The Wegener’s granulomatosis quantitative trait locus on chromosome 6p21.3 as characterised by tagSNP genotyping
- M Heckmann1,
- J U Holle2,
- L Arning1,
- S Knaup1,
- B Hellmich2,
- M Nothnagel4,
- P Jagiello1,3,
- W L Gross2,
- J T Epplen1,
- S Wieczorek1
- 1Ruhr University, Human Genetics, Bochum, Germany
- 2University Hospital of Schleswig-Holstein, Campus Luebeck, and Rheumaklinik Bad Bramstedt, Department of Rheumatology, Luebeck, Germany
- 3Centre of Medical Genetics, Osnabrueck, Germany
- 4Institute of Medical Informatics and Statistics (IMIS), Medical Faculty, Christian Albrechts University, Kiel, Germany
- Dr S Wieczorek, Ruhr University, Human Genetics, MA5/39, 44780 Bochum, Germany;
- Accepted 13 October 2007
- Published Online First 29 October 2007
Background: A genomic region on chromosome 6p21.3, including HLA-DPB1, has been linked to Wegener’s granulomatosis (WG). The basis of this association is difficult to evaluate because of the complex haplotype block architecture of this region.
Objective: To identify the causative molecular genetic variation(s) using a detailed HapMap based fine-mapping approach.
Methods: 282 patients with WG and 380 healthy controls were genotyped for HLA-DPB1 as well as for 35 informative single nucleotide polymorphisms (SNPs) within the respective region. 25 of these SNPs have been selected as tagging SNPs for another 219 associated SNPs. Allele and genotype frequencies were analysed separately by contingency tables and logistic regression. Finally, the coding region of RING1 was directly sequenced in subjects who carried haplotypes that were correlated with contrasting WG risks.
Results: The previously reported strong association of WG with the HLA-DPB1*0401 allele was confirmed in an independent WG sample (n = 108, pc = 6.4×10−8). When the complete cohort (n = 282) was considered, the association remained highly significant in ANCA-positive (pc = 1.26×10−22), but not in ANCA-negative patients. An SNP 3′ of HLA-DPB1 yielded the smallest p value and was associated with WG partly independently from the HLA-DPB1 alleles. Another informative SNP in the vicinity of RING1 showed significant WG association that was also partly independent of HLA-DPB1. RING1 sequencing, however, did not show any variation potentially predisposing to WG.
Conclusions: The HLA-DPB1/RING1 region is strongly associated with WG in ANCA-positive subjects. Further analyses of potential cis regulatory sequences of candidate genes HLA-DPB1, RING1 and RXRB appear warranted.
Wegener's granulomatosis (WG) is an autoimmune disorder of unknown aetiology characterised by granulomatous inflammation of the respiratory tract and systemic necrotising small vessel vasculitis. WG is rare with a prevalence of only 2.3/100 000.1 A hallmark of the disease is the presence of antineutrophil cytoplasmic antibodies (ANCA) to a defined target antigen, proteinase 3 (PR3-ANCA), in ∼95% of all patients with generalised WG.2 In addition to their diagnostic value, PR3-ANCA appear to play a direct pathogenic role in WG as they can trigger activation of neutrophils via interaction with membrane (m) PR3.3 4 Leucocyte mPR3 expression correlates with disease activity in WG, and high mPR3 expression has been shown to be a risk factor for relapse.5 6
Whole-genome association studies have not been published for WG. Rather, numerous genes and loci have been tested individually for statistical association of certain alleles, but many of those studies had small cohort sizes.7 Reproducibly certain alleles of α1-antitrypsin (proteinase inhibitor 1) were associated with WG.8 9 Recently, we identified the PTPN22 620W allele, which is of pathogenetic relevance in numerous inflammatory conditions, as a WG risk factor.10 Additionally, a strong association of WG with a 280 kb-spanning haplotype on chromosome 6p21.3 was detected performing an extended microsatellite screen.11 We subsequently specified the boundaries of this candidate region by additional microsatellite markers within 3.6 MB of the genomic environment.12 Interesting candidate genes for WG are located within this area—for example, HLA-DPB1 and the genes encoding ring finger protein 1 (RING1) and retinoid X receptor β (RXRB). The HLA-DPB1*0401 allele was shown to correlate with increased risk for WG, yet it is still unclear whether the origin of the association is the DPB1 allele itself or a linked variation, a problem of many HLA linked conditions. Linkage disequilibrium (LD) with HLA-DPB1 has been demonstrated for certain RXRB haplotypes earlier.13 While screening coding and promoter regions of RXRB did not disclose WG-predisposing variations,12 such analyses are missing for RING1.
