HEGESMA: genome search meta-analysis and heterogeneity testing

Meta-analysis of genome-wide scans provides evidence for sex- and site-specific regulation of bone mass

Several genome-wide scans have been performed to detect loci that regulate BMD, but these have yielded inconsistent results, with limited replication of linkage peaks in different studies. In an effort to improve statistical power for detection of these loci, we performed a meta-analysis of genome-wide scans in which spine or hip BMD were studied. Evidence was gained to suggest that several chromosomal loci regulate BMD in a site-specific and sex-specific manner.

INTRODUCTION

BMD is a heritable trait and an important predictor of osteoporotic fracture risk. Several genome-wide scans have been performed in an attempt to detect loci that regulate BMD, but there has been limited replication of linkage peaks between studies. In an attempt to resolve these inconsistencies, we conducted a collaborative meta-analysis of genome-wide linkage scans in which femoral neck BMD (FN-BMD) or lumbar spine BMD (LS-BMD) had been studied.

MATERIALS AND METHODS

Data were accumulated from nine genome-wide scans involving 11,842 subjects. Data were analyzed separately for LS-BMD and FN-BMD and by sex. For each study, genomic bins of 30 cM were defined and ranked according to the maximum LOD score they contained. While various densitometers were used in different studies, the ranking approach that we used means that the results are not confounded by the fact that different measurement devices were used. Significance for high average rank and heterogeneity was obtained through Monte Carlo testing.

RESULTS

For LS-BMD, the quantitative trait locus (QTL) with greatest significance was on chromosome 1p13.3-q23.3 (p = 0.004), but this exhibited high heterogeneity and the effect was specific for women. Other significant LS-BMD QTLs were on chromosomes 12q24.31-qter, 3p25.3-p22.1, 11p12-q13.3, and 1q32-q42.3, including one on 18p11-q12.3 that had not been detected by individual studies. For FN-BMD, the strongest QTL was on chromosome 9q31.1-q33.3 (p = 0.002). Other significant QTLs were identified on chromosomes 17p12-q21.33, 14q13.1-q24.1, 9q21.32-q31.1, and 5q14.3-q23.2. There was no correlation in average ranks of bins between men and women and the loci that regulated BMD in men and women and at different sites were largely distinct.

CONCLUSIONS

This large-scale meta-analysis provided evidence for replication of several QTLs identified in previous studies and also identified a QTL on chromosome 18p11-q12.3, which had not been detected by individual studies. However, despite the large sample size, none of the individual loci identified reached genome-wide significance.

Identification of chromosomal regions linked to premature myocardial infarction: a meta-analysis of whole-genome searches

Myocardial infarction (MI) is a complication of coronary artery disease and the leading cause of death in the Western world. MI is considered a distinct phenotype with an increased genetic component for its premature type. MI's exact inheritance pattern is still unknown. Genome searches for identifying susceptibility loci for premature MI produced inconclusive or inconsistent results. Thus, a genome search meta-analysis (GSMA) was applied to available genome search data on premature MI. GSMA is a non-parametric method to identify genetic regions that rank high, on average in terms of linkage statistics across genome searches unweighted or weighted by study size. The significance of each region's average and heterogeneity, unadjusted or adjusted by neighbouring average simulated ranks, was calculated using a Monte Carlo test. The meta-analysis involved five genome searches in Caucasians. Eight regions (6p22.3-6p21.1, 14p13-14q13.1, 13q33.1-13q34, 5p15.33-5p15.1, 8q13.2-8q22.2, 1p36.21-1p35.2, 12q24.31-12q24.33, 8q24.21-8q24.3) were found to have significant average rank by either unweighted or weighted analyses. In addition, region 8q24.21-8q24.3 produced significant low heterogeneity (P (unadjusted)=0.03 and P (adjusted)=0.05). Four regions (6p22.3-6p21.1, 14p13-14q13.1, 8q13.2-8q22.2, 8q24.21-8q24.3) were not identified by the individual studies. The meta-analysis suggests that these four regions should be further investigated for genes that confer susceptibility to MI.

