Hypogalactosylation of immunoglobulin G in rheumatoid arthritis: relationship to HLA-DRB1 shared epitope, anticitrullinated protein antibodies, rheumatoid factor, and correlation with inflammatory activity

Patients with RA were attending the Department of Rheumatology and Clinical Immunology (Berlin). All patients fulfilled American College of Rheumatology criteria for classification of RA [37]. Serum samples from 178 patients with RA were selected according to ACPA and/or SE status with other clinical characteristics blinded. Additionally, 126 patients with ankylosing spondylitis (AS; n = 64) or nonradiographic axial spondyloarthritis (nr-axSpA; n = 62) were included from the GErman SPondyloarthritis Inception Cohort, which was described previously [38]. Samples from blood donors at the Institute of Transfusion Medicine (Berlin) with a similar age and sex distribution as RA were used as healthy control subjects (HC). None of them presented with symptomatic autoimmune disorders or clinically detectable inflammatory or infectious disease. Baseline characteristics of the study population are summarized in Table 1 and Additional file 1: Table S1. All individuals were included in the study after they provided informed consent. The study was approved by the Charité institutional ethics committee.

ACPA were quantified at the time of enrollment with a second-generation enzyme-linked immunosorbent assay (ELISA) using the CCPlus Immunoscan kit (Euro-Diagnostica, Malmö, Sweden) for detection of RA [39]. A cutoff value of 25 AU/ml was used. RF was quantified using an ELISA-based IgM-specific technique (cutoff value, 24 IU/ml) as previously described (DLD Diagnostika, Hamburg, Germany) [39]. CRP and erythrocyte sedimentation rate (ESR) were characterized using standard techniques [39]. SE was defined by HLA-DRB1 alleles with HLA-DRB1 chain residues 67Leu–69Glu–71Lys/Arg–74Ala–86Gly/Val and characterized using standard techniques (Dynal, Oslo, Norway; GenoVision, Vienna, Austria; PROTRANS, Ketsch, Germany) [39, 40].

Analysis of IgG-Fc N-glycans was performed as described previously [41] with minor modifications. Briefly, IgG was isolated from human serum using protein A Sepharose. IgG-Fc glycopeptides were separated from IgG antigen-binding fragment by pepsin digestion and isolated by ultrafiltration. After denaturation, IgG-Fc N-glycans were released by peptide:N-glycosidase F (Roche Applied Science, Indianapolis, IN, USA) digestion, separated with reversed-phase Supelco SP20SS resin microcolumns (Sigma-Aldrich, St. Louis, MO, USA) and subsequently desalted with graphite microcolumns (Agilent Technologies, Santa Clara, CA, USA). N-glycans were derivatized by methylation of carboxylic acid groups of terminal sialic acids to neutralize their negative charges, followed by purification to remove excess reagents. Subsequently, dried eluates were labeled with 8-aminopyrene-1,3,6-trisulfonic acid (APTS) (Sigma-Aldrich) at 55 °C for 2 h in darkness. Capillary electrophoresis of labeled N-glycans was performed on a Beckman P/ACE MDQ system equipped with laser-induced fluorescence detection (Beckman Coulter, Brea, CA, USA). Separation was achieved with reversed polarity on polyvinyl alcohol-coated capillaries (50 μm inner diameter, 50-cm effective separation length from injection to detector; Beckman Coulter) using the background electrolyte (Beckman Coulter). APTS-labeled maltose was used as an internal standard to normalize detected migration time. Migration time was expressed in glucose units using an APTS-labeled dextran hydrolysate ladder. Electropherogram quality was evaluated by intensities and signal-to-noise ratio of peaks detected in all samples. Data analysis was performed with 32 Karat 8.0 software (Beckman Coulter). Structural assignment of IgG-Fc glycans was performed as reported previously. In this study 22 peaks were detected, which represent 29 glycan structures. Relative intensities were calculated after normalization as the ratio of areas between individual peaks to all peaks. In peaks with relevant comigration, the N-glycan structure of the dominant peak defined the galactosylation type. Glycosylation traits were quantified by summing the corresponding peak areas, where the abbreviations depict the number of antennary N-acetylglucosamine (GlcNAc) (A), bisecting GlcNAc (B), core fucose (F), galactose (G1, G2), and sialic acids (S1, S2) (Fig. 1): agalactosylation (G0) = A2 + A2B + FA2 + FA2B; monogalactosylation (G1) = A2[3]G1 + A2B[3]G1 + FA2[6]G1 + FA2[3]G1 + FA2B[6]G1 + FA2B[3]G1 + A2[3]G1S1 + FA2[6]G1S1 + A2B[3]G1S1 + FA2[3]G1S1; and G0/G1 ratio = (G0)/(G1). Because G0 and G1 represent over 70% of total IgG-Fc glycosylation, the IgG G0/G1 ratio was used to characterize aberrant IgG galactosylation as described previously for analyses on whole serum N-glycans as well as purified IgG [27]. Additionally, data on the main findings are presented for G0%, which represents the percentage of G0 among total IgG glycans.

