Introduction

Adult height is a complex genetic trait with high heritability (>80%).1,2,3,4,5 Genetic studies of adult height have revealed that it is normally distributed, suggesting that height could be affected by the interaction of multiple genetic and environmental factors. Compared with other complex traits, adult height can be measured easily and accurately, reducing error variance in the phenotype. In most genetic studies for other complex human traits, height has been collected along with other data. Obviously, this raises the possibility of pooling data. Several recent genome scans for adult height revealed significant and suggestive evidence for linkage on several chromosomes.4,5,6,7 However, few of these regions were replicated among different studies. This may not be surprising considering that height is a complex genetic trait affected by multiple genes, each having a small effect, as well as by environmental factors. An obvious step to increase power and decrease false-positive results in linkage analysis of complex traits is to pool data across studies, so long as one is careful regarding the heterogeneity among different data sets.8,9 Here we report the results of our combined analysis of genome scans for adult height in the ongoing NHLBI Family Blood Pressure Program (FBPP). We first performed the genome scan for adult height within each ethnic group of the four networks in FBPP separately. We then performed a genome scan using information from all the groups by combining IBD sharing estimated from each group separately.

Materials and methods

Subjects were selected from the four component networks (GenNet, GENOA, HyperGEN, and SAPPHIRe) of the FBPP. All the families were ascertained by hypertensive probands. The ascertainment scheme and program design have been reported elsewhere.10 In brief, GenNet sampled African-American and European-American nuclear families through identification of a young-middle-aged proband with elevated blood pressure (BP). Both GENOA and HyperGEN sampled African-American and European-American sibships containing sibpairs with essential hypertension. GENOA also sampled Mexican-American sibships containing sibpairs with hypertension, as well as some non hypertensive sibs in all three ethnic groups. SAPPHIRe recruited groups of Japanese and Chinese sibpairs that were concordant for hypertension or extremely discordant. In each network, we combined the data according to ethnic background, including European Americans and African Americans from GenNet; European Americans, African Americans and Mexican Americans from GENOA; European Americans and African Americans from HyperGEN; and Asians from SAPPHIRe.

DNA was extracted from whole blood by standard methods at each of the four networks and was sent to the Mammalian Genotyping Service in Marshfield, WI for analysis. Screening Set 8 (372 highly polymorphic microsatellite markers) was used for all four networks. This screening set has an average heterozygosity of 80%, an average inter-marker distance of 10 cM, and covers 95% of the human genome. Some aspects of the marker data upon which this paper is based are under review by members of the FBPP group. However, no edits of the data contemplated thus far have any bearing on the inferences of this paper.

Individuals younger than 20 years old were excluded from the analysis because growth may not have reached a maximum. Height was adjusted by age for each gender separately in each of the eight groups and the residuals were standardized to obtain Z scores. These were the phenotypes in our analyses. Heritability for height was estimated for each group separately using the variance component method. The genome scans were performed using the multipoint variance component method in SOLAR.11 The VC method specifies the expected genetic covariances between relatives as a function of the estimated proportion of alleles sharing IBD at a marker locus. The IBD probabilities were estimated using a multipoint approach that considers all available genotypes. The likelihood ratio test was applied to test the null hypothesis of no additive genetic variance due to a quantitative trait locus (QTL) at a particular location. To obtain the empiric genomewide significance of this result, we performed simulations by randomly generating genotypes while retaining the phenotypes and pedigrees under the hypothesis of no linkage. Maximum LOD score (MLS) from each simulation was recorded and the P-value for our genome scan was obtained by calculating the frequency of MLS larger than the tested LOD score.

It is well known that allele frequencies between different human populations can be quite different. The estimates of allele frequencies could be biased if we used the combined data set directly; in turn, this could bias the estimates of IBD. To get an unbiased estimate of IBD, we first estimated multiple IBD sharing within each of eight groups separately using SOLAR and then combined all the IBD sharing for further analysis using the variance components method implemented in SOLAR for the combined data.

Results

Data on height were available from 6752 individuals in 2508 families from the FBPP study. The characteristics of the family members for each group are presented in Table 1. The height for females (161.3 cM) and males (174.5) was significantly different (P < 0.0001) from each other in the combined sample. The number of families in each group is from 605 to 1410 and the total number of families is 2508 families in the combined data. The mean sibship size in each group is from 1.7 to 3.6, and the mean sibship size is 2.8 in the combined data. The heritability of height for the eight groups varied from 0.75 to 0.98, which is compatible with previous reports.4,5,6 For individual genome scans, the region showing the strongest evidence for linkage was found in GENOA European Americans, centered at 14q21.1(MLS=3.67 at marker D14S592 ) (Table 2). The same region also showed evidence of linkage in SAPPHIRe Asians with a LOD=1.60. To obtain the genomewide empirical P-value, we performed 250 simulations. We obtained MLS>3.67 19 times, giving an empirical genomewide P-value of 0.08. Four other regions with a LOD>2.0 (nominal P=0.0012) were found in individual genome scans. These regions are at 1p11, 3q25, 5q23, and 6q13 (Table 2).

