PRIMe: a method for characterization and evaluation of pleiotropic regions from multiple genome-wide association studies (original) (raw)

Bioinformatics. 2011 May 1; 27(9): 1201–1206.

Jie Huang

1National Heart, Lung, and Blood Institute's Framingham Heart Study, Framingham, MA 01702, 2Division of Intramural Research, National Heart, Lung and Blood Institute, Bethesda, MD 20824 and 3Cardiology Division, Department of Medicine, Massachusetts General Hospital, Harvard Medical School, Boston, MA 02114, USA

Andrew D. Johnson

Christopher J. O'Donnell

*To whom correspondence should be addressed.

Associate Editor: Martin Bishop

Received 2010 Oct 27; Revised 2011 Feb 8; Accepted 2011 Feb 24.

Supplementary Materials

Supplementary Data

GUID: AC7BEEC7-4532-4ED9-832F-9C3DF250EECF

GUID: 27F7080E-23DF-4FC1-8D10-F208A3F515E8

GUID: 47306F37-013D-4B7A-84F9-AE7BA11EE80E

Abstract

Motivation: The concept of pleiotropy was proposed a century ago, though up to now there have been insufficient efforts to design robust statistics and software aimed at visualizing and evaluating pleiotropy at a regional level. The Pleiotropic Region Identification Method (PRIMe) was developed to evaluate potentially pleiotropic loci based upon data from multiple genome-wide association studies (GWAS).

Methods: We first provide a software tool to systematically identify and characterize genomic regions where low association _P_-values are observed with multiple traits. We use the term Pleiotropy Index to denote the number of traits with low association _P_-values at a particular genomic region. For GWAS assumed to be uncorrelated, we adopted the binomial distribution to approximate the statistical significance of the Pleiotropy Index. For GWAS conducted on traits with known correlation coefficients, simulations are performed to derive the statistical distribution of the Pleiotropy Index under the null hypothesis of no genotype–phenotype association. For six hematologic and three blood pressure traits where full GWAS results were available from the Cohorts for Heart and Aging Research in Genomic Epidemiology (CHARGE) Consortium, we estimated the trait correlations and applied the simulation approach to examine genomic regions with statistical evidence of pleiotropy. We then applied the approximation approach to explore GWAS summarized in the National Human Genome Research Institute (NHGRI) GWAS Catalog.

Results: By simulation, we identified pleiotropic regions including SH2B3 and BRAP (12q24.12) for hematologic and blood pressure traits. By approximation, we confirmed the genome-wide significant pleiotropy of these two regions based on the GWAS Catalog data, together with an exploration on other regions which highlights the FTO, GCKR and ABO regions.

Availability and Implementation: The Perl and R scripts are available at http://www.framinghamheartstudy.org/research/gwas_pleiotropictool.html.

Contact: vog.hin.iblhn@cllennodo

Supplementary information: Supplementary data are available at Bioinformatics online.

1 INTRODUCTION

Pleiotropy describes the effect of a single genetic region on multiple phenotypic traits. The concept of pleiotropy was first defined a century ago as one mutation resulting in multiple distinct phenotypes (Stearns, 2010). Among the proposed mechanisms, the genetic region may encode a product that is used by multiple cell types, or it may have a signaling function affecting multiple targets, or the traits under study may themselves be highly inter-related at a physiological level (Hodgkin, 1998; Pyeritz, 1989). Pleiotropic genes have been shown in other species to affect environmental adaptation and tend to reside in central node positions in protein–protein interaction networks (Foster et al., 2004; Zou et al., 2008). The identification and characterization of pleiotropic genes and regions offers a unique window into the complexities of biological molecular interaction networks, and may potentially indicate evidence for epistasis (Tyler et al., 2009).

