A Large-Scale Genome-Wide Linkage Analysis to Map Loci Linked to Stature in Chinese Population

Abstract

Context: A number of genome-wide scans of stature have been reported previously, but with inconsistent results. The inconsistency may be partly due to differential population characteristics and gender- and/or age-specific effects on this trait.

Objective: This study aimed to identify the quantitative trait loci (QTLs) underlying the variation of stature in Chinese population, and to evaluate age- and gender-specific linkage for stature.

Methods: We conducted a large-scale, genome-wide linkage scan using the data from three independent samples (a total of 7112 subjects from 1811 pedigrees) enrolled from the same geographical region in China. Linkage analyses were performed in the pooled samples and in subgroups defined by age (≤25 vs. >25 yr), gender, or both, using the model-free regression method implemented in MERLIN-REGRESS.

Results: The strongest linkage signal was obtained on 17q24 (LOD = 3.82) in the pooled samples. Age-specific analysis revealed two additional significant QTLs on 13q34 and 18p11.3 among subjects 25 yr or younger. In gender-specific analyses, males showed suggestive QTLs on 12q21 (LOD = 2.31) and 17q22 (LOD = 2.60), and females showed a suggestive QTL on 13q31.1 (LOD = 2.68). Age- and gender-specific linkage analyses suggested that males older than 25 yr contributed more signals to QTLs on 12q21 and 17q22, with a LOD score of 3.00 and 2.26, respectively, whereas females older than 25 yr presented a suggestive QTL on 8q24.3 (LOD = 2.57).

Conclusion: Our study identified a strong linkage of chromosome 17q24 to stature in this Chinese population, and indicated that it may be informative to consider differential age and gender effects in the genetic dissection of stature.

There is a strong linkage of chromosome 17q24 to stature in the Chinese population. It is informative to consider age and gender effects in the genetic dissection of stature.

Human height, or stature, is a complex trait resulting from a complex growth procedure in which a variety of systemic and paracrine or autocrine factors are involved, including GH (1), IGF-I (2), and glucocorticoid (3). Epidemiological studies have shown that stature is correlated with a number of important medical conditions, including cardiovascular diseases (4), hip fractures (5), and cancers (6), which suggest that a better understanding of the genesis of stature not only benefits to the diagnosis, prognosis, and possible reassurance for children with short or tall stature, but also sheds light on investigating the mechanisms of other related traits or diseases.

The genetics of stature has been studied since 1903, and stature has been found to be highly heritable with estimated heritability above 68% (7). A number of genome-wide linkage studies of stature have been reported, which have identified genome-wide significant quantitative trait loci (QTLs) with LOD scores of 3.6 or more on 1p21 (8), 6p25 (9), 9q22 (10), 14q23 (11), and Xq24 (10), whereas the QTL on 1p21 was male specific (8). Furthermore, more than 20 suggestive QTLs with LOD scores of 2.2 or more scattered across the genome were reported (supplemental Table E1, which is published as supplemental data on The Endocrine Society’s Journals Online web site at http://jcem.endojournals.org). However, only suggestive or significant signals on 1p21 (8,11), 3p14 (9,12), 6q12–14 (11,13), 6q25 (9,14), 7q36 (11,15), 9q34 (15,16), and 12q11–13 (9,17) were validated at least once in an independent sample after the initial identification. Most of these linkage studies were conducted in adult populations of European origin. To our knowledge, only Dempfle et al. (17) identified a suggestive QTL on 12q11 for stature in childhood.

Stature is known as sexually dimorphic, and it has been demonstrated that there is significant gender heterogeneity in the environmental and genetic contribution to the variation in stature at takeoff, peak height velocity, and also in adult height (18). A recent study, which evaluated sex-specific heritability and genome-wide linkage for 17 quantitative traits in the Hutterites, showed evidence of sex difference in heritability and linkage on 12 traits, including adult stature (19). This report suggested that some genetic variations may act differently between males and females. Therefore, it may be informative to consider sex-specific models in the genetic analysis of stature. However, limited genome-scan studies of stature conducted sex-specific analyses (8,17,20). Among those, several male-specific suggestive QTLs for stature were identified on 1p21, 9p24, 18q21, and 21q21, as well as a female-specific suggestive QTL on 13q12.3.

