A massive infiltration of the tissues by inflammatory cells is characteristic of a number of tissue-specific autoimmune diseases including rheumatoid arthritis. The modulation of this response occurs through a combination of factors, including the number and type of receptor expressed by different leukocyte populations, and localised and systemic concentrations of regulatory cytokines. Genetic association of rheumatoid arthritis (RA) with alleles of the class II antigen-presenting molecule HLA-DRB1 on chromosome 6p has been established for decades.1 Recently, the 620W allele of PTPN22 (which encodes the lymphoid tyrosine phosphatase (LYP)) has been confirmed as a determinant of RA by extensive replication of association in Caucasian patient cohorts.2 Other genes are implicated in RA susceptibility with CTLA4 and PADI4 the closest to being “confirmed”, although their effect (odds ratio (OR) <1.3) is considerably less than that of PTPN22 (OR 1.5–2.0).3 4

Many lines of evidence suggest chemokine-receptor mediated inflammatory responses have a role in the pathogenesis of autoimmunity. In RA and type 1 diabetes (T1D), for example, CCR5-expressing leukocytes are associated with disease progression.5^–7 Excessive or inappropriate production of chemokine (CC)-receptor ligands (β chemokines) such as RANTES (CCL5), ΜIP1–α (CCL3) and ΜIP1–β (CCL4) are a characteristic feature of RA and T1D.8^–12 In animal models, autoimmune diabetes5 and experimentally-induced arthritis13 14 can be partially blocked using selective CCR5 antagonists or anti-CCR5 antibodies. Anti-RANTES and anti-CCL3 antibodies decrease autoimmune symptoms,5 15 and expression of the murine homologue of chemokine ligand 3-like 1 (CCL3L1), MIP-1α, is correlated with diabetes progression in the non-obese diabetic (NOD) mouse model (administration of anti MIP-1 antibodies delaying disease onset).10 In humans, antirheumatic drugs such as dexamethasone and KE-298 inhibit production of RANTES,16 and a naturally occurring variant of CCR5, CCR5Δ32, (which has a deletion within exon 4 that results in premature termination of translation) is associated with protection against RA.17

Given such evidence, it is reasonable to hypothesise that genetic variations which alter the expression level of chemokine genes may influence susceptibility to RA and other autoimmune disease. One such candidate, the β-chemokine CCL3L1, is encoded by a variable copy-number gene that arose from a segmental duplication of CCL3 on chromosome 17q12. A potent ligand for chemokine receptor 1 (CCR1) and CCR5,18 gene dosage correlates to the ratio of CCL3L1: CCL3 transcription and to chemokine production.19 This has previously been demonstrated to have pathogenic significance; because of the binding capacity of CCL3L1 for CCR5, CCL3L1 can inhibit the entry of human immunodeficiency virus (HIV)–1 into T-lymphocytes.20 21 Low CCL3L1 copy number (CN) is a significant risk factor for HIV infection and disease progression, while high CN is protective,22 and CCL3L1 CN has also been implicated in Kawasaki disease (KD), a childhood vasculitis.23 Genetic interaction between CCL3L1 and CCR5 has been reported in the aetiology of HIV20^–22 and KD.23 Given genetic (and functional) evidence for the contributions of β-chemokines to autoimmunity and the knowledge that the CCL3L1 receptor (CCR5) is associated with RA,17 we tested for association between CCL3L1 CN and RA in two independent Caucasian cohorts, and with a second autoimmune disease (T1D) in a third independent cohort.

METHODS

Study populations

All study subjects were Caucasian. The New Zealand (NZ) RA cohort consisted of 834 RA patients recruited from outpatient clinics in Auckland, Bay of Plenty, Wellington, Canterbury, Otago and Southland. A total of 28% were male and 72% female. Of the RA patients for whom clinical data were available, 82% (571/697) were rheumatoid factor (RF) positive and 65% (303/465) were anti-cyclic citrullinated peptide (aCCP) antibody positive. The control group (n = 933) consisted of healthy subjects recruited from Otago and Auckland. The UK RA cohort consisted of 302 RA patients recruited at Lewisham, and Guy’s and St. Thomas’s Hospitals. A total of 77% of cases were female, 79% were RF positive and 87% had erosive disease.24 All RA patients satisfied the 1987 American College of Rheumatology (ACR) criteria for RA. A total of 255 UK control samples were purchased from the European Collection of Cell Cultures (http://www.ecacc.org). The T1D cohort comprised 252 cases recruited from the Auckland, Canterbury and Otago regions of NZ and had an average age of onset of 12.5 years. All cases were diagnosed by a physician and commenced insulin therapy at diagnosis. The controls for the T1D cohort (separate from the 933 controls for the NZ RA cohort) consisted of 282 healthy recruits from Dunedin and Auckland, NZ. Ethical approval for the study in NZ was given by the MultiRegion Ethics Committee and in the UK by the Lewisham Hospital and Guy’s and St. Thomas’ Hospitals local research ethics committees. All subjects gave written informed consent.

