 Research
 Open Access
 Published:
Modeling genetic imprinting effects of DNA sequences with multilocus polymorphism data
Algorithms for Molecular Biology volume 4, Article number: 11 (2009)
Abstract
Single nucleotide polymorphisms (SNPs) represent the most widespread type of DNA sequence variation in the human genome and they have recently emerged as valuable genetic markers for revealing the genetic architecture of complex traits in terms of nucleotide combination and sequence. Here, we extend an algorithmic model for the haplotype analysis of SNPs to estimate the effects of genetic imprinting expressed at the DNA sequence level. The model provides a general procedure for identifying the number and types of optimal DNA sequence variants that are expressed differently due to their parental origin. The model is used to analyze a genetic data set collected from a pain genetics project. We find that DNA haplotype GAC from three SNPs, OPRKG36T (with two alleles G and T), OPRKA843G (with alleles A and G), and OPRKC846T (with alleles C and T), at the kappaopioid receptor, triggers a significant effect on pain sensitivity, but with expression significantly depending on the parent from which it is inherited (p = 0.008). With a tremendous advance in SNP identification and automated screening, the model founded on haplotype discovery and statistical inference may provide a useful tool for genetic analysis of any quantitative trait with complex inheritance.
Background
In diploid organisms, there are two copies at every autosomal gene, one inherited from the maternal parent and the other from the paternal parent. Both copies are expressed to affect a trait for a majority of these genes. Yet, there is also a small subset of genes for which one copy from a particular parent is turned off. These genes, whose expression depends on the parent of origin due to the epigenetic or imprinted mark of one copy in either the egg or the sperm, have been thought to play an important role in complex diseases and traits, although imprinted expression can also vary between tissues, developmental stages, and species [1]. Anomalies derived from imprinted genes are often manifested as developmental and neurological disorders during early development and as cancer later in life [2–5].
The mechanisms for genetic imprinting are still incompletely known, but they involve epigenetic modifications that are erased and then reset during the formation of eggs and sperm. Recent research shows that the parentoforigin dependent expression of imprinted genes is related with environmental interactions with the genome [5]. The phenomenon of genomic imprinting is explained from an evolutionary perspective [6]. Genomic imprinting evolves in mammals with the advent of live birth through a parental battle between the sexes to control the maternal expenditure of resources to the offspring [7].
Paternally expressed imprinted genes tend to promote offspring growth by extracting nutrients from the mother during pregnancy while it is suppressed by those genes that are maternally expressed. This genetic battle between the paternal and maternal parents appears to continue even after birth [8, 9].
The genetic mechanisms for imprinting can be made clear if the genomic distribution of imprinted genes and their actions and interactions are studied. Genetic mapping with molecular markers and linkage maps has been used to map quantitative trait loci (QTLs) that show parentoforigin effects [10–12]. Using an outbred strategy appropriate for plants and animals, significant imprinting QTLs were detected for body composition and body weight in pigs [13, 14], chickens [15] and sheep [16]. Cui et al. [12] proposed an F_{2}based strategy to map imprinting QTLs by capitalizing on the difference in the recombination fraction between different sexes. More explorations on the development of imprinting models are given in Cui and others [12, 17]. Liu et al. [18] developed a randomeffect model for estimating the parentdependent genetic variance of complex traits at imprinting QTLs.
We will propose a statistical model for estimating the imprinting effects of DNA sequence variants that encode a complex trait. This model uses widely available single nucleotide polymorphisms (SNPs) that reside within a coding sequence of the human genome. The central idea of this model is to separate maternally and paternallyderived haplotypes (i.e., a linear combination of alleles at different SNPs on a single chromosome) from observed genotypes. By specifying one risk haplotype, i.e., one that operates differently from the rest of haplotypes (called nonrisk haplotypes), Liu et al. [19] proposed a statistical method for detecting risk haplotypes for a complex trait with a random sample drawn from a natural population. Liu et al.'s approach can be used to characterize DNA sequence variants that encode the phenotypic value of a trait. Wu et al. [20] constructed a general multiallelic model in which any number of risk haplotypes can be assumed. The best number and combination of risk haplotypes can be estimated by using the likelihoods and AIC or BIC values. We will derive a computational algorithm for estimating the imprinting effects of SNPconstructed haplotypes with multilocus genetic data based on these previous workings. The new algorithmic model provides a general framework for hypothesis tests on the pattern of genetic imprinting expressed by haplotypes. A real example from a pain genetic study is used to demonstrate the application of the model.