In this study we performed linkage analysis to map loci affecting stature by combining three large-scale independent genome scans, which consist of 7112 subjects from 1811 pedigrees enrolled from the same region in China. We performed linkage analyses in the pooled samples and in subgroups stratified by age and gender to explore age effect and sexually dimorphic effect on stature.

Subjects and Methods

Study subjects

This study included subjects from three independent genome-wide scan studies that were designed to dissect genetics of asthma (21), hypertension (22), and osteoporosis (23), respectively. All the subjects were recruited from the same region of Anhui province in China. The sampling scheme and exclusion criteria for each study were described elsewhere (21,22,23). Briefly, the asthma study included 2551 individuals from 533 asthma index families with at least two siblings aged 8 yr or older; the hypertension study included 1468 individuals from 337 families with siblings aged 15–55 yr and with at least a sibling pair having extreme blood pressure; and the osteoporosis study included 3093 individuals aged 25–64 yr from 941 families with at least a sibling pair having extreme total hip bone mineral density. Table 1 summarized the number of families, individuals, and markers genotyped corresponding to each study. All three studies were approved by Harvard School of Public Health institutional review boards and Biomedical Institute of Anhui Medical University institutional review boards. Written informed consent was obtained from each participant.

Table 1.

Overview of the data sets from the three genome-scan studies

	Asthma	Hypertension	Osteoporosis
No. of pedigrees	533	337	941
Pedigree type	Nuclear families	Nuclear families	Sibpair
Pedigree structure
Both parents	485	336
One parent	45	1
Sibships only	3		941
No. of genotyped individuals	2551	1468	3093
No. of markers screened	422	390	400
No. of markers analyzed	418	366	398

Genotyping

Markers from the Weber screening set (350–400 markers) were genotyped in each study, but a specific set of markers varied by studies. Detailed information about the genotyping method and genetic map for each data set were reported previously (21,22,23). Markers in an average span of approximately 10 centimorgans (cMs) across 22 autosomal chromosomes were genotyped for each data set.

Construction of standardized genetic map

Because different sets of genomic markers were genotyped in the three studies, we constructed a standardized genetic map to pool the three samples for linkage analysis. Using the deCODE map as the backbone, we applied the similar procedure reported by McQueen et al. (24) to generate a common sex-averaged map for the markers from each study. If the genetic location of a particular marker could be found on the deCODE map, then this location was used. Otherwise, we determined the physical location (in base pairs) of that marker using the National Center for Biotechnology Information (NCBI) Build 35.1, and used the physical locations of two flanking markers found on the deCODE map as the referenced markers. By assuming that the ratio of the distance between markers on the physical map was the same as the ratio of the distance on the genetic map, we interpolated the genetic location of the marker, which was not available on the deCODE map. If we could not identify a marker either on the deCODE map or in the National Center for Biotechnology Information Build 35.1, the marker was discarded. More than 98% of the markers from each study sample were mapped to the standardized common map.

Data pooling procedures

In consideration of the variation in allele coding across different platforms, we created unique marker names for each study sample. If a common marker was genotyped in two study samples, the marker in one of the study sample was renamed and treated as a new marker. The genetic location for this “new” marker was coded as its original location plus 0.02 cM. Accordingly, the standardized genetic map was updated by adding this “new” marker. As a result, no identical marker was found in any two of the three study samples. We then pooled the raw genotyping data from each sample and generated missing genotyping data for markers from the other two samples.