Endothelial NO synthase gene polymorphisms and hypertension: a meta-analysis

Studies investigated the association between endothelial NO synthase gene polymorphisms and hypertension-reported contradicted or nonconclusive results. A meta-analysis of 35 genetic association studies that examined the relation between hypertension and the G894T, 4a/b, T786C, and G23T polymorphisms of the endothelial nitric oxide synthase gene was carried out. Subgroup analysis by ethnicity and potential sources of heterogeneity and bias were explored. The meta-analysis included genotype data on 7779/10 498, 2216/3222, 2491/3913, and 833/587 cases/controls for G894T, 4b/a, T786C, and G23T, respectively. For the 4b/a polymorphism, overall, the heterogeneity between studies was not significant (P=0.82), and the allele b was associated with a 15% decreased risk of hypertension relative to allele a (odds ratio: 0.85; 95% CI: 0.74 to 0.98). Overall and in whites, the recessive model for allele b produced significant results (odds ratios: 0.78; 95% CI: 0.68 to 0.90 and OR: 0.76 95% CI: 0.62 to 0.92, respectively), whereas the dominant model produced nonsignificant results. In studies involved East Asians and blacks, an association was not demonstrated. Regarding the G894T, T768C, and G23T polymorphisms, in no case (ie, overall, in whites, or in East Asians) was a statistically significant association and heterogeneity found. There was no substantial source of bias in the selected studies. In conclusion, there is evidence of association only between 4b/a polymorphism and hypertension; however, studies exploring combinations of the polymorphisms may help us better understand the genetics of hypertension.

A meta-analysis of genotypes and haplotypes of methylenetetrahydrofolate reductase gene polymorphisms in acute lymphoblastic leukemia

A meta-analysis of case-control studies that investigated the association between the C677T and/or A1298C polymorphisms in the methylenetetrahydrofolate reductase (MTHFR) gene and acute lymphoblastic leukemia (ALL) was carried out. Pooled odds ratios (OR) of various genetic contrasts of each polymorphism were estimated using random (RE) and fixed effects (FE) models. Pooled ORs for combined genotypes and haplotypes were estimated after adjustment for study effect using a log-linear model and the expectation-maximization algorithm in combination with log-linear modeling, respectively. The recessive model for allele 1298C produced a rather marginal association: RE OR: 0.67; 95% confidence interval (CI): 0.46-0.99 and FE OR: 0.64; 95% CI: 0.49-0.84. In Caucasians, the results of the recessive model for allele 1298C was consisted with a protective effect of ALL development: FE OR: 0.63; 95% CI: 0.46-0.87. In childhood ALL, according to the results of the allele contrast and the recessive model for 677T allele it was conceivable that a protective effect exist: RE OR = 0.74; 95% CI: 0.57-0.96 and RE OR: 0.69; 95% CI: 0.51-0.94, respectively. The combined genotypes produced significant pooled OR for the 677CC/1298CC relative to 677CC/1298AA (OR: 0.54; 95% CI: 0.36-0.80). The haplotype 677C/1298C might be more protective to ALL relative to haplotype 677C/1298A (OR: 0.77; 95% CI: 0.61-0.97). When studies not in Hardy-Weinberg equilibrium (HWE) were corrected to account for departures from HWE, then, the pattern of results remained the same. Overall, there is high heterogeneity between the studies in both polymorphisms. A differential magnitude of effect in large versus small studies and alteration of early extremes effects existed.

Heterogeneity-based genome search meta-analysis for preeclampsia

Preeclampsia is a pregnancy-related disorder that causes maternal and fetal morbidity and mortality. Its exact inheritance pattern is still unknown, and genome searches for identifying susceptibility loci for preeclampsia have thus far produced inconclusive or inconsistent results. We performed a heterogeneity-based genome search meta-analysis (HEGESMA) that synthesized the available genome scan data on preeclampsia. HEGESMA identifies genetic regions (bins) that rank highly on average in terms of linkage statistics across genome scans (searches). The significance of each bin's average rank and heterogeneity across scans was calculated using Monte Carlo tests. The meta-analysis involved four genome-scans on general preeclampsia and five scans on severe preeclampsia. In general preeclampsia, 13 bins had significantly high average rank (Prank< 0.05) by either unweighted or weighted analyses, while four of them (2p11.2-2q21.1, 9q21.32-9q31.2, 2p15-2p11.2, 2q32.1-2q35) were formally significant by both analyses. Heterogeneity of bin 2.8 (2q32.1-2q35) was significantly low in both unweighted and weighted analysis (PQ< 0.01). In severe preeclampsia, 10 bins had significantly high average rank by either unweighted or weighted analyses and five of them (3q11.1-3q21.2, 2q37.1-2q37.3, 18p11.32-18p11.22, 2p15-2p11.2, 7q34-7q36.3) were significant by both analyses. Bin 2q37.1-2q37.3 showed marginal low heterogeneity in unweighted and weighted analysis (PQ= 0.06). Results should be interpreted with caution as the p values were modest. Further investigation of these regions by genotyping with additional markers and families may help to direct the identification of candidate genes for preeclampsia.