Beta 1–4 galactosyltransferase (B4GALT3) activity was estimated from transcriptome data for B-cell development (Gene Expression Omnibus accession number [GEO:GSE19599]) and differentiation to plasma cells (Gene Expression Omnibus accession number [GEO:GSE12366]) using data from the Gene Expression Omnibus (GEO) online repository [42, 43]. The respective data of this repository were derived from two independently pooled samples of each cell population that have subsequently been hybridized onto Applied Biosystems GeneChip Human Genome U133 Plus 2.0 microarrays (Thermo Fisher Scientific, Waltham, MA, USA). Array data have been uploaded into the BioRetis database (www.​bioretis.​de), and normalized signal intensities as well as group analyses have been tested.

Statistical analyses were performed using IBM SPSS Statistics 23.0 (IBM, Armonk, NY, USA) and Prism 6.0 software (GraphPad Software, La Jolla, CA, USA). The Kruskal-Wallis test followed by Dunn’s multiple comparisons test, the Wilcoxon rank-sum test, and the Mann-Whitney U test were used to analyze differences between groups. Associations between IgG G0/G1 ratio and disease-related parameters were evaluated by regression analysis. For secondary analyses (i.e., analyses of different IgG hypogalactosylation between RA subgroups defined by ACPA, RF, and SE as well as different relationships between IgG hypogalactosylation and activity markers in dependence of ACPA, RF, and SE), p values after Bonferroni correction for multiple testing assuming 12 secondary hypotheses are given.

In patients with RA, IgG G0/G1 (median, 1.25; IQR, 1.09–1.56) was significantly higher than in HC (0.95; 0.77–1.13) (p < 0.0001) and axSpA (0.89; 0.79–1.09) (p < 0.0001) (Fig. 2). Whereas axSpA exhibited inflammatory activity as measured by elevated CRP level, in HC acute or recent inflammatory processes were excluded. The patients with axSpA did not differ from HC in terms of IgG G0/G1 ratios, supporting the hypothesis that inflammation is not a sufficient precondition for IgG hypogalactosylation. The RA cohort exhibited age and sex distributions similar to those of HC. However, the axSpA cohort differed considerably from RA and HC with respect to age and sex. Thus, analyses stratified by age as well as logistic regression analyses adjusting for age and sex as possible confounders were performed. We analyzed the differences in IgG G0/G1 ratio between RA and axSpA in age categories (≤ 25 years, 25–45 years, 45–65 years, > 65 years). In all age categories, RA exhibited higher IgG G0/G1 ratios than axSpA. These differences in IgG G0/G1 ratios reached statistical significance in patients aged 25–45 years (median [IQR], RA, 1.17 [1.00–1.29]; axSpA, 0.83 [0.73–0.95]; p < 0.0001) and 45–65 years (RA, 1.25 [1.09–1.59]; axSpA, 1.03 [0.85–1.21]; p < 0.0001). In the small group of patients aged ≤ 25 years (five patients with RA, two patients with axSpA), a nonsignificant trend was found (RA, 1.27 [1.23–1.40]; axSpA, individual IgG G0/G1 ratios 0.75 and 0.82; p = 0.053). In patients aged > 65 years (46 patients with RA, 2 patients with axSpA), the nonparametric comparison was nonsignificant (RA, 1.40 [1.16–1.79]; axSpA, individual IgG G0/G1 ratios 1.19 and 1.37; p = 0.66). When we compared RA, axSpA, and HC in a regression model, IgG hypogalactosylation was significantly more pronounced in RA than in axSpA and HC even after correction for age and sex as possible confounders (p < 0.001 for both comparisons). Plots for IgG G01/G1 ratio over age in all study cohorts are shown in Fig. 3.

