Comparison and assessment of family- and population-based genotype imputation methods in large pedigrees.

Genotype imputation is widely used in genome-wide association studies to boost variant density, allowing increased power in association testing. Many studies currently include pedigree data due to increasing interest in rare variants coupled with the availability of appropriate analysis tools. The performance of population-based (subjects are unrelated) imputation methods is well established. However, the performance of family- and population-based imputation methods on family data has been subject to much less scrutiny. Here, we extensively compare several family- and population-based imputation methods on family data of large pedigrees with both European and African ancestry. Our comparison includes many widely used family- and population-based tools and another method, Ped_Pop, which combines family- and population-based imputation results. We also compare four subject selection strategies for full sequencing to serve as the reference panel for imputation: GIGI-Pick, ExomePicks, PRIMUS, and random selection. Moreover, we compare two imputation accuracy metrics: the Imputation Quality Score and Pearson's correlation R 2 for predicting power of association analysis using imputation results. Our results show that (1) GIGI outperforms Merlin; (2) family-based imputation outperforms population-based imputation for rare variants but not for common ones; (3) combining family- and population-based imputation outperforms all imputation approaches for all minor allele frequencies; (4) GIGI-Pick gives the best selection strategy based on the R 2 criterion; and (5) R 2 is the best measure of imputation accuracy. Our study is the first to extensively evaluate the imputation performance of many available family- and population-based tools on the same family data and provides guidelines for future studies.

Detection of Mendelian consistent genotyping errors in pedigrees.

Detection of genotyping errors is a necessary step to minimize false results in genetic analysis. This is especially important when the rate of genotyping errors is high, as has been reported for high-throughput sequence data. To detect genotyping errors in pedigrees, Mendelian inconsistent (MI) error checks exist, as do multi-point methods that flag Mendelian consistent (MC) errors for sparse multi-allelic markers. However, few methods exist for detecting MC genotyping errors, particularly for dense variants on large pedigrees. Here, we introduce an efficient method to detect MC errors even for very dense variants (e.g., SNPs and sequencing data) on pedigrees that may be large. Our method first samples inheritance vectors (IVs) using a moderately sparse but informative set of markers using a Markov chain Monte Carlo-based sampler. Using sampled IVs, we considered two test statistics to detect MC genotyping errors: the percentage of IVs inconsistent with observed genotypes (A1) or the posterior probability of error configurations (A2). Using simulations, we show that this method, even with the simpler A1 statistic, is effective for detecting MC genotyping errors in dense variants, with sensitivity almost as high as the theoretical best sensitivity possible. We also evaluate the effectiveness of this method as a function of parameters, when including the observed pattern for genotype, density of framework markers, error rate, allele frequencies, and number of sampled inheritance vectors. Our approach provides a line of defense against false findings based on the use of dense variants in pedigrees.

A statistical framework to guide sequencing choices in pedigrees.

The use of large pedigrees is an effective design for identifying rare functional variants affecting heritable traits. Cost-effective studies using sequence data can be achieved via pedigree-based genotype imputation in which some subjects are sequenced and missing genotypes are inferred on the remaining subjects. Because of high cost, it is important to carefully prioritize subjects for sequencing. Here, we introduce a statistical framework that enables systematic comparison among subject-selection choices for sequencing. We introduce a metric "local coverage," which allows the use of inferred inheritance vectors to measure genotype-imputation ability specifically in a region of interest, such as one with prior evidence of linkage. In the absence of linkage information, we can instead use a "genome-wide coverage" metric computed with the pedigree structure. These metrics enable the development of a method that identifies efficient selection choices for sequencing. As implemented in GIGI-Pick, this method also flexibly allows initial manual selection of subjects and optimizes selections within the constraint that only some subjects might be available for sequencing. In the present study, we used simulations to compare GIGI-Pick with PRIMUS, ExomePicks, and common ad hoc methods of selecting subjects. In genotype imputation of both common and rare alleles, GIGI-Pick substantially outperformed all other methods considered and had the added advantage of incorporating prior linkage information. We also used a real pedigree to demonstrate the utility of our approach in identifying causal mutations. Our work enables prioritization of subjects for sequencing to facilitate dissection of the genetic basis of heritable traits.

