Search results

AbstractThe classical twin design relies on a number of strong number of assumptions in order to yield unbiased estimates of heritability. This includes the equal environments assumption — that monozygotic and dizygotic twins experience similar degrees of environmental similarity — an assumption that is likely to be violated in practice for many traits of interest. An alternative method of estimating heritability that does not suffer from many of these limitations is to model trait similarity between sibling pairs as a function of their empirical genome-wide identity by descent sharing, estimated from genetic markers. In this review, I recount the story behind Nick Martin's and my development of this method, our first attempts at applying it in a human population and more recent studies using the original and related methods to estimate trait heritability.

AbstractWhen using maximum likelihood methods to estimate genetic and environmental components of (co)variance, it is common to test hypotheses using likelihood ratio tests, since such tests have desirable asymptotic properties. In particular, the standard likelihood ratio test statistic is assumed asymptotically to follow a χ2distribution with degrees of freedom equal to the number of parameters tested. Using the relationship between least squares and maximum likelihood estimators for balanced designs, it is shown why the asymptotic distribution of the likelihood ratio test for variance components does not follow a χ2distribution with degrees of freedom equal to the number of parameters tested when the null hypothesis is true. Instead, the distribution of the likelihood ratio test is a mixture of χ2distributions with different degrees of freedom. Implications for testing variance components in twin designs and for quantitative trait loci mapping are discussed. The appropriate distribution of the likelihood ratio test statistic should be used in hypothesis testing and model selection.

AbstractThis article analyzes the effect of classroom separation of twins on their cognitive abilities, measured at different ages in Dutch primary education. We use a large longitudinal school-based sample of twins and their classmates. The analysis tries to reduce the bias by unobserved factors due to the nonrandom assignment of twins by taking into account differences in school environment, previous test scores and variation in class assignment between years. We find that classroom separation matters for language in Grade 2. Nonseparated twins score higher on language, and the difference is larger for same-sex pairs. This finding is robust for various methods that take unobserved effects into account. In addition, there is some evidence for higher scores in arithmetic in Grade 2. For the higher grades we find no effect of classroom separation on cognitive ability. In the analysis of the effect of a separation of at least 3 years we find that separation increases language performance between Grade 6 and 8 for opposite-sex pairs.

AbstractThe basis of most twin studies is the ascertainment of twins, often through twin registries, and determination of zygosity. The current rate of twin births in many industrialized countries implies that in the near future around 3% or more of individuals will be a twin. Hence, there are and will be a lot of twins around and many of those will not participate in twin studies. However, if large population-based samples are available that include appropriate identifiers, then twins can be detected and twin studies performed, even in the absence of zygosity information. We quantified the number of twin pairs that could be detected from a longitudinal survey in the Netherlands, which aims to answer questions about educational strategies and performance in primary education in the Netherlands. We detected 2865 twin pairs if we used a coded name identifier, date of birth, school, grade and year of survey, which is 2.01% of 284,945 pupils in five cohorts. Relaxing our selection criteria increased the number of apparent twin pairs identified, most of which are false positives due to chance matching of identification criteria. We show that the intraclass correlation on measured phenotypes can be used as a quality control measure for twin identification, and quantify the proportion of false negatives (true twin pairs not identified) due to missing data and data coding errors. We compared our estimated rate of twins in the sample to census data and estimate that with our most stringent selection criteria we detect more than 80% of all twin pairs in the sample. We conclude that the identification of twin pairs from large population-based samples is feasible, rapid and accurate if the appropriate identifiers are available, and that twin pairs from such sources are a valuable resource for studies to answer scientific question about twins versus nontwins and about genetic and environmental factors of twin resemblance.

Recently a paper authored by ourselves and a number of co-authors about the proportion of phenotypic variation in height that is explained by common SNPs was published inNature Genetics(Yang et al., 2010). Common SNPs explain a large proportion of the heritability for human height (Yang et al.). During the refereeing process (the paper was rejected by two other journals before publication inNature Genetics) and following the publication of Yang et al. (2010) it became clear to us that the methodology we applied, the interpretation of the results and the consequences of the findings on the genetic architecture of human height and that for other traits such as complex disease are not well understood or appreciated. Here we explain some of these issues in a style that is different from the primary publication, that is, in the form of a number of comments and questions and answers. We also report a number of additional results that show that the estimates of additive genetic variation are not driven by population structure.