Data cleaning procedures

Markers with significant deviation from Hardy-Weinberg equilibrium (P < 0.001) were removed from the analysis. We examined the relationship within each family using the RELPAIR (25) program, and excluded MZ twins and individuals with genotyping data inconsistent with their other family members. The PedCheck (26) program was used to check for mendelian inconsistencies. Any inconsistent genotypes were set to missing for the entire family at that particular locus. In addition, MERLIN was used to identify unlikely genotypes (27). Any unlikely genotypes detected by MERLIN were set to missing. The cleaned combined genotyping data were used in the subsequent analyses.

Phenotype

Stature in each project was measured without shoes to the nearest 0.1 cm on a portable stadiometer. Standardized residuals of stature were calculated and used as the main phenotype. Specifically, for each study sample, we first stratified study subjects into subgroups based on age (<15 yr, 15–25 yr, >25 yr) and gender, and then built a predictive model for stature using age (continuous), occupation (only for individuals ≥15 yr), smoking status (only for individuals ≥15 yr), education levels, and asthma medication history (0 = no, 1 = yes) as the covariates within each group. Residuals of stature were obtained from group-specific predictive models. The residuals of stature for all subjects were then standardized and subsequently used in linkage analysis. For age- and gender-specific linkage analysis, all available genotyping data were used and set the phenotype for subjects who did not fall into the category of interest as missing data. To minimize the possible confounding effect of vertebral fractures on stature, the phenotypes of 33 subjects who had a history of vertebral fractures were set as missing data.

Linkage analysis

Linkage analysis was performed on the standardized residuals of stature using the model-free regression method MERLIN-REGRESS (28) implemented in the MERLIN package. The model-free regression method is based on a revised Haseman-Elston method that performs regression of estimated identity-by-descent sharing between relative pairs on the squared sums and squared differences of trait values of the relative pairs. In this study, the results with LOD scores of 2.2 or more were considered as suggestive linkage evidence, and the results with LOD scores of 3.6 or more were considered as genome-wide significant linkage evidence, which was suggested in previous reports (29). Moreover, because we performed linkage analysis using the pooled sample from three independent studies, we performed the heterogeneity test of linkage results using Homo program (version 0.2), which was kindly provided by Dr. Harald Goring. (Department of Genetics, Southwest Foundation for Biomedical Research, San Antonio, Texas).

Results

By pooling all three study samples, a total of 7112 subjects from 1811 families was genotyped, and 7043 subjects had available phenotypes. The phenotypical characteristics of these subjects were summarized in Table 2. The mean stature and corresponding sd were 156.4 ± 11.4 cm. As shown in Fig. 1, stature increased with age till 25 yr old in the asthma study, and after that, it decreased with age. Therefore, we chose 25 yr as a cutoff in the following age-specific analysis. There were 1151 subjects aged 25 yr or younger (mean ± sd = 15.35 ± 5.68), of which 744 (64.6%) were younger than 18 yr old.

Table 2.

Population characteristics in each study and the pooled samples

Variables	Asthma	Hypertension	Osteoporosis	Pooled samples
na	2541	1445	3057	7043
Age (yr)	30.7 ± 16.4	43.1 ± 16.0	45.4 ± 7.4	39.6 ± 14.8
Weight (kg)	47.1 ± 13.3	55.6 ± 9.7	54.9 ± 8.1	52.2 ± 11.2
Stature (cm)	152.2 ± 14.9	159.3 ± 8.8	158.6 ± 7.5	156.4 ± 11.4
BMI (kg/m2)	19.8 ± 3.2	22.0 ± 5.2	21.8 ± 2.7	21.1 ± 3.6
Age group (yr)
≤25	974 (38.3)	177 (12.3)	0	1151 (16.3)
>25	1567 (61.7)	1268 (87.7)	3057 (100)	5894 (83.7)
Male	1237 (48.7)	798 (55.2)	1577 (51.6)	3612 (51.3)
Farmerb	1418 (74.2)	1171 (81.0)	2686 (87.9)	5275 (82.3)
Current smokerb	594 (31.0)	524 (36.2)	1120 (36.6)	2238 (34.9)
Educationb
Elementary school	1435 (75.2)	1038 (71.9)	2243 (73.6)	4716 (73.7)

Values are expressed as mean ± sd or number (%). BMI, Body mass index. 