C677T and A1298C methylenetetrahydrofolate reductase gene polymorphisms in schizophrenia, bipolar disorder and depression: a meta-analysis of genetic association studies

A meta-analysis of the previous studies of allelic association between schizophrenia, bipolar disorder and depression with the C677T and A1298C polymorphisms of the methylenetetrahydrofolate reductase (MTHFR) gene was carried out. Attention was paid to genetic differences in ancestral groups. Ten studies of the C677T MTHFR polymorphism and schizophrenia were included, along with four studies with the A1298C MTHFR polymorphism and schizophrenia. Four association studies of the C677T MTHFR polymorphism with bipolar disorder, and five studies of the C677T base pair change with depression were included. The studies contained Caucasian and east Asian samples. For schizophrenia, the main analysis for investigating the association of the C677T allele T and the risk of developing schizophrenia relative to the allele C showed lack of heterogeneity (P = 0.22, I(2) = 24%) between samples. In these samples, the fixed effects odds ratio was marginally significant [odds ratio 1.13 (95% confidence interval, 1.04-1.23)]. In an analysis of subgroups, the odds ratio were not significant for the Caucasian sample [random effects odds ratio 1.14 (0.95-1.36) and fixed effects odds ratio 1.13 (1.00-1.27)](P = 0.14, I(2) = 43%) and in Asians [random effects odds ratio 1.15 (0.98-1.35) and fixed effects odds ratio 1.15 (1.01-1.31)] (P = 0.19, I(2) = 37). The recessive model for allele T produced significant results in the main analysis [fixed effects odds ratio 1.32 (1.12-1.56)] in east Asians [fixed effects odds ratio 1.45 (1.13-1.85)] and female participants [fixed effects odds ratio 1.70 (1.07-2.71)], whereas for Caucasians the results were nonsignificant. For the A1298C polymorphism, in the main analysis the allele frequency difference between C and A, and the dominant model for allele C produced marginally significant associations [fixed effects odds ratio 1.16 (1.03-1.31) and fixed effects odds ratio 1.19 (1.02-1.40), respectively]. In bipolar disorder, overall, there was no significant association between either allele of the C677T polymorphism and the risk of developing bipolar disorder. Only in the east Asian population was the C677T association of marginal significance, with fixed effects odds ratio 1.23 (1.00-1.52). No sources of potential bias were found in the selected studies. The meta-analysis results suggested that east Asians have a greater genetic risk from the MTHFR gene in developing schizophrenia and depression, and that the genetic effects in bipolar disorder and depression are different. A further exploration of the involvement of the MTHFR gene in the susceptibility to schizophrenia and affective disorders, with a greater number of studies with larger sample sizes, however, are needed to fully establish the role of the MTHFR gene.

Testing for genetic heterogeneity in the genome search meta-analysis method

The Genome Search Meta-Analysis (GSMA) method is widely used to detect linkage by pooling results of previously published genome-wide linkage studies. The GSMA uses a non-parametric summed rank statistic in 30 cM bins of the genome. Zintzaras and Ioannidis ([2005] Genet. Epidemiol. 28:123-137) developed a method of testing for heterogeneity of evidence for linkage in the GSMA, with three heterogeneity statistics (Q, Ha, B). They implement two testing procedures, restricted versus unrestricted for the summed rank within the bin. We show here that the rank-unrestricted test provides a conservative test for high heterogeneity and liberal test for low heterogeneity in linked regions. The rank-restricted test should therefore be used, despite the extensive simulations needed. In a simulation study, we show that the power to detect heterogeneity is low. For 20 studies of affected sib pairs, simulated assuming linkage in all studies to a gene with sibling relative risk of 1.3, the power to detect low heterogeneity using the Q statistic was 14%. With linkage present in 50% of the studies (to a gene with sibling relative risk of 1.4), the Q heterogeneity statistic had power of 29% to detect high heterogeneity. The power to detect linkage using the summed rank was high in both of these situations, at 98% and 79%, respectively. Although testing for heterogeneity in the GSMA is of interest, the currently available method provides little additional information to that provided by the summed rank statistic.

