To examine the association between the IL1 gene cluster and susceptibility to ankylosing spondylitis (AS) in 3 independent case–control cohorts.
To examine the association between the IL1 gene cluster and susceptibility to ankylosing spondylitis (AS) in 3 independent case–control cohorts.
We analyzed 394 patients and 446 controls from Alberta, Newfoundland, and Toronto, Canada. Samples were genotyped using a panel of 38 single-nucleotide polymorphism (SNP) markers within the IL1 gene cluster. Data from 20 informative and nonredundant SNP markers were analyzed using several association test strategies. First, we used the program WHAP to identify single-marker associations. Second, we used WHAP to analyze “sliding windows” of 3 contiguous markers along the entire extent of the IL1 gene cluster in order to identify haplotypic associations. Third, we used the linkage disequilibrium mapping program DMLE to estimate the posterior probability distribution of a disease locus.
A total of 14 SNP markers showed significant single-locus disease associations, the most significant being rs3783526 (IL1A) (P = 0.0009 in the Alberta cohort, P = 0.04 in the Newfoundland cohort) and rs1143627 (IL1B) (P = 0.0005 in the Alberta cohort, P = 0.02 in the Newfoundland cohort). Analysis of 3-marker sliding windows revealed significant and consistent associations with all of the haplotypes in the IL1A and IL1B loci in the Alberta cohort and with IL1B in the Newfoundland cohort, especially haplotypes rs1143634/rs1143630/rs3917356 and rs1143630/rs3917356/rs3917354 (P = 0.006–0.0001). With DMLE, a strong peak in the probability distribution was estimated near IL1A in both the Alberta and the Newfoundland populations.
These results indicate that the IL1 locus, or a locus close to IL1, is associated with susceptibility to AS.
Ankylosing spondylitis (AS) is one of the most common causes of inflammatory arthritis, with an estimated prevalence of 0.1–0.9%. Genetic factors have been strongly implicated in its etiology, and heritability as assessed by twin studies has been estimated to be >90% (1). Although HLA–B27 is the major gene associated with AS and is present in >95% of white individuals with this disease, only 1–5% of random B27 carriers develop disease, and the contribution of B27 to the overall genetic predisposition has been estimated at only 20–30% (1). Although there is evidence supporting the notion of contribution from an additional gene(s) in the HLA region with a total estimated contribution of 40–50% for this region, it is apparent that non–major histocompatibility complex (non-MHC) genes contribute a major portion of genetic susceptibility to AS (2). Thus, although the sibling risk ratio (λS) has been estimated to be 82, the λHLA value ranges only from 5 to 6 while the λnon-MHC has been estimated to be 14, which is substantially greater than the entire genetic component of rheumatoid arthritis (λS = 6) (3).
Genome-wide scans have demonstrated several areas of suggestive or significant linkage in non-MHC regions, including 1p, 1q, 2p, 3p, 2q, 5q, 6q, 9q, 10q, 11p, 11q, 16q, 17p, 17q, and 19q, although consistent linkages have been identified infrequently and the individual contributions of non-MHC regions are small (overall risk ratio [λnon-MHC] <2.0) (4–6). Candidate gene studies for non-MHC genes are limited, and the only association that has been reported in 2 independent disease cohorts is with the CYP2D6 locus on chromosome 22, which has not been implicated in genome-wide scans (7, 8). In addition, several case–control studies have examined the interleukin-1 (IL-1) family gene cluster, which lies 123–126 cM from the p-telomere of chromosome 2 in a region linked to disease susceptibility in a genome-wide scan (4). This gene cluster includes, within a 360-kb region, the IL-1α (IL1A), IL-1β (IL1B), IL1F7, IL1F9, IL1F6, IL1F8, IL1F5, IL1F10, and IL-1 receptor antagonist (IL1RN) genes. The IL1F5–IL1F10 genes in this cluster are widely expressed, including on activated monocytes and B cells, and signal through a range of IL-1 receptors (9, 10). Several may function as IL-1 antagonists due to their sequence similarity to IL1RN, although their function is largely unknown.