Power of family-based association designs to detect rare variants in large pedigrees using imputed genotypes.

Recently, the "Common Disease-Multiple Rare Variants" hypothesis has received much attention, especially with current availability of next-generation sequencing. Family-based designs are well suited for discovery of rare variants, with large and carefully selected pedigrees enriching for multiple copies of such variants. However, sequencing a large number of samples is still prohibitive. Here, we evaluate a cost-effective strategy (pseudosequencing) to detect association with rare variants in large pedigrees. This strategy consists of sequencing a small subset of subjects, genotyping the remaining sampled subjects on a set of sparse markers, and imputing the untyped markers in the remaining subjects conditional on the sequenced subjects and pedigree information. We used a recent pedigree imputation method (GIGI), which is able to efficiently handle large pedigrees and accurately impute rare variants. We used burden and kernel association tests, famWS and famSKAT, which both account for family relationships and heterogeneity of allelic effect for famSKAT only. We simulated pedigree sequence data and compared the power of association tests for pseudosequence data, a subset of sequence data used for imputation, and all subjects sequenced. We also compared, within the pseudosequence data, the power of association test using best-guess genotypes and allelic dosages. Our results show that the pseudosequencing strategy considerably improves the power to detect association with rare variants. They also show that the use of allelic dosages results in much higher power than use of best-guess genotypes in these family-based data. Moreover, famSKAT shows greater power than famWS in most of scenarios we considered.

GIGI: an approach to effective imputation of dense genotypes on large pedigrees.

Recent emergence of the common-disease-rare-variant hypothesis has renewed interest in the use of large pedigrees for identifying rare causal variants. Genotyping with modern sequencing platforms is increasingly common in the search for such variants but remains expensive and often is limited to only a few subjects per pedigree. In population-based samples, genotype imputation is widely used so that additional genotyping is not needed. We now introduce an analogous approach that enables computationally efficient imputation in large pedigrees. Our approach samples inheritance vectors (IVs) from a Markov Chain Monte Carlo sampler by conditioning on genotypes from a sparse set of framework markers. Missing genotypes are probabilistically inferred from these IVs along with observed dense genotypes that are available on a subset of subjects. We implemented our approach in the Genotype Imputation Given Inheritance (GIGI) program and evaluated the approach on both simulated and real large pedigrees. With a real pedigree, we also compared imputed results obtained from this approach with those from the population-based imputation program BEAGLE. We demonstrated that our pedigree-based approach imputes many alleles with high accuracy. It is much more accurate for calling rare alleles than is population-based imputation and does not require an outside reference sample. We also evaluated the effect of varying other parameters, including the marker type and density of the framework panel, threshold for calling genotypes, and population allele frequencies. By leveraging information from existing genotypes already assayed on large pedigrees, our approach can facilitate cost-effective use of sequence data in the pursuit of rare causal variants.

Genome scan of age-at-onset in the NIMH Alzheimer disease sample uncovers multiple loci, along with evidence of both genetic and sample heterogeneity.

Alzheimer's disease (AD) is a common neurodegenerative disorder of late life with a complex genetic basis. Although several genes are known to play a role in rare early onset AD, only the APOE gene is known to have a high contribution to risk of the common late-onset form of the disease (LOAD, onset >60 years). APOE genotypes vary in their AD risk as well as age-at-onset distributions, and it is likely that other loci will similarly affect AD age-at-onset. Here we present the first analysis of age-at-onset in the NIMH LOAD sample that allows for both a multilocus trait model and genetic heterogeneity among the contributing sites, while at the same time accommodating age censoring, effects of known genetic covariates, and full pedigree and marker information. The results provide evidence for genomic regions not previously implicated in this data set, including regions on chromosomes 7q, 15, and 19p. They also affirm evidence for loci on chromosomes 1q, 6p, 9q, 11, and, of course, the APOE locus on 19q, all of which have been reported previously in the same sample. The analyses failed to find evidence for linkage to chromosome 10 with inclusion of unaffected subjects and extended pedigrees. Several regions implicated in these analyses in the NIMH sample have been previously reported in genome scans of other AD samples. These results, therefore, provide independent confirmation of AD loci in family-based samples on chromosomes 1q, 7q, 19p, and suggest that further efforts towards identifying the underlying causal loci are warranted.