Table 1 Characteristics of families
Full size table
Table 2 Results of linkage analysis of individual genome scans
Full size table

The genome scan result for the combined IBD analysis is presented in Figure 1. Seven regions with LOD>1 (nominal P=0.016) are also listed in Table 3 as demonstrating suggestive evidence for linkage. The strongest support for linkage was found at 7q36.1 region (174 cM, marker D7S3058) with multipoint LOD=2.46.

Figure 1
figure 1

Genome-scan results for combined linkage analysis. The X-axis is the chromosome location and the Y-axis is the LOD score. Each chromosome is scaled according to its length from the genetic map.

Full size image
Table 3 Evidence of linkage analysis in combined samples
Full size table

Discussion

Hirschhorn et al4 and Perola et al5 reported strong evidence for linkage on chromosome 7 (150 cM, marker D7S2195, LOD=3. 40; and marker 164 cM, D7S1826, LOD=2.91). Encouragingly, the most significant result in our combined analysis was also found around the same region (174 cM, marker D7S3058, LOD=2.46), which makes it likely a true positive result.

Another two regions showing suggestive evidence in our study have been reported previously. Our chromosome 2 region (2p12, 92 cM, marker D2S1394, LOD=1.27) has been reported by Hirschhorn et al4 (104 cM, marker D2S113, LOD=2.23). Thompson et al12 have reported linkage on chromosome 20 (20p11, 34.22 cM, marker D20S66, LOD=3.0). Although none of the linkage studies reported previously replicated this finding, we did find suggestive evidence at this region (29 cM, marker D20S604, LOD=1.77).

The strongest finding in all the individual studies was on 14q21.1 (MLS=3.67) with a genomewide empirical P-value of 0.08, which is not significant at the α=0.05 level. None of the linkage studies reported previously has shown evidence for linkage at this region. More evidence from future studies is needed to clarify this result.

It has been shown by simulation studies that the power to detect a contributing locus with a small effect in a genome-scan study with moderate sample size is very limited when an acceptable type I error rate is maintained.4 In a recent review paper, Altmuller et al13 compared 31 whole-genome scans for different human complex diseases and found that the most obvious difference influencing success in finding linkage across studies was sample size. Pooling data from multiple studies will greatly increase power assuming the variations among studies are due to random sampling error. Pooling data also will reduce the type I error rates resulting from many independent tests of common hypotheses. By performing linkage analysis in our combined data, we found suggestive evidence for linkage at three regions reported previously for linkage with adult height.

Power calculations for this study have been undertaken under a simplified scenario using a method developed by Sham et al14 For a diallelic QTL with equal allele frequency, we assumed a model with 5% QTL additive variance, 70% residual shared variance and a recombination rate between marker and QTL of 5%. Considering a sample size of 6700 with sibship size of 3 with, the power to achieve LOD=3 is 97%. If the sibship size decreases to 2, the power will decrease to only 25% in the same sample. If we increase QTL additive variance from 5 to 10% and keep other parameters the same, the power to detect the QTL with LOD=3 will increase to 97% even with sibship size of 2. Although the average sibship size of the FBPP study is close to 3 (2.8), the variation of pedigree structure among different networks could decrease the power under an ideal situation. Based on the above discussion, it is reasonable to expect a high power to detect a QTL with an effect size of 10% in the combined sample.

No obvious candidate gene for adult height was found for the chromosome 7 region (7q31) in an initial search in NCBI databases. For the chromosome 2 region (2p12), a candidate gene is bone morphogenetic protein 10 (BMP10). This gene is a member of TGF-β family of growth factors. For the region at chromosome 20 (20p11), an obvious candidate gene is BMP2. Like BMP10, BMP2 also belongs to the TGF-β family and induces bone formation. In an association study by Thompson et al,12 no significant association was found with BMP2 in a subset of 20 tallest and 20 shortest individuals selected from a sample of over 500 Pima Indians. However, the power to detect any association is low with such a small sample size.

In summary, by combining the data sets from eight studies in the FBPP, we replicated evidence for a QTL influencing adult height in chromosome 7 (7q31) (LOD=2.46), which has been reported in two studies previously. Suggestive linkage (LOD>1) was found in another six regions in our combined analysis. Evidence for linkage for two of these regions (2p12, 20p11) has also been reported previously.