Results
The new imprinting model was used to detect the difference of gene expression between the maternal and paternal parents at the haplotype level. Genetic and phenotypic data were from a pain genetics project in which 237 subjects (including 143 men and 94 women) from five different races were sampled. All these subjects were genotyped for three SNPs, OPRKG36T (rs1051160), with two alleles G and T, OPRKA843G (rs702764), with alleles A and G, and OPRKC846T (rs16918875), with alleles C and T, at the kappaopioid receptor [21]. Three traits of pain sensitivity to heat were tested with a procedure described by Fillingim [22], and they are the average pain rating for heat stimuli at 49°C (PreInt49tot), the increase in heat pain threshold following administration of 0.5 mg/kg of pentazocine (an mixed action opioid agonistantagonist with activity at the kappa receptor) (Hpthpent), and the increase in heat pain tolerance following administration of 0.5 mg/kg of pentazocine (an mixed action opioid agonistantagonist with activity at the kappa receptor) (Hptopent). PreInt49tot is a baseline pain measure before any drug administration. Although the model allows the estimation of any covariate effect, we will remove the effect on the pain traits due to different races because of too few samples for some races. Our final imprinting analysis was based on 237 subjects for PreInt49tot but on 85 subjects for Hpthpent and Hptopent. Sexspecific haplotype frequencies for the three SNPs were estimated for males and females, from which linkage disequilibria were estimated and tested with results given in Table 1. Of the eight haplotypes, haplotype GAC occupies an overwhelming proportion in both sexes (>75%). Some haplotypes, like GAT, TAT, and TGC, are very rare, with population frequencies tending to be zero. Linkage disequilibria between these SNPs are generally strong: those between OPRKA843G and OPRKC846T and between OPRKG36T and OPRKC846T are highly significant (p < 0.01), although that between OPRKG36T and OPRKA843G is much less significant. There is a highly significant highorder linkage disequilibrium among these three SNPs. It is interesting to see significant difference in haplotype distribution between the two sexes (Table 1). This sexspecific difference is due to the difference in linkage disequilibria because allele frequencies seem to be mostly sexinvariant (Table 1).
A "biallelic" model assuming that there is only one haplotype is used to estimate haplotype effects at the kappaopioid receptor on pain traits.
Significant haplotype effects were observed on the three pain sensitivity traits studied. By assuming a risk haplotype from all possible haplotypes, we calculated the resultant likelihoods of haplotype effects, which are given in Table 2.
It can be seen that an optimal risk haplotype is GAC for preint49tot and TAC for Hpthpent and Hptopent. Statistical tests indicate that these risk haplotypes trigger a significant effect on PreInt49tot (p = 0.004) and Hpthpent and Hptopent (p = 0.025), respectively (Table 3). Risk haplotype GAC displays a significant additive effect (p = 0.005) on preint49tot, but there is no dominant effect due to its interaction with nonrisk haplotypes. It is interesting to find that risk haplotype GAC contributes to variation in preint49tot in a parentoforigin manner. The subjects with risk haplotype GAC inherited from the maternal parent will be significantly different in preint49tot than those with the risk haplotype inherited from the paternal parent. No significant imprinting effects are detected on traits Hpthpent and Hptopent, although risk haplotype TAC displays significant additive and dominant effects on these two traits.
