Venous blood (30 ml) was collected in commercial sample test tubes containing a gel separator. The blood samples were kept at room temperature to clot and then spun at 3,000 rpm for 10 minutes), followed by collection of the serum, aliquoting and storage at -20°C until needed for analysis.

RF (composite immunoglobulin M (IgM), IgG and IgA), CRP and serum amyloid A (SAA) were assayed by nephelometry (BN ProSpec System; Siemens Healthcare Diagnostics, Deerfield, IL, USA). As per the manufacturer's supplier controls, RF, CRP and SAA results were considered positive when their individual concentrations exceeded 11 IU/ml, 5 μg/ml and 6.8 μg/ml, respectively. aCCP were measured using a second-generation immunofluorometric procedure using the ImmunoCAP Phadia 250 assay system with reagents and controls provided by the manufacturer (Phadia AB, Uppsala, Sweden). A concentration greater than 10 U/ml was deemed positive.

Serum cytokines, chemokines and growth factors were measured using the Bio-Plex multiplex suspension array system with xMAP technology (Bio-Rad Laboratories, Hercules, CA, USA), which enables simultaneous detection and quantitation of multiple different analytes in a single sample. The system uses an array of microspheres in liquid suspension that are conjugated with a mAb specific for a target protein. A wide range of standards (0.38 to 91,756.00 pg/ml) was used to enable quantitation of the individual cytokines by using a Bio-Plex array reader (Bio-Rad Laboratories) with a dual laser detector and real-time digital signal processing. The following analytes were measured: IL-1β, IL-1Ra, IL-2, IL-4, IL-6, IL-7, IL-8, IL-10, IL-12, IL-17, IFN-γ, TNF, granulocyte colony-stimulating factor (G-CSF), granulocyte macrophage colony-stimulating factor (GM-CSF), chemokine (C-C motif) ligand 2 (CCL2), CCL4 and vascular endothelial growth factor (VEGF). The upper limits of normal for these analytes were calculated as the mean +2 SD for 10 healthy control subjects (five female and five male, average age 44.4 ± 15.0 years, age range 26 to 65 years).

Genomic DNA was obtained by using the Maxwell 16 System with the Maxwell 16 Blood DNA Purification Kit (Promega, Madison, WI, USA) to extract DNA from whole blood, which was collected in ethylenediaminetetraacetic acid (EDTA) sample tubes. The patients' HLA-DRB1 alleles were determined by using a DNA-based high-resolution typing method, the LABType HD Class II DRB1 Typing Test (One Lambda Inc, Canoga Park, CA, USA), with reverse sequence-specific oligonucleotide probes and Luminex technology. The target DNA (HLA-DRB1 gene) was amplified by PCR using group-specific primers biotinylated for detection with streptavidin R-phycoerythrin conjugate. The PCR product was then denatured and hybridised to complementary DNA probes conjugated to fluorophores. Bound DNA was detected on the Luminex system, and software (HLA Fusion&#8482 Software Version 2.0, One Lambda, Inc.,21001 Kittridge Street, Canoga Park, CA, USA) was used to map the reaction patterns to those associated with published HLA gene sequences and assign the represented HLA-DRB1 alleles. HLA alleles were assigned according to the RAA amino acid sequence at positions 72-74 as SE (S) and non-SE (X), and the S group was further subdivided according to amino acid groups at positions 70 and 71 as described by du Montcel et al. 7 (see Table 2). The International Immunogenetics Information System database 31 32 and the Allele*Frequencies in Worldwide Populations Net Database 33 were used to establish regional and ethnicity-matched HLA-DRB1 frequencies in sub-Saharan Africa. The non-SE patients with the no SE allele (XX) genotype (n = 12) were used as controls in the calculation of contingency tables.

Descriptive and inferential statistics techniques were used in the analyses. Tests for association of contingency tables were performed using either the two-tailed Fisher's exact test or the χ² test with the Yates correction. One-way analysis of variance (ANOVA) was performed using the Kruskal-Wallis test for nonparametric data for more than two groups, or the Mann-Whitney U test when two groups were compared. Logistic regression was used for binary outcome variables, and the OR and corresponding 95% confidence interval were reported. P < 0.05 was considered statistically significant. The analyses were done using Stata software (Stata Corp, College Station, TX, USA).

All of the "classical" RA-associated HLA-DRB1*01, HLA-DRB1*04 and HLA-DRB1*10 alleles showed statistically significant ORs, with HLA-DRB1*04 the most prominent (Table 3), underscoring the significance of the HLA-DRB1 SE previously reported in the African population 34 35 36 37 38 . The HLA-DRB1 alleles were categorized according to the new classification as being homozygous for the SE allele (SS), heterozygous for the SE allele (SX) or XX. The majority (92%) of the patients had at least one allele associated with the amino acid motif. More than half (57%) of the patients typed were homozygous (SS), 35% were heterozygous (SX) and only 8% had no associated HLA-DRB1 allele (XX). The SS, SX and XX allele distribution among the African subgroup of the cohort (n = 124) was 58% (n = 72), 35% (n = 43) and 7% (n = 9), respectively. The other South African population groups (Caucasian, Asian and mixed ancestry) were not meaningfully represented, however.