Here we present our data of a detailed single nucleotide polymorphism (SNP) association screen of the WG-linked region based on LD structure information of this locus from HapMap.14 Furthermore, we have screened exonic and neighbouring sequences of RING1 in order to identify the origin of the WG association.
PATIENTS AND METHODS
Two hundred and eighty-two unrelated patients of German origin diagnosed with WG were included in this study. All of them met the American College of Rheumatology classification criteria for WG according to international standards.15 The presence of ANCA was determined as described before.16 Three hundred and eighty healthy German blood donors served as controls. Stratification had been ruled out for the control group earlier.17
The patient cohort comprised 143 of the 150 ANCA-positive patients with WG analysed earlier.11 One hundred and thirty-nine additional patients have now been included, 109 of whom were ANCA-positive, 27 ANCA-negative and no ANCA status was available for three patients. ANCA-negative patients had never had positive ANCA titres, even before any WG treatment was started.
All subjects gave informed consent and ethical principals for medical research involving human subjects according to the Declaration of Helsinki have been followed. The study was approved by the local ethics committee.
HLA-DPB1 alleles were established as described before.18 Alleles with a minor allele frequency (MAF) <5.0% were combined into one category and treated as one allele. p Values for differences in allele distributions between cases and controls were obtained from χ2 tests on contingency tables. Power calculations were performed using Quanto 22.214.171.124
Selection of SNP
We assumed that the European-ancestry population sample (CEU) of the HapMap project was sufficiently representative for the German population. We selected 25 tagging SNPs, or tagSNPs, (SNP1–25, table 2, fig 2) from the HapMap CEU database (data release 20 on NCBI B35 assembly) to capture 244 SNPs in the region 6p21.3. With this aim, we employed the tagger algorithm implemented in the Haploview 3.3.2 software package.20 Parameters for tagSNPs selection were chosen according to well-accepted protocols21 using a multimarker tagging approach (2- and 3-marker haplotype tags, LOD threshold 3.0) and MAF ⩾0.2. The mean correlation (r2) between tagSNPs and captured SNPs was 0.966, with a minimum r2 of 0.8.
Additional SNPs were selected for fine mapping of the genomic vicinity of HLA-DPB1 (SNP26–29), the centromeric edge of the WG-associated locus (SNP30, 31) and the region surrounding RING1 (SNP32–35). The average distance between all 35 SNPs was 6.4 kb, with 27.8 kb (between SNP30 and SNP31) being the largest distance between two neighbouring SNPs (supplementary table 1).
Sixteen SNPs were analysed by polymerase chain reaction-restriction fragment length polymorphism (PCR-RFLP) methods. Primers were selected from the USCS database using the Primer3 software.22 PCRs were run according to standard protocols using HotStarTaq polymerase (Qiagen, Hilden, Germany). For primer sequences and annealing temperatures see supplementary table 2.
TaqMan genotyping kits for the remaining 19 SNPs were selected using the SNPbrowser software 3.5 (Applied Biosystems, Foster City, USA). PCR were performed with TaqMan universal PCR Master Mix, No AmpErase UNG (Applied Biosystems) on an iCycler device (BioRad, Munich, Germany). One of these SNPs (SNP22; rs213213) had already been analysed earlier by RFLP in parts of our patient cohort11 and could therefore be used as a method control. No discrepancies of genotypes were detected between the two methods.
Statistical analysis and haplotype block definition
Deviations from Hardy–Weinberg equilibrium of each SNP marker were evaluated by χ2 tests with one degree of freedom. SNP marker genotypes were tested for association with WG using χ2 tests with two degrees of freedom, as implemented in Haploview. A priori p values have not been corrected for multiple comparisons, since this study comprises a detailed evaluation of a previously established association and p values from individual markers are expected to be (partly) dependent among each other.
We defined haplotype blocks in order to investigate haplotype frequency differences between patients and controls in regions of high LD. We used the definition by Gabriel et al.23 Only subjects with complete genotype data for all 35 SNPs (n = 297) have been included for LD estimation and haplotype block definition. Maximum-likelihood haplotype frequency estimates were obtained separately in patients with WG and controls using Haploview.