The statistical properties of the imprinting model are investigated through simulation studies. The first simulation mimics the data structure of the real example (with 237 subjects) above based on its estimates of sexspecific allele frequencies and linkage disequilibria among three SNPs (Table 1) and of the additive, dominant, and imprinting effects arising from different haplotypes (for PreInt49tot, Table 3). The second simulation includes sample size from 237 to 400, 800, and 2000, keeping the other parameters unchanged. Each simulation scheme is repeated for 1000 runs.
Results from simulation studies are summarized as follows:

(1)
Population genetic parameters including threeSNP haplotype frequencies, allele frequencies, and linkage disequilibria of different orders can be precisely estimated even when a smaller sample size (237) is used. As expected, the estimation precision can be improved consistently when the sample size increases from 237 to 2000.

(2)
Quantitative genetic parameters can also be well estimated, but the better estimation of the dominant and imprinting effects needs a larger sample size (400 or more). With a sample size of 2000, the precision of all parameter estimates are remarkably high, with sampling errors of each estimate being low than 10% of the estimate.

(3)
The power to detect imprinting effects reaches 75% for a sample size of 237, but it can increase dramatically when increasing the sample size to 400.

(4)
The type I error rate of detecting the imprinting effect, i.e, a probability for a false positive discovery of that effect, is quite low (< 10%) for a small sample size and can be lowered when sample size increases.
The simulation for testing the type I error rate in (4) was based on the same parameters as used in (1)–(3), except that no imprinting effect is assumed. Because we have derived a series of closed forms for the estimation of parameters within the EM framework, parameter estimation is very efficient. For a single simulation run, it will take a few dozen of seconds to obtain all estimates on a PC laptop. Also, estimates do not depend heavily on initial values of parameters, showing that the estimates achieve a global maxima for the likelihood.
Discussion
Although a traditional view assumes that the maternally and paternally derived alleles of each gene are expressed simultaneously at a similar level, there are many exceptions where alleles are expressed from only one of the two parental chromosomes [1, 23]. This socalled genetic imprinting or parentoforigin effect has been thought to play a pivotal role in regulating the phenotypic variation of a complex trait [24–27]. With the discovery of more imprinting genes involved in trait control through molecular and bioinformatics approaches, we will be in a position to elucidate the genetic architecture of quantitative variation for various organisms including humans. Genetic mapping in controlled crosses has opened up a great opportunity for a genomewide search of imprinting effects by identifying imprinted quantitative trait loci (iQTLs). This approach has successfully detected iQTLs that are responsible for body mass and diseases [10, 11, 14, 19, 28, 29]. Cloning of these iQTLs will require highresolution mapping of genes, which is hardly met for traditional linkage analysis based on the production of recombinants in experimental crosses. Single nucleotide polymorphisms (SNPs) are powerful markers that can explain interindividual differences. Multiple adjacent SNPs are especially useful to associate phenotypic variability with haplotypes [30–34]. The quantitative effect of haplotypes on a complex trait was modeled by Liu et al. [19] and subsequently refined by Wu et al. [20].
In this article, we incorporate genetic imprinting into Wu et al.'s [20] multiallelic model to estimate the number and combination of multiple functional haplotypes that are expressed differently depending on the parental origin of these haplotypes. Because of the modeling of any possible distinct haplotypes, the multiallelic model will have more power for detecting significant haplotypes and their imprinting effects than biallelic models. The imprinting model was shown to work well in a wide range of parameter space for a modest sample size. However, a considerably large sample size is needed if there are multiple risk haplotypes that contribute to trait variation. By analyzing a real example for pain genetics, the new model detects significant haplotypes composed of three SNPs within the kappaopioid receptor, which may play an important role in affecting pain sensitivity to heat.