The presence and impact of pleiotropy in genome data for normal human characteristics and human disease traits merits further investigation, but thus far the efforts at statistical development are insufficient. Methods and software for analyzing multivariate phenotypes have been proposed (Ferreira and Purcell, 2009; Lange et al., 2003; Liu et al., 2009; Yang et al., 2010). However, the available methods, such as those by Ferreira and Purcell (2009), Lange et al. (2003) and Liu et al. (2009), all require use of individual-level phenotype data and thus cannot be used to study pleiotropy using only existing summarized genome-wide association studies (GWAS) results. The method of Yang et al. (2010) can be used on existing summarized GWAS results; however this method only considers single nucleotide polymorphism (SNP) level but not region-level pleiotropy.

A novel analytic approach was recently demonstrated to examine pleiotropic genes in psychiatric phenotypes (Huang et al., 2010). The current abundance of GWAS results provides an unprecedented opportunity to fully examine this phenomenon in a systematic manner. We previously analyzed results across 118 GWAS articles published from 2005 through 2008, creating a comprehensive database of 56 411 SNP-phenotype associations at a significance level of ≤0.001 (Johnson and O'Donnell, 2009). This study highlighted potential pleiotropic regions, presenting the 61 densest regions of associations from 118 GWAS. Polymorphisms associated with multiple traits in genes including APOE and the MHC region were identified, as well as novel candidates (PIGU, RAPGEF1, COL4A1/2 and OAS1) that were subsequently replicated in other studies (Johnson and O'Donnell, 2009). Another resource for the collection of significant GWAS associations is the National Human Genome Research Institute's Catalog of Published Genome-Wide Association Studies (GWAS Catalog), which is regularly updated. A summary of results from 151 (of 237) published GWAS through December 2008 described 531 replicated SNP-trait associations and highlighted 18 regions of association with two or more traits (Hindorff et al., 2009). Since that time GWAS have continued to be published at a rapid pace, as of October 16, 2010, including 584 GWAS and 4054 SNP-trait associations reported in the GWAS Catalog. Other repositories of GWAS-related data exist such as the database of Genotypes and Phenotypes (dbGAP) but thus far these have not been widely applied to address hypotheses relating to pleiotropy (Mailman et al., 2007).

We developed the Pleiotropic Region Identification Method (PRIMe) and applied it to cardiovascular-related traits from the Cohorts for Heart and Aging Research in Genomic Epidemiology (CHARGE) Consortium (Psaty et al., 2009). We provide an initial example involving six hematologic traits (hemoglobin concentration, hematocrit, mean corpuscular hemoglobin, mean corpuscular hemoglobin concentration, mean corpuscular volume and red blood cell count) and three blood pressure traits (systolic blood pressure, diastolic blood pressure and hypertension) used in two independent GWAS meta-analysis reports from the CHARGE Consortium (Ganesh et al., 2009; Levy et al., 2009). We further evaluated an approximation approach to estimate the statistical distribution of the Pleiotropy Index for uncorrelated traits, using data reported in the publicly accessible GWAS Catalog.

2 METHODS

2.1 PRIMe visualization

We implemented the PRIMe tool to systematically identify and characterize regions across multiple GWAS/traits. Here we use the terms ‘GWAS’ and ‘trait’ interchangeably when one GWAS reports association results for one phenotypic trait. However, the CHARGE datasets we used for pleiotropic analysis include two overall GWAS studies with a total of nine traits. The PRIMe visualization method defines a genomic region of interest out of the whole genome as follows: let P S denote the threshold for association significance of SNPs, which can be user defined. Let r denote the correlation coefficient between an SNP pair, measured as the square root of the linkage disequilibrium (LD) measure of _r_2. PRIMe iteratively finds SNPs with the lowest association _P_-value among all traits as the driver, and SNPs whose _r_2 with the driver is above the user-specified threshold (≥0.8 by default) as passengers. In order to define distinct regions out of a genome with extensive LD patterns, once a SNP is designated as a passenger, it will not be considered again as a new driver or passenger. After completion of this iterative process, regions are defined by one driver and zero or more passengers.