To model the contribution of particular SNPs to the association signal, we employed logistic regression considering main effects only. Significance was evaluated using the Wald and the likelihood-ratio test (LRT). In particular, we investigated whether SNP3, 4 and 32–34 as well as HLA-DPB1*0401 would make independent contributions. Logistic regression was carried out in the R statistical software package.24 The HLA-DPB1*0401 allele was compared with all other alleles jointly. Each SNP and HLA-DPB1*0401 were modelled to act either in a dominant, a recessive or in a multiplicative mode.
Correlation of ANCA status or certain genotypes, respectively, and clinical WG manifestation was analysed by contingency table and χ2/Fisher exact test (supplementary tables 6, 7a–c). Age of onset was compared by analysis of variance.
Direct sequencing of RING1
Five ANCA-positive patients homozygous for the highly associated haplotype (SNP18–21 and SNP33–35; TCTCCCT, see “Results”) were subsequently analysed by direct sequencing of the seven RING1 exons and 0.5 kb of the promoter region. For comparison five control subjects carrying the conversely associated haplotype (CCCTACT) in homozygous state were also analysed (for primer sequences see supplementary table 3). PCR products were purified using AMPure (Agencourt/Beckman Coulter, Schiedam, the Netherlands). Sequence reactions were subsequently performed with DYEnamic ET dye terminator kits (Amersham Biosciences, GE Healthcare, Freiburg, Germany). Products were run on a MegaBACE 1000 capillary sequencer (Amersham Biosciences) after purification with CleanSEQ (Agencourt).
Nineteen alleles of HLA-DPB1 were genotyped. Allele frequencies from our control cohort were virtually identical with those published earlier11 and did not differ significantly from those of middle European populations.25 In addition, allele frequencies within each respective group were in Hardy–Weinberg equilibrium (data not shown).
The allele distributions differed highly significantly between patients and controls for the complete additional sample (p = 1.12×10−8), with the HLA-DPB1*0401 allele being strongly over-represented in the patient group. This effect was even stronger when only ANCA-positive patients with WG were considered (p = 5.88×10−10). ANCA-negative patients did not show significant differences in comparison with controls (p = 0.84, table 1, fig 1). Allele frequencies were quite similar between the ANCA-positive group investigated earlier11 and the ANCA-positive patients additionally included in this study. As this observation was also present for virtually all SNP markers (supplementary fig 1), we considered these two subpopulations of ANCA-positive patients with WG as one cohort in subsequent analyses, thereby increasing statistical power.
Since the ANCA-negative group represents only a small fraction of the case cohort, the lack of HLA-DPB1*0401 association could apparently be explained by a loss of statistical power. Yet, power calculations make it quite likely that an association would have been detected: On the basis of the HLA-DPB1*0401 data from ANCA-positive patients with WG and controls (OR = 3.38, MAF = 0.442, table 1), the comparison of 27 ANCA-negative patients and 369 controls yielded a power of 0.9835 for a multiplicative risk model (supplementary table 4). Power for recessive (0.8204) and dominant (0.6324) risk models was considerably reduced. Yet, at the same time, when genotype distributions from the complete sample were considered for HLA-DPB1*0401 by logistic regression, a multiplicative model revealed the strongest p value (p = 2.40×10−17) compared with dominant (p = 5.02×10−10) and recessive (p = 1.62×10−15) models (supplementary table 5).
Call rates were >90% (⩾605 subjects) for all SNPs, at an average of 94.35%. Genotypes of all SNPs were in Hardy–Weinberg equilibrium. SNP allele frequencies in controls were well approximate to the HapMap CEU sample (table 2). Moreover, allele frequencies were quite similar between the patients from the group investigated earlier11 and the ANCA-positive patients additionally included in this study. All SNPs revealing clearly significant p values in one group also showed significant allele distribution differences in the other group and vice versa (supplementary table 1).