Joint linkage and segregation analysis under multiallelic trait inheritance: simplifying interpretations for complex traits.

Identification of the genetic basis of common traits may be hindered by underlying complex genetic architectures that are inadequately captured by existing models, including both multiallelic and multilocus modes of inheritance (MOI). One useful approach for localizing genes underlying continuous complex traits is the joint oligogenic linkage and segregation analysis implemented in the package Loki. The method uses reversible jump Markov chain Monte Carlo to eliminate the need to prespecify the number of quantitative trait loci (QTLs) in the trait model, thus providing posterior distributions for the number of QTLs in a Bayesian framework. The current implementation assumes QTLs are diallelic, and therefore can overestimate the number of linked QTLs in the presence of a multiallelic QTL. To address the possibility of multiple alleles, we extended the QTL model to allow for a variable number of additive alleles at each locus. Application to simulated data shows that, under a diallelic MOI, the multiallelic and diallelic analysis models give similar results. Under a multiallelic MOI, the multiallelic analysis model provides better mixing and improved convergence, and leads to a more accurate estimate of the underlying trait MOI and model parameter values, than does the diallelic model. Application to real data shows the multiallelic model results in fewer estimated linked QTLs and that the predominant QTL model is similar to one of two predominant models estimated from the diallelic analysis. Our results indicate that use of a multiallelic analysis model can lead to better understanding of the genetic architecture underlying complex traits.

Empirical significance values for linkage analysis: trait simulation using posterior model distributions from MCMC oligogenic segregation analysis.

Variance-components (VC) linkage analysis is a powerful model-free method for assessing linkage, but the distribution of VC logarithm of the odds ratio (LOD) scores may deviate substantially from the assumed asymptotic distribution. Typically, the null distribution of the VC-LOD score and other linkage statistics has been estimated by generating new genotype data independently of the trait data, and computing a linkage statistic for many such marker-simulated data sets. However, marker simulation is susceptible to errors in the assumed marker and map model and is computationally intensive. Here, we describe a method for generating posterior distributions of linkage statistics through simulation of trait data based on the original sample and on results from an initial scan using a Bayesian Markov-chain Monte Carlo (MCMC) approach for oligogenic segregation analysis. We use samples of oligogenic trait models taken from the posterior distribution to generate new samples of trait data, which were paired with the original marker data for analysis. Empirical P-values obtained from trait and marker simulation were similar when derived for several strong linkage signals from published linkage scans, and for analysis of data with a known, simulated, trait model. Furthermore, trait simulation produces the expected null distribution of VC-LOD scores and is computationally fast when marker identity-by-descent estimates from the original data could be reused. These results suggest that trait simulation gives valid estimates of statistical significance of linkage signals. Finally, these results also demonstrate the feasibility of obtaining empirical significance levels for evaluating Bayesian oligogenic linkage signals with either marker or trait simulation.

MCMC-based linkage analysis for complex traits on general pedigrees: multipoint analysis with a two-locus model and a polygenic component.

We describe a new program lm_twoqtl, part of the MORGAN package, for parametric linkage analysis with a quantitative trait locus (QTL) model having one or two QTLs and a polygenic component, which models additional familial correlation from other unlinked QTLs. The program has no restriction on number of markers or complexity of pedigrees, facilitating use of more complex models with general pedigrees. This is the first available program that can handle a model with both two QTLs and a polygenic component. Competing programs use only simpler models: one QTL, one QTL plus a polygenic component, or variance components (VC). Use of simple models when they are incorrect, as for complex traits that are influenced by multiple genes, can bias estimates of QTL location or reduce power to detect linkage. We compute the likelihood with Markov Chain Monte Carlo (MCMC) realization of segregation indicators at the hypothesized QTL locations conditional on marker data, summation over phased multilocus genotypes of founders, and peeling of the polygenic component. Simulated examples, with various sized pedigrees, show that two-QTL analysis correctly identifies the location of both QTLs, even when they are closely linked, whereas other analyses, including the VC approach, fail to identify the location of QTLs with modest contribution. Our examples illustrate the advantage of parametric linkage analysis with two QTLs, which provides higher power for linkage detection and better localization than use of simpler models.