2.2 Statistical evaluation

To follow-up regions identified by the PRIMe visualization tool, we conducted statistical evaluation for scenarios ranging from uncorrelated to highly correlated traits. We use the term Pleiotropy Index to denote the number of traits with low association _P_-values at a particular genomic region. We estimated the statistical distribution of the Pleiotropy Index under the null hypothesis of no genotype–phenotype association for any of the traits. Let P T denote the probability of the reaching a certain Pleiotropy Index value in a genomic region. Since the calculation of P T depends on the particular LD pattern of a genomic region, which varies substantially across the genome, the same Pleiotropy Index value at different genomic regions would yield different P T, and thus are not comparable. Therefore, we do not attempt to address the probability of attaining a Pleiotropy Index for any region across the genome per se. Instead, we assume that a region of interest has been identified through basic PRIMe queries or other means, and the main objective is to evaluate statistical significance for that particular region. We evaluated the simulated or approximated P T value against the genome-wide significance threshold of 5 × 10−8 to assess pleiotropy (Pe'er et al., 2008).

2.2.1 For uncorrelated traits: the binomial approximation

When GWAS are conducted across different cohorts with non-overlapping samples, they can be assumed to be uncorrelated, under the assumption that there is no genotype–phenotype association for each individual GWAS. The number of independent SNPs within a genomic region could also be derived using the pairwise tagging approach (by default, _r_2 threshold of 0.8) (de Bakker et al., 2005). For K uncorrelated traits and M independent SNPs, the statistical distribution of the Pleiotropy Index can be mathematically derived as follows. In a region with M independent SNPs, the number of SNPs with association _P_-values below P S follows a binomial distribution B(n, p), with n equal to M and p equal to P S. Therefore, the probability that at least one out of M SNPs with association _P_-value below P S is equal to 1 − (1 − P S)M because (1 − P S)M is the probability that none of the SNPs have a _P_-value below the threshold. For K uncorrelated traits each with M SNPs in the region, the number of traits with at least one SNP below P S also follows a binomial distribution B(n, p), this time with n equal to K and p equal to 1 − (1 − P S)M. Therefore, for uncorrelated traits and independent SNPs, P T can be approximated from a simple binomial approach.

2.2.2 For correlated traits: the multivariate simulation

In the backdrop of deep phenotyping and large consortia, GWAS scans could be correlated due to both overlapping samples and similar phenotypic measurements. We estimated the overall trait correlation by calculating the correlation of _z_-statistics (β/SE) for all common SNPs (e.g. ∼2.5 million SNPs based on HapMap2 imputation) in a number of GWAS from consortia (Table 1, Supplementary Tables S1 and S2). For K correlated traits in a region of M SNPs with known LD, the test statistics (z) of each SNP for each trait follows approximately a multivariate normal distribution (Conneely and Boehnke, 2007).

Table 1.

Correlation of six hematologic and three blood pressure traits

P	Hb	HCT	MCH	MCHC	MCV	RBC	DBP	SBP	HTN
Hb	1.00	0.88	0.17	0.18	0.12	0.54	0.11	0.09	0.07
HCT	1.00	0.06	−0.03	0.12	0.55	0.1	0.08	0.06
MCH	1.00	0.37	0.64	−0.23	0.02	0.00	0.00
MCHC	1.00	−0.01	0.07	0.04	0.03	0.02
MCV	1.00	−0.2	−0.02	−0.02	−0.03
RBC	1.00	0.07	0.05	0.04
DBP	1.00	0.72	0.56
SBP	1.00	0.64
HTN	1.00

The probability that a trait indexed by k has at least one SNP with a _P_-value below the threshold of P S is equivalent to the probability that the maximum of the absolute value of z _k_1, z _k_2,… z kM statistics is greater than z s (the _z_-value corresponding to the P S/2 two-sided threshold). Theoretically, this probability can be derived from a multiple dimension integral of the multivariate normal density function of the _z_-statistics. However, this multiple dimension integral is challenging to evaluate numerically when the number of SNPs is large. Moreover, using the multivariate normal approximation from one GWAS to derive the null distribution of the Pleiotropy Index would require taking the correlation among the GWAS into consideration. Therefore, we take a simulation approach to evaluate the distribution of the Pleiotropy Index. We simulate the test statistics for all traits and all SNPs simultaneously, and to do so we use the fact that the correlation matrix between all K × M test statistics is the (kronecker) product of the M dimensional correlation matrix between SNPs (r) and the K dimensional correlation matrix of the traits analyzed (ρ) (Fig. 1).