^aOnly individuals with available phenotype information were included. 

^bOnly individuals 15 yr or older were included in the asthma study or in pooled samples. 

Figure 1.

Change of stature with age in samples from each individual study, stratified by gender.

Linkage analyses for each study separately and for the pooled samples

A total of 1182 markers was included in this study, and the averaged heterozygosity was 72.8%. We first conducted linkage analysis for each study separately. A significant LOD score of 4.69 (P < 0.00001) on 17q24 at 98.5 cM and a suggestive LOD score of 2.60 on 18p11 (P = 0.0003) at 5.4 cM were detected for stature in the asthma study. Two suggestive QTLs, one on 12q21 at 100 cM (LOD = 2.34; P = 0.0005) and the other on 13q12 at 26.3 cM (LOD = 2.47; P = 0.0004), were detected in the osteoporosis study. No significant or suggestive signal was detected in the hypertension study (Fig. 2).

Figure 2.

QTL plots for stature in the pooled samples as well as in each individual study.

In the pooled samples, we identified the highest LOD score of 3.82 (P = 0.00001) on 17q24 at 98.5 cM, which met the stringent genome-wide significance level (Table 3 and Fig. 2). No other significant or suggestive linkage was detected.

Table 3.

QTLs (LOD ≥ 2.2) for stature in the pooled samples and each subgroup

Population	Chromosome	Location (cM)	LOD score	P value
Pooled samples	17q24	98.5	3.82	0.00001
Age stratification
≤ 25 yr	13q34	116.3	6.02	<0.00001
	17q24	98.5	2.30	0.0006
	18p11.3	22.4	5.56	<0.00001
Gender stratification
Male	12q21	101.5	2.31	0.0006
	17q22	86.5	2.60	0.0003
Female	13q31.1	78.3	2.68	0.0002
Age and gender stratification
Male > 25 yr	12q21	94.5	3.00	0.0001
	17q22	86.5	2.26	0.0006
Female > 25 yr	8q24.3	158.4	2.57	0.0003

Age-specific linkage analyses

We stratified the pooled samples by age (≤25 yr vs. >25 yr). A total number of 408 families with at least two members 25 yr or younger and 1329 families with at least two members older than 25 yr, respectively, were included in the analyses. A peak on chromosome 17q24 was found in both age groups, with a LOD score of 2.30 (P = 0.0006) and 1.20 (P = 0.009) in subjects 25 yr or younger and in subjects older than 25 yr, respectively, suggesting that both groups contributed to the significant QTL on chromosome 17q. In addition, we detected two significant QTLs on chromosome 13q34 (LOD = 6.02; P < 0.00001) and 18p11.3 (LOD = 5.56; P < 0.00001) in subjects 25 yr or younger only. Although in subjects older than 25 yr, no suggestive or significant loci were detected.

Gender-specific linkage analyses

We further conducted gender-specific linkage analyses of stature. As shown in Fig. 3, when examining the linkage results on chromosome 17q, males contributed more signal than females on the QTL at 17q24 (LOD = 1.76, P = 0.002 in males; LOD = 0.65, P = 0.05 in females). In addition, a LOD score of 2.60 on 17q22 at 86.5 cM was found in males, which was about 12 cM upstream of the QTL detected in the pooled samples. The other suggestive gender-specific loci were detected on chromosome 12q21 in males (LOD = 2.31; P = 0.0006), as well as on 13q31.1 in females (LOD = 2.68; P = 0.0002) (Table 3).

Figure 3.

Age- and gender-specific QTL plots for stature on chromosomes 8, 12, 13, and 17.

Age- and gender-specific linkage analyses

As presented in Table 3, no suggestive or significant QTL was detected for either males or females who were 25 yr or younger.