Considering the complete WG cohort (n = 282), 23 of the 35 SNPs showed a statistically significant difference in the allele distribution between patients and controls (table 2, fig 2). p Values were even smaller for the majority of SNPs when only alleles from ANCA-positive patients with WG were considered (fig 2). On the other hand, and comparable with HLA-DPB1, none of the SNPs reached significant p values (p<0.01) in the ANCA-negative subgroup (supplementary table 1). Calculations from 2- and 3-tagSNP haplotypes for the indirectly tested (ie, tagged) SNP did not show any p value that was smaller than p values obtained from direct analysis of the 25-tagSNP (data not shown). Comparing allele frequencies between patients and controls, SNP3 (rs3117228), located 1.5 kb downstream from HLA-DPB1, had the smallest p value in the complete sample (p = 3.6×10−21) of all SNPs (figs 2 and 3).
SNP3 was only in weak LD with HLA-DPB1*0401 (r2 = 0.22). Logistic regression subsequently showed an independent contribution of HLA-DPB1*0401 to the WG association of this locus. For this analysis, we considered the genetic model yielding the smallest p value in the single-marker analysis—that is, a recessive model for SNP3 and a multiplicative model for HLA-DPB1*0401. Depending on the order of exclusion from the model, p values obtained from an LRT in the complete sample were 2.6×10−24 for SNP3 and 3.3×10−5 for HLA-DPB1*0401 or, vice versa, 5.5×10−17 for HLA-DPB1*0401 and 1.2×10−12 for SNP3. p Values obtained from the conservative Wald test were considerably higher, equalling 1.6×10−11 for SNP3 and 4.2×10−5 for HLA-DPB1*0401.
SNP34, being in moderate to strong LD with SNP33 (r2 = 0.62) but not with SNP32 (r2 = 0.13), was the most informative marker of the RING1 region, with the smallest p value under a recessive model. SNP34 was only slightly correlated with HLA-DPB1 (r2 = 0.16). Logistic regression again showed an independent contribution of both variations: LRT p values equalled, depending on the order of exclusion from the model, 4.2×10−15/1.1×10−6 for SNP34 and 9.6×10−8/3.8×10−16 for HLA-DPB1*0401, while Wald test p values equalled 2.4×10−6 for SNP34 and 1.9×10−7 for HLA-DPB1*0401. Thus, association with the RING1 locus is not attributable to association of the HLA-DPB1 locus.
Haplotype block analysis
Figure 2 shows the haplotype block structure. While block 1 and block 7 did not show any significant haplotype distribution between patients with WG and controls (p = 0.69 and 0.08, respectively), the other haplotype blocks showed marginally (block 3: p = 0.03; block 4: p = 0.01) or highly significant (block 2: p = 2.4×10−7; block 5: p = 2.95×10−5; block 6: p = 0.002) WG associations. Block 5, covering (amongst others) candidate genes RXRB and RING1 also showed highly significant distribution differences of the two most frequent haplotypes, one of which is WG protective (CCCTACT; p = 1.84×10−10) and the other one (TCTCCCT; p = 1.07×10−6) vice versa predisposing to WG (fig 4). Block 6 showed a similarly significant haplotype distribution difference (data not shown).
Direct sequencing did not show any previously unknown coding variation. Neither a variation with potential effect on splicing mechanisms nor a nucleotide exchange within the sequences of the promoter area (500 bp 5′ of exon 1) was detected. Genotypes of the SNPs located within the analysed sequences were confirmed throughout.
Correlation of ANCA status and genotype data with clinical WG manifestation
Age of disease onset did not correlate with ANCA status or genetic markers. Yet, renal WG manifestation was significantly more common in ANCA-positive patients, a phenomenon that has been described before.26 Moreover, peripheral neuropathy and constitutional symptoms were more common in the ANCA-positive cohort (supplementary table 6). When adjusting for the genotypic data HLA-DPB1*0401, SNP3 and SNP34 were significantly associated with the presence of renal involvement in patients with WG (supplementary table 7a–c). When only ANCA-positive patients were considered, this genotypic effect generally persisted but was lost after conservative Bonferroni correction.
We have previously demonstrated the presence of a WG-associated locus on chromosome 6p21.3, including HLA-DPB1. Here, we refine its location in a considerably larger number of patients and controls using well selected tagSNPs to narrow down the causative molecular genetic variation. In this respect, this study comprises one of the first concrete tagSNP-based studies to overcome the problem of evaluating HLA associations found in many disease or traits, respectively, following an approach whose power has been highlighted previously.27 28
When comparing allele and genotype data from the original and the additional patient group separately, similar frequencies were obtained for the two subgroups. In contrast to ANCA-negative patients with WG (n = 27), the association is restricted to ANCA-positive patients only. Although the number of ANCA-negative patients with WG remained limited under the strict criteria applied, this observation supports similar findings in PTPN22.10 Therefore, we propose that the group of ANCA-negative patients with WG constitutes an (genetic) entity of its own and should be evaluated separately, especially in genetic association studies. This is supported by the finding that, both, ANCA status and genotypes of the most informative genetic markers disclosed by this study contribute to the clinical manifestation of WG, especially renal disease.