Multipoint linkage analysis with many multiallelic or dense diallelic markers: Markov chain-Monte Carlo provides practical approaches for genome scans on general pedigrees.

Computations for genome scans need to adapt to the increasing use of dense diallelic markers as well as of full-chromosome multipoint linkage analysis with either diallelic or multiallelic markers. Whereas suitable exact-computation tools are available for use with small pedigrees, equivalent exact computation for larger pedigrees remains infeasible. Markov chain-Monte Carlo (MCMC)-based methods currently provide the only computationally practical option. To date, no systematic comparison of the performance of MCMC-based programs is available, nor have these programs been systematically evaluated for use with dense diallelic markers. Using simulated data, we evaluate the performance of two MCMC-based linkage-analysis programs--lm_markers from the MORGAN package and SimWalk2--under a variety of analysis conditions. Pedigrees consisted of 14, 52, or 98 individuals in 3, 5, or 6 generations, respectively, with increasing amounts of missing data in larger pedigrees. One hundred replicates of markers and trait data were simulated on a 100-cM chromosome, with up to 10 multiallelic and up to 200 diallelic markers used simultaneously for computation of multipoint LOD scores. Exact computation was available for comparison in most situations, and comparison with a perfectly informative marker or interprogram comparison was available in the remaining situations. Our results confirm the accuracy of both programs in multipoint analysis with multiallelic markers on pedigrees of varied sizes and missing-data patterns, but there are some computational differences. In contrast, for large numbers of dense diallelic markers, only the lm_markers program was able to provide accurate results within a computationally practical time. Thus, programs in the MORGAN package are the first available to provide a computationally practical option for accurate linkage analyses in genome scans with both large numbers of diallelic markers and large pedigrees.

Segregation analysis of a complex quantitative trait: approaches for identifying influential data points.

BACKGROUND/AIMS: Complex traits pose a particular challenge to standard methods for segregation analysis (SA), and for such traits it is difficult to assess the ability of complex SA (CSA) to approximate the true mode of inheritance. Here we use an oligogenic Bayesian Markov chain Monte Carlo method for SA (OSA) to verify results from a single-locus likelihood-based CSA for data on a quantitative measure of reading ability. METHODS: We compared the profile likelihood from CSA, maximized over the trait allele frequency, to the posterior distribution of genotype effects from OSA to explore differences in the overall parameter estimates from SA on the original phenotype data and the same data Winsorized to reduce the potential influence of three outlying data points. RESULTS: Bayesian OSA revealed two modes of inheritance, one of which coincided with the QTL model from CSA. Winsorizing abolished the model originally estimated by CSA; both CSA and OSA identified only the second OSA model. CONCLUSION: Differences between the results from the two methods alerted us to the presence of influential data points, and identified the QTL model best supported by the data. Thus, the Bayesian OSA proved a valuable tool for assessing and verifying inheritance models from CSA.

Genome scan for quantitative trait loci influencing HDL levels: evidence for multilocus inheritance in familial combined hyperlipidemia.

Several genome scans in search of high-density lipoprotein (HDL) quantitative trait loci (QTLs) have been performed. However, to date the actual identification of genes implicated in the regulation of common forms of HDL abnormalities remains unsuccessful. This may be due, in part, to the oligogenic and multivariate nature of HDL regulation, and potentially, pleiotropy affecting HDL and other lipid-related traits. Using a Bayesian Markov Chain Monte Carlo (MCMC) approach, we recently provided evidence of linkage of HDL level variation to the APOA1-C3-A4-A5 gene complex, in familial combined hyperlipidemia pedigrees, with an estimated number of two to three large QTLs remaining to be identified. We also presented results consistent with pleiotropy affecting HDL and triglycerides at the APOA1-C3-A4-A5 gene complex. Here we use the same MCMC analytic strategy, which allows for oligogenic trait models, as well as simultaneous incorporation of covariates, in the context of multipoint analysis. We now present results from a genome scan in search for the additional HDL QTLs in these pedigrees. We provide evidence of linkage for additional HDL QTLs on chromosomes 3p14 and 13q32, with results on chromosome 3 further supported by maximum parametric and variance component LOD scores of 3.0 and 2.6, respectively. Weaker evidence of linkage was also obtained for 7q32, 12q12, 14q31-32 and 16q23-24.

