Statistical analysis was performed with Stata statistical software, version 6.0, 1999 (Stata Corporation, College Station, TX). OR = odds ratio; 95% CI = 95% confidence interval.
Article first published online: 5 FEB 2002
Copyright © 2002 by the American College of Rheumatology
Arthritis & Rheumatism
Volume 46, Issue 2, pages 557–559, February 2002
How to Cite
Morgan, A. W., Griffiths, B., Barrett, J. H., Markham, A. F., Emery, P. and Isaacs, J. D. (2002), Reply. Arthritis & Rheumatism, 46: 557–559. doi: 10.1002/art.10123
- Issue published online: 5 FEB 2002
- Article first published online: 5 FEB 2002
- Arthritis Research Campaign
- Medical Research Council, UK
To the Editor:
We thank Nieto et al for their interest in our study, which demonstrated an association between FcγRIIIA and RA in both UK Caucasian and North Indian/Pakistani cohorts (1). We welcome the opportunity to discuss the differences in analytic and genotyping approaches used in our respective studies (2), which may contribute to the different results.
In our study, the 2 cohorts of RA patients were recruited from the same district general hospitals, and local population control subjects were recruited from the same geographic area as the RA patients. We presented the results for each ethnic group separately and in combination. Because of the major ethnic differences in the various Indian/Pakistani populations, recruitment of our patients and controls was restricted to individuals from the states of Kashmir, Punjab, Haryana, Himachial, Pradesh, or Delhi. Individuals from these areas are of Aryan descent and differ from southern Indians, who are of Dravidian descent. The Arya arrived in India in approximately 1500 BC and are believed to be from the Caspian region or the southern steppes of Russia, and, therefore, they have a degree of genetic similarity with white Europeans (3). The FcγRIIIA allele frequencies of the UK Caucasian and North Indian/Pakistani cohorts varied by only 1%. We therefore also combined data across the 2 ethnic groups, which increased the power for detecting an association between the FcγRIIIA-158F/V polymorphism and RA. Table 1 presents genotypic data for the combined group, both with and without adjustment for ethnicity, using the Mantel-Haenszel method for combining risk estimates across strata (4). There was no evidence against homogeneity of the odds ratios, and one can see that the results with and without adjustment are almost identical, confirming the acceptability of combining these data. Nieto and colleagues thought that the frequency of the “shared epitope” in our North Indian/Pakistani population was low at 17%. However, this percentage actually compares well with previously reported frequencies of 19.6–24% in this ethnic group (5–7). Therefore, the “shared epitope” frequency of 35% quoted by Nieto et al is (8) at variance with much of the published literature.
|Population, genotype||OR (95% CI, P)||P for trend of odds|
|FcγRIIIA-158FV||1.50 (0.89–2.52, 0.13)||0.03|
|FcγRIIIA-158VV||2.18 (0.96–4.96, 0.06)|
|FcγRIIIA-158FV||1.86 (1.06–3.27, 0.03)||0.04|
|FcγRIIIA-158VV||1.84 (0.66–5.13, 0.23)|
|Combined ethnic group†|
|FcγRIIIA-158FV||1.65 (1.12–2.41, 0.01)||0.003|
|FcγRIIIA-158VV||2.09 (1.10–3.95, 0.02)|
|Combined, adjusting for ethnicity†|
|FcγRIIIA-158FV||1.65 (1.13–2.43, 0.01)||0.003|
|FcγRIIIA-158VV||2.04 (1.08–3.88, 0.03)|
|FcγRIIIA-158FV||0.79 (0.37–1.70, 0.55)||0.04|
|FcγRIIIA-158FF||1.65 (0.76–3.61, 0.20)|
Nieto and colleagues analyzed our data using a method different from the one that we used and obtained distinctly different results. First, they calculated a borderline odds ratio for the association of FcγRIIIA alleles in our North Indian/Pakistani cohort (P = 0.05). We chose not to calculate odds ratios from our allele data, however, because it is not immediately obvious how the relative odds of an allele occurring in patients and controls translate into a statement about the risk of a disease (9). In contrast, odds ratios derived directly from genotypic data are easier to interpret. Our report and that of Nieto et al also highlight the fact that 2 × 3 genotypic table analyses are not straightforward in that a variety of methods are available for their interpretation (9). Neither Nieto et al nor our group tested an a priori hypothesis specifying the FcγRIIIA-158F or V allele as the putative risk allele in RA, and data were therefore analyzed according to the genotyping results.
Sasieni (9) has suggested that an appropriate method for analyzing genetic case–control data is to calculate odds ratios for homozygotes for the putative disease-associated allele (FcγRIIIA-158V from our data, FcγRIIIA-158F from the data of Nieto et al) and for heterozygotes, using homozygotes for the alternative allele as the reference group. Under a dominant model the odds ratios for homozygotes and heterozygotes would be equal, under a codominant model (in which each allele contributes to disease risk) a trend in odds ratios would be expected, and under a recessive model the odds ratio for heterozygotes would be 1.0.
