Rheumatoid arthritis (RA) is a chronic inflammatory disease that leads to erosion and deformation of the joints. The prevalence in the general population is 0.5% to 1%, and women are at two to three times greater risk for developing the disease. Twin studies show a concordance rate of 12% to 15% in monozygotic twins and 4% in dizygotic twins, and the genetic heritability is estimated at 60% 1. Despite much effort to identify arthritis causing genes, only few genetic loci have been confirmed to be associated with RA, among which are the human leucocyte antigen (HLA) locus and the protein tyrosine phosphatase non-receptor 22 gene (PTPN22) 234.

RA is a heterogeneous disease with symptoms and disease progression that vary between patients. Classification of typical RA requires fulfilment of four out of seven criteria established by the American College of Rheumatology 5. The heterogeneity of the symptoms in RA probably mirrors the underlying genetic contribution; hence, the various combinations of symptoms observed in patients are probably caused by different combinations of risk alleles.

In the search for markers that can predict development of the disease before its onset, several autoantibodies, including rheumatoid factors (RFs) and antibodies against cyclic citrullinated peptides (anti-CCP antibodies), have been detected 678. RFs, autoantibodies that recognize the Fc part of immunoglobulins 9, are present in 60% to 70% of RA patients and have been found to be associated with more severe clinical manifestations 1011. Anti-CCP antibodies are present in 50% to 60% of RA patients and have also been shown to predict more severe disease 10121314. Interestingly, recent data suggest that the HLA-DRB1 locus, which has been shown to be associated with RA, is only associated with the presence of anti-CCP antibodies, and this association is independent of both RA development and the presence of RFs 15. Together with recent data describing the interaction between environmental factors and genetic predisposition 1617, these findings support the hypothesis that several independent disease mechanisms may lead to development of RA. They also emphasize the need to use relevant subgroups of RA patients in genetic association studies.

Various approaches have been used to identify genes that contribute to common diseases such as RA 1819. Because of increasing knowledge of gene functions and immunological pathways, a candidate gene approach can be more efficient in terms of both cost and time than traditional linkage analysis or genome-wide approaches. The selection of genes in a candidate gene study can be based on previous knowledge of gene functions as well as on disease or immunological mechanisms. It can also be based on findings in animal models.

The Pia4 locus in rats has been shown to reduce the severity of pristane-induced arthritis (an arthritis model) 20. A single nucleotide polymorphism (SNP) in the Ncf1 gene was found to be responsible for the protective effect 21. Ncf1 encodes the protein p47^phox, which is part of the NADPH-oxidase (NOX) complex that produces reactive oxygen species (ROS) in response to infectious stimuli. Rats carrying the risk allele have reduced capacity to produce ROS, which is linked to increased risk for development of arthritis 2122. Apart from NCF1, the NOX complex is composed of five other proteins encoded by the genes NCF2, NCF4, CYBB, CYBA and RAC2 2324. In phagocytes massive ROS production, called the oxidative burst, takes place in the phagosome or endosome (intracellular) or in the plasma membrane (extracellular) after ingestion of invading pathogens or after stimulation of innate immune receptors 252627. However, the ability to generate ROS extends to cells other than classical phagocytes, such as dendritic cells, suggesting that the NOX complex has additional functions in the immune system 28. Interestingly, recent findings from our group show that the redox balance of T-cell membranes has an important effect on the activation and proliferation of T cells 29.

Results obtain in animal models make NCF1 and the other genes of the NOX complex candidate genes for human RA. However, transferring animal data to the human setting is not straightforward in this case. In contrast to rats, the genomic organization of NCF1 in humans is very complex, which makes SNP-based association analysis difficult 3031. However, because of the complex interplay between the proteins of the NOX complex, it is likely that genetic changes in any of the genes could have the same effect on ROS production, as is seen for Ncf1 in rats.

We used a candidate gene approach to study the association with RA of 51 SNPs in five genes included in the NOX complex. We found a SNP, located in intron 4 in NCF4, to be associated with RA in a Swedish case-control cohort. This supports the hypotheses that the NOX complex is involved in the development of RA and that ROS could be an important regulator of the immune system.