Mathematical derivation of the test statistics correlation for K traits and M SNPs. This is based on the assumption that there is no genotype–phenotype association for any of the traits.

3 RESULTS

3.1 PRIMe visualization

Figure 2 shows adjacent regions within the 12q24 region, which were highlighted by PRIMe for association with blood pressure and hematologic traits in the CHARGE Consortium. In Figure 2, the SNP in blue is designated as the driver SNP due to its lowest association _P_-value among all traits. The driver SNP can also be associated with other traits (shown in blue with less significant _P_-values). The SNPs shown in bright red are designated as passengers. The SNPs shown in weaker red are those whose _r_2 with the driver SNP do not reach the user-specified threshold (0.8 by default).

PRIMe plot of the SH2B3 region and the BRAP region for GWAS of six hematologic traits and three blood pressure traits. The SH2B3 region (a) and the BRAP region (b). The driver SNPs (shown in blue) are rs10774625 and rs11065987, respectively. A horizontal red line indicates the log of genome-wide significance _P_-value of 5 × 10−8. The longest isoforms of RefSeq genes in the region are annotated including introns (green) and exons (black). The legend lists traits in the format of first_author(PubMed ID)(trait_name)[group_name].

3.2 For correlated traits: the multivariate simulation

Table 1 shows the trait correlation matrix for nine hematologic and blood pressure traits from the CHARGE Consortium data. Correlation coefficients for these nine traits range widely from 0 to 0.88. To further illustrate the cause and extent of trait correlation, we calculated the pairwise correlation for two other publically available sets of data: the Wellcome Trust Case Control Consortium (WTCCC) and Global Lipids, as shown in Supplementary Tables S1 and S2 respectively (WTCCC 2007; Teslovich et al., 2010). The latter two datasets are used only as references on trait correlation and not for pleiotropy analysis. We recommend that users of PRIMe consider any pair of traits with correlations <0.1 as largely uncorrelated.

For the nine hematologic and blood pressure traits, the two regions shown in Figure 2 (referred to as the SH2B3 region and BRAP region hereafter) have the highest Pleiotropy Index of 4 among all identified regions with a Pleiotropy Index ≥2. For these nine traits in the SH2B3 region, we performed 200 million simulations, and found only one instance where four of the nine traits all had at least one SNP with simulated association _P_-value (derived from the _z_-statistic) less than the pre-specified threshold of 1 × 10−5. Therefore, we conclude that for these nine traits and the SH2B3 region, based on a P S of 1 × 10−5, the simulated P T value for the Pleiotropy Index of 4 is equal to 1/(2 × 108) = 5.0 × 10−9 (Supplementary Figure S1, left panel). Similarly, for these nine traits and using GWAS results from the BRAP region to estimate the SNP correlation, we performed 2 × 108 simulations and found six instances where four of the nine traits all had at least one SNP with simulated association _P_-value below the threshold. Therefore, the simulated P T value for the Pleiotropy Index of 4 is equal to 6/(2 × 108) = 3.0 × 10−8 for the BRAP region (Supplementary Figure S1, right panel). Although the Pleiotropy Index is the same (i.e. 4), for both regions (out of a total of nine traits analyzed), the simulated P T for the BRAP region is less significant than that for the SH2B3 region, because the former resides in a longer haplotype block and, therefore, has a higher probability of observing more traits associated with SNPs in the region by chance. We conclude there is evidence that the proposed pleiotropy for these two regions is significant, because P T is estimated to be < 5.0 × 10−8.