Haplotype GAC derived from this gene appears to be imprinted for PreInt49tot, a pain sensitivity trait to heat stimuli at 49°C before drug administration, leading to different levels of pain sensitivity between the patients when they inherit this haplotype from maternal and paternal parents. In this example, no imprinting was detected for Hpthpent and Hptopent, two pain sensitivity traits measured after the patients were administrated with pentazocine. This result should be, however, explained with caution. First, the risk haplotype, TAC, detected for Hpthpent and Hptopent is a rare haplotype in the admixed population studied, although it is quite common in African Americans and Hispanics. The significance of this rare haplotype detected could be a sample size artifact, or it could be indicating a powerful haplotype effect. Second, in a different analysis, no significant genetic association was detected for the same heat pain test at heat stimuli at 52°C (data not shown). Nonetheless, the method provides a powerful tool for detecting possible associations and imprinting effects, which provide a starting point for future work to pursue the positive results with larger sample sizes and family studies. There have not been any previous reports suggesting an imprinting effect at an opioid receptor locus, or related to pain measures.
In practice, although the human genome contains millions of SNPs, it is not possible and also not necessary to model and analyze these SNPs simultaneously. These SNPs are often distributed in different haplotype blocks [35], within each of which a particular (small) number of representative SNPs or htSNPs can uniquely explain most of the haplotype variation. A minimal subset of htSNPs, identified by several computing algorithms, can be implemented into our imprinting model to detect their imprinting effects at the haplotype level. In addition, our model can be extended to model imprinting effects in a network of interactive architecture, including haplotypehaplotype interactions from different genomic regions, haplotypeenvironment interactions, and haplotype effects regulating pharmacodynamic reactions of drugs. It can be expected that all extensions will require expensive computation, but this computing can be made possible if combinatorial mathematics, graphical models, and machine learning are incorporated into closed forms of parameter estimation.
This imprinting model assumes that if the SNPs constituting haplotypes are tightly linked, haplotype frequencies estimated from the current generation can be used to approximate haplotype frequencies in the parental generation. To relax this assumption, a strategy of sampling a panel of random families from a population is required, in which a known family structure allows the tracing and estimation of maternally and paternallyderived haplotypes. Such a strategy will help to precisely estimate and test imprinting effects of haplotypes, providing a new gateway for studying the genetic architecture of complex traits.
Methods
Imprinting Model
Consider a set of three ordered SNPs, each with two alleles 1 and 0, genotyped from a candidate gene. These three SNPs form eight haplotypes, 111, 110, 101, 100, 011, 010, 001, and 000. A risk haplotype group is defined as a set of haplotypes that are in manner distinct from the other haplotypes in affecting a complex trait. For example, if a risk haplotype group only consists of the haplotype 111, the remaining seven haplotypes form the nonrisk haplotype group; this means that the diplotypes composed of 111 will have different genotypic values from those composed of 110, 101, 100, 011, 010, 001, or 000. There may be multiple risk haplotype groups, and we let R_{ k }denote any risk haplotype from the k^{th} risk haplotype group (k = 1,...,K; K < 8, where K is the number of risk haplotype groups) and R_{0} denote any of the remaining nonrisk haplotypes in the nonrisk haplotype group. The combinations between the risk and nonrisk haplotypes are called the composite diplotypes, including R_{ k }R_{ k' }(k ≤ k' = 1,...,K), R_{ k }R_{0} and R_{0}R_{0} (any two nonrisk haplotypes). Here we do not distinguish between R_{ k }R_{ k' }and R_{ k' }R_{ k }as we do not know parental origin of the haplotypes, however genetic imprinting implies that the same composite diplotype may function differently, depending on the parental origin of its underlying haplotypes. To reflect the parental origin of haplotypes in the composite diplotype, the following notation is used: R_{ k }R_{ k' }(k, k' = 1,....,K), R_{ k }R_{0}, R_{0}R_{ k }, and R_{0}R_{0}, where the vertical lines are used to separate the haplotypes derived from the maternal parent (left) and paternal parent (right).