Two independent studies demonstrated associations with allele 2 of a variable-number tandem repeat (VNTR) in intron 2 of the IL1RN gene but not with polymorphisms in IL1A or IL1B (11, 12), while another case–control study demonstrated single-marker as well as haplotypic associations with single-nucleotide polymorphisms (SNPs) in exon 6 of IL1RN (13). This haplotypic association was confirmed in a family-based association study of western Canadian families with AS (13). However, 1 family-based association study that included 37 families (14) and another that included 244 affected sibpairs (15) showed no evidence of linkage to genes in the IL1 cluster. Moreover, a recent case–control study showed no associations with markers in IL1RN but a highly significant association with markers in IL1B and IL1F10, which was confirmed in a family-based association study (16).
We examined 3 independent case–control cohorts of white Canadians in different geographic locations (western, central, and eastern Canada) that have been phenotypically well-characterized in a systematic manner by investigators collaborating in the Spondyloarthritis Research Consortium of Canada disease network. The number of SNPs required to cover a genomic region depends on the linkage disequilibrium (LD) structure of the region, and this knowledge has been used to analyze haplotypes of SNP markers for associations with several diseases, based on the assumption that haplotypes capture more of the LD structure and that such association studies would therefore be more powerful. Recent studies of LD in the human genome demonstrate that the structure of pairwise LD in the genome is quite variable, implying that the number of SNPs required to cover regions is also highly variable (17–19). Therefore, it seems unlikely that there is a single best approach to association testing and, although some investigations illustrate that haplotype-based tests of association are more powerful than tests based on multilocus genotypes or a series of single-locus tests (20), others indicate the opposite conclusion (21). We addressed these concerns by using both haplotype- and single-locus–based tests. In addition, we used fine-scale multipoint LD mapping to seek further evidence of association with AS in the region.
Three cohorts of patients and controls from different rheumatic disease centers in Canada were studied. All patients were diagnosed as having AS according to the modified New York criteria (22). A total of 394 patients and 446 controls were analyzed.
The first cohort consisted of 200 unrelated white AS patients attending the outpatient department of the University of Alberta Hospital. Of these, 13 had inflammatory bowel disease (IBD), 12 had psoriasis, and 1 had reactive arthritis. This population included 143 men and 57 women with a mean age of 42.9 years (range 22–76) and a mean disease duration of 18.7 years (range 2–39), of whom 92.7% were HLA–B27 positive. Controls were 200 healthy unrelated white blood donors from Edmonton. Ethical approval for the study was obtained from the University of Alberta Bioethics Research Committee. This case–control cohort represents a random selection from the 400 cases and controls we described in a previous report (13). The ethnic composition of this cohort is primarily northern and eastern European.
The second cohort comprised 82 unrelated white AS patients attending the outpatient department at the Toronto Western Hospital. Seven had IBD, 5 had psoriasis, and 1 had reactive arthritis. This population included 56 men and 26 women with a mean age of 41.7 years (range 23–74) and a mean disease duration of 18.2 years (range 2–38); 89% were B27 positive. Controls were 96 healthy unrelated white individuals from Toronto. Ethical approval for the study was obtained from the University Health Network Research Committee. The ethnic composition of this cohort is broadly European.
The third cohort included 112 unrelated white AS patients attending the outpatient rheumatology department of Memorial University, Newfoundland. Ten of these patients had IBD, 20 had psoriasis, and 4 had reactive arthritis. There were 82 men and 30 women with a mean age of 42.3 years (range 20–76) and a mean disease duration of 17.1 years (range 2–41), of whom 87% were B27 positive. Controls were 150 healthy unrelated white blood donors from St. John's, Newfoundland. Ethical approval for the study was obtained from the Memorial University Bioethics Research Committee. The ethnic composition of this cohort is relatively restricted to individuals of Irish/southwest English descent.