When patients with RA were differentiated by previously untreated and treated patients, no difference with respect to IgG hypogalactosylation could be found between groups (p = 0.74). When we looked at specific drugs, those patients treated with sulfasalazine (n = 47) exhibited significantly less IgG hypogalactosylation than patients treated without sulfasalazine (n = 131) (IgG G0/G1, median [IQR], treated with sulfasalazine, 1.16 [1.07–1.29]; treated without sulfasalazine, 1.31 [1.1–1.66], p = 0.001; G0%, treated with sulfasalazine, 40.1 [37.3–42.5]; treated without sulfasalazine, 42.9 [38.6–49.5], p = 0.003). This finding remained statistically significant after correction for age, sex, and CRP level (p = 0.009 for IgG G0/G1 ratio as well as G0%). In contrast, methotrexate, leflunomide, hydroxychloroquine, and tumor necrosis factor inhibitors did not exhibit altered IgG galactosylation.

We compared RA and axSpA with respect to IgG G0/G1 ratio after stratifying for different CRP levels (< 5 mg/L, 5–10 mg/L, 10–15 mg/L, and > 15 mg/L) (Fig. 4). This analysis showed that across all CRP categories in patients with RA, significantly higher IgG G0/G1 ratios were present than in patients with axSpA (CRP < 5 mg/L, RA, 1.18 [1.04–1.29], vs axSpA, 0.86 [0.78–0.97], p < 0.0001; CRP 5–10 mg/L, RA, 1.17 [1.07–1.37], vs axSpA, 0.96 [0.82–1.09], p = 0.0002; CRP 10–15 mg/L, RA, 1.33 [1.08–1.79], vs axSpA, 1.05 [0.70–1.19], p = 0.0056; CRP > 15 mg/L, RA, 1.61 [1.27–2.20], vs axSpA, 1.10 [0.89–1.24], p = 0.0003). Furthermore, a significant increase in IgG G0/G1 ratio with increasing CRP levels could be observed in RA but not in axSpA (Fig. 4). The difference in IgG hypogalactosylation as measured by IgG G0/G1 ratio as well as G0% between RA and axSpA was significant after adjustment for possible confounders (CRP, age, and sex) (p < 0.001) (Fig. 5).

The IgG G0/G1 ratio was tested in RA in dependency of the presence or absence of ACPA, RF, and SE. As shown in Fig. 6a, in ACPA-positive RA, the IgG G0/G1 ratio was significantly higher (1.32 [1.10–1.74]) than in ACPA-negative RA (1.22 [1.08–1.42]) (p = 0.034). This relatively small difference was significant in multivariate analyses adjusting for possible confounders (CRP, age, and sex) for IgG G0/G1 ratio (p < 0.02) and for G0% (p < 0.05; median [IQR], ACPA-positive, 42.9 [38.1–50.6]; ACPA-negative, 40.5 [38.3–44.9]). There was only a nonsignificant trend toward higher IgG G0/G1 ratio in HLA-DRB1 SE carriers (1.27 [1.12–1.68]) compared with noncarriers (1.22 [1.07–1.44]) (p = 0.09). IgG galactosylation in RF-positive (1.27 [1.08–1.57]) and RF-negative RA (1.23 [1.11–1.58]) was essentially the same (p = 0.81).