Classification of RA-associated HLA-DRB1 alleles according to the du Montcel et al. classification system 7 revealed that S2 and S3P had significantly higher allele frequencies than those found in the general sub-Saharan population 32 33 . S3D and the non-epitope-associated alleles (X) were not significantly different in the RA groups compared to the general sub-Saharan population. As shown in Table 4, S1 was significantly undertransmitted in our population.

Analysis (by one-way ANOVA and Mann-Whitney U test) of the median values of autoantibodies revealed that they were generally higher in the SS group than in the SX group. Significant associations were noted for aCCP (SS vs SX vs XX; P = 0.0298) and for RF (SS vs XX, P = 0.0129; SX vs XX, P = 0.0273).

As X alleles have been shown to be undertransmitted in the RA cohort (that is, noncarriers of risk alleles) (see Table 4), they were used as the control group within the RA cohort to determine the possible associations of the HLA-DRB1 genotype with RA-associated autoantibodies by comparing the total positive and negative results for the RA-associated antibodies (RF and aCCP) in the risk groups with those for the control group (XX). As shown in Table 5, both RF and aCCP seropositivity were significantly higher in both the SS and SX groups of the total cohort in comparison with the XX control group. The data are as follows: RF (SS group vs XX control group: OR = 7.0, P = 0.0059, and SX group vs XX control group: OR = 6.3, P = 0.0126) and aCCP (SS group vs XX control group: OR = 10.2, P = 0.0010, and SX group vs XX control group: OR = 9.2, P = 0.0028).

The SS and SX groups of the African patients showed a statistically significant predisposition to aCCP seropositivity compared with the XX group (SS group vs XX control group: OR = 7.01, P = 0.0121; SX group vs XX control group: OR = 6.4, P = 0.0223), and RF was also more prevalent in the SS and SX groups compared to the XX control group, although the difference was not significant (Table 5). Stratifying the patients according to du Montcel et al.'s allele classification 7 as SS, SX or XX therefore seems to strengthen the genetic association with the presence of aCCP, an observation which is in agreement with the findings of Huizinga et al. 14 and Klareskog et al. 39 .

Further analysis was undertaken to probe the associations with the amino acids at positions 70 and 71 (S1, S2, S3D and S3P) with aCCP seropositivity by comparing noncarriers with the carrier allele groups. As shown in Table 6, this analysis revealed significantly higher ORs in the case of the S2 and S3P allele groups and weak, albeit significant, associations with the S1, S3D and X allele groups. RF seropositivity was weakly associated with all allele groups.

The association between SE genotype and aCCP or RF seropositivity was further probed by categorising patients into high-risk or low-risk allele groups according to the du Montcel classification system 7 with the XX genotype used as a comparator. The low-risk genotype groups were (1) S1, consisting of genotypes S1/S1 + S1/X; (2) S3D, consisting of S3D/S3D + S3D/X; and (3) S1/S3D, consisting of S1/S1 + S1/X + S3D/S3D + S3D/X + S1/S3D. The high-risk genotype groups were (1) S2, consisting of genotypes S2/S2 + S2/X; (2) S3P, consisting of S3P/S3P + S3P/X; and (3) S2/S3P, consisting of S2/S2 + S2/X + S3P/S3P + S3P/X + S2/S3P. As shown in Table 7, these comparisons revealed a highly significant risk associated with the S2/S3P, S3P and S1/S3D groups and, to a lesser extent, S1 and S3D as described in previous studies 21 40 . No predisposing effect associated with RF seropositivity was identified, which confirms that the SE alleles are primarily associated with aCCP.

Analysis (one-way ANOVA and Mann-Whitney U test) of the median values of the other circulating biomarkers and clinical markers of disease activity revealed significant associations with CRP (SS vs SX vs XX: P = 0.0041; SS vs SX: P = 0.0018), SAA (SS vs SX vs XX: P = 0.0018; SS vs SX: P = 0.0005), TNF (SS vs SX vs XX: P = 0.0295), IL-1Ra (SS vs XX: P = 0.0311) and DAS28 (SS vs SX vs XX: P = 0.0324; SS vs XX: P = 0.0103; SX vs XX: P = 0.0210). These trends were also noted with regard to most of the other cytokines, although they were not statistically significant (IL-1β, IL-2, IL-4, IL-6, IL-8, IL-12, CCL4, G-CSF, GM-GSF and IFN-γ).

Statistical comparison of the S1, S2, S3D and S3P groups did not reveal any significant differences, but in general the highest means for acute phase reactants, proinflammatory or anti-inflammatory cytokines, chemokines and VEGF were located predominantly in the SS-associated groups (data not shown).