Cases and controls were genotyped for a panel of 38 SNPs located throughout the IL1 gene cluster but specifically located within the IL1B, IL1A, and IL1RN genes, based on previous association studies (13, 16). SNPs were identified from the literature and by searching public SNP databases. DNA was extracted from peripheral blood and the final concentration adjusted to 5 ng/μl. Samples were genotyped by time-of-flight mass spectrometry using the Sequenom platform (Sequenom, San Diego, CA). First, 2.5 ng of genomic DNA was amplified under standard conditions using forward and reverse primer pairs. Reactions were multiplexed where possible. After DNA amplification, all unincorporated nucleotides in the polymerase chain reaction (PCR) product were deactivated using shrimp alkaline phosphatase. A primer extension reaction was then carried out using the mass extend primer and the appropriate termination mix. The primer extension products were then cleaned and spotted onto a SpectroChip. The chip was scanned using a mass spectrometry workstation (Bruker Daltonics, Billerica, MA), and the resulting spectra were analyzed with Sequenom SpectroTyper-RT software. Alleles of the IL1RN 86-bp VNTR were characterized following PCR amplification and separation by electrophoresis on 2% agarose gels.
Of the 38 SNPs originally typed, 20 were kept for the analyses. The SNPs and the primers used for their characterization are listed in Table 1. Eight of the 38 (rs3783572, rs3917360, rs3917348, rs3087258, rs2708943, rs2121335, rs2289934, and rs1562304) were removed because their minor allele frequencies were <0.05 in both cases and controls in at least 1 of the populations, and 1 (rs1143633) was removed because of violations of Hardy-Weinberg equilibrium. For this latter marker, all 3 populations showed a heterozygote deficiency in both cases and controls. The program DHW was used to help determine whether this was more likely to be caused by a disease association or by random chance/genotyping errors (23). This analysis showed that the probability of the violation being due to an underlying disease association was <10−8 in all 3 populations (data not shown).
|Marker region*||SNP ID (rs no.)||Position, kb||Primer|
|IL1Bexon 4 (3953)||1143634||58.31||5′-ACGTTGGATGCAGTTCAGTGATCGTACAGG-3′||5′-ACGTTGGATGGTGCTCCACATTTCAGAACC-3′||5′-CACATTTCAGAACCTATCTTCTT-3′|
|IL1F7 exon 2||2708943‡||142.63||5′-ACGTTGGATGGTGAAGTGTTGATATCTGGTG-3′||5′-ACGTTGGATGGAATGCTGAATTTCTTCGGG-3′||5′-GGTTTAAGTTCTTCACCTTT-3′|
|IL1F6 exon 3||2289934‡||232.20||5′-ACGTTGGATGGAATGGACTCAATCTCTGCC-3′||5′-ACGTTGGATGAAATGGCAGGACCTGTCTTC-3′||5′-TGCAGTGTGGGCTGGTCCCC-3′|
|IL1RN11100 exon 4||315952†||358.22||5′-ACGTTGGATGGGCAGACTCAAAACTGGTGG-3′||5′-ACGTTGGATGAACAGAAAGCAGGACAAGCG-3′||5′-AAACTGGTGGTGGGGCC-3′|
Measures of pairwise LD were determined using Haploview (Whitehead Institute for Biomedical Research, Cambridge, MA). This program uses an expectation-maximization algorithm to calculate maximum-likelihood estimates of haplotype frequencies, given genotype measurements (24). All groups of adjacent high-LD markers (defined as having a D′ of 1.0 and an r2 of >0.80) were reduced to a single marker for the final analyses. In such multimarker groups, the marker with the most significant association with AS was kept. Nine markers (rs1533463, rs2071373, rs1800587, rs1143637, rs16944, rs990524, rs1794065, rs315952, and rs895495) were removed based on these criteria, leaving a core 20-SNP marker set (designated markers 1–20) (Table 2). The haplotype block organization of the IL1 gene cluster was determined using the default settings in Haploview version 3.11.