Subsequently, we studied IgG G0/G1 ratio in RA subsets defined by combinations of SE and ACPA status (Fig. 6b). Interestingly, in SE-positive/ACPA-negative RA, the IgG G0/G1 ratio (1.25 [1.17–1.66]) was largely comparable to both ACPA-positive subsets (ACPA-positive/SE-negative, 1.32 [1.17–1.64]; ACPA-positive/SE-positive, 1.30 [1.07–1.77]). In SE-negative/ACPA-negative RA, the IgG G0/G1 ratio (1.13 [1.05–1.35]) was significantly lower than in all other RA subsets. RA subsets defined by SE and RF status exhibited quite comparable IgG G0/G1 ratios (Fig. 6c). If considering a multiple testing strategy, Bonferroni correction would indicate only a trend.

The axSpA cohort was also evaluated considering two subgroups, AS and nr-axSpA (Fig. 7). In AS, we observed significantly higher IgG G0/G1 ratios (0.93 [0.81–1.16]) than in nr-axSpA (0.85 [0.75–0.97]) (p = 0.02). However, when adjusted for potentially confounding factors (sex and age), the difference in IgG G0/G1 ratio between AS and nr-axSpA was not statistically significant (p = 0.18). In addition, patients with nr-axSpA showed a significantly lower IgG G0/G1 ratio than HC (p < 0.02). Again, this difference did not survive adjustment for sex and age. There was a statistically nonsignificant trend toward lower IgG G0/G1 ratios in HLA-B27 carriers (0.87 [0.77–1.13]) than in noncarriers (0.96 [0.79–1.18]) (p = 0.07) (Fig. 7).

Representative transcriptomic data for comparison of the enzymatic activity involved in sugar modification of immunoglobulins were selected from the GEO database. Table 2 shows not only that transcripts for B4GALT3, the enzyme responsible for the beta 1–4 galactosylation of IgG, is expressed about threefold higher in plasma cells than in naive B cells but also that other transcripts of glycosyltransferases related to posttranslational modification of immunoglobulins are significantly increased in plasma cells compared with B cells.

When we compared the correlation between IgG G0/G1 ratio and RA disease activity (DAS28, CRP, and ESR) stratified by SE, ACPA, or RF, we identified a significant heterogeneity with respect to the presence or absence of SE and RF (Fig. 8). Stratified by SE, the correlations between IgG G0/G1 ratio and DAS28 (p = 0.04, p_Bonferroni = 0.48), CRP (p = 0.0003, p_Bonferroni = 0.0036), or ESR (p = 0.03, p_Bonferroni = 0.36) were significantly different in SE carriers and noncarriers. In general, HLA-DRB1 SE carriers exhibited a stronger correlation of IgG G0/G1 ratio with disease activity than noncarriers (Fig. 8a–c). Surprisingly, when we assessed these relationships between ACPA-positive and ACPA-negative RA, only a nonsignificant trend for heterogeneity could be detected (Fig. 8d–f). When we considered RF status, the correlation between IgG G0/G1 ratio and CRP was significantly stronger in RF-negative than in RF-positive RA (p < 0.0001, p_Bonferroni < 0.0012) (Fig. 8g–i). When we classified patients as seronegative (RF-negative and ACPA-negative) or seropositive (RF-positive and/or ACPA-positive), the relationship between IgG G0/G1 ratio and CRP was significantly stronger in seronegative than in seropositive RA (p < 0.0001, p_Bonferroni < 0.0012) (Fig. 9).