LD and recombination profiles for the ATM genomic locus (142 kb). These haplotypes were used successfully in association studies, to detect coding polymorphisms in the ATM gene. We conclude that reasonably spaced, highly informative SNPs have the ability to define a larger number of ancestral chromosomes and have increased power for population-based association studies.

Appendix A: Primers Used for PCR and Sequencing

f1.atm, ATGGTCATCTCGTTACAGGCAATGC

r1.atm, CCCCAAGTGACTGAAGGCATCTAGG

f2.atm, TGGTGGAACCATTTCCGTTTAACG

r2.atm, GCGCCCTTCTAATAACCCGCC

f3.atm, GCCCAGAACCTCCGAATGACG

r3.atm, CGACTTAGCGTTTGCGGCTCG

f4.atm, TGGCTGGCAACATTACCAACTGC

r4.atm, TGCATCTTTTTCTGCCTGGAGGC

f5.atm, TGTGTGCTAGGGAGGAATCTGGTGG

r5.atm, GGCTGTCTCTAGGCTTGTTGAGGGC

f6.atm, CCATCATCCGAAAGGAGCCAAAAC

r6.atm, GCAGCAATTTCCCTGTTTCTGCC

f7.atm, AAATTGGCAGGATGATGAGGATGC

r7.atm, GCTGTCAAGCTGCATCAGCGTTAG

f8.atm, CCAAAGCGTGCCAGAATGGTATG

r8.atm, CCAAAGCGTGCCAGAATGGTATG

f9.atm, GGTATGCGTAGCGGGGCTAGTGAG

r9.atm, CGCAGGAAAAAGCCAGATGCAATC

f10.atm, GCCCTAGCCCCAGTGTATGTGGAG

r10.atm, GGCAGCCAGTTTCCGAGAACTACC

f11.atm, TTTTTGGCAAGGTGAGTATGTTGGC

r11.atm, TGCGAACTTGGTGATGATTGTCAGC

f12.atm, AGATTGTTCCAGGACACGAAGGGAG

r12.atm, TTTCTTCCCATTGTCACCTGTTCCC

f13.atm, TGCGAAAAACAGGCTTTGTTTGC

r13.atm, GGTGATGGAAAAGAGACGGGGC

f14.atm, GCAAGTCCCCTCACCAGCAACAC

r14.atm, GATGCCTTCCCATCATCCTGATACC

f15.atm, TCTGGGAAGAAGTTACGCAGGGAAC

r15.atm, CTGACTGGCACTAGAATTTGCTGGC

f16.atm, GGCGGAATGAATGTGAGTTATGCG

r16.atm, CCAGGTGATTTCTCCATCCCGTG

f17.atm, CTGCCTAAAGCAGCAGTTTTTGCC

r17.atm, TGTTGCTATCCCGAAGCTGAAACC

f18.atm, GGTGTGTAAGCAAGAATGCCTGGG

r18.atm, GCCACAGATTTTGAGACCACTGCAC

f19.atm, TAGTTTGTATGGCTGTGGTGGAGGG

r19.atm, CATCCCTCTGCTTCAGGAGTATCCC

f20.atm, CCAGTAGGGGGTCCCTCATTTCC

r20.atm, TGAGAAGCTGGGAGTGTTTCTGCC

f21.atm, CCCCGTACATGAAGGGCAGTTG

r21.atm, TGGGTGGCTGGGCTAATGAAGAG

f22.atm, GGTTCAGCGAGAGCTGGAGTTGG

r22.atm, GCAGCAGGGGGAAAACCCAC

f23.atm, CCACAGATTAGCAACAAGTTGGGGC