|Cases||Controls||P||OR (95% CI)||Cases||Controls||P||OR (95% CI)||Cases||Controls||P||OR (95% CI)|
|1. rs2856836||T||0.69||0.73||0.27||0.84 (0.62–1.14)||0.68||0.71||0.49||0.87 (0.60–1.27)||0.71||0.7||0.87||1.04 (0.66–1.66)|
|2. rs3783550||A||0.72||0.62||0.002||1.61 (1.20–2.18)||0.70||0.65||0.3||1.23 (0.85–1.80)||0.65||0.68||0.57||0.87 (0.56–1.37)|
|3. rs3783547||T||0.60||0.67||0.09||0.75 (0.55–1.02)||0.59||0.65||0.24||0.77 (0.51–1.16)||0.63||0.62||0.89||1.03 (0.66–1.62)|
|4. rs3783543||C||0.72||0.62||0.002||1.60 (1.18–2.15)||0.69||0.65||0.34||1.21 (0.83–1.75)||0.61||0.64||0.67||0.91 (0.59–1.40)|
|5. rs17561||G||0.69||0.73||0.16||0.80 (0.59–1.09)||0.67||0.69||0.69||0.92 (0.64–1.35)||0.73||0.70||0.66||1.12 (0.71–1.78)|
|6. rs3783526||G||0.73||0.62||0.0009||1.68 (1.24–2.28)||0.75||0.66||0.04||1.53 (1.03–2.29)||0.65||0.68||0.58||0.88 (0.56–1.38)|
|7. rs1800794||C||0.69||0.74||0.14||0.80 (0.59–1.08)||0.68||0.71||0.46||0.86 (0.59–1.26)||0.72||0.73||0.81||0.94 (0.59–1.51)|
|8. rs1143643||G||0.56||0.65||0.016||0.70 (0.52–0.95)||0.60||0.65||0.26||0.81 (0.56–1.17)||0.64||0.53||0.04||1.57 (1.00–2.47)|
|9. rs1143634||C||0.75||0.83||0.018||0.63 (0.44–0.91)||0.80||0.80||0.86||1.04 (0.66–1.65)||0.78||0.83||0.23||0.71 (0.41–1.24)|
|10. rs1143630||C||0.93||0.93||0.73||0.90 (0.52–1.56)||0.95||0.95||0.94||1.04 (0.46–2.37)||0.93||0.93||0.78||1.13 (0.50–2.56)|
|11. rs3917356||G||0.52||0.58||0.09||0.78 (0.59–1.04)||0.50||0.57||0.08||0.73 (0.52–1.04)||0.59||0.56||0.66||1.10 (0.72–1.68)|
|12. rs3917354||T||0.77||0.83||0.03||0.67 (0.48–0.96)||0.79||0.80||0.77||0.94 (0.61–1.43)||0.78||0.82||0.36||0.77 (0.46–1.30)|
|13. rs1143627||T||0.72||0.60||0.0005||1.72 (1.28–2.33)||0.75||0.64||0.02||1.68 (1.13–2.50)||0.67||0.61||0.28||1.29 (0.83–2.00)|
|14. rs3811047||G||0.67||0.73||0.1||0.77 (0.57–1.04)||0.66||0.76||0.02||0.61 (0.41–0.90)||0.74||0.76||0.76||0.93 (0.57–1.51)|
|15. rs2723187||C||0.92||0.93||0.54||0.85 (0.50–1.43)||0.91||0.94||0.17||0.61 (0.31–1.20)||0.92||0.96||0.19||0.53 (0.21–1.34)|
|16. rs895497||C||0.75||0.75||0.97||0.99 (0.71–1.38)||0.71||0.75||0.25||0.80 (0.53–1.19)||0.72||0.73||0.88||0.97 (0.60–1.56)|
|17. rs1900287||A||0.68||0.75||0.049||0.74 (0.54–1.00)||0.67||0.66||0.85||1.04 (0.71–1.50)||0.65||0.70||0.29||0.78 (0.50–1.23)|
|18. rs3811058||T||0.94||0.91||0.24||1.38 (0.81–2.35)||0.89||0.89||0.9||0.97 (0.55–1.70)||0.94||0.84||0.002||3.16 (1.46–6.86)|
|19. rs419598||T||0.73||0.66||0.05||1.37 (1.01–1.86)||0.68||0.71||0.37||0.85 (0.59–1.23)||0.69||0.70||0.87||0.96 (0.63–1.47)|
|20. rs315951||G||0.66||0.72||0.11||0.78 (0.58–1.05)||0.67||0.74||0.07||0.71 (0.50–1.04)||0.79||0.66||0.004||1.91 (1.24–2.94)|
Since the 3 samples were from geographically distinct areas, the populations were analyzed separately and not pooled. All P values are shown without correction for multiple comparisons. Individual P values (for markers or haplotypes) were obtained by permutation (10,000 randomizations). Since many marker pairs are in high LD with one another, a Bonferroni correction for multiple testing could be overly conservative. Corrected P values for multiple comparisons were obtained by the permutation method of Churchill and Doerge (25). In general, the global significance level calculated in this manner was very similar to that obtained by Bonferroni correction.