Determination of CCL3L1 copy number

CCL3L1 copy number was measured using reverse transcriptase (RT)-PCR as previously described.22 All samples within each run were assayed in duplicate and averaged to determine copy number. Conventional PCR and gel visualisation was used to verify the absence of the gene when no copies of CCL3L1 were detected by Taqman, using the following CCL3L1 primer pair; sense 5′-GTTCTCTTAGCTCTCTTCATGGAATTT-3′, antisense 5′-TTACTTCCCAGTGGGGTCTG-3′. CCR5 was amplified as an internal control using the primer pair sense 5′-TGCTTGGCCAAAAAGAGAGT-3′, antisense 5′-CCCGATGTATAATAATTGATGTCATCG-3′. All samples for which no copies of CCL3L1 were detected by Taqman (Applied Biosystems, Foster City, California, USA) or conventional PCR were retested at least once.

CCL3L1 copy number quality control

Inter-sample variation (Vi) between duplicates was determined using cchart from the STATA 7.0 suite (Stata corp., College Station, Texas, USA) as previously described22 and all samples for which Vi exceeded 3 standard deviations were repeated. To test for run-to-run repeatability, 45% of cases and 52% of controls were re-assayed. All samples for which copy number varied by more than 1 were removed from the analysis (1% of cases, 0.3% of controls).

RESULTS

In all control cohorts, CCL3L1 CN varied from 0–7 per diploid genome with the most common CN being 2. A total of 51% of the NZ controls and 53% of the UK controls carried two copies, consistent with reports for other Caucasian populations; 49% of 100 European Americans and 51% of 675 white US subjects.19 22 To test the hypothesis that CCL3L1 CN influences risk of autoimmunity, subjects were initially divided into three groups; those with two copies, and those with lower (<2) or higher (>2) CN (table 1). This subdivision was performed to genetically assess these biological consequences; decreased, median and increased CCL3L1 levels. Although there was evidence for association between RA and low (<2) CN in the NZ RA cohort (OR 1.40, 95% CI 1.10–1.79, p = 0.007), this was not replicated in either the UK RA or the NZ T1D cohort (OR 0.89, 95% CI 0.56–1.41, p = 0.61 and OR 0.90, 95% CI 0.58–1.40, p = 0.64, respectively). There was increased risk of developing RA conferred by higher (>2) CCL3L1 CN in the NZ cohort (OR 1.34, 95% CI 1.08–1.66, p = 0.009), although the increased risk was not observed in the UK RA cohort (OR 1.09, 95% CI 0.75–1.60, p = 0.64). However, evidence for association of CN>2 with disease was observed in the T1D cohort (OR 1.46, 95% CI 0.98–2.20, p = 0.064). Thus there was evidence for association of CCL3L1 CN>2 with increased risk of autoimmunity in two of the three cohorts studied. In the combined RA and T1D cohort CN>2 conferred a significantly increased risk of disease of OR 1.30, 95% CI 1.09–1.54, p = 0.003.

Given the negative association between the CCR5Δ32 polymorphism and RA,17 we tested for genetic interaction between CCL3L1 and CCR5. In the RA cohort, only CCR5Δ32 homozygosity is protective (table 2; OR 0.37, 95% CI 0.18–0.79, p = 0.001). Based on this, CCR5 genotypes were divided into CCR5Δ32/Δ32 (protective; CCR5p) and CCR5Δ32/CCR5WT or CCR5WT/CCR5WT (non-protective; CCR5n), where WT = wild type. CCL3L1 genotype was divided into two groups; CN⩽2 (low risk) or CN>2 (higher risk). This was performed on the basis of the genetic association data (table 1); in the combined RA analysis strongest association was observed to the CN >2 group (OR 1.25, 95% CI 1.05–1.50, p = 0.015). Thus, our working hypothesis was that elevated CCL3L1 CN conferred increased risk relative to normal or low CCL3L1 CN. Having two CCL3L1 groups also reduced the number of categories in the interaction analysis. There were four genetic risk groups: individuals with non-protective CCR5 genotype and two or fewer copies of CCL3L1 (CCR5n/CCL3L1⩽2; reference group), those with non-protective CCR5 genotype and risk-associated CCL3L1 CN (CCR5n/CCL3L1>2), CCR5Δ32 homozygotes with fewer than three copies of CCL3L1 (CCR5p/CCL3L1⩽2), and CCR5Δ32 homozygotes with risk-associated CCL3L1 CN (CCR5p/CCL3L1>2). Our data support the presence of interaction between CCR5 and CCL3L1 in the RA cohort (table 3). The risk conferred by CCL3L1>2 is ablated by the protective CCR5Δ32/Δ32 genotype (OR 0.10, 95% CI 0.01–0.76, p = 0.026 for CCR5p/CCL3L1>2 compared to OR 1.19, 95% CI 1.00–1.43, p = 0.049 for CCR5n/CCL3L1>2). Similar results were observed when all RA and T1D cases and controls were combined (model 3; OR 0.20, 95% CI 0.04–0.91, p = 0.037 for CCR5p/CCL3L1>2 compared to OR 1.24, 95% CI 1.06–1.46, p = 0.008 for CCR5n/CCL3L1>2).