r23.atm, TGGCATAAGCACACGGAAACTCTCC

f24.atm, AGGTTCCGATGGCAAGGAGAGG

r24.atm, CTGTGTCTTTCCACCACTCCCCAG

f25.atm, CAGTCATGGTTCTGGGGAGAGAAGC

r25.atm, GCCTTTCTGATTTCCCTTCCTGCC

f26.atm, CTTGATGGTGGGAGGGACTTAGGG

r26.atm, TGCCTAGATGTTTGAGAGCCTGCC

f27.atm, CAGGGCACACAGGGTACAGTGTAGG

r27.atm, TCAGTTCAGACCATCTCATGCCTCC

f28.atm, CAGGGGGATGATAGTGATGATGTGG

r28.atm, TTCAAAACATACATGCCCTGCCTTC

f29.atm, CAAAGACTGAGAGCTGAGCCCAGTG

r29.atm, GCACAATCTCCTCCTTTCTGCTGC

f30.atm, TGGTTTAGAAATGCCTTCAGCCCC

r30.atm, TGCACTCTACCTGCCATGCTTCC

f31.atm, GCCATGTCAGTGCCCAACTTGAAG

r31.atm, TTGGTGCTGCGTTTGGAATCTTG

f32.atm, GATTCCAAACGCAGCACCAAACC

r32.atm, GGTAGTTGATGGGGGAGGGGAAC

f33.atm, GTTCCCCTCCCCCATCAACTACC

r33.atm, GAGCACAGTGCCTTCTTCCACTCC

f34.atm, CCCTGACAATCTGGGGCACAAAC

r34.atm, CCGTGGCTTTTGCTGGCATTC

f35.atm, GTCCTGTGGCATTGTGCATAACTCC

r35.atm, GCAGACATTAGGCATAAGCCCCTTC

f36.atm, GATGACTGCCCTTGTTCCCCAAG

r36.atm, TGGTTAAGTTGCTTTTCCCCCCAG

Appendix B: Primer Sequences and Concentrations Used for Multiplex PCR

Group 8:

3F ATM, 5′-GCCCAGAACCTCCGAATGACG-3′; and 3R-2 ATM, 5′-GCCGTGAAGCGAAAGAGGCG-3′ (0.25 μM)

11F ATM,5′-TTTTTGGCAAGGTGAGTATGTTGGC-3′; and 11R ATM, 5′-TGCGAACTTGGTGATGATTGTCAGC-3′ (0.25 μM)

14F ATM, 5′-GCAAGTCCCCTCACCAGCAACAC-3′; and 14R ATM, 5′-GATGCCTTCCCATCATCCTGATACC-3′ (0.25 μM)

23F-2 ATM, 5′-GGTGGAATCTGGTCTAGTTACCC-3′; and 23R ATM, 5′-TGGCATAAGCACACGGAAACTCTCC-3′ (0.25 μM)

27F ATM, 5′-CAGGGCACACAGGGTACAGTGTAGG-3′; and 27R ATM, 5′-TCAGTTCAGACCATCTCATGCCTCC-3′ (0.188 μM)

29R ATM, 5′-CAAAGACTGAGAGCTGAGCCCAGTG-3′; and 29R ATM, 5′-GCACAATCTCCTCCTTTCTGCTGC-3′ (0.125 μM)

30F ATM, 5′-TGGTTTAGAAATGCCTTCAGCCCC-3′; and 30R-2 ATM, 5′-CAGCCAGTCCAACATAAATCAG-3′ (0.25 μM)

31F ATM, 5′-GCCATGTCAGTGCCCAACTTGAAG-3′; and 31R ATM, 5′-TTGGTGCTGCGTTTGGAATCTTG-3′ (0.25 μM)

Group 7:

7R ATM, 5′-GCTGTCAAGCTGCATCAGCGTTAG-3′; and 7F-2 ATM, 5′-GTTGGATTACCATGTTCACCAG-3′ (.188 μM)

10F ATM, 5′-GCCCTAGCCCCAGTGTATGTGGAG-3′and 10R-2 ATM, 5′-GCAGAGATAATCATGGGCAGG-3′ (0.25 μM)