MCMC multilocus lod scores: application of a new approach.

On extended pedigrees with extensive missing data, the calculation of multilocus likelihoods for linkage analysis is often beyond the computational bounds of exact methods. Growing interest therefore surrounds the implementation of Monte Carlo estimation methods. In this paper, we demonstrate the speed and accuracy of a new Markov chain Monte Carlo method for the estimation of linkage likelihoods through an analysis of real data from a study of early-onset Alzheimer's disease. For those data sets where comparison with exact analysis is possible, we achieved up to a 100-fold increase in speed. Our approach is implemented in the program lm_bayes within the framework of the freely available MORGAN 2.6 package for Monte Carlo genetic analysis (http://www.stat.washington.edu/thompson/Genepi/MORGAN/Morgan.shtml).

Comparison of marker types and map assumptions using Markov chain Monte Carlo-based linkage analysis of COGA data.

We performed multipoint linkage analysis of the electrophysiological trait ECB21 on chromosome 4 in the full pedigrees provided by the Collaborative Study on the Genetics of Alcoholism (COGA). Three Markov chain Monte Carlo (MCMC)-based approaches were applied to the provided and re-estimated genetic maps and to five different marker panels consisting of microsatellite (STRP) and/or SNP markers at various densities. We found evidence of linkage near the GABRB1 STRP using all methods, maps, and marker panels. Difficulties encountered with SNP panels included convergence problems and demanding computations.

Joint oligogenic segregation and linkage analysis using bayesian Markov chain Monte Carlo methods.

One of the most challenging areas in human genetics is the dissection of quantitative traits. In this context, the efficient use of available data is important, including, when possible, use of large pedigrees and many markers for gene mapping. In addition, methods that jointly perform linkage analysis and estimation of the trait model are appealing because they combine the advantages of a model-based analysis with the advantages of methods that do not require prespecification of model parameters for linkage analysis. Here we review a Markov chain Monte Carlo approach for such joint linkage and segregation analysis, which allows analysis of oligogenic traits in the context of multipoint linkage analysis of large pedigrees. We provide an outline for practitioners of the salient features of the method, interpretation of the results, effect of violation of assumptions, and an example analysis of a two-locus trait to illustrate the method.

Linkage analyses of four regions previously implicated in dyslexia: confirmation of a locus on chromosome 15q.

Dyslexia is a common, complex disorder, which is thought to have a genetic component. There have been numerous reports of linkage to several regions of the genome for dyslexia and continuous dyslexia-related phenotypes. We attempted to confirm linkage of continuous measures of (1) accuracy and efficiency of phonological decoding; and (2) accuracy of single word reading (WID) to regions on chromosomes 2p, 6p, 15q, and 18p, using 111 families with a total of 898 members. We used both single-marker and multipoint variance components linkage analysis and Markov Chain Monte Carlo (MCMC) joint segregation and linkage analysis for initial inspection of these regions. Positive results were followed with traditional parametric lod score analysis using a model estimated by MCMC segregation analysis. No positive linkage signals were found on chromosomes 2p, 6p, or 18p. Evidence of linkage of WID to chromosome 15q was found with both methods of analysis. The maximum single-marker parametric lod score of 2.34 was obtained at a distance of 3 cM from D15S143. Multipoint analyses localized the putative susceptibility gene to the interval between markers GATA50C03 and D15S143, which falls between a region implicated in a recent genome screen for attention-deficit/hyperactivity disorder, and DYX1C1, a candidate gene for dyslexia. This apparent multiplicity of linkage signals in the region for developmental disorders may be the result of errors in map and/or model specification obscuring the pleiotropic effect of a single gene on different phenotypes, or it may reflect the presence of multiple genes.