DISCUSSION

Here, we have provided evidence that CCL3L1 copy number influences disease susceptibility to RA in a large NZ cohort, with possession of higher than the population-specific median CN = 2 conferring disease risk (OR 1.34, 95% CI 1.08–1.66, p = 0.009). This association, however, was not replicated in a second smaller RA cohort from the UK (OR 1.09, 95% CI 0.75–1.60, p = 0.643), although evidence for association CN>2 was observed to T1D (OR 1.46, 95% CI 0.98–2.20, p = 0.064). In the combined RA/T1D cohort, there was increased risk associated with CCL3L1 CN>2 (OR 1.30, 95% CI 1.09–1.54, p = 0.003). This is consistent with a biological model whereby high CCL3L1 CN would enhance the inflammatory response of leukocytes bearing cognate receptors, including CCR5+ T-cells. For a person carrying a higher than average CCL3L1 CN, the recruitment of T-cells and macrophages to a site of early inflammation (such as a damaged joint) would be elevated compared to an individual with two copies, increasing the initial response and the strength of the pro-inflammatory feedback loop that develops. This will, at least in part, be mediated by CCR5, which explains why the risk conferred by high CCL3L1 CN requires functional CCR5 expression (table 3). This also suggests that the CCR5Δ32 allele has a dominant effect over increased CCL3L1 CN.

There are several possible reasons why association of CCL3L1 CN>2 was not replicated in the UK RA cohort. First, the evidence for association in the NZ RA cohort may represent a false positive finding. As discussed below, however, the observation of a biologically plausible genetic interaction between CCL3L1 and CCR5 supports the hypothesis that CCL3L1 CN>2 is genuinely associated with RA. Nevertheless it is important that CCL3L1 CN and CCR5Δ32 are tested for association with RA in other large cohorts and that genetic interaction is assessed. Second, the inconsistency between the NZ and UK RA cohorts may represent genetic and/or environmental differences between the two populations. A third reason is the reduced power of the UK cohort to detect association. The UK RA and NZ T1D case-control cohorts were of similar size. Assuming that CCL3L1 CN>2 exerts a genetic effect of OR 1.30 for autoimmunity, a CN>2 frequency of 0.28 and α = 0.05, cohorts of this size have 52% power to detect association (power calculation performed as described previously).28 Thus, if CCL3L1 CN>2 is genuinely associated with autoimmunity, it is not surprising that evidence for association of this genotype with disease was observed in only one of the two smaller cohorts (the NZ T1D cohort).

One of the difficulties in identifying genes involved in complex traits such as RA is involvement of multiple loci, most of which confer only moderate disease risk, and which may be influenced by the genotype at another locus.29 There is a strong argument for testing for genetic interaction even in the absence of single-locus main effects,31^–33 particularly when a biological interaction is established. In the presence of epistasis one polymorphism must be considered in the context of a (possibly) interacting second unlinked polymorphism; a significant association may only be apparent when individual allele frequencies are conducive to detecting a main effect; undetected epistasis may be a significant reason why single-locus tests of association fail to replicate across separate cohorts. As a corollary, demonstrating genetic epistasis between genes for which a plausible biological model of interaction may be hypothesised, can provide independent evidence that the main effects seen for the individual loci are biologically relevant and real.31 Although genetic epistasis does not necessarily imply biological interaction,27 genuine epistatic interactions should be more readily replicable by others with large cohorts than when analysing the interacting loci in isolation.31 32 Here, we provide evidence by genetic interaction analysis (table 3) that the pro-inflammatory effects of increased CCL3L1 CN operate through the CCR5 signalling pathway. This supports our data reporting association of increased CCL3L1 CN being associated with RA and T1D representing genuine, rather than false positive, association (table 1).

In conclusion, two pieces of evidence presented here—association at the single-locus level between CCL3L1 CN and RA (table 1), and a further example of statistical interaction between CCL3L1 and CCR5 (table 3)—support the hypothesis that CN variation in CCL3L1 influences susceptibility to RA and that CCL3L1-initiated signalling through CCR5 is a checkpoint in RA.