Introduction Complex traits, including aging-associated conditions, can be influenced by a multiplicity of genetic and environmental factors. Because each factor is expected to make only a small contribution to trait variability, and this contribution may itself be influenced by interactions with other susceptibility factors, identifying the genetic basis of complex traits is challenging and requires large sample sizes [1]. Isolated founder populations, which have already proven useful in the study of many Mendelian disorders [2], provide an attractive setting for the study of complex traits [3,4] because they typically exhibit greater genetic and environmental homogeneity than more cosmopolitan populations. Sardinia is the second largest island in the Mediterranean. Its modern population numbers approximately 1.65 million and constitutes a genetically isolated founder population [5–7], which has already aided in the identification of genes involved in several Mendelian disorders [8–12]. In addition to its status as an isolated founder population and its relatively large size, the Sardinian population is attractive for genetic studies due to its organization into long-established settlements [13]. Here, we use a large cohort of 6,148 Sardinians to study the heritability of a spectrum of 98 quantitative traits. Studying broad groups of traits, we could assess the generality of any trends, such as changes in heritability with aging. To increase the potential clinical utility of the results, we focused on traits that affect major domains of clinical interest. For example, in addition to anthropometric features, we quantified levels of plasma and serum markers, including total cholesterol, high-density lipoprotein (HDL), and low-density lipoprotein (LDL) levels, and measured subclinical vascular alterations [14–18] that are of intrinsic interest and are also useful predictors of cardiovascular disease [19]. Similarly, we assessed individual differences in personality using the five-factor model [20,21], which quantifies recurring dimensions of personality. Again, in addition to their intrinsic interest, these personality traits are important in understanding a variety of important life outcomes, including mental disorders. Our study uses the full range of phenotypic variation in the population to dissect the genetic contribution and provide a quantitative assessment of the impact of inherited variation on each trait. In addition, we report evidence for heterogeneity in the genetic and environmental contributions to variation, by comparing variances and covariances between males and females and between the younger and older individuals in our cohort. Finally, we examine evidence for an overlap in the genetic determinants of multiple traits, identifying clusters of traits that appear to be influenced by the same genes. The joint study of cardiovascular and personality traits afforded us an opportunity to look for a genetic factor that might contribute to the association of certain personality traits and cardiovascular problems [22]. Overall, our results should be useful to investigators interested in identifying the genetic determinants of quantitative trait variation, especially for clinically relevant quantitative traits affecting cardiovascular function and personality. Results Cohort Recruitment We recruited and phenotyped 6,148 individuals, male and female, age 14 y and above (Figure 1A) from a cluster of four towns in the Lanusei Valley in the Ogliastra region of the Sardinian province of Nuoro. This corresponds to approximately 62% of the population eligible for recruitment in the area, which totaled 9,841 individuals in the 2001 census. Compared to the census population, our sample is enriched for females at all ages (3,523 individuals, or 57%, of our sample, compared to 5,089, or 52%, of the census population). Ascertainment was less complete for individuals more than 74 y of age, among whom only approximately 29% of the population was recruited (238 individuals more than 74 y recruited, but 813 were reported in the 2001 census). Figure 1 Age, Sex, and Birthplace Distribution for Participants (A) Shows the number of recruited females (black bars) and males (white bars) from the four clustered towns. (B) Shows the birthplace distribution of participants, in progressively larger geographic units: Lanusei, L.I.E.A. (Lanusei and the three surrounding towns of Ilbono, Elini, and Arzana), the Lanusei valley, the region of Ogliastra, the province of Nuoro, and all of Sardinia. (C) Shows the birthplace distribution for grandparents of participants in the same progressively larger geographic units. Nearly all subjects were born in Sardinia (5,857 [95%]) and, specifically, in the Ogliastra region (5,442 [89%]; Figure 1B shows the birth places of participants in the restricted geographical region). Emphasizing the stability of the population, all grandparents were born in Sardinia for 95% of participants (Figure 1C). The cohort is organized into multiple complex pedigrees. Information collected at recruitment allowed us to organize 5,610 individuals into 711 connected pedigrees, each up to five generations deep. The largest pedigree connects 625 phenotyped individuals. In total the sample includes 34,469 relative pairs, with an average kinship coefficient of 0.1628. These relative pairs include 4,933 sibling pairs, 180 half-sibling pairs, 4,014 first cousins, 4,256 parent–child pairs, 675 grandparent–grandchild pairs, and 6,400 avuncular pairs in addition to other more distant relatives. Our sample also includes 11 monozygotic twins (identified by genotyping approximately 10,000 single nucleotide polymorphisms in all individuals). Because monozygotic twins are often more similar to each other than predicted by a simple genetic model (even with genetic dominance included), we included only one individual from each of these twin pairs in the analysis reported below. Summary of Quantitative Trait Variation To examine the effect of age and sex on each trait, we first generated and reviewed summary plots for each trait. The complete set of plots is available online (http://www.sph.umich.edu/csg/chen/public/sardinia) together with detailed results for all our analysis. Figure 