Allelic and haplotypic associations were calculated using the WHAP software package (26, 27), which uses SNPHAP (28) to estimate haplotypes via a standard expectation-maximization algorithm. WHAP performs regression-based single-marker and haplotype association tests. Standard chi-square tests were used for comparing genotype frequencies in cases and controls. Odds ratios and 95% confidence intervals were calculated. WHAP can be used for overall (“omnibus”) haplotype frequency tests as well as for haplotype-specific tests. The power of the omnibus test can be increased by an independent secondary test, based on the pairwise correlation between haplotype genetic similarity and haplotype effect similarity. This combined probability, or “improved” omnibus statistic, can be calculated automatically using WHAP. All WHAP-based P values were determined by permutation. Analysis of haplotypic associations was done systematically across the entire extent of the IL1 gene cluster using a “sliding window” approach (29), in which haplotypes comprising 3 SNPs were analyzed consecutively, beginning with markers 1 (rs2856836), 2 (rs3783550), and 3 (rs3783547) in IL1A.
We also performed fine-scale multipoint LD mapping of each population using DMLE+ (30). This program uses a Bayesian Markov chain Monte Carlo procedure to estimate the posterior probability of the disease mutation's location, given genotype data from a sample of unrelated cases and controls. The scale of resolution is much finer than that which is possible using linkage mapping, due to the increased number of recombination events available using population data. This program makes a number of assumptions, the most important of which are that 1) the majority of cases share the same susceptibility allele, identical by descent, and 2) the susceptibility allele is relatively rare in the general population. Simulation analyses have shown that good results are obtainable even with a fair degree of genetic heterogeneity and phenocopy, and incomplete penetrance.
The average pairwise LD across all markers in the Alberta, Newfoundland, and Toronto cohorts, respectively, was 0.64, 0.60, and 0.55 in controls and 0.63, 0.61, and 0.57 in cases. LD was significantly greater in the Alberta and Newfoundland groups compared with the Toronto group, in both cases and controls (P < 0.02). There were no significant differences between cases and controls in any individual cohort. LD was particularly evident in both the Alberta and Newfoundland populations for alleles of SNPs located in the IL1A and IL1B regions of the IL1 gene cluster (Figure 1). This was less evident in Toronto controls. LD was less evident between markers in the IL1A or IL1B regions and markers in the IL1RN gene.
Analysis of SNP markers using the WHAP program showed that 14 SNP markers had significant disease associations in at least 1 population cohort (Table 2). Single-marker associations in the Alberta cohort were primarily noted in the IL1A and IL1B regions of the IL1 gene cluster, with 4 markers being significantly associated after experiment-wise correction for multiple comparisons: rs3783550 (IL1A intron), rs3783543 (IL1A intron), rs3783526 (IL1A untranslated), and rs1143627 (IL1B untranslated). The association with the latter 2 SNP markers was also significant in the Newfoundland cohort before correction. Moreover, the direction of case–control allele frequency bias was the same in the Alberta and Newfoundland cohorts for 12 of the 13 SNP markers in IL1A and IL1B, but for only 5 of the 13 markers when Alberta and Toronto were compared. The most significant associations noted in the Toronto cohort were with SNP markers in IL1F10 (rs3811058) and IL1RN (rs315951). Analysis of the IL1RN VNTR revealed no significant disease associations in any of the cohorts.
