Mark Abney

The sex-specific genetic architecture of quantitative traits in humans.

Mapping genetically complex traits remains one of the greatest challenges in human genetics today. In particular, gene-environment and gene-gene interactions, genetic heterogeneity and incomplete penetrance make thorough genetic dissection of complex traits difficult, if not impossible. Sex could be considered an environmental factor that can modify both penetrance and expressivity of a wide variety of traits. Sex is easily determined and has measurable effects on recognizable morphology; neurobiological circuits; susceptibility to autoimmune disease, diabetes, asthma, cardiovascular and psychiatric disease; and quantitative traits like blood pressure, obesity and lipid levels, among others. In this study, we evaluated sex-specific heritability and genome-wide linkages for 17 quantitative traits in the Hutterites. The results of this study could have important implications for mapping complex trait genes.

The Genetic Dissection of Complex Traits in a Founder Population

We estimated broad heritabilities (H(2)) and narrow heritabilities (h(2)) and conducted genomewide screens, using a novel association-based mapping approach for 20 quantitative trait loci (QTLs) among the Hutterites, a founder population that practices a communal lifestyle. Heritability estimates ranged from.21 for diastolic blood pressure (DBP) to.99 for whole-blood serotonin levels. Using a multipoint method to detect association under a recessive model we found evidence of major QTLs for six traits: low-density lipoprotein (LDL), triglycerides, lipoprotein (a) (Lp[a]), systolic blood pressure (SBP), serum cortisol, and whole-blood serotonin. Second major QTLs for Lp(a) and for cortisol were identified using a single-point method to detect association under a general two-allele model. The heritabilities for these six traits ranged from.37 for triglycerides to.99 for serotonin, and three traits (LDL, SBP, and serotonin) had significant dominance variances (i.e., H(2) > h(2)). Surprisingly, there was little correlation between measures of heritability and the strength of association on a genomewide screen (P>.50), suggesting that heritability estimates per se do not identify phenotypes that are influenced by genes with major effects. The present study demonstrates the feasibility of genomewide association studies for QTL mapping. However, even in this young founder population that has extensive linkage disequilibrium, map densities <<5 cM may be required to detect all major QTLs.

Estimating the human mutation rate using autozygosity in a founder population

Knowledge of the rate and pattern of new mutation is critical to the understanding of human disease and evolution. We used extensive autozygosity in a genealogically well-defined population of Hutterites to estimate the human sequence mutation rate over multiple generations. We sequenced whole genomes from 5 parent-offspring trios and identified 44 segments of autozygosity. Using the number of meioses separating each pair of autozygous alleles and the 72 validated heterozygous single-nucleotide variants (SNVs) from 512 Mb of autozygous DNA, we obtained an SNV mutation rate of 1.20 × 10−8 (95% confidence interval 0.89–1.43 × 10−8) mutations per base pair per generation. The mutation rate for bases within CpG dinucleotides (9.72 × 10−8) was 9.5-fold that of non-CpG bases, and there was strong evidence (P = 2.67 × 10−4) for a paternal bias in the origin of new mutations (85% paternal). We observed a non-uniform distribution of heterozygous SNVs (both newly identified and known) in the autozygous segments (P = 0.001), which is suggestive of mutational hotspots or sites of long-range gene conversion.

Estimation of variance components of quantitative traits in inbred populations.

Use of variance-component estimation for mapping of quantitative-trait loci in humans is a subject of great current interest. When only trait values, not genotypic information, are considered, variance-component estimation can also be used to estimate heritability of a quantitative trait. Inbred pedigrees present special challenges for variance-component estimation. First, there are more variance components to be estimated in the inbred case, even for a relatively simple model including additive, dominance, and environmental effects. Second, more identity coefficients need to be calculated from an inbred pedigree in order to perform the estimation, and these are computationally more difficult to obtain in the inbred than in the outbred case. As a result, inbreeding effects have generally been ignored in practice. We describe here the calculation of identity coefficients and estimation of variance components of quantitative traits in large inbred pedigrees, using the example of HDL in the Hutterites. We use a multivariate normal model for the genetic effects, extending the central-limit theorem of Lange to allow for both inbreeding and dominance under the assumptions of our variance-component model. We use simulated examples to give an indication of under what conditions one has the power to detect the additional variance components and to examine their impact on variance-component estimation. We discuss the implications for mapping and heritability estimation by use of variance components in inbred populations.