2 displays the distribution of six illustrative traits for males and females. It is clear that for many traits there are marked differences between the sexes, affecting not only trait means, but also the overall pattern of variability around these means. Figure 3 illustrates the effect of age on the same six traits. For each trait, observed measurements are plotted against age at enrollment, and two quadratic regression lines (blue for females and red for males) are presented to summarize the impact of age on the traits. These plots allowed us to identify outliers in each trait and to compare trait distributions with other studies. Figure 2 Distribution of Six Illustrative Traits in Male and Female Participants Relative densities are plotted for males (solid lines) and females (dashed lines) for two serum values (cholesterol levels [A] and HDL [B]), two measures of cardiovascular function (IMT of the carotid artery [C] and PWV [D]), and two personality facets (NEO_N3 [E] and NEO_O5 [F]). A complete set of plots, including all traits, is available online (http://www.sph.umich.edu/csg/chen/public/sardinia). Figure 3 Illustrative Quantitative Traits Plotted as a Function of Age These are the same traits as in Figure 2. All values are plotted, and polynomial regression curves fitted to the data show inferred trends for males (solid red lines) and females (dashed blue lines) with increasing age. A complete set of plots, allowing for all traits, is available online (http://www.sph.umich.edu/csg/chen/public/sardinia). We next calculated the mean and standard deviation for all traits, both in the entire cohort and after stratifying the sample by sex and age. When stratifying the sample by age, we considered four age bands (14–29, 30–44, 45–59, and 60–102 y of age), each including approximately 25% of sampled individuals. The results are summarized in Table S1, with traits organized as blood test results (38 traits), anthropometric measures (five traits), cardiovascular measures (20 traits), and personality traits (five factors and 30 facets of personality). Nearly all traits showed highly significant evidence (analysis of variance p 0.05, indicating no significant degradation in fit when using the parsimonious models). Thus, there was clear evidence for heterogeneity in variance components by sex, but it was difficult to decide whether the heterogeneity was due to genes, environment, or both. Heterogeneity in Variance Components, by Age To look for heterogeneity in variance components by age, we divided individuals into two groups. The “younger” group included individuals less than 42 y of age (the median age in our sample), whereas the “older” group included individuals 42 y of age and older. We found significant evidence for heterogeneity in variance components by age in 62 of the 98 traits examined (the results are summarized in Table 4). This included a majority of traits in all categories, including anthropometric traits (three of five), blood test results (24 of 38), cardiovascular traits (13 of 20), and personality factors and facets (22 of 35). Again, we considered a series of intermediate models, including only heterogeneity in environmental or genetic variance components, or in which variance components differed by a constant factor between the young and old, and used the BIC to select the best-fitting model. For 26 traits, a model in which only the environmental variance differed between young and old was selected, and for 20 of these traits, environmental variance was greater among older individuals (so that heritability was lower). Heritability was higher in older individuals for IMT and five personality traits. Table 4 Model Comparisons between Young and Old For 21 traits, a model in which only genetic variance differed between the young and old was selected, and heritability was higher in the young for 15 traits (12 personality traits and three blood test results). It is noteworthy that the six traits more heritable in the old included several blood pressure–related traits (SBP, DBP, mean blood pressure, and pulse pressure). For these cardiovascular traits, heritability increased an average of 18% among older individuals, from approximately 8% for younger individuals to approximately 26% in older individuals. For 15 other traits, a model in which heritabilities between the young and old differed by a constant factor provided the best fit to the data, whereas for one trait (fractionated bilirubin), both environmental and genetic variance components appeared to differ between the young and old. Bivariate Analysis We calculated genetic correlation coefficients for all pairings of 93 traits (including the 38 blood phenotypes, five anthropometric measures, 20 cardiovascular traits, and 30 facets of personality, but excluding the five factors of personality, which are derived from the 30 facets). This corresponds to a total of 8,556 genetic correlation coefficients, of which 118 coefficients were greater than 0.50. In contrast, only 36 of the overall correlation coefficients were greater than 0.50. A full matrix of pairwise correlation coefficients is available http://www.sph.umich.edu/csg/chen/public/sardinia). We identified 18 clusters of traits with a genetic correlation greater than 0.50 (Table S2). To summarize the full pairwise correlation matrix, we used a hierarchical clustering approach that successively groups traits with large genetic correlations (see Figure 4). In the figure, traits connected by short branches share more of their genetic correlation, whereas traits that join up only near the root of the tree have only a small genetic correlation. Some of the clusters occur because traits are related by definition (for example, pulse pressure and SBP), or by physiology (for example, diastolic diameter [diam_D] and systolic diameter [diam_S], and IMT and wall lumen). Other clusters are quite interesting. For example, hip circumference, waist circumference, body mass index (BMI), and weight all cluster close together and near insulin levels. These traits are all related to the metabolic syndrome [27], and the result supports a genetic underpinning for the syndrome. As another example, the clustering of facets for the NEO O, NEO N, NEO C, and