Evidence for a novel late-onset Alzheimer disease locus on chromosome 19p13.2.

Late-onset familial Alzheimer disease (LOFAD) is a genetically heterogeneous and complex disease for which only one locus, APOE, has been definitively identified. Difficulties in identifying additional loci are likely to stem from inadequate linkage analysis methods. Nonparametric methods suffer from low power because of limited use of the data, and traditional parametric methods suffer from limitations in the complexity of the genetic model that can be feasibly used in analysis. Alternative methods that have recently been developed include Bayesian Markov chain-Monte Carlo methods. These methods allow multipoint linkage analysis under oligogenic trait models in pedigrees of arbitrary size; at the same time, they allow for inclusion of covariates in the analysis. We applied this approach to an analysis of LOFAD on five chromosomes with previous reports of linkage. We identified strong evidence of a second LOFAD gene on chromosome 19p13.2, which is distinct from APOE on 19q. We also obtained weak evidence of linkage to chromosome 10 at the same location as a previous report of linkage but found no evidence for linkage of LOFAD age-at-onset loci to chromosomes 9, 12, or 21.

Summary of Group 8: Development and extension of linkage methods.

The diverse contributions to Group 8 focus on the development and extension of linkage methods, and share themes that are organized around approaches to analysis. The themes discussed include issues of the accuracy of estimates of marker identity-by-descent (IBD) information, the influence of such IBD information on linkage detection, and methods for dealing with genetic heterogeneity and multiple testing. In addition, challenges were identified and solutions were offered for coping with some of the common problems of complex trait analysis, including trait model selection and computational compromises. Analytic approaches based on Bayesian and Monte Carlo methods were prominent, and provided optimistic results.

Oligogenic segregation analysis of hereditary prostate cancer pedigrees: evidence for multiple loci affecting age at onset.

Previous studies have suggested strong evidence for a hereditary component to prostate cancer (PC) susceptibility. Here, we analyze 3,796 individuals in 263 PC families recruited as part of the ongoing Prostate Cancer Genetic Research Study (PROGRESS). We use Markov chain Monte Carlo (MCMC) oligogenic segregation analysis to estimate the number of quantitative trait loci (QTLs) and their contribution to the variance in age at onset of hereditary PC (HPC). We estimate 2 covariate effects: diagnosis of PC before and after prostate-specific antigen (PSA) test availability, and presence/absence of at least 1 blood relative with primary neuroepithelial brain cancer (BC). We find evidence that 2 to 3 QTLs contribute to the variance in age at onset of HPC. The 2 QTLs with the largest contribution to the total variance are both effectively dominant loci. We find that the covariate for diagnosis before and after PSA test availability is important. Our findings for the number of QTLs contributing to HPC and the variance contribution of these QTLs will be instructive in mapping and identifying these genes.

A score for Bayesian genome screening.

Bayesian Monte Carlo Markov chain (MCMC) techniques have shown promise in dissecting complex genetic traits. The methods introduced by Heath ([1997], Am. J. Hum. Genet. 61:748-760), and implemented in the program Loki, have been able to localize genes for complex traits in both real and simulated data sets. Loki estimates the posterior probability of quantitative trait loci (QTL) at locations on a chromosome in an iterative MCMC process. Unfortunately, interpretation of the results and assessment of their significance have been difficult. Here, we introduce a score, the log of the posterior placement probability ratio (LOP), for assessing oligogenic QTL detection and localization. The LOP is the log of the posterior probability of linkage to the real chromosome divided by the posterior probability of linkage to an unlinked pseudochromosome, with marker informativeness similar to the marker data on the real chromosome. Since the LOP cannot be calculated exactly, we estimate it in simultaneous MCMC on both real and pseudochromosomes. We investigate empirically the distributional properties of the LOP in the presence and absence of trait genes. The LOP is not subject to trait model misspecification in the way a lod score may be, and we show that the LOP can detect linkage for loci of small effect when the lod score cannot. We show how, in the absence of linkage, an empirical distribution of the LOP may be estimated by simulation and used to provide an assessment of linkage detection significance.