Methylenetetrahydrofolate reductase gene and susceptibility to breast cancer: a meta-analysis

The methylenetetrahydrofolate reductase (MTHFR) gene polymorphisms have been linked to the risk of developing breast cancer. A meta-analysis of 18 case-control studies investigating the association between the C677T and the A1298C polymorphisms of the MTHFR gene and breast cancer (BC) was carried out. The meta-analysis included genotype data on 5467/7336 and 3768/5276 cases/controls for C677T and A1298C, respectively. In the meta-analysis, the consistency of genetic effects across different ethnicities and the effect of menopausal status for various genetic contrasts were investigated. The overall analysis for investigating the association between the C677T allele T and the risk of developing BC showed significant heterogeneity (p = 0.08, I2 = 34%) and non-significant association [odds ratio (OR) 1.02; 95% confidence interval (0.95-1.10)]. The allele contrast was not significant in Caucasians (nine studies) and in East Asians (four studies) [OR 1.03 (0.93-1.14) and OR 0.96 (0.81-1.15), respectively] or in pre-menopausal (five studies) and post-menopausal (four studies) groups [OR 1.10 (0.94-1.29) and OR 1.06 (0.95-1.18), respectively]. The genotype contrast of the homozygotes (TT vs CC) produced significant results only for pre-menopausal cases [OR 1.46 (1.05-2.03)]. The recessive model for allele T produced significant association only in pre-menopausal cases [OR 1.49 (1.09-2.03)]. The dominant model for the effect of allele T produced no significant results, overall and in each subgroup. For the A1298C polymorphism, all genotype contrasts showed lack of association, overall and in Caucasians. In summary, the accumulated evidence supports an association in pre-menopausal women. BC is a complex disease with multifactorial etiology, and therefore, case-control studies that investigate gene-environment interaction might elucidate further the genetics of the disease.

Meta-analysis of genome-wide linkage studies for bone mineral density

Genome-wide linkage studies have shown several chromosome loci that may harbor genes that regulate bone mineral density (BMD), but results have been inconsistent. A meta-analysis was performed to assess evidence for linkage of BMD across whole genome scan studies. Eleven whole-genome scans of BMD or osteoporosis containing 3,097 families with 12,685 individuals were included in this genome scan meta-analysis (GSMA). For each study, 120 genomic bins of approximately 30 cM were defined and ranked according to maximum evidence for linkage within each bin. Bin ranks were weighted and summed across all studies. The summed rank for each bin was assessed empirically for significance using permutation methods. A total of seven bins lie above the 95% confidence level (P=0.05) and one bin was above the 99% confidence level (P=0.01) in the GSMA of eleven linkage studies: bins 16.1 (16pter-16p12.3, Psumrnk <0.01), 3.3 (3p22.2-3p14.1), 1.1 (1pter-1p36.22), 18.2 (18p11.23-18q12.2), 6.3 (6p21.1-6q15), 20.1 (20pter-20p12.3), and 18.1 (18pter-18p11.23). GSMA was performed with seven studies with linkage scores of LOD >1-1.85 for sensitivity test, confirming the linkage on chromosome 16p and 3p and revealing evidence of new linkage in bins 10.2 (10p14-10q11.21) and 22.2 (22q12.3-22pter). In conclusion, the meta-analysis of whole-genome linkage studies of BMD has shown chromosome 16pter-16p12.3 to have the greatest evidence of linkage as well as revealing evidence of linkage in chromosomes 1p, 3p, 6, 10, 18, 20p, and 22q across studies. This data may provide a basis with which to carry out targeted linkage and candidate gene studies particularly in these regions.

Evidence for a gene influencing heart rate on chromosome 5p13-14 in a meta-analysis of genome-wide scans from the NHLBI Family Blood Pressure Program

BACKGROUND

Elevated resting heart rate has been shown in multiple studies to be a strong predictor of cardiovascular disease. Previous family studies have shown a significant heritable component to heart rate with several groups conducting genomic linkage scans to identify quantitative trait loci.