Here we report an association of the SNP rs729749, located in the gene NCF4, with RA in Swedish RF/anti-CCP antibody negative male patients. NCF4 is part of the NOX complex, and our results support recent findings that ROS play a role in the development of RA. However, because this is the first report of an association between a gene in the NOX complex and RA, genetic replication as well as functional studies will be required before it is possible to determine conclusively whether NCF4 is an RA susceptibility gene. Furthermore, the results from this study do not exclude the possibility that the other genes in the NOX complex could affect RA susceptibility. There could, for example, be epistatic effects between the different genes in the complex.

There is a lack of consensus in the field about statistical significance levels in association studies. In this study we initially used an α level of 0.05 to reduce the risk for overlooking associations that are specific for a subgroup of RA patients. There are strong medical and biological arguments in favour of subgrouping RA patients, and an analysis of nonstratified samples might conceal a true association. In the initial analysis of the genotype frequencies we found indications of association with three SNPs, and stratification of the samples indicates that the association for rs729749 is male specific and predominates in RF-negative disease. We also found a weaker association in anti-CCP antibody negative disease, but at this stage it is not possible to say whether this is due to the fact that the presence of RFs and anti-CCP antibodies is dependent 41 or whether the anti-CCP antibodies independently influence genetic risk.

The detected association of rs729749 is strongest in RF-negative males, and therefore it is possible that this particular SNP modulates the affected disease pathway so that it mainly influences the clinical disease in RF-negative RA, specifically in men. However, it is not impossible that this pathway play a role also in other subtypes of RA. The results from this study strengthen the view that different combinations of genes are involved in the disease progression in different RA subclasses defined, for instance, by sex or autoantibody status. The most striking aspect of the identified association of the rs729749 SNP is the clear sex specificity, indicating that the disease pathway affected by the rs729749 SNP is specific to men. The fact that RA is three times more common in women than in men suggests that men and women respond differently to factors that trigger the onset of RA and that some disease pathways could be sex specific. The RF-negative specificity of the association is not as striking as the sex specificity, but it might still point toward which RA subgroup is affected by the rs729749 SNP. Because both RFs and anti-CCP antibodies can be detected before the appearance of any disease symptoms 742, they might reflect or be part of initial disease mechanisms that are only present in anti-CCP antibody or RF positive patients. Furthermore, it has repeatedly been shown that patients who are negative for these autoantibodies have a less severe disease outcome than do anti-CCP antibody or RF positive RA patients 101112. This further supports the view that autoantibody negative RA is another variant of RA. Also, the few genes confirmed to be associated with RA have been shown mainly to affect subclasses of RA.

The well established association of the HLA-DRB1 alleles, which encode the so-called shared epitope (SE), was recently suggested to be restricted to anti-CCP antibody but not RF positive RA 151641. However, it has also been shown that SE alleles are associated with higher titres of autoantibodies against type II collagen 43. Interestingly, the HLA-DR3 locus, which does not encode the SE, appears to be associated only with anti-CCP negative RA 4144. The recently discovered PTPN22 risk allele has been shown to be associated with both RF and anti-CCP antibody positive RA 434546, but there are conflicting data regarding the autoantibody restriction of this association 47. Furthermore, the reported association of the PD-1.2A allele of the PDCD1 gene is restricted to RF negative as well as SE negative RA 48. Hence, even though little is known about the precise disease mechanisms in RA, it is obvious that the symptoms can be caused by several different distinct disease pathways.

Because of the lack of certain autoantibodies, it is tempting to speculate that RF or anti-CCP antibody negative RA patients have a lesser B-cell component, and reports indicating that B-cell depleting therapy is a less effective treatment for RF-negative RA supports such a view 4950. Several recent clinical studies, however, have yielded conflicting results on B-cell depletion efficiency 51, and more extensive studies are required before any conclusions can be drawn about the B-cell dependence in RF and anti-CCP antibody negative RA. In addition, it is fair to assume that the autoantibody negative forms of RA, in themselves, represent a heterogeneous collection of pathways related, for example, to osteoarthritis, arthropathies, or lupus.

