Considerable attention is currently being focused on the genotyping of single nucleotide polymorphisms (SNPs) to search for complex disease susceptibility alleles. The success to detect association between marker alleles and disease critically depends on the extent of linkage disequilibrium (LD) between functional alleles and surrounding markers. Studies on some chromosomes and genome regions have shown that LD is highly variable across human genome and is structured into discrete blocks of sequences separated by hotspots of recombination and/or LD breakdown 123456. The LD heterogeneity across the genome is a crucial aspect for genome-wide association studies, which has been emphasized as a powerful approach for complex disease study 789. Due to practical limitations of the expenses, the sample size used on the whole genome scale may not be sufficiently large. For example, the ongoing HapMap represents one of the most comprehensive and largest efforts ever attempted in biomedical research, in which a total of 270 subjects from African, Caucasian and Japanese/Chinese were recruited. However, sample size is an important factor influencing the accuracy of LD evaluation. Small sample size will bias the accuracy of LD estimation 210.

A limited but hypothesis-driven approach for association studies is to focus on the polymorphisms located in or around the candidate genes known to be of potentially functional importance for complex traits (based on their cellular and molecular studies). LD patterns in specific candidate genes need be empirically assessed to facilitate efficient experimental designs and execution of association studies 111. So far, only limited information is available about the LD patterns and haplotype characteristics in candidate genes for complex diseases 1213.

The purpose of this study was to determine the LD and haplotype characteristics in eight candidate genes for the two common complex disorders, osteoporosis and obesity. These eight genes are apolipoprotein E (APOE), type I collagen α1 (COL1A1), estrogen receptor-α (ER-α), leptin receptor (LEPR), parathyroid hormone (PTH)/PTH-related peptide receptor type 1 (PTHR1), transforming growth factor-β1 (TGF-β1), uncoupling protein 3 (UCP3), and vitamin D (1,25-dihydroxyvitamin D₃) receptor (VDR).

Determining the haplotype and LD characteristics of candidate genes and/or genomic regions has important implications for strategies of complex diseases gene mapping 19. In the present study, we illustrated the haplotype distribution and LD patterns in eight candidate genes for two complex diseases, obesity and osteoporosis, in 1873 subjects from 405 Caucasian nuclear families.

In this study, the yin yang haplotype pairs, mismatching at every SNP, were observed in two genes, PTHR1 and UCP3, but not in the other six. There are several possible reasons for such differences among genes. Other things being the same, lower LD levels result in smaller yin yang haplotype size, the genomic span per yin yang haplotype region 15. The different SNP density and LD patterns among genes may be partly responsible for the discrepancies about yin yang haplotypes among genes. Lower minor allele frequencies in some genes may be another reason since minor allele frequency ≥10% is a threshold for yin yang haplotype analysis 15. Moreover, yin yang haplotypes were affected by the factors that might be different among genes, such as the mutation and recombination rates 15.

So far, LD was inferred either by using unrelated random sample or family data. The LD values evaluated from these two different data sets were correlated, but in some instances they were quite different. Family subjects contain more information than unrelated random individuals, and the inference of LD using family data is more accurate 4. Sample size is another important factor influencing the accuracy of LD evaluation. Generally, larger sample size can minimize sampling error and produce more accurate evaluation of LD 20. In this study, haplotypes were reconstructed based on the nuclear families in a larger sample rather than unrelated individuals as the other studies 12182122. This increased the precision of LD evaluation.

In the present study, two most common measurements of LD, D' and r², were used, both measurements varying between 0 and 1. According to their calculating formula 23, they are positively correlated, as observed in this study, but sometimes they are quite different. For example, D' value for SNP6-SNP8 was 1.000, while the value for r² was only 0.004. Such a discrepancy between the two measures was also observed by Tiret et al 17. One possible explanation for the discrepancy is that the rare allele frequency of SNP8 was very low, only 1.9%, and r² is more sensitive to allele frequencies than D' 23.

Although a trend of decay of LD with increasing physical distance was observed in the present study, there was a great variation in LD with distance. For example, the physical distances for SNP33-SNP34 in VDR and SNP20-SNP21 in PTHR1 were both about 5 kb, however, the D' values were 0.062 and 0.964, and the r² were 0.004 and 0.826, respectively. In the ER-α gene, the physical distance of SNP14-SNP15 was about 2.6 times that of SNP13-SNP14, while the D' values were 0.675 and 0.028, and r² were 0.599 and 0.019, respectively. Such irregularity has been observed in the other studies 6171824 and is expected, since LD is also affected by other factors, such as natural selection, the rates of recombination and mutation, and gene conversion 25.

Several limitations of this study should be acknowledged. Compared with some recent analyses 41426, we only selected a limited number of SNPs in each gene, which may bring some difficulty to define the LD characteristics. Though 15-kb chromosome-wide resolution for LD pattern and haplotype structure study is acceptable but not ideal 27. In addition, the yin yang haplotype analysis might be affected by the limited SNPs and genes examined. The SNPs were chosen based on the information from available literature and public databases rather than evenly distributed within gene. Another limitation is that this study was restricted to Caucasians of European descent, which limited the ability to perform cross-population comparisons for yin yang haplotype coverage and haplotype diversities etc. Since genetic diversity and LD extent differ between populations, studying contrasted populations is informative. However, if the purpose is to provide guidelines for the design of association studies in a given population, information drawn from that population should be preferable 5, especially because Caucasians are more susceptible to osteoporosis and obesity compared with other populations 2829.

In conclusion, the haplotype distribution was different among the eight studied candidate genes for osteoporosis and/or obesity. The LD characteristics were different between genes or even at different regions within gene, though LD decays with the increase of physical distance. The differences in haplotype distributions and LD patterns among genes underscore the importance of characterizing each locus of interest prior to association studies.

JRL performed SNP genotyping, combined the results of analysis and drafted the manuscript. LJZ performed SNP genotyping and data analysis. PYL and YL performed statistical analysis. VD contributed to the manuscript preparation. HS, YJL, YYZ, DHX and PX performed SNP genotyping. HWD conceived of the study, and participated in its design and coordination. All authors read and approved the final manuscript.