The SH2B3 region was recently associated in a single report with platelet count, coronary artery disease, type 1 diabetes and celiac disease validating its likely pleiotropic effects (Soranzo et al., 2009). In this region, the SNP rs3185404 is non-synonymous and predicted to disrupt protein function. Furthermore, SH2B3 is an important signaling protein and thus it is likely to affect cellular functions in a variety of human tissues (Barrett et al., 2009; Hunt et al., 2008).

3.3 For uncorrelated traits: the binomial approximation

We applied the PRIMe tool to the largest GWAS repository publicly available, the GWAS Catalog. For the same SH2B3 and BRAP regions, Pleiotropy Indices of nine were observed for both regions (Supplementary Figures S2 and S3, respectively). There are no full-scale genome-wide z statistics to calculate trait correlations. Instead, we assume traits are not correlated for the 584 GWAS reported as of October 16, 2010. Given the independent number of SNPs (4 for the SH2B3 region and 11 for the BRAP region), the calculated P T is equal to 5.24 × 10−21 and 4.53 × 10−17, respectively, based on the binomial approximation approach described above. Therefore, both regions are also deemed to demonstrate significant pleiotropy based on the GWAS Catalog data. A systematic analysis of data in the entire GWAS Catalog found that a total of 57 regions have a Pleiotropy Index of ≥5. The top regions include the FTO gene and the GCKR region (Supplementary Figures S4 and S5), both with a Pleiotropy Index of 18, while the well-known pleiotropic ABO region has a Pleiotropy Index of 9 (Supplementary Figure S6).

4 DISCUSSION

PRIMe does not rely on prespecified methods for genomic region grouping such as gene boundaries or physical position bins. Instead, association _P_-values for individual SNPs and LD between SNPs are used to define genomic regions where multiple traits show significant association with SNPs. Since pairwise LD (r) among SNPs is a key determinant of the width of a region; we used PLINK to precalculate LD based on the HapMap II + III samples from the CEU (Utah residents with ancestry from northern and western Europe) and TSI (Toscans in Italy) populations (Frazer et al., 2007; Purcell et al., 2007). We did not use the LD data downloadable from HapMap website, because it calculates LD only for SNPs up to 250 kb apart (Thorisson et al., 2005). To capture long-range haplotype blocks, we include LD for SNPs up to 1500 kb apart. Of note, the LD threshold chosen by investigators will not only affect the physical boundaries of a putative pleiotropic region, but also the statistical significance of the estimated P T.

For the SH2B3 and BRAP regions, if a LD threshold of _r_2 ≥ 0.5 rather than 0.8 is used, only one large region instead of two separate ones will be identified. This change of LD threshold will further complicate the choice of 5 × 10−8 as the threshold for evaluating genome-wide significance, since the total number of independent regions across the genome would be different when ‘high LD’ is defined as _r_2 ≥ 0.5 instead of 0.8. Furthermore, with a P S threshold of 1 × 10−3 instead of 1 × 10−5 prespecified, the same SH2B3 region is noted with a Pleiotropy Index of 6 but the simulated P T would not have reached a genome-wide significant threshold. From a statistical testing perspective, higher values of Pleiotropy Index under less stringent P S thresholds tend to result in a less significant P T. On the other hand, a highly stringent P S threshold will pose technical challenges for simulation. We performed 200 million simulations (i.e. 10/(5 × 10−8)) in an attempt to simulate the distribution for P S up to 5 × 10−8. However, as Supplementary Figure S1 shows, despite the high number of simulations, we still did not observe a Pleiotropy Index > 2 when P S thresholds of 1 × 10−7 and 5 × 10−8 were used. Therefore, an exact P T could not be specified other than P T < 1/(2 × 108) = 5 × 10−9. We suggest a P S of 1 × 10−5, which corresponds to the significance threshold currently used by the GWAS Catalog. Currently the direction of association effect is not considered for computing the Pleiotropy Index. Studies have suggested there may be biologically plausible mechanisms for different effect directions even among closely related phenotypes (Sirota et al., 2009).