Broad and Narrow Heritabilities of Quantitative Traits in a Founder Population

Estimation of the components of variance for a quantitative trait allows one to evaluate both the degree to which genetics influences the trait and the traits underlying genetic architecture. For particular traits, the estimates also may have implications for discriminating between potential models of selection and for choosing an appropriate model for linkage analysis. Using a recently developed method, we estimate the additive and dominance components of variance--or, equivalently, the narrow and broad sense heritabilities--of several traits in the Hutterites, a founder population with extensive genealogical records. As a result of inbreeding and because Hutterite individuals are typically related through multiple lines of descent, we expect that power to detect dominance variance will be increased relative to that in outbred studies. Furthermore, the communal lifestyle of the Hutterites allows us to evaluate the genetic influences in a relatively homogeneous environment. Four phenotypes had a significant dominance variance, resulting in a relatively high broad heritability. We estimated the narrow and broad heritabilities as being, respectively,.36 and.96 for LDL,.51 and 1.0 for serotonin levels, and.45 and.76 for fat free mass (FFM). There was no significant additive component for systolic blood pressure (SBP), resulting in a narrow heritability of 0 and a broad heritability of.45. There were several traits for which we found no significant dominance component, resulting in equal broad and narrow heritability estimates. These traits and their heritabilities are as follows: HDL,.63; triglycerides,.37; diastolic blood pressure,.21; immunoglobulin E,.63; lipoprotein(a),.77; and body-mass index,.54. The large difference between broad and narrow heritabilities for LDL, serotonin, FFM, and SBP are indicative of strong dominance effects in these phenotypes. To our knowledge, this is the first study to report an estimate of heritability for serotonin and to detect a dominance variance for LDL, FFM, and SBP.

Quantitative-Trait Homozygosity and Association Mapping and Empirical Genomewide Significance in Large, Complex Pedigrees: Fasting Serum-Insulin Level in the Hutterites

We present methods for linkage and association mapping of quantitative traits for a founder population with a large, known genealogy. We detect linkage to quantitative-trait loci (QTLs) through a multipoint homozygosity-mapping method. We propose two association methods, one of which is single point and uses a general two-allele model and the other of which is multipoint and uses homozygosity by descent for a particular allele. In all three methods, we make extensive use of the pedigree and genotype information, while keeping the computations simple and efficient. To assess significance, we have developed a permutation-based test that takes into account the covariance structure due to relatedness of individuals and can be used to determine empirical genomewide and locus-specific P values. In the case of multivariate-normally distributed trait data, the permutation-based test is asymptotically exact. The test is broadly applicable to a variety of mapping methods that fall within the class of linear statistical models (e.g., variance-component methods), under the assumption of random ascertainment with respect to the phenotype. For obtaining genomewide P values, our proposed method is appropriate when positions of markers are independent of the observed linkage signal, under the null hypothesis. We apply our methods to a genome screen for fasting insulin level in the Hutterites. We detect significant genomewide linkage on chromosome 19 and suggestive evidence of QTLs on chromosomes 1 and 16.

Sex-Specific Genetic Architecture of Whole Blood Serotonin Levels

Recently, a quantitative-trait locus (QTL) for whole blood serotonin level was identified in a genomewide linkage and association study in a founder population. Because serotonin level is a sexually dimorphic trait, in the present study, we evaluated the sex-specific genetic architecture of whole blood serotonin level in the same population. Here, we use an extended homozygosity-by-descent linkage method that is suitable for large complex pedigrees. Although both males and females have high broad heritability (H2=0.99), females have a higher additive component (h2=0.63 in females; h2=0.27 in males). Furthermore, the serotonin QTL on 17q that was identified previously in this population, integrin beta 3 (ITGB3), and a novel locus on 2q influence serotonin levels only in males, whereas linkage to a region on chromosome 6q is specific to females. Both sexes contribute to linkage signals on 12q and 16p. There were, overall, more associations meeting criteria for suggestive significance in males than in females, including those of ITGB3 and the serotonin transporter gene (5HTT). This analysis is consistent with heritable sexual dimorphism in whole blood serotonin levels resulting from the effects of a combination of sex-specific and sex-independent loci.