The HLA-DPB1/RXRB/RING1 region constitutes a quantitative trait locus for ANCA-positive WG. Although LD has been described for HLA-DPB1/RXRB13 we were able to demonstrate HLA-DPB1-independent contributions of RXRB/RING1 SNPs. There are three clusters of highly significant p values (fig 2), representing the genomic regions 3′ of HLA-DPB1 (SNP3–5), COL11A2 (SNP18, 19) and RING1/RXRB (SNP20, SNP33, 34). Haplotypes of the genomic regions containing these genes show highly significant distribution differences between cases and controls.
COL11A2, predominantly expressed in hyaline cartilage, in the vitreous body, in the nucleus pulposus of intervertebral discs and in the inner ear, encodes a component of the collagen XI trimer. Mutations in COL11A2 are associated with oto-spondylo-megaepiphyseal dysplasia, Stickler syndrome29 and Weissenbacher–Zweymuller syndrome. Moreover, some forms of non-syndromic hearing loss (DFNB53,30) and sensorineural deafness (DFNA13) are linked to COL11A2. Although sensorineural hearing loss can be a feature of WG,31 we do not regard COL11A2 as a primary candidate gene.
RING1 constitutes one of the E3 ubiquitin ligase proteins that mediate monoubiquitination of histone H2A, thereby playing a central role in histone code and gene regulation. It is associated with the human polycomb group protein complex and can act as a transcriptional repressor and modulator of RING2 activity.32 Recent studies have produced evidence that members of the RING-type ubiquitin ligase family, such as RING1, are required to repress follicular helper T cells and autoimmunity in a mouse model.33
By direct sequencing of DNA from preselected patients carrying the WG associated RING1 haplotype and controls homozygous for the conversely associated haplotype, we have ruled out a variation within the coding region with high probability. Yet, non-coding allelic variations of RING1 have not been analysed. Thus, variations in cis regulatory sequences of RING1 need to be evaluated further.
A potential role of RXRB in WG pathogenesis had been discussed earlier11 due to the involvement of the retinoid receptor X β protein in vitamin D receptor signalling.34 Vitamin D receptors regulate calcium homoeostasis as well as immune and endocrine functions, vitamin D metabolism and cellular differentiation. The coding and promoter regions have already been screened in a cohort of 48 ANCA-positive patients with WG12 without revealing nucleotide exchanges with potential impact on WG. Yet, analyses of regulative sequences of RXRB appear mandatory.
HLA-DPB1 is associated with chronic beryllium disease,35 another granulomatous disorder. Therefore, it has been suggested that the development of granulomatous inflammatory states may be generally influenced by an impact of certain HLA-DPB1 alleles. Yet, while chronic beryllium disease is strongly associated with the Glu69 allele of the Glu69/Lys69 polymorphism within HLA-DPB1,36 only a very weak association of WG (and at that with the contrary Lys69 allele) could be demonstrated.11 The specific region around HLA-DPB1 is characterised by an area (∼37.5 kb) of strong LD, and it is therefore very difficult to narrow down the origin more precisely.
The most significant allele distribution difference of all the SNPs of this study (SNP3) is located in the region 3′ of HLA-DPB1, which is not included in the HLA-DPB1 haplotype block. Its association is at least in part independent of the HLA-DPB1 alleles determined by exon 2 sequences. Therefore, one could speculate that a non-coding sequence variation in this area may have an impact on WG pathogenesis. This would subsequently include secondary effects on genetic factors in cis or trans, with the latter pointing to an even more complex genetic background in WG.
We thank I Alheite, D Falkenstein and G Schlueter for excellent technical assistance. Especially, we thank DA Akkad for methodical support and optimisation of TaqMan assays.
Competing interests: None.
Funding: This work was supported by FoRUM grant F483–2005 to SW.
Ethics approval: Approved by the local ethics committee.
▸ An additional figure and tables are published online only at http://ard.bmj.com/content/vol67/issue7