We applied Sasieni's approach to both our own data and those of Nieto et al (Table 1). Our own data best fit a codominant model and, in our original publication, we presented data for carriage of the FcγRIIIA-158V allele and for homozygosity of the FcγRIIIA-158V allele. The data presented by Nieto et al are most consistent with a recessive model, with the apparent decreased risk of disease in the heterozygotes interpreted as being attributable to sampling variation. The odds ratios presented by Nieto et al therefore compare the FcγRIIIA-158FF genotype with a reference group consisting of heterozygotes and FcγRIIIA-158VV homozygotes, showing a difference that is just significant at the 1% level. The fact that a recessive model was used to analyze their data must be considered when interpreting their results.
In our UK Caucasian group with nodular RA, both carriage of the FcγRIIIA-158V allele (OR 2.2, 95% CI 1.0–5.1, P = 0.004) and homozygosity for the FcγRIIIA-158V allele (OR 4.4, 95% CI 1.5–12.9, P = 0.004) were associated with RA. We accept that our results for the North Indian and Pakistani cohort just failed to reach significance for an association of homozygosity of the FcγRIIIA-158V allele with nodular RA (OR 14.8, 95% CI 1.2–179.7, P = 0.05). As we stated in our report, however, of the 4 patients in this cohort who had nodules, 3 carried the FcγRIIIA-158V allele, 2 of whom were homozygous.
The most appropriate control group for this type of analysis depends on the hypothesis being tested. The primary objective of our study was to determine if the FcγRIIIA-158V/F polymorphism was associated with RA, not to study parameters of RA severity. In this context, we thought that a comparison between the subset of patients with nodular RA, which may represent a different etiopathogenic RA group, and the control population was most appropriate. In fact, analysis of RA patients with and those without nodules demonstrated that homozygosity for FcγRIIIA-158V was associated with the presence of nodules in the UK Caucasian RA patients (OR 3.6, 95% CI 1.1–11.8, P = 0.03) and showed a trend toward significance in the North Indian/Pakistani cohort (OR 10.3, 95% CI 0.8–129.6, P = 0.09 [data not shown]).
Last, we would like to draw attention to the differing FcγRIIIA allele frequencies in our control populations. Although this difference could be attributable to the distinct ethnic groups studied, Table 2 summarizes the literature on this point. FcγRIIIA-158F/V genotyping is problematic because of the high sequence homology with FcγRIIIB. We chose to genotype this polymorphism by direct sequencing, because, in our hands, conventional assays produced inconsistent results, particularly when the DNA quality was poor, as reported by another group (10). One striking feature when comparing the FcγRIIIA-158F/V allele frequencies in published control populations (Table 1) is that there are 2 clusters, with the frequency of the V allele being in the range of either 0.27–0.33 or 0.41–0.47. This clustering could represent genuine interethnic differences, although whenever the same group previously analyzed 2 ethnically diverse populations, similar allele frequencies for each ethnic group were obtained (1, 10–14). Loss of assay specificity for FcγRIIIA over FcγRIIIB would result in an overrepresentation of the FcγRIIIA-158V allele caused by sequence homology. We are the only group that has directly sequenced large cohorts for this polymorphism, and we have obtained the lowest V allele frequencies. Indeed, when we genotyped a subset of our controls using conventional amplification-refractory mutation screening–polymerase chain reaction, this method resulted in an apparent V allele frequency of 0.40.
|Population studied||Allele frequency||Genotype frequency||Assay technique||Ref.|
|North Indian and Pakistani||0.73||0.27||0.52||0.41||0.07||Direct sequencing||Morgan et al (1)|
|UK Caucasian||0.72||0.28||0.52||0.39||0.09||Direct sequencing||Morgan et al (1)|
|Japanese||0.72||0.28||0.54||0.36||0.10||ARMS-PCR||Sugita et al (10)|
|US Caucasian||0.70||0.30||0.50||0.39||0.11||ASO||Lehrnbecher et al (11)|
|Dutch Caucasian||0.67||0.33||0.43||0.47||0.10||ARMS-PCR||Dijstelbloem et al (14)|
|US African||0.67||0.33||0.42||0.50||0.08||ASO||Lehrnbecher et al (11)|
|Spanish Caucasian||0.59||0.41||0.32||0.54||0.14||RFLP: Rsa I + Sty I||Nieto et al (2)|
|Dutch Caucasian||0.57||0.43||0.32||0.51||0.17||RFLP: Nla III||Koene et al (15)|
|US ethnically diverse||0.56||0.44||0.26||0.61||0.13||ARMS-PCR||Wu et al (12)|
|US African||0.56||0.44||0.34||0.44||0.22||RFLP: Nla IV||Oh et al (16)|
|Korean||0.53||0.47||0.20||0.66||0.14||ARMS-PCR||Salmon et al (13)|
In summary, the differences between our results and those obtained by Nieto et al may reflect the different analytic strategies and models used and possibly also the genotyping methods. Further studies in our laboratory are currently under way to determine the role of FcγRIIIA in RA susceptibility in larger RA populations and in different aspects of RA pathogenesis and severity. We are also addressing the possible influence of other FcγR located in the same gene cluster on chromosome 1.
Supported by the Arthritis Research Campaign and the Medical Research Council, UK.
- 3The peoples of India. Cambridge: Cambridge University Press; 1913..
Ann W. Morgan MRCP*, Bridget Griffiths MD, MRCP*, Jennifer H. Barrett PhD*, Alexander F. Markham DSc, FRCPath, MRCP*, Paul Emery MD, FRCP*, John D. Isaacs PhD, FRCP*, * University of Leeds Leeds, UK