For the approximation approach, we make the assumption that traits are not correlated. When traits studied in separate GWAS scans are not truly uncorrelated, the distribution of the Pleiotropy Index deviates from the binomial distribution, with higher proportions of correlated traits leading to greater deviations. As the correlation coefficient increases, the expected Pleiotropy Index increases and therefore is more likely (at any given significant P T) to achieve larger values of the Pleiotropy Index (see Supplementary Figure S7). For the two CHARGE GWAS, five common cohorts contributed to a total of six cohorts for both GWAS, with the degree of sample overlap for these two GWAS being ∼80%. However, a high degree of sample overlap alone does not imply that there will be high correlation for quantitative traits [e.g. 0.00 between mean corpuscular hemoglobin and systolic blood pressure for the CHARGE traits (Table 1), and −0.08 between HDL and LDL for the Global Lipids traits (Supplementary Table S2)]. For dichotomous traits illustrated by the WTCCC data, the use of a common control dataset generates an explicit overlap of phenotypic measurement, therefore leading to some degree of correlation among GWAS results for the seven traits (from 0.29 to 0.43, Supplementary Table S1). The highest correlation of 0.43 between type 1 diabetes and rheumatoid arthritis could also indicate common pathways for auto-immune diseases (Supplementary Table S1). For both quantitative and qualitative traits, under the null hypothesis of no genotype–phenotype association for each individual GWAS, even identically ascertained phenotypes will not yield trait correlation given no sample overlap.

The finding of the FTO region highlights how care must be taken in interpretation of conclusions on pleiotropy, since the commonly associated traits for the FTO region include many weight-related measures or for type 2 diabetes. Although a few of the association signals in the GCKR region are for similar serum lipids and glucose measures as well, there are distinct traits such as hematologic traits, serum uric acid and C-reactive protein. The significant test statistic in the case of FTO may be flagging either pleiotropy or further independent replication for adiposity traits, while the case of GCKR may reflect classic pleiotropy. For similar but not identical traits, suspected pleiotropy could be due to independent effects or through a chain of mediation. The identification of mediation can still provide useful biological information. A recent study on the genetics of FTO provides the first direct evidence that increased FTO expression causes obesity in mice. Mice with increased FTO expression on a high-fat diet develop glucose intolerance. The causal mechanism of FTO on human phenotypes warrants further gene expression profiling studies (Church et al., 2010). The null hypothesis of no genotype–phenotype association for each individual GWAS has been used as a basis for our simulation and approximation approaches, which could be invalid when traits with strong association with particular genomic regions have been well established and extensively replicated. Deriving the distribution under the alternative will depend on many variables (number of associated loci, strength of association with each trait, etc.) and could require a markedly different approach. We compared the power between our method and that of Ferreira and Purcell (2009), and found that their method is slightly more powerful under scenarios with various association _P_-value thresholds and type I error. However, our tool and method can be rapidly applied to summary GWAS results, together with other features including visualization.

In summary, we have developed a computational tool and statistical method to systematically identify, characterize and evaluate pleiotropic regions. PRIMe can be downloaded and run on local machines without uploading potentially sensitive GWAS data to public servers. The scripts are designed in a user friendly manner so that all customizable parameters can be specified in an interface script, which calls subsequent scripts that require no user manipulation. The availability of PRIMe may encourage sharing of more detailed GWAS data in a secure manner within collaborating groups in order to conduct discovery of potential pleiotropic regions. We anticipate that fine-mapping and re-sequencing will enable a higher resolution view of pleiotropy, and further refine whether the same or distinct alleles in targeted regions contribute to multiple traits (Kitsios and Zintzaras, 2009). We have recently used PRIMe to help prioritize the selection of regions for sequencing in following up GWAS, gaining more potential return for smaller resource investments by focusing on regions with evidence for multiple traits. The value of this approach would be greatly improved by broader sharing of complete GWAS results among members of the scientific community with appropriate safeguards (Lumley and Rice, 2010), and the improvements in resources with more dense maps of SNPs provided by the 1000 Genomes and other sequencing projects that are underway. This new approach, together with efforts that pull together GWAS data into publicly accessible centralized repositories, is well suited to propel the study of the genetic architecture of complex diseases beyond individual investigators or individual datasets (Hindorff et al., 2009; Johnson and O'Donnell, 2009).