NEO A factors reinforces the structure of the five-factor personality model. Other results are more unexpected. For example, the personality facet NEO E4 (activity) clusters closer to components of NEO C (conscientiousness) than it does to other facets of NEO E. To further investigate the genetic relationship between different personality facets, we also carried out a factor analysis of genetic correlations (Table S3). This factor analysis confirms that the genetic structure of personality replicates its phenotypic structure quite well, but again places NEO E4 closer to components of NEO C. Figure 4 Clustering of Genetic Correlations The 98 quantative traits are classified into clusters inferred from genetic correlations between any two traits, with an “average” distance measure used in the clustering algorithm. Classes of traits are color-coded as personality (red), serum composition (blue), cardiovascular (black), and anthropometric (green). Overlap of the apparent genetic contribution to variance is indicated on the ordinate, with larger overlaps towards the bottom. Eighteen values exceed 50% overlap (see text). We looked specifically for a genetic link between personality traits and cardiovascular disease [22]. Hostility, depression, anger, and anxiety have been associated with cardiovascular risk factors, including arterial stiffness and thickness (see [28] and references therein), and are independent predictors of incident cardiovascular disease and mortality [29]. Several mechanistic links have been proposed to explain the relationship between personality traits and cardiovascular diseases and outcomes [30]. However, the basis for the association has been conjectural. We find no substantive sharing of a genetic basis for cardiovascular traits and any psychological traits. For example, genetic correlation between N2 (hostility and anger) or A4 (low compliance/aggression) and IMT, PWV, SBP, DBP, or heart rate was not significantly different from zero. Discussion The cohort of Sardinians described here provided us with a valuable opportunity to investigate the heritability of multiple traits simultaneously. For some traits, the size of our cohort exceeds the total number of individuals examined in all previously published studies of their heritability. The large size of the cohort and the diversity of the relationships sampled enabled us not only to consider the overall heritability of each trait, but also to investigate the possibility of heterogeneity in genetic effects by age or sex, as well as the evidence for shared genetic determinants between different traits. To facilitate downstream studies, complete results of all our analyses (including likelihoods and parameter estimates for each model fitted) are available online (http://www.sph.umich.edu/csg/chen/public/sardinia). Overall, we estimated heritabilities of approximately 0.40 on average for individual blood test results, approximately 0.51 for anthropometric measures, approximately 0.25 for measures of cardiovascular function, and approximately 0.19 for personality factors and facets. In general, our results appear to be consistent with previous studies (see, for example, [31–34]), and particularly with previous studies based on extended pedigrees, (e.g., in the Hutterites [35] and another Sardinian village [36]). Our estimates of heritability are smaller than in previous studies of twins and siblings, both for cardiovascular traits [37,38] and for personality traits [39–43]. Extended pedigree samples such as ours allow specific assessment of narrow heritability potentially, and it is possible that non-additive effects inflated estimates of heritability in studies of twins and small families [44,45]. In our cohort, four of five components of the five-factor model (NEO N, E, O, and C) and most cardiovascular traits showed evidence for genetic dominance. Our broad estimates of heritability, which allow for genetic dominance, are more similar to results in studies of twins and siblings. Nearly all traits showed highly significant evidence (p 0, σs 2 = 0), models with only shared environment (σd 2 = 0, σs 2 > 0), and other intermediate models (σd 2 > 0, σs 2 > 0), comparisons of parameter estimates from these models are informative. In the model with genetic dominance, the quantity H2 = (σd 2 + σg 2)/(σd 2 + σg 2 + σe 2) provides a liberal estimate of the overall impact of genes on the phenotype at hand, whereas in the model attributing any excess similarity between siblings to shared environment, the quantity h2 = σg 2/(σs 2 + σg 2 + σe 2) provides a very conservative estimate of the overall impact of genes. Whenever there was significant evidence (p j implies that i is not an ancestor of j (any ordering where ancestors precede their descendants is suitable). Then, we defined the kinship coefficient for X-linked genes, ϕij (X) , as follows: Although this definition only covers the situation in which i ≥ j, it can be used to estimate any kinship coefficient because ϕij (X) = ϕji (X) . The definition reflects the fact that males carry only one allele for X-linked genes, inherited from their mother. Females carry two alleles, one inherited from each parent. The functions mother(i) and father(i) return indexes for the parents of i. Supporting Information Protocol S1 Supplementary Methodology: Protocol Details for Measuring Cardiovascular Traits This section provides a detailed protocol for the assessment of cardiovascular traits. (18 KB PDF) Click here for additional data file. Table S1 Detailed Descriptive Statistics for 98 Traits This table includes trait means and variances. Trait means are stratified by sex and into four age bands. (37 KB PDF) Click here for additional data file. Table S2 Clusters of Traits for Which Genetic Correlation Is More Than 0.5 Highlights subsets of traits identified in the clustering analysis, for which the genetic correlation exceeds 0.5. (7 KB PDF) Click here for additional data file. Table S3 Genetic Factor Structure of Personality Traits The table presents Procrustes-rotated principal components from the genetic correlations among the 30 facets of the NEO-PI-R, targeted to the American normative factor structure. (11 KB PDF) Click here for additional data file.