The rodent arthritis models are heterogeneous as well but they may relate to more defined arthritis conditions than the human disease. Interestingly, the pristane-induced arthritis model used to identify the polymorphism in Ncf1 in rats is B-cell independent. Studies have shown that transfer of T-cells from genetically susceptible and immunized rats into genetically resistant and un-immunized rats is enough to induce arthritis in the resistant rat 52.

Both the HLA-DRB1 locus and PTPN22 are believed to affect T-cell activation. HLA-DRB1 is expressed by antigen-presenting cells and restricts the presentation of antigens to T cells. PTPN22 on the other hand has an intracellular effect, and the amino acid shift to the disease-associated tryptophan impairs the function of the Lyp protein, encoded by PTPN22, rendering it a less potent negative regulator of T-cell activation 53. Studies in animal models as well as recent findings suggest that the NOX complex could also be involved in T-cell activation 2954. ROS production has been found in antigen-presenting cells, and it has also been shown that T-cell receptor signalling is affected by oxidation of the T-cell membrane as well as by intracellular levels of ROS 5556. Interestingly, hydrogen peroxide generated from ^•O₂^- produced by the NOX complex can readily cross the cell membrane and inactivate protein tyrosine phosphatases, including that encoded by PTPN22, through oxidation of a specific cysteine residue 54. ROS production of antigen-presenting cells, such as dendritic cells, has been found to be crucial for the activation of T-cells 28 and could therefore determine the immune response to an antigen. Furthermore, recent work from our group highlights the importance of the redox balance on the surface of T cells for activation. T-cells originating from the Ncf1 mutated rat DA have an increased number of reduced thiol groups on the cell surface, which increases proliferation and activation 29. Gelderman and coworkers 29 showed that T-cells originating from the nonmutated E3 rats could be made arthritogenic by increasing the number of reduced thiol groups on the cell surface. Because NOX production could not be detected in T-cells, it was concluded that the redox balance of T-cell surfaces is determined by other cells, such as macrophages or dendritic cells.

In the rat a clear difference in the capacity to produce oxidative burst is the result of a single SNP in Ncf1, resulting in the shift from threonine to the disease-promoting methionine at position 153 in the p47^phox protein 21. The consequences of the amino acid shift have not yet been identified, although position 153 does not coincide with any critical binding sites. It is most likely that the shift affects the three-dimensional structure of the protein, impairing the binding capabilities to the other proteins in the complex. The activation of the NOX complex is initiated through phosphorylation of the three cytoplasmic proteins p47^phox (NCF1), p67^phox (NCF2) and p40^phox (NCF4) 232657. The phosphorylations lead to conformational changes of p47^phox, and subsequent translocation of the three proteins to the membrane, where it interacts with and activates the membrane bound complex flavocytochrome b₅₅₈, which is composed of gp91^phox (CYBB) and p22^phox (CYBA) 2458. RAC2 also translocates to the membrane after dissociation from Rho-guanine nucleotide dissociation inhibitor and interacts with flavocytochrome b₅₅₈ 2359. Studies show that p67^phox and Rac2 are critical for the function of the complex 2560, whereas p40^phox and p47^phox function as adaptor proteins and are responsible for the assembly of the complex 61. The precise role of p40^phox is still debated, and there is evidence of both positive and negative regulation of the NOX complex 616263. Recent studies have suggested that p40^phox, as well as p47^phox, functions by 'carrying' p67^phox to the membrane through the interaction with phospholipids 6264. Studies of mutations causing human chronic granulomatous disease show that mutations in NCF1, CYBB, CYBA, or NCF2 lead to a complete lack of function of the NOX complex 6566. However, SNPs in less critical regions, like the 153 SNP in Ncf1 found in rat, apparently reduce but do not completely abolish ROS production, which leads to an increased susceptibility to arthritis 21. These facts indicate that the SNPs of interest for examination in an RA association study of the NOX complex will probably be located at noncritical positions and will not completely abolish but only slightly modify binding to the other proteins.