According to traditional quantitative genetic principles, the genotypic value of a given composite diplotype is partitioned into different components due to additive and dominance genetic effects. For an imprinting model, an additional component is the imprinting genetic effect due to different contributions of haplotypes from the maternal and paternal parents. Mathematically, the genotypic value of a composite diplotype of known parental origins is expressed as
where μ is the overall mean, a_{ k }is the additive effect due to the substitution of risk haplotype k by the nonrisk haplotype, d_{ kk' }is the dominance effect due to the interaction between risk haplotypes k and k'(d_{ kk' }= d_{k'k}), d_{k 0}is the dominance effect due to the interaction between risk haplotypes k and the nonrisk haplotype (d_{k 0}= d_{0k}), i_{ kk' }is the imprinting effects due to different parental origins of risk haplotypes k and k'(i_{ kk' }= i_{ k'k }), and i_{k 0}is the imprinting effect due to different parental origins of risk haplotype k and the nonrisk haplotype (i_{k 0}= i_{0k}). The sizes and signs of i_{ kk' }and i_{k 0}determine the extent and direction of imprinting effects at the haplotype level.
A set of genetic parameters, including the additive, dominance, and imprinting effects, define the genetic architecture of a quantitative trait. By estimating and testing these parameters, the genetic architecture of a trait can well be studied. Specific genetic effects of haplotypes can be estimated from genotypic values of the composite diplotypes with the formulas as follows:
Estimating Model
Genetic Structure
Suppose the three SNPs are genotyped from a natural human population at HardyWeinberg equilibrium (HWE). Let and denote the frequencies of two alternative alleles r_{ l }(r_{ l }= 1 or 0) at SNP l in the population of sex s (s = M for females and P for males). Sexspecific frequencies of eight haplotypes produced by the three SNPs are generally expressed as . For each sex s, genotypes consisting of three SNPs are denoted as , totaling 27 distinct genotypes. Let denote the observation of a threeSNP genotype for sex s, which sums over the two sexes to . Some genotypes are consistent with diplotypes, whereas those that are heterozygous at two or more SNPs are not. Each double heterozygote contains two different diplotypes, and the triple heterozygote, i.e., 10/10/10, contains four different diplotypes: 111000 (with probability for sex s), 110001 (with probability for sex s), 101010 (with probability for sex s), and 100011 (with probability for sex s). Note that slashes are used to separate genotypes at different SNPs and vertical lines are used to separate haplotypes derived from the maternal parent (left) and paternal parent (right). The observed genotypes, the underlying diplotypes, and diplotype frequencies are provided in Additional file 1. From the HWE assumption, diplotype frequencies are simply expressed as the products of the underlyinghaplotype frequencies derived from different parents.
For a random sample from a natural population, it is impossible to estimate the frequencies of maternally and paternallyderived haplotypes. However, parentspecific haplotype frequencies can be approximated by sexspecific haplotype frequencies in the current generation if the SNPs studied are highly linked together. For example, we will argue below that which measures frequency for the haplotype 100 among females in the current generation can be accurately approximated by the haplotype frequency 100 in the "maternal generation" or previous generation of females. This is proven as follows. Let (t) and (t + 1) be the frequencies of a representative threeSNP haplotype r_{1}r_{2}r_{3} for a monosexual population in generations t and t + 1, respectively. The relationship of haplotype frequencies between the two generations is expressed as
where (t), (t), and (t) are the allele frequencies of three SNPs in generation t, and D_{12}(t), D_{13}(t), and (D_{23}(t) are the linkage disequilibria between the first and second SNPs, the first and third SNPs, and the second and third SNPs in generation t. Under HardyWeinberg equilibrium, the allele frequencies remain constant from generation to generation and the linkage disequilibria decay in proportion with the recombination fractions r_{ ij }in that
and
Because the three SNPs are genotyped from the same region of a candidate gene, their recombination fractions should be very small and can be thought to be close to zero. Thus, the frequencies of maternally and paternallyderived haplotypes can be approximated by the estimates of these haplotype frequencies in the female and male populations of current generation, respectively.
Likelihoods
With a random sample from a natural population, in which each genotyped subject is measured for a phenotypic trait of interest, we will develop a model to estimate population genetic parameters, including the eight maternallyderived haplotype frequencies , the eight paternallyderived haplotype frequencies , and the quantitative genetic parameters (Ω_{ q }) that include haplotype effects () and the residual variance of the trait (σ^{2}). The haplotype effects are derived uniquely from genotypic values of composite diplotypes () as provided in equations (2)–(6).