Analysis of the entire 20-SNP marker set using WHAP revealed a significant overall haplotypic association in the Alberta cohort (omnibus P = 0.04). Overall haplotypic association was particularly strong for the combined marker set of 13 SNPs in IL1A and IL1B (omnibus P = 0.0002), and 3 specific 13-marker haplotypes were identified as being disease associated: TACCGGCACCATT (susceptible) (P = 0.02), TCTTGACGCCGTC (protective) (P = 0.00003), and CATCTGTGTCGCT (susceptible) (P = 0.02).
Analysis of 3-marker sliding windows using WHAP revealed significant associations with all of the haplotypes in the IL1A and IL1B loci in the Alberta cohort and in IL1B in the Newfoundland cohort (Table 3). In contrast, weak haplotypic associations with haplotypes spanning the IL1F6, IL1F8, IL1F10, and IL1RN loci were noted in the Toronto cohort. Several specific 3-marker haplotypes were significantly associated with disease in both the Alberta and the Newfoundland cohorts (Table 4). The most significant of these associations were noted with 2 haplotypes comprising markers 9, 10, and 11 (rs1143634, rs1143630, rs3917356) and markers 10, 11, and 12 (rs1143630, rs3917356, rs3917354) in the IL1B gene. Weaker associations were observed in both the Alberta and the Toronto cohorts for the haplotype comprising markers 16, 17, and 18 (rs895497, rs1900287, rs3811058) spanning the IL1F6, IL1F8, and IL1F10 loci and for the haplotype comprising markers 18, 19, and 20 (rs3811058, rs419598, rs315951) spanning the IL1F10 and IL1RN loci; these were not significant after experiment-wise correction.
|3-marker sliding window||P|
|3-marker sliding window||Association*||Specific haplotype||P†|
We analyzed a subset of 13 SNPs demonstrating significant single-marker associations plus 2 SNPs in IL1RN, using DMLE to identify any peaks in probability distribution for disease associations in the IL1 gene cluster (Figure 2). Map positions are reported relative to the first marker, rs2856836 (0.0 cM), with the most distant marker (rs315951) at position 0.395 cM. Under a dominant model, a strong peak in the probability distribution for the Alberta and Newfoundland populations, located at approximately −0.010 to −0.020 cM (approximately −10 to −20 kb) from marker 1 (rs2856836 [IL1A]), was estimated with DMLE. A secondary peak in the probability distribution in the IL1A/B region was identified in the Alberta cohort. The Toronto population showed a much broader, lower likelihood peak in the IL1F6–IL1F10 region, indicating a less informative data set. When the same data were tested assuming a recessive model, there was less agreement among populations, and the peak probabilities were not as high as in the dominant model.
Our analyses of a panel of SNPs in the IL1 gene cluster support the notion that there is an AS disease susceptibility locus near the IL1A or IL1AB region in the Alberta population. The Newfoundland cohort showed very similar patterns of allelic and haplotypic associations, although none were significant after correction for multiple comparisons. LD mapping also indicated a strong similarity between the Alberta and Newfoundland populations, indicating a high probability of a disease locus near the IL1A/B region. This similarity is somewhat surprising given the substantial geographic separation and ethnic disparity between the 2 populations. The Toronto cohort generally exhibited opposite and weaker patterns of association.
We adopted, a priori, a structured approach in which SNPs were first chosen on the basis of having a minor allele frequency of >5%. Associations with alleles that have a frequency of <5% could simply reflect random events and/or genotyping error. Moreover, the expectation-maximization algorithm for inferring haplotypes is known to be inaccurate with low-frequency haplotypes (28). In a second step, we determined pairwise LD, and all groups of adjacent high-LD markers were reduced to a single marker for the final analyses.