Supplementary Material

ACKNOWLEDGEMENTS

We acknowledge the invaluable expertise on the statistical simulation and power evaluation from Dr Josée Dupuis and Dr Qiong Yang at the Boston University Department of Biostatistics. This research was conducted in part using data and resources from the Framingham Heart Study of the National Heart Lung and Blood Institute of the National Institutes of Health and Boston University School of Medicine. The analyses reflect intellectual input and resource development from the Framingham Heart Study investigators participating in the SNP Health Association Resource (SHARe) project. The authors acknowledge the essential role of the Cohorts for Heart and Aging Research in Genome Epidemiology (CHARGE) Consortium in development and support of this manuscript, especially acknowledging members from the CHARGE Consortium Blood Pressure and Hematological Working Groups for making their GWAS data fully available (Acknowledgement Section in Supplementary Material).

Funding: This project is supported by the National Heart, Lung and Blood Institute, Division of Intramural Research.

Conflict of Interest: none declared.

REFERENCES

Barrett J.C., et al. Genome-wide association study and meta-analysis find that over 40 loci affect risk of type 1 diabetes. Nat Genet. 2009;41:703–707. [PMC free article] [PubMed] [Google Scholar]
Church C., et al. Overexpression of Fto leads to increased food intake and results in obesity. Nat. Genet. 2010;42:1086–1092. [PMC free article] [PubMed] [Google Scholar]
Conneely K.N., Boehnke M. So Many Correlated Tests, So Little Time! Rapid Adjustment of P Values for Multiple Correlated Tests. Am. J. Hum. Genet. 2007;81:1158–1168. [PMC free article] [PubMed] [Google Scholar]
de Bakker P.I., et al. Efficiency and power in genetic association studies. Nat. Genet. 2005;37:1217–1223. [PubMed] [Google Scholar]
Ferreira M.A., Purcell S. M. A multivariate test of association. Bioinformatics. 2009;25:132–133. [PubMed] [Google Scholar]
Foster K.R., et al. Pleiotropy as a mechanism to stabilize cooperation. Nature. 2004;431:693–696. [PubMed] [Google Scholar]
Frazer K.A., et al. A second generation human haplotype map of over 3.1 million SNPs. Nature. 2007;449:851–861. [PMC free article] [PubMed] [Google Scholar]
Ganesh S.K., et al. Multiple loci influence erythrocyte phenotypes in the CHARGE Consortium. Nat. Genet. 2009;41:1191–1198. [PMC free article] [PubMed] [Google Scholar]
Hindorff L.A., et al. Potential etiologic and functional implications of genome-wide association loci for human diseases and traits. Proc. Natl Acad. Sci. USA. 2009;106:9362–9367. [PMC free article] [PubMed] [Google Scholar]
Hodgkin J. Seven types of pleiotropy. Int. J. Dev. Biol. 1998;42:501–505. [PubMed] [Google Scholar]
Huang J., et al. Cross-disorder genomewide analysis of schizophrenia, bipolar disorder, and depression. Am. J. Psychiatry. 2010;167:1254–1263. [PMC free article] [PubMed] [Google Scholar]
Hunt K.A., et al. Newly identified genetic risk variants for celiac disease related to the immune response. Nat. Genet. 2008;40:395–402. [PMC free article] [PubMed] [Google Scholar]
Johnson A.D., O'Donnell C.J. An open access database of genome-wide association results. BMC Med. Genet. 2009;10:6. [PMC free article] [PubMed] [Google Scholar]