Given sexspecific genotypic observations (M_{ M }and M_{ P }) and phenotypic data (y), a joint loglikelihood is constructed as
Thus, maximizing the likelihood in (1) is equivalent to individually maximizing the three terms on the right hand side of (1).
A polynomial likelihood is constructed for the marker data of a sex (s) to estimate and . For notational convenience, we define a function h(r) on genotypes to be
So, for example, h(r) = 2 if r is a double heterozygote. The first two log likelihoods on the righthand side of equation (7) are then expressed as
A closed system of the EM algorithm was derived to estimate these haplotype frequencies (see the Text S1). The estimates of sexspecific haplotype frequencies are then used to estimate haplotype effects by constructing a mixture modelbased likelihood. The construction of this likelihood requires knowledge of which haplotypes are risk haplotype(s). By assuming that haplotype 111 is only one risk heplotype, we construct the likelihood as
and
The EM algorithm is derived to estimate quantitative genetic parameters with a detailed procedure given in Additional file 2. The estimated genotypic values of composite diplotypes are used to estimate the additive, dominant and imprinting effects of haplotypes using equations (2)–(6).
Model Selection
For an observed marker (M) and phenotypic (y) data set, we do not know which are the risk haplotypes nor how many there are. Standard model selection criteria such as the AIC and BIC are used to determine the optimal number and type of risk haplotypes. Among a total of eight haplotypes formed by three SNPs, up to seven ones can be risk haplotypes. The modeling of one to seven risk haplotypes is equivalent to the analysis of the genetic data by biallelic, triallelic, quadrialleli, pentaallelic, hexaallelic, septemallelic and octoallelic models, respectively. The biallelic model dissolves all the haplotypes into two distinct groups, a single risk haplotype group and a nonrisk haplotype group. The risk haplotype group may be composed of one (8 choices), two (28 choices), three (56 choices) or four haplotypes (70 choices). The triallelic, quadrialleli, pentaallelic, hexaallelic, septemallelic and octoallelic models contains 28, 56, 170, 56, 24 and 8 cases, respectively. The likelihoods and model selection criteria, AIC or BIC, are then calculated and compared among different models and assumptions. This is summarized in Table 4.
Hypothesis Tests
For a given data set, testing the existence of functional haplotypes is a first step. This can be done by formulating the following hypotheses:
H_{1} : At least one of equality in H_{0} does not hold The loglikelihood ratio (LR) is then calculated by plugging the estimated parameters into the likelihood under the H_{0} and H_{1}, respectively. The LR can be viewed as being asymptotically χ^{2}distributed with (k + 1)^{2}  1 degrees of freedom.
After a significant haplotype effect is detected, a series of further tests are performed for the significance of additive, dominance and imprinting effects triggered by haplotypes. The null hypotheses under each of these tests can be formulated by setting the effect being tested to be equal to zero. For example, under the triallelic model, the null hypothesis for testing the imprinting effect of the haplotypesis expressed as
In practice, it is also interesting to test each of the additive genetic effects, each of the dominance effects and each of the imprinting effects for the triand quadriallelic models. The estimates of the parameters under the null hypotheses can be obtained with the same EM algorithm derived for the alternative hypotheses but with a constraint of the tested effect equal to zero. The loglikelihood ratio test statistics for each hypothesis is thought to asymptotically follow a χ^{2}distributed with the degree of freedom equal to the difference of the numbers of the parameters being tested under the null and alternative hypotheses.
References
 1.
Reik W, Walter J: Genomic imprinting: parental influence on the genome. Nature Reviews Genetics. 2001, 2: 2132.
 2.
Falls J, Pulford D, Wylie A, Jirtle R: Genomic imprinting: implications for human disease. American Journal of Pathology. 1999, 154 (3): 635647.
 3.
Jirtle R: Genomic imprinting and cancer. Experimental cell research. 1999, 248: 1824.