Several groups have now reported case–control studies in which extended haplotypes in the IL1 gene cluster have been associated with specific features of osteoarthritis (31–33). Given the low levels of LD seen between IL1A/B and ILRN in our control populations, there is little reason to expect that an extended haplotype spanning this region would be strongly associated with disease. For this reason, we chose to use a 3-marker “sliding window” analysis of the region. Small haplotype segments such as these can sometimes be more strongly associated with the causal variant than would be single markers. At the same time, they are less likely to have had their disease association broken down by the large number of recombinations, as indicated by the low observed LD, than would be haplotypes spanning larger regions.
The differences between the Alberta and Newfoundland populations in the significance levels of the associations are possibly due to insufficient sample sizes in the latter. The similarity between the 2 groups in both association and LD mapping results is evidence in support of this hypothesis. Pooling of the Newfoundland and Alberta data generally resulted in more significant P values for single-marker and haplotypic associations than were seen in the analyses of the individual cohorts, and a higher DMLE peak was found in the same region (data not shown). This suggests that low sample size was the primary factor accounting for the relative lack of statistical significance in the Newfoundland cohort.
With respect to the discrepancy in the results in the Toronto population as compared with the Alberta and Newfoundland populations, several possible reasons may be cited, including false-positive results due to population stratification, false-negative findings due to small effect size coupled with inadequate sample size, and population-specific differences (34). In our view, a major factor accounting for the discrepancy is the multiple founders of various ethnicities that comprise the white population in Toronto. This can lead to population stratification or population-specific effects. Alberta and Newfoundland represent more homogenous northeastern European white populations. The benefits of pursuing gene identification studies in such populations include an increase in allele and locus homogeneity, as well as the often-discussed potential increase in linkage disequilibrium. The former is of importance given the etiologic heterogeneity that characterizes complex diseases. Thus, detection of a significant signal in the Toronto population is more challenging since modest genetic signals may be overlooked. This notion is further supported by the recent identification of a disease association in the Newfoundland psoriatic arthritis population that was not replicated in the Toronto psoriatic arthritis cohort (35).
Thus, there appear to be population-specific differences between Toronto and the other 2 white populations. We would argue that all reports of association studies should include information on the LD structure of the case and control groups, which differed between the Alberta/Newfoundland populations and the Toronto population in the present study. Future work is likely to define more quantifiable aspects of this phenomenon. Such analysis was not performed in previous AS case–control studies, and associations with alleles in IL1A and IL1B were not reported. This could account for some of the observed discrepancies with our results. Nevertheless, the data suggest that these associations may not be applicable to other populations.
Phenotypic heterogeneity might also account for population differences, and several loci, including 4 on chromosome 2, have been associated with age at onset, disease activity, and severity of functional impairment (36). However, these regions are distinct from those implicated in susceptibility, particularly the IL1 gene cluster. Furthermore, there were no obvious differences in disease severity (as assessed by the Bath AS Disease Activity Index and the Bath AS Functional Index [37,38]) between the 3 population cohorts in the present study (data not shown). A preliminary report has also implicated a locus in IL1A/B in susceptibility to psoriatic arthritis (35), while a family study has suggested familial aggregation of 2 distinct phenotypes differentiated by the presence of peripheral inflammation, psoriasis, and IBD (39). However, there was no significant difference in the prevalence of peripheral arthritis or concomitant psoriasis or IBD between the AS cohorts in our study (data not shown). Furthermore, analysis of subgroups stratified by carriage of B27 or by presence of extraarticular features such as psoriasis or IBD revealed similar case and control allelic frequencies for the primary disease-associated SNPs identified in this study (data not shown).