In males older than 25 yr, no additional suggestive or significant QTLs were detected. However, we found that in this older age group, LOD score on 17q22 (LOD = 2.26; P = 0.0006) still met the genome-wide suggestive linkage level, and the linkage signal on 12q21 was even stronger (LOD = 3.0; P = 0.0001) than in the pooled male sample. Our data suggested that the older male age group contributes more signals to these two male-specific regions than the younger age group (Fig. 3).

In females older than 25 yr, we identified another suggestive linkage signal on 8q24.3 with a LOD score of 2.57, whereas the linkage signal at this region in females 25 yr or younger was extremely low (LOD = 0.0) (Fig. 3). As to the female-specific QTL on 13q31.1 detected in the pooled female sample, we detected LOD scores of 1.62 (P = 0.003) and 0.72 (P = 0.03), respectively, in females older than 25 yr and those 25 yr or younger.

Assessment of linkage heterogeneity

In the regions that encompassed the significant or suggestive linkage signals on chromosomes 8, 12, 13, 17, and 18, we further performed tests of linkage heterogeneity among the study population using the Homo program. No significant evidence of linkage heterogeneity was detected.

Discussion

Our study was the largest linkage study investigating genetics of stature in an Asian-ancestry population. By pooling and analyzing data from three independent genome-scan studies, we found novel linkage regions for stature in the Chinese population. The most significant evidence for linkage to stature was detected on 17q24. In particular, male subjects contributed more signals to this region than female subjects, with a maximum LOD score of 2.60 on 17q22, about 12 cM upstream of the peak on 17q24. In addition, we identified two significant QTLs for subjects 25 yr or younger on 13q34 and 18p11.3, a suggestive QTL on 12q21 in males, a suggestive QTL on 13q31.1 in females, and a suggestive QTL on 8q24.3 in females older than 25 yr. However, we did not find any suggestive linkage at the suggestive or significant QTLs previously identified among the European population (supplemental Table E1).

Although there was no overlap in the significant or suggestive linkage signals across the three independent studies, our data indicated that it was appropriate to pool the three study samples for the following reasons. First, all the subjects in the three studies were enrolled from the same region, and they were fairly homogeneous with respect to ethnicity, socioeconomic status, lifestyle, and genetic background. Second, we used the Homo program to examine linkage heterogeneity for the suggestive or significant QTLs identified in this study and did not find linkage heterogeneity across the three samples. Third, stature was not the primary phenotype used for sample enrollment in each study, and the association between stature and the primary phenotype was fairly weak. The sample enrolled in each study should be random with respect to stature. As suggested by Hirschhorn et al. (9), the lack of overlapped linkage regions across different studies may be partly due to the effect of statistical fluctuation (sampling variation). Another explanation was that the power to identify linkage signals for stature within each study was limited due to inadequate sample size. In comparison, linkage analyses in appropriately pooled samples should be much more powerful and provide more robust results.

Although the strongest signal we detected on 17q24 in the pooled samples and also in the asthma study was not overlapped with any major QTLs reported previously, Liu et al. (16) found a weak signal at 17q23.3 with a LOD score of 1.71 in the Caucasian population, which is about 5 cM upstream of our identified significant region. In addition, three other previous studies reported linkage signals with LOD more than 1.5 on chromosome 17q at 74.32 cM (9), 91.10cM (30), and 115cM (31), respectively. Because the confidence intervals for QTLs implicated in genome-wide scans may cover a relatively wide range of chromosomal regions (32), the results reported by our and other studies strongly indicated that a susceptibility gene or genes regulating stature may harbor on chromosome 17q. For example, the GH 1 gene, located at about 9 cM away from this identified QTL, might be one of the candidates, given the well-known biological role of this hormone in the postnatal growth of skeletal and soft tissue (33,34). Until now, some mutations in the GH 1 gene have been identified in individuals with short stature, and these mutations may contribute to the variation in stature (35,36).