Genome-Wide Association Studies and the Problem of Relatedness Among Advanced Intercross Lines and Other Highly Recombinant Populations

Model organisms offer many advantages for the genetic analysis of complex traits. However, identification of specific genes is often hampered by a lack of recombination between the genomes of inbred progenitors. Recently, genome-wide association studies (GWAS) in humans have offered gene-level mapping resolution that is possible because of the large number of accumulated recombinations among unrelated human subjects. To obtain analogous improvements in mapping resolution in mice, we used a 34th generation advanced intercross line (AIL) derived from two inbred strains (SM/J and LG/J). We used simulations to show that familial relationships among subjects must be accounted for when analyzing these data; we then used a mixed model that included polygenic effects to address this problem in our own analysis. Using a combination of F2 and AIL mice derived from the same inbred progenitors, we identified genome-wide significant, subcentimorgan loci that were associated with methamphetamine sensitivity, (e.g., chromosome 18; LOD = 10.5) and non-drug-induced locomotor activity (e.g., chromosome 8; LOD = 18.9). The 2-LOD support interval for the former locus contains no known genes while the latter contains only one gene (Csmd1). This approach is broadly applicable in terms of phenotypes and model organisms and allows GWAS to be performed in multigenerational crosses between and among inbred strains where familial relatedness is often unavoidable.

The Importance of Genealogy in Determining Genetic Associations with Complex Traits