AbstractWe expand our previous deterministic power calculations by calculating the required sample size to detect C in ACE models. The theoretical expected value of the maximum log-likelihood for the AE model was derived using two optimisation methods and these gave near-identical results. Theoretical predictions were verified by computer simulation and the results agreed very well. We have developed a user-friendly web-based tool, TwinPower, to perform power calculations to detect either A or C for the classical twin design. This new tool can be found at http://genepi.qimr.edu.au/cgi-bin/twinpower.cgi

AbstractThe prioritization of genes within a candidate genomic region is an important step in the identification of causal gene variants affecting complex traits. Surprisingly, there have been very few reports of bioinformatics tools to perform such prioritization. The purpose of this article is to investigate the performance of 3 positional candidate gene software tools available, PosMed, GeneSniffer and SUSPECTS. The comparison was made for 40, 20 and 10 Mb regions in the human genome centred around known susceptibility genes for the common diseases breast cancer, Crohn's disease, age-related macular degeneration and schizophrenia. The known susceptibility gene was not always ranked highly, or not ranked at all, by 1 or more of the software tools. There was a large variation between the 3 tools regarding which genes were prioritized, and their rank order. PosMed and GeneSniffer were most similar in their prioritization gene list, whereas SUSPECTS identified the same candidate genes only for the narrowest (10 Mb) regions. Combining 2 or all of the candidate gene finding tools was superior in terms of ranking positional candidates. It is possible to reduce the number of candidate genes from a starting set in a region of interest by combining a variety of candidate gene finding tools. Conversely, we recommend caution in relying solely on single positional candidate gene prioritization tools. Our results confirm the obvious, that is, that starting with a narrower positional region gives a higher likelihood that the true susceptibility gene is selected, and that it is ranked highly. A narrow confidence interval for the mapping of complex trait genes by linkage can be achieved by maximizing marker informativeness and by having large samples. Our results suggest that the best approach to classify a minimum set of candidate genes is to take those genes that are prioritized by multiple prioritization tools.

Monozygotic (MZ) twins form an important system for the study of biological plasticity in humans. While MZ twins are generally considered to be genetically identical, a number of studies have emerged that have demonstrated copy-number differences within a twin pair, particularly in those discordant for disease. The rate of autosomal copy-number variation (CNV) discordance within MZ twin pairs was investigated using a population sample of 376 twin pairs genotyped on Illumina Human610-Quad arrays. After CNV calling using both QuantiSNP and PennCNV followed by manual annotation, only a single CNV difference was observed within the MZ twin pairs, being a 130 KB duplication of chromosome 5. Five other potential discordant CNV were called by the software, but excluded based on manual annotation of the regions. It is concluded that large CNV discordance is rare within MZ twin pairs, indicating that any CNV difference found within phenotypically discordant MZ twin pairs has a high probability of containing the causal gene(s) involved.

AbstractPeople meeting diagnostic criteria for anxiety or depressive disorders tend to score high on the personality scale of neuroticism. Studying this dimension of personality can therefore give insights into the etiology of important psychiatric disorders. Neuroticism can be assessed easily via self-report questionnaires in large population samples. We have examined the genetic and phenotypic stability of neuroticism, measured up to 4 times over 22 years, on different scales, on a data set of 4999 families with over 20,000 individuals completing at least 1 neuroticism questionnaire. The neuroticism scales used were the Eysenck Personality Questionnaire revised (EPQ-R), the EPQ-R shortened form, and the NEO 5 factor inventory personality questionnaire. The estimates of heritability of the individual measures ranged from .26 ± .04 to .36 ± .03. Genetic, environmental, and phenotypic correlations averaged .91, .42, and .57 respectively. Despite the range in heritabilities, a more parsimonious 'repeatability model' of equal additive genetic variances and genetic correlations of unity could not be rejected. Use of multiple measures increases the effective heritability from .33 for a single measure to .43 for mean score because of the reduction in the estimate of the environmental variance, and this will increase power in genetic linkage or association studies of neuroticism.

AbstractHuman height is a highly heritable trait, with genetic factors explaining up to 90% of phenotypic variation. Vitamin D levels are known to influence several physiological processes, including skeletal growth. The vitamin D receptor (VDR) gene has been reported as contributing to variation in height. A meta-analysis of 13607 adult individuals found a small but significant association with the rs1544410 (BsmI) polymorphism. In contrast, the meta-analysis found no effect in a sample of 550 children. Two recent studies reported variants with large effect on height elsewhere in VDR (rs10735810 [FokI] and rs7139166 [-1521] polymorphisms). We genotyped large Caucasian samples from Australia (N= 3906) and the Netherlands (N= 1689) for polymorphisms in VDR. The Australian samples were twin families with height measures from 3 time points throughout adolescence. The Dutch samples were adult twins. We use the available family data to perform both within and between family tests of association. We found no significant associations for any of the genotyped variants after multiple testing correction. The (non-significant) effect of rs1544410 in the Australian adolescent cohort was in the same direction and of similar magnitude (additive effect 0.3cm) to the effect observed in the published adult meta-analysis. An effect of this size explains ~0.1% of the phenotypic variance in height — this implies that many, probably hundreds, of such variants are responsible for the observed genetic variation. Our results did not support any role for two other regions (rs10735810, rs7139166) of VDR in explaining variation in height.