 4.
Luedi P, Hartemink A, Jirtle R: Genomewide prediction of imprinted murine genes. Genome Research. 2005, 15 (6): 875884.
 5.
Waterland R, Lin J, Smith C, Jirtle R: Postweaning diet affects genomic imprinting at the insulinlike growth factor 2(Igf2) locus. Human Molecular Genetics. 2006, 15 (5): 705716.
 6.
Killian J, Byrd J, Jirtle J, Munday B, Stoskopf M, MacDonald R, Jirtle R: M6P/IGF2R imprinting evolution in mammals. Molecular Cell. 2000, 5 (4): 707716.
 7.
Haig D: Altercation of generations: genetic conflicts of pregnancy. American journal of reproductive immunology (New York, NY: 1989). 1996, 35 (3): 226
 8.
Lefebvre L, Viville S, Barton S, Ishino F, Keverne E, Surani M: Abnormal maternal behaviour and growth retardation associated with loss of the imprinted gene Mest. Nature genetics. 1998, 20: 163169.
 9.
Li L, Keverne E, Aparicio S, Ishino F, Barton S, Surani M: Regulation of maternal behavior and offspring growth by paternally expressed Peg3. Science. 1999, 284 (5412): 330
 10.
de Koning D, Rattink A, Harlizius B, Van Arendonk J, Brascamp E, Groenen M: Genomewide scan for body composition in pigs reveals important role of imprinting. Proceedings of the National Academy of Sciences. 2000, 97 (14): 79477950. 10.1073/pnas.140216397.
 11.
de Koning D, Bovenhuis H, van Arendonk J: On the detection of imprinted quantitative trait loci in experimental crosses of outbred species. Genetics. 2002, 161 (2): 931938.
 12.
Cui Y, Lu Q, Cheverud J, Littell R, Wu R: Model for mapping imprinted quantitative trait loci in an inbred F2 design. Genomics. 2006, 87 (4): 543551.
 13.
Jeon J, Carlborg O, Törnsten A, Giuffra E, Amarger V, Chardon P, AndersenEklund L: A paternally expressed QTL affecting skeletal and cardiac muscle mass in pig maps to the IGF2 locus. Nat Genet. 1999, 21: 157158.
 14.
Nezer C, Moreau L, Brouwers B, Coppieters W, Detilleux J, Hanset R, Karim L, Kvasz A, Leroy P, Georges M: An imprinted QTL with major effect on muscle mass and fat deposition maps to the IGF2 locus in pigs. Nature Genetics. 1999, 21 (2): 155156.
 15.
TuiskulaHaavisto M, De Koning D, Honkatukia M, Schulman N, Mäkitanila A, Vilkki J: Quantitative trait loci with parentoforigin effects in chicken. Genetics Research. 2004, 84 (01): 5766.
 16.
Lewis A, Redrup L: Genetic imprinting: conflict at the Callipyge locus. Current Biology. 2005, 15 (8): 291294. 10.1016/j.cub.2005.04.003.
 17.
Cui Y: A statistical framework for genomewide scanning and testing of imprinted quantitative trait loci. Journal of theoretical biology. 2007, 244: 115126.
 18.
Liu T, Todhunter R, Wu S, Hou W, Mateescu R, Zhang Z, BurtonWurster N, Acland G, Lust G, Wu R: A random model for mapping imprinted quantitative trait loci in a structured pedigree: An implication for mapping canine hip dysplasia. Genomics. 2007, 90 (2): 276284.
 19.
Liu T, Johnson J, Casella G, Wu R: Sequencing complex diseases with HapMap. Genetics. 2004, 168: 503511.
 20.
Wu S, Yang J, Wang C, Wu R: A General Quantitative Genetic Model for Haplotyping a Complex Trait in Humans. Current Genomics. 2007, 8 (5): 343350.
 21.