Another promising QTL was identified on 12q21 in males. Specifically, males older than 25 yr contributed most signals to this QTL. This region is about 30 and 12 cM away from the previously reported QTLs in Finland (9) and Caucasian populations (14), respectively. At this region, IGF-I and the suppressor of cytokine signaling 2 (SOCS2) might be the two promising candidate genes controlling variation of stature. It has been demonstrated that IGF-I mediates growth-promoting effects of GH (37), and SOCS2 regulates IGF-I receptor medicated cell signaling, so both could play an important role in mammalian growth and development. In addition, the polymorphisms in the IGF-I gene (38) and SOCS2 gene (39) have been linked to human stature. Rietveld et al. (40) also observed an association between the IGF-I alleles and stature in males aged 55 yr and older. However, further studies are needed to validate whether these two genes or other genes nearby this region may affect the variation of human height and whether these genes had sex-specific effects.

In females we identified two suggestive QTLs on 13q31.1 and 8q24.3, whereas the latter was identified in females older than 25 yr only. These two regions were reported previously in the Finland population, with a LOD score of 2.52 on 8q24 and a LOD score of 3.56 on 13q33 (9). However, they analyzed the data in the combined male and female population, rather than the gender-specific subgroup. Sammalisto et al. (8) reported a suggestive female-specific QTL on 13q12.3, which showed a weak linkage with a LOD score of 1.40 in our females older than 25 yr and was about 62 cM away from our identified peak on 13q31.1.

Although there is considerable sex difference in final adult stature, the question if there is a sex-specific effect in linkage on stature remains inconclusive. Our data, along with others (8), found suggestive or significant evidence for linkage in either the male-only or female-only sample, but not in the pooled samples. It implies that in linkage analysis, it may be informative to examine linkage in both pooled and sex-specific samples.

We identified significant QTLs on 13q34 and 18p11.3 for stature in subjects 25 yr or younger. The underlying rationale for the observed age-specific QTLs remains unclear. One possible explanation is gene-environmental interaction because it is possible that genetic influence on stature may be modified by different environmental exposures at different growth stages. Another possibility is that the two signals were actually the loci directly related to growth rather than stature because maturation delay is often found in children with short stature (41), which indicate that genes affect pubertal onset, and/or age of growth spurt may also affect childhood or adolescent stature indirectly. However, in this study we were unable to distinguish the linkage signals for stature from those for growth rates, given the absence of the information reflecting pubertal stage, growth rate, or hormone status, and also the inadequate power to perform age-stratified analysis among subjects 25 yr or younger. Finally, the two QTLs on 13q34 and 18p11.3 are close to the telomere of chromosomes 13 and 18, respectively. We could not exclude the possibility that these two signals were false-positive results. Future studies are needed to confirm and clarify these two QTLs identified here.

Our study had some other limitations. First, the participating pedigrees were ascertained via probands that manifested a specific phenotype of interest (asthma phenotype, extreme blood pressure, or extreme bone mineral density). In our population the association of stature with the original trait of interest in each individual study was fairly weak, and the samples selected for various diseases were pooled together, which may reduce the potential ascertainment bias (15). However, we still could not exclude the possibility that such designs may result in false-positive results. Second, we did not adjust some environmental factors, including nutrition intakes, physical exercise, as well as unmeasured socioeconomic factors, in the linkage analyses, which may be important for further investigation. And in this study, the statistical power to identify linkage signals for stature within each subgroup was limited due to the inadequate sample size.

In summary, our study has provided strong evidence that a gene or genes on chromosome 17q may play an important role in the variation of stature. In addition, we identified one suggestive QTL on 12q21 in males, two suggestive QTLs on 13q31.1 and 8q24.3 in females, as well as two significant QTLs on 13q34 and 18p11.3 in subjects 25 yr or younger, which indicate that it may be informative to consider differential age and gender effects in the genetic dissection of stature. Further investigation of these suggestive chromosomal regions may lead to the discovery of causative genes of stature.