Kitsios G.D., Zintzaras E. Genomic convergence of genome-wide investigations for complex traits. Ann. Hum. Genet. 2009;73(Pt 5):514–519. [PMC free article] [PubMed] [Google Scholar]
Lange C., et al. A multivariate family-based association test using generalized estimating equations: FBAT-GEE. Biostatistics. 2003;4:195–206. [PubMed] [Google Scholar]
Levy D., et al. Genome-wide association study of blood pressure and hypertension. Nat. Genet. 2009;41:677–687. [PMC free article] [PubMed] [Google Scholar]
Liu J., et al. Bivariate association analyses for the mixture of continuous and binary traits with the use of extended generalized estimating equations. Genet. Epidemiol. 2009;33:217–227. [PMC free article] [PubMed] [Google Scholar]
Lumley T., Rice K. Potential for revealing individual-level information in genome-wide association studies. JAMA. 2010;303:659–660. [PubMed] [Google Scholar]
Mailman M.D., et al. The NCBI dbGaP database of genotypes and phenotypes. Nat. Genet. 2007;39:1181–1186. [PMC free article] [PubMed] [Google Scholar]
Pe'er I., et al. Estimation of the multiple testing burden for genomewide association studies of nearly all common variants. Genet. Epidemiol. 2008;32:381–385. [PubMed] [Google Scholar]
Psaty B.M., et al. Cohorts for Heart and Aging Research in Genomic Epidemiology (CHARGE) Consortium: Design of prospective meta-analyses of genome-wide association studies from 5 cohorts. Circ. Cardiovasc. Genet. 2009;2:73–80. [PMC free article] [PubMed] [Google Scholar]
Purcell S., et al. PLINK: a tool set for whole-genome association and population-based linkage analyses. Am. J. Hum. Genet. 2007;81:559–575. [PMC free article] [PubMed] [Google Scholar]
Pyeritz R.E. Pleiotropy revisited: molecular explanations of a classic concept. Am. J. Med. Genet. 1989;34:124–134. [PubMed] [Google Scholar]
Sirota M., et al. Autoimmune disease classification by inverse association with SNP alleles. PLoS Genet. 2009;5:e1000792. [PMC free article] [PubMed] [Google Scholar]
Soranzo N., et al. A genome-wide meta-analysis identifies 22 loci associated with eight hematological parameters in the HaemGen consortium. Nat. Genet. 2009;41:1182–1190. [PMC free article] [PubMed] [Google Scholar]
Stearns F.W. One hundred years of pleiotropy: a retrospective. Genetics. 2010;186:767–773. [PMC free article] [PubMed] [Google Scholar]
Teslovich T.M., et al. Biological, clinical and population relevance of 95 loci for blood lipids. Nature. 2010;466:707–713. [PMC free article] [PubMed] [Google Scholar]
Thorisson G.A., et al. The International HapMap Project Web site. Genome Res. 2005;15:1592–1593. [PMC free article] [PubMed] [Google Scholar]
Tyler A.L., et al. Shadows of complexity: what biological networks reveal about epistasis and pleiotropy. Bioessays. 2009;31:220–227. [PMC free article] [PubMed] [Google Scholar]
WTCCC. Genome-wide association study of 14,000 cases of seven common diseases and 3,000 shared controls. Nature. 2007;447:661–678. [PMC free article] [PubMed] [Google Scholar]
Yang Q., et al. Analyze multivariate phenotypes in genetic association studies by combining univariate association tests. Genet Epidemiol. 2010;34:444–454. [PMC free article] [PubMed] [Google Scholar]
Zou L., et al. Systematic analysis of pleiotropy in C. elegans early embryogenesis. PLoS Comput. Biol. 2008;4:e1000003. [PMC free article] [PubMed] [Google Scholar]

Articles from Bioinformatics are provided here courtesy of Oxford University Press