Takasaki I, Suzuki T, Sasaki A, Nakao K, Hirakata M, Okano K, Tanaka T, Nagase H, Shiraki K, Nojima H: Suppression of acute herpetic painrelated responses by the kappaopioid receptor agonist ()17cyclopropylmethyl3, 14betadihydroxy4, 5alphaepoxybeta[nmethyl3trans3(3furyl) acrylamido] morphinan hydrochloride (TRK820) in mice. The Journal of pharmacology and experimental therapeutics. 2004, 309: 36
 22.
Fillingim R, Doleys D, Edwards R, Lowery D: Spousal Responses Are Differentially Associated With Clinical Variables in Women and Men With Chronic Pain. Clinical Journal of Pain. 2003, 19 (4): 217
 23.
Wilkins J, Haig D: What good is genomic imprinting: the function of parentspecific gene expression. Nature Reviews Genetics. 2003, 4 (5): 359368.
 24.
Wood A, Oakey R: Genomic imprinting in mammals: emerging themes and established theories. PLoS genetics. 2006, 2 (11):
 25.
Lewis A, Reik W: How imprinting centres work. Cytogenet Genome Res. 2006, 113 (1–4): 8189.
 26.
Jirtle R, Skinner M: Environmental epigenomics and disease susceptibility. Nature Reviews Genetics. 2007, 8 (4): 253262.
 27.
Feil R, Berger F: Convergent evolution of genomic imprinting in plants and mammals. Trends in Genetics. 2007, 23 (4): 192199.
 28.
Nezer C, Collette C, Moreau L, Brouwers B, Kim J, Giuffra E, Buys N, Andersson L, Georges M: Haplotype sharing refines the location of an imprinted quantitative trait locus with major effect on muscle mass to a 250kb chromosome segment containing the porcine IGF2 gene. Genetics. 2003, 165: 277285.
 29.
Cheverud J, Hager R, Roseman C, Fawcett G, Wang B, Wolf J: Genomic imprinting effects on adult body composition in mice. Proceedings of the National Academy of Sciences. 2008, 105 (11): 425310.1073/pnas.0706562105.
 30.
Judson R, Stephens J, Windemuth A: The predictive power of haplotypes in clinical response. Pharmacogenomics. 2000, 1: 1526.
 31.
Bader J: The relative power of SNPs and haplotype as genetic markers for association tests. Pharmacogenomics. 2001, 2: 1124.
 32.
Winkelmann B, Hoffmann M, Nauck M, Kumar A, Nandabalan K, Judson R, Boehm B, Tall A, Ruano G, März W: Haplotypes of the cholesteryl ester transfer protein gene predict lipidmodifying response to statin therapy. The Pharmacogenomics Journal. 2003, 3 (5): 284296.
 33.
Clark A: The role of haplotypes in candidate gene studies. Genetic epidemiology. 2004, 27 (4): 321333.
 34.
Jin G, Miao R, Deng Y, Hu Z, Zhou Y, Tan Y, Wang J, Hua Z, Ding W, Wang L: Variant genotypes and haplotypes of the epidermal growth factor gene promoter are associated with a decreased risk of gastric cancer in a highrisk Chinese population. Cancer science. 2007, 98 (6): 864868.
 35.
Altshuler D, Brooks L, Chakravarti A, Collins F, Daly M, Donnelly P: A haplotype map of the human genome. Nature. 2005, 437 (7063): 12991320.
Acknowledgements
This work is supported by Joint grant DMS/NIGMS0540745 and NIH RO1 grant NS41670.
Author information
Additional information
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
SW, CW, and AB contributed equally to this manuscript. SW, CW, AB, YL derived the algorithm and performed the data analysis. RBF, MRW, RS, LK designed the experiment and collected the data. MMC provided statistical advice and help. RW conceived the model and wrote the paper. AB provided final modifications to the paper. All authors have read and approved the final manuscript.
Electronic supplementary material
Rights and permissions
About this article
Received
Accepted
Published
DOI
Keywords
 Haplotype Frequency
 Imprint Gene
 Pentazocine
 Maternal Parent
 Risk Haplotype