METHODS

We performed a genome-wide linkage scan to identify quantitative trait loci influencing resting heart rate among 3,282 Caucasians and 3,989 African-Americans in three independent networks comprising the Family Blood Pressure Program (FBPP) using 368 microsatellite markers. Mean heart rate measurements were used in a regression model including covariates for age, body mass index, pack-years, currently drinking alcohol (yes/no), hypertension status and medication usage to create a standardized residual for each gender/ethnic group within each study network. This residual was used in a nonparametric variance component model to generate a LOD score and a corresponding P value for each ethnic group within each study network. P values from each ethnic group and study network were merged using an adjusted Fisher's combining P values method and the resulting P values were converted to LOD scores. The entire analysis was redone after individuals currently taking beta-blocker medication were removed.

RESULTS

We identified significant evidence of linkage (LOD = 4.62) to chromosome 10 near 142.78 cM in the Caucasian group of HyperGEN. Between race and network groups we identified a LOD score of 1.86 on chromosome 5 (between 39.99 and 45.34 cM) in African-Americans in the GENOA network and the same region produced a LOD score of 1.12 among Caucasians within a different network (HyperGEN). Combining all network and race groups we identified a LOD score of 1.92 (P = 0.0013) on chromosome 5p13-14. We assessed heterogeneity for this locus between networks and ethnic groups and found significant evidence for low heterogeneity (P < or = 0.05).

CONCLUSION

We found replication (LOD > 1) between ethnic groups and between study networks with low heterogeneity on chromosome 5p13-14 suggesting that a gene in this region influences resting heart rate.

A heterogeneity-based genome search meta-analysis for autism-spectrum disorders

Autism and autism-spectrum disorders exhibit high heritability, although specific susceptibility genes still remain largely elusive. We performed a heterogeneity-based genome search meta-analysis (HEGESMA) of nine genome scans on autism or autism-spectrum disorders. Each genome scan was separated in 30 cM bins and the maximum linkage statistic from each bin was ranked. Significance for each bin's average rank and for between-scan heterogeneity (dis-similarity in the average ranks) was obtained through Monte Carlo tests. For autism, data from 771 affected sibpairs were synthesized across six separate genome scans. Region 7q22-q32 reached genome-wide significance both in weighted and unweighted analyses, with evidence for significantly low between-scan heterogeneity. The flanking chromosomal region 7q32-qter reached the less stringent threshold of suggestive significance, with no evidence for low between-scan heterogeneity. For autism-spectrum disorders (634 affected sibpairs from five separate scans), no chromosomal region reached genome-wide significance. However, suggestive significance was reached for the chromosomal regions 17p11.2-q12 and 10p12-q11.1 in weighted analyses. There was evidence for significantly high between-scan heterogeneity for the former region. The meta-analysis suggests that the 7q22-q32 region should be further scrutinized for autism susceptibility genes, while autism-spectrum disorders seem to have quite diverse linkage signals across scans, possibly suggesting genetic heterogeneity across subsyndromes and subpopulations.

Meta-analysis of genome-wide linkage studies for quantitative lipid traits in African Americans

Genetic influences on lipid traits have been suggested by numerous studies. In addition to heritability studies, over 50 genome scans have been performed to identify regions of linkage for quantitative lipid levels. Five of these scans have been performed in African Americans (four univariate and one bivariate linkage analysis), but with results that have been largely inconclusive. Linkage analyses are often limited by both sample size and heterogeneity, which may lead to nominal LOD scores or lack of evidence for linkage; the use of meta-analysis to combine linkage results from populations with similar ethnic backgrounds may help overcome some of these limitations. Thus, we performed a meta-analysis using data from four genome scans conducted in African American families to identify chromosomal regions showing evidence of linkage for total cholesterol (TC), triglycerides (TG), low density lipoprotein cholesterol (LDL) and high density lipoprotein cholesterol (HDL). Significant evidence (i.e. P<0.00042) for linkage was found for LDL on chromosome 1q32.1-q41 (Pweighted=0.00014 and Punweighted=0.00007) and 1q41-q44 (Pweighted=0.00017 and Punweighted=0.00014). We found suggestive evidence (i.e. P<0.00847) for TG on 16p12.1-q11.2 and for HDL on 4p15.1-p11. We also assessed heterogeneity between studies and found significant evidence for low heterogeneity for both regions on chromosome 1q (P=0.0300 and P=0.0279, respectively) for LDL and chromosome 16 (P=0.0429) for TG. Statistically significant evidence for linkage and low heterogeneity on chromosome 1q therefore suggest that this region may harbor a gene underlying the inheritance of LDL in African Americans.