15F ATM, 5′-TCTGGGAAGAAGTTACGCAGGGAAC-3′; and 15R-2 ATM, 5′-TGGGGAGACTATGGTAAAAGAGG-3′ (0.31 μM)

16F ATM, 5′-GGCGGAATGAATGTGAGTTATGCG-3′; and 16R ATM, 5′-CCAGGTGATTTCTCCATCCCGTG-3′ (0.25 μM)

20F ATM, 5′-CCAGTAGGGGGTCCCTCATTTCC-3′; and 20R ATM, 5′-TGAGAAGCTGGGAGTGTTTCTGCC-3′ (0.25 μM)

25F ATM, 5′-CAGTCATGGTTCTGGGGAGAGAAGC-3′; and 25R-2 ATM, 5′-CTATCAATATCTAGCTCTGGGGC-3′ (0.15 μM)

28F ATM, 5′-CAGGGGGATGATAGTGATGATGTGG-3′; and 28R ATM, 5′-TTCAAAACATACATGCCCTGCCTTC-3′ (0.5 μM)

Appendix C: Probes Used for ASO Hybridization

ATMAso 3T, 5′-TAACCCTCCTTCCCGC-3′

ATMAso 3a, 5′-TAACCCTCCATCCCGC-3′

ATMAso 7T, 5′-AAGGAACTTGTAATATTTTTC-3′

ATMAso 7c, 5′-AGGAACTCGTAATATTTTTC-3′

ATMAso 10T, 5′-TGGGAAACATGACCAGGG-3′

ATMAso 10c, 5′-GGGAAACACGACCAGGG-3′

ATMAso 11T, 5′-GTAACTTATAATAACCTTTC-3′

ATMAso 11c, 5′-GAAGTAACTTACAATAACC-3′

ATMAso 14C, 5′-TCTGTACAAGAAAAATTTG -3′

ATMAso 14g, 5′-TCTGTAGAAGAAAAATTTG-3′

ATMAso 15C, 5′-TTTCCTCTCAGTCTACAGG-3′

ATMAso 15t, 5′-TTTTTCCTCTTAGTCTACAGG-3′

ATMAso 16C, 5′-TAGAGATGATGTCGGCTTC-3′

ATMAso 16t, 5′-CTAGAGATGATGTTGGCTTC-3′

ATMAso 20A, 5′-GTAATGTCAGAGTATTAAA-3′

ATMAso 20c, 5′-TAATGTCAGCGTATTAAA-3′

ATMAso 23T, 5′-CAAAAGCTTCTCTTGCTTT-3′

ATMAso 23c, 5′-AAAAGCTTCTCCTGCTTTC-3′

ATMAso 25C, 5′-TTTTTTGTGGCATTCACAC-3′

ATMAso 25t, 5′-TTTTTTGTGGTATTCACAC-3′

ATMAso 27C, 5′-CTGCTCATGCCTCCTCTC-3′

ATMAso 27t, 5′-CTGCTCATGTCTCCTCTCC-3′

ATMAso 28C, 5′-TTCTATTAAACAGTATTA-3′

ATMAso 28a, 5′-TTCTATTAAAAAGTATTA-3′

ATMAso 29.1T, 5′-GATAAAGATATGTTGACAA-3′

ATMAso 29.1c, 5′-GATAAAGATACGTTGACAA-3′

ATMAso 29.2C, 5′-ACTTCCTGACGAGATACAC-3′

ATMAso 29.2t, 5′-ACTTCCTGATGAGATACAC-3′

ATMAso 30c, 5′-CCTAAGCCACGTTCCTCTA-3′

ATMAso 30t, 5′-CCTAAGCCATGTTCCTCTA-3′

ATMAso 31.1C, 5′-AAATAGAGCGATTTTGGTT-3′

ATMAso 31.1t, 5′-AAATAGAGAGATTTTGGTTC-3′

ATMAso 31.2C, 5′-AGAAATTCCTCATGAACTC-3′

ATMAso 31.2a, 5′-AGAAATTCATCATGAACTC-3′