To the Editor: Most common diseases, such as asthma, type 2 diabetes, bipolar disorder, and cardiovascular disease, are known to have genetic components, but the susceptibility genes have been notoriously difficult to localize and to identify. These complex diseases likely have a large number of genetic and nongenetic risk factors that together have varying effects on phenotype. Many investigators have recommended founder populations for complex-trait mapping, with the expectation that fewer susceptibility alleles will be segregating in these restricted gene pools (Lander and Schork 1994; Wright et al. 1999; Shifman and Darvasi 2001). Some or all individuals in these populations are inbred, but often the exact relationships between all members are either unknown or not taken into account. It is tempting to use such populations for their presumed homogeneity, even in the absence of accurate pedigree information. The failure to take full pedigree information into account can either reduce the power to detect linkage (Dyer et al., in press) or inflate LOD scores (Miano et al. 2000). The failure to account for relatedness among individuals will also affect association studies. In particular, many statistical tests of association are not strictly valid, owing to the lack of true independence between individuals. Nonetheless, such populations have been used widely in association studies (e.g., de Silva et al. 1999; Laprise et al. 2000; Ospina-Duque et al. 2000; Summerhill et al. 2000; Bitti et al. 2001; Hegele et al. 2001), and some authors have even recommended the inclusion of founder populations in case-control studies, owing to their decreased heterogeneity (Shifman and Darvasi 2001). However, the impact that ignoring pedigree relationships has on tests of association has not been evaluated. The Hutterites are an extreme example of a large, complex pedigree with multiple inbreeding loops. We are in the unique position of having complete genealogical information on this 12,903-person, 13-generation pedigree (Abney et al. 2000). Additionally, we have extensive phenotype characterization and a dense microsatellite map (of 568 short-tandem-repeat–polymorphism markers) of ∼750 members of this population, who are descendants of just 64 Hutterite founders (Ober et al. 2000). Thus, we were able to assess the effect that ignoring pedigree information has on statistical tests of association. We performed two separate genomewide scans of association on each of three quantitative phenotypes: serum immunoglobulin E (IgE), serum LDL, and body-mass index (BMI). These phenotypes were chosen to represent quantitative traits associated with diverse complex diseases (asthma, cardiovascular disease, and diabetes, respectively). All phenotypes were adjusted for age and sex and were transformed so that the residuals were approximately normally distributed (Abney et al. 2001). The heritabilities of IgE and BMI were completely accounted for by additive genetic variance, with heritabilities of .63 and .54, respectively; the heritability of LDL had a strong dominance component in addition to additive genetic variance, with a broad heritability of .96 (discussed in detail by Abney et al. [2001]). To estimate the effect that each allele at each locus has on the trait values, we used a statistical test of association, developed specifically for use in large, inbred pedigrees (Ober et al. 2001 [in this issue]). Pedigree structure is taken into account by the use of variance components to model the polygenic background (Abney et al. 2000, 2001). When pedigree structure is ignored, the method is equivalent to a linear regression of the trait on age, sex, and genotype, with a Bonferroni correction applied to the P value for the t test for significance of genotype. In the first scan, we included the variance components and therefore took into account the relatedness between individuals. In the second scan, we did not include any pedigree information (additive and dominance variances were 0). In both scans, the observed P values were adjusted for multiple comparisons, by use of a Bonferroni correction. The two methods yielded dramatically different results. In general, the significance of association with a given marker was considerably inflated when pedigree structure was not included, although in some cases the reverse was true (see fig. 1). Only two loci were among the five most significant results, by both methods, for the three traits (the same marker at 7p21 was associated with IgE, by both methods, and the same marker at 8q12 was associated with LDL, by both methods). In addition, many more loci showed evidence of association when the pedigree structure was not included (see fig. 2). In fact, 10%–22% of all markers appeared to have a strong association (P<.01) with the phenotype, when pedigree structure was not included. Results for 58 single-nucleotide polymorphisms showed similar trends (data not shown). Figure 1 Results of two genome scans—one including pedigree structure (lighter bars) and one not including pedigree structure (darker bars)—for IgE (a), LDL (b), and BMI (c). The five most significant loci when structure is included (left sides ... Figure 2 Number of significantly associated (P<.01) loci when pedigree structure is included (lighter bars) and when pedigree structure is not included (darker bars). Although the Hutterites are an extreme example of a complex pedigree, there are several other populations known to have similar structures (Badner et al. 1990; Slutsky et al. 1997; Hsueh et al. 2000). Moreover, individuals from a variety of smaller, island populations who either have been or are currently being studied may be more related to each other than can be discerned from the recently collected pedigree data (de Silva et al. 1999; Mathias et al. 2000; Bitti et al. 2001). In fact, even presumably outbred populations may contain hidden consanguinity (Broman and Weber 1999), and cryptic relatedness may be a problem in association studies of rare disorders (Bacanu et al. 2000). This problem may be avoided by the use of statistical tools designed to detect misspecified or cryptic relationships (McPeek and Sun 2000; Sun et al., in press). We cannot prove that the inclusion of the pedigree structure in the method results in true associations, until the alleles contributing to these quantitative traits are found; however, we believe that the number of associations found when structure is ignored is unrealistic. Presumably, a profound failure of the assumption of independence between individuals, in method 2, results in a dramatically increased number of type 1 errors. Overall, our data suggest that failing to take into account extended-familial relationships can result in a large number of false-positive results, and some “true” associations may be missed. In addition, the level of significance could be overestimated by several orders of magnitude. In an association study in which it is not possible to take into account all familial relationships, as we have done with the Hutterites, another option is to use genomic controls (Devlin and Roeder 1999). Otherwise, naive approaches to genetic-association analysis could result in an enormous amount of time and of money spent in following up artifactual associations.

Genome-wide association study identifies ITGB3 as a QTL for whole blood serotonin

Serotonin has been implicated in common disorders involving the central nervous, gastrointestinal, cardiovascular, and pulmonary systems. We describe the first genome-wide screen to identify quantitative trait loci (QTLs) influencing whole blood serotonin in 567 members of a single large pedigree, using a novel association-based mapping approach. We identified an association between the β3 integrin (ITGB3) Leu33Pro polymorphism on 17q21 and whole blood serotonin levels (P-value=9.8 × 10−5). This variant explained the evidence for linkage in this region when included as a covariate in the linkage analysis (change in LOD from 1.87 to 0.16), indicating that ITGB3 may be an important serotonin QTL.