Statistics from Altmetric.com
In a recently published study1 involving 79 ankylosing spondylitis (AS) patients and 132 unrelated healthy blood donors, Zhu et al report association of a tumour necrosis factor α (TNFα) single nucleotide polymorphism (SNP), rs1799724, with AS.1 I have many concerns with this paper and its conclusions.
The authors state that only SNP rs1799724 is not in Hardy–Weinberg equilibrium. However, inspection of the genotypes presented in table 2 of the paper reveals that rs2284178 is also not in Hardy–Weinberg equilibrium in the controls (p = 1.3×10−5). This can occur because of hidden relationship between genotyped individuals or disease association or selection, but most commonly occurs because of genotyping error.
The two SNPs not in Hardy–Weinberg equilibrium are critical to the findings of the study. The authors suggest that as no linkage disequilibrium was observed between human leucocyte antigen (HLA) gene HLA-B27 and rs2284178, which lies between rs1799724 and HLA-B27, or between rs2284178 and rs1799724, that the association of rs1799724 with AS is not due to linkage disequilibrium with HLA-B27. There are three concerns regarding this assertion.
Firstly, the SNP involved, rs2284178, is itself not in Hardy–Weinberg equilibrium, raising the concerns outlined above.
Secondly, the argument assumes that linkage disequilibrium is continuous, which is known not to be the case. It is quite possible for strong linkage disequilibrium to exist between two markers, where an intervening marker shows no linkage disequilibrium with either. For example, using the HapMap data for Chinese (build 35, http://www.hapmap.org), marker rs2284178 is not in linkage disequilibrium with either markers rs11752262 (r2 = 0.06) or rs11757602 (r2 = 0.017), which lie on either side of rs2284178. However, rs11752262 and rs11757602 are themselves in quite strong linkage disequilibrium (r2 = 1).
Finally, to calculate Lewontin’s D′ statistic (used by Zhu et al) requires that the marker genotypes are known, rather than simply carriage of an allele, to determine the haplotypic phase of the alleles of interest.2 Although it is not absolutely clear, the HLA-B27 genotyping method employed by Zhu and colleagues for the control samples does not appear to provide the requisite data,3 appearing only to determine if samples carry HLA-B27 rather than the HLA-B genotype (ie, it does not distinguish between HLA-B27 heterozygotes and homozygotes). It is also not clear, but likely, that the case genotyping is similarly limited. If so, then how did the authors calculate D′? Indeed the markedly different frequency of rs1799724 in B27-positive and -negative AS cases (47.9 vs 62.5%) suggests significant linkage disequilibrium between HLA-B27 and this SNP. However, as only nine HLA-B27 positive cases were studied, the sample size is inadequate to make any meaningful conclusion about linkage disequilibrium with HLA-B27.
The association findings are also reported uncorrected for the number of SNPs studied, and none of the SNPs remain associated following appropriate Bonferroni correction.
Whilst there is strong evidence of the existence of non-B27 major histocompatibility complex (MHC) genes in AS, notably HLA-B60,4–6 but also probably others,7 identification of those genes will require much larger studies with complete control for linkage disequilibrium with HLA-B27. Linkage disequilibrium across the MHC is extreme and complex,8 9 and failure to control for it properly leads to an inability to differentiate true association from linkage disequilibrium.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.