A previous report from Alberta described an association with SNPs in IL1RN in a much larger cohort (400 cases and 400 controls) (13). Analysis of a randomly selected subgroup of these patients revealed only weak associations with IL1RN, that likely represented LD with a disease locus in the vicinity of IL1A and IL1B. This reasoning is supported by the findings from LD mapping indicating a disease locus in the vicinity of IL1A. Our data are consistent with findings of a previous study that showed no linkage with IL1RN (15) and with a family-based association study in which it was concluded, based on conditional logistic regression analysis, that IL1B-511 (rs16944) and IL1F10-3 (rs3811058) were primarily associated with disease (16). In the latter study, significant residual association was still noted when either of these markers was accounted for, providing evidence of the presence of more than 1 disease-associated haplotype.
IL1B-511 was significantly associated with disease in the Alberta and Newfoundland cohorts (data not shown) but was not selected for the 20-marker set because of its strong LD with marker rs3783526, while marker rs3811058 was associated with disease only in the Toronto cohort. It is therefore possible that either there are several haplotypes that carry the disease-associated allele or there is more than 1 disease susceptibility locus for AS in the IL1 gene cluster. Alternatively, disease-associated loci within IL1 may interact with respect to susceptibility and/or phenotype. This might also account for previously reported disease associations with polymorphisms in IL1RN but not IL1A/B (11, 12). A further possibility is that certain IL1 polymorphisms might have a more prominent role in influencing the rate of development of ankylosis, which might differ across populations. No prospective studies have yet addressed this issue.
Association studies have revealed a diverse and largely unrelated group of diseases that may have associations with polymorphisms in genes within this cluster. Allele 2 of the IL1RN VNTR has been associated with type 1 diabetes mellitus, lichen sclerosus, psoriasis, ulcerative colitis, systemic lupus erythematosus, diabetic nephropathy, multiple sclerosis, osteoporotic fractures, essential hypertension, attention deficit hyperactivity disorder, and alopecia areata, as well as survival following severe sepsis (40–52). Polymorphisms within IL1B have been associated with an increased risk for gastric cancer and Parkinson's disease (53, 54). It has been suggested that genetic polymorphisms at both IL1A (rs1800587) and IL1B (rs1143634) are associated with early-onset adult periodontitis (55). Several studies have also demonstrated associations between the IL1A promoter polymorphism, rs1800587, and Alzheimer's disease (56, 57).
There have been reports that interindividual differences in IL1 production may have a genetic basis. One study demonstrated a haplotype based on SNPs rs16944 (IL1B-511) and rs1143634 (IL1B3953) that was associated with increased whole blood secretion of IL1B after lipopolysaccharide (LPS) stimulation, in 2 independent populations (58). As pointed out by those authors, polymorphism at rs16944 altered an activator protein 2 promoter element. On the other hand, others were unable to demonstrate any effect of several polymorphisms in the IL1 gene cluster, including rs16944 and rs1143634, on LPS-stimulated IL1B secretion in a controlled in vitro assay using freshly isolated monocytes (59); rather, the IL1RN VNTR was shown to be associated with differences in release of both IL1B and IL1RN. However, there is little evidence to suggest that the IL1RN VNTR directly influences the mechanism of IL1B secretion.
The role of IL1 in the pathogenesis of AS is presently unclear. Limited analyses of sacroiliac joint biopsy specimens by in situ hybridization and of peripheral synovial tissue by microarray analysis have not revealed preferential expression of IL1 (60), while analyses of serum samples from AS patients have yielded conflicting findings (61, 62). Similarly, open-label studies with anakinra, a biosynthetic IL1RN, showed therapeutic benefit in one study but not in another (63, 64). However, examination of peripheral joint synovial tissue by microarray analysis has shown levels of IL-1β expression comparable with that observed in rheumatoid arthritis (65).
Our results support the notion that there is at least 1 disease susceptibility locus in the IL1 gene cluster. Consistent associations with polymorphisms in the IL1A/B region were evident in 2 of 3 case–control cohorts despite substantial ethnic separation. LD mapping provided supporting evidence for the presence of a disease-associated allele in the IL1 region.
We thank Dr. Shaun Purcell for helpful advice in the application of the WHAP software program.