The rs729749 SNP is located in the middle part of intron 4 in NCF4. By studying the genome maps using the UCSC genome browser, we could not find any evidence of it affecting a transcription factor binding site or a known regulatory element. However, there is a conserved regulatory region predicted approximately 300 base pairs downstream of rs729749. The haplotype analysis indicates that rs729749, or a SNP in very high LD, is responsible for the association detected for the haplotype. Also, because the predicted regulatory region is very close to rs729749, it could very well contain the causative SNP.

The genetic analysis of the association shows that the homozygous model gives the best significance, indicating that this could be the genetic effect. However, the low number of samples in the subgroup of RF-negative males together with the low frequency of the T allele make this assumption somewhat uncertain. Furthermore, it is not obvious what the functional consequences of a homozygous effect would be. We therefore chose to present all models tested in order to provide the full picture.

In the logistic regression analysis, weak associations were also detected for RF and anti-CCP antibody positive disease. One possible explanation for these findings is that some patients could have been 'misdiagnosed' regarding autoantibody status. The diagnosis is made by measuring the autoantibody titres, and certain threshold are used to determine positive versus negative status. However, sometimes autoantibody status changes as the disease progresses, and therefore there is a degree of uncertainty regarding autoantibody status.

A method or a strategy to apply multiple testing corrections accurately in a case-control candidate gene association study has not yet been established 6768. The conventional multiple correction methods, such as Bonferroni, are considered by some to be too stringent for large-scale studies 6970. The Bonferroni correction method is used to correct for the number of independent tests performed. However, in an association analysis neither the SNPs tested nor the different stratified analyses are truly independent. The allele frequencies of the SNPs are affected by the LD of the region, and the stratified analyses are based on the same SNPs and samples as the initial analysis and can therefore not be considered to be independent from each other. Nonetheless, in order to obtain an indication of the validity of the P values obtained in these analyses, we used the Bonferroni correction method to estimate an α level correcting for the number of tests (159) performed in the frequency analysis. The 159 tests reflect the sex-stratified analysis of all 51 SNPs (153) plus the stratified analysis of RF and anti-CCP antibody status performed for rs729749, rs789181 and rs1476002 in the male subset (6). Correcting for 159 tests gives a new α level of 0.0003, which means that only the association of rs729749 in RF-negative males passed the significance threshold using this stringent method.

The EIRA study included mainly individuals born in Sweden. Taking those born outside Sweden into consideration also, 97% were of Caucasian origin. In order to minimize potential bias resulting from population stratification, we performed logistic regression analyses in which ORs were adjusted for age and residential area.

We found evidence of an RA-associated SNP in the NCF4 gene of the NOX complex. The association is specific for male patients, with the strongest association found in RF-negative RA. This finding supports the notion of RA being a heterogeneous disease caused by a variety of disease pathways that are regulated by a variety of contributing risk genes. The detected association with a component of the NOX complex also strengthens previous evidence obtained in animal models that suggests that the NOX complex and ROS have a major impact on inflammation and autoimmunity. It is hoped that this finding will help to elucidate the complex genetics that underlie RA and autoimmunity.

CCP = cyclic citrullinated peptide; CEPH = Centre d'Etude du Polymorphisme Humain; CI = confidence interval; EIRA = Epidemiological Investigation of Rheumatoid Arthritis; HWE = Hardy-Weinberg equilibrium; NOX = NADPH oxidase; OR = odds ratio; RA = rheumatoid arthritis; RF = rheumatoid factor; ROS = reactive oxygen species; SE = shared epitope; SNP = single nucleotide polymorphism.

The authors declare that they have no competing interests.

LMO has been the responsible investigator for the project and has carried out SNP selection, genotype preparation, statistical analyses and drafted the manuscript. A-KL participated in the design of the study and the statistical analyses. HK and LA performed the logistic regression analysis. LP and LK were responsible for the distribution of the EIRA cohort. HB has contributed with intellectual property to the study. RH conceived of the study, participated in its design and helped to draft the manuscript.