The current research is embedded in the PARA study, a nationwide prospective cohort on pregnancy and RA (21, 31). The cohort consisted of serum samples of 253 pregnancies from 219 RA patients, collected between 2002 and 2009 (31). In addition, 32 pregnancies of healthy Caucasian volunteers without adverse obstetric histories were included and followed from the first trimester of pregnancy. Of each patient, at least one sample was obtained during pregnancy and one postpartum, with a minimum of three samples per patient. Only completed pregnancies were included and all patients fulfilled the 1987 ACR criteria for RA. Disease activity was assessed using the disease activity score (DAS) in 28 joints, incorporating the swollen joint count, the tender joint count and the C-reactive protein (CRP) level [DAS28(3)-CRP]. The study was in compliance with the Helsinki Declaration and was approved by the Ethics Review Board at the Erasmus University Medical Center, Rotterdam, the Netherlands. Written and informed consent was obtained from all research participants included in this study. For quality control purposes, 111 plasma standards (Visucon-F frozen normal control plasma, pooled from 20 human donors, citrated, and buffered with 0.02 M HEPES, obtained from Affinity Biologicals, Ancaster, ON, Canada) and 40 PBS blanks were distributed across the 21 96-well sample plates.

SDS and analytical grade ethanol were obtained from Merck (Darmstadt, Germany). Disodium hydrogen phosphate dihydrate (Na₂HPO₄ × 2H₂O), potassium dihydrogenphosphate (KH₂PO₄), sodium chloride, Nonidet P-40 substitute (NP-40), 1-hydroxybenzotriazole monohydrate 97% (HOBt), 50% sodium hydroxide (NaOH), and super-DHB were obtained from Sigma-Aldrich (Steinheim, Germany). HPLC-grade acetonitrile (ACN) was purchased from Biosolve (Valkenswaard, the Netherlands), and 1-ethyl-3-(3-(dimethylamino)propyl)carbodiimide (EDC) hydrochloride was obtained from Fluorochem (Hadfield, UK). Peptide-N-glycosidase F (PNGase F) was obtained from Roche Diagnostics (Mannheim, Germany) and ultrapure water (MQ) was generated from a Purelab Ultra system (Veolia Water Technologies, Ede, the Netherlands), which was maintained at 18.2 MΩ at 25°C.

N-Glycans were released from serum proteins using PNGase F, as was previously described (32). In summary, 6 µL serum of each sample was denatured by addition of 12 µL of 2% SDS, followed by incubation for 10 min at 60°C. The release step was performed by the addition of 12.6 µL releasing mixture, which consisted of 2.5× PBS containing 2% NP-40 and 0.4 mU PNGase F, followed by 16 h incubation at 37°C. After release, the samples were stored at −20°C or transferred to the robot platform to be used directly.

Further sample preparation was performed on an automated system, as previously described (30). In short, linkage-specific sialic acid stabilization (ethyl esterification) was performed by adding 60 µL 250 mM EDC 250 mM HOBt in ethanol to 3 µL released glycan sample. This mixture was incubated for 75 min at room temperature, after which 120 µL ACN was added. To purify the samples, a GHP membrane (GHP plate, Pall AcroPrep Advance 96 Filter plate, Pall Corporation, Ann Arbor, MI, USA) was prewetted with 70% ethanol, activated with MQ and equilibrated with 100% ACN. The complete sample was transferred to the GHP plate and incubated for 5 min. Subsequently the GHP plate was washed three times with 100 µL 96% ACN. Elution was performed by adding 30 µL of MQ to the GHP plate, incubating at room temperature for 5 min, followed by centrifugation into a PCR plate.

From each purified sample, 10 µL was premixed with 10 µL matrix consisting of 5 mg/mL super-DHB in 99% ACN with 1 mM NaOH. Of this mixture, 2 µL was spotted onto a MALDI target plate (800/384 MTP AnchorChip, Bruker Daltonics, Bremen, Germany) and was left to dry. MALDI-TOF-MS spectra were recorded on an UltrafleXtreme mass spectrometer with a Smartbeam-II laser (Bruker Daltonics) in reflectron positive mode, controlled by flexControl 3.4 (Build 135). Measurements were performed within a range from m/z 1,000 to 5,000, accumulating 10,000 laser shots at a frequency of 1,000 Hz and with 100 shots per raster spot using a random walking pattern.

Average spectra were separately created for all healthy samples and for all RA samples. Within these two averages, 98 signals were manually assigned to putatively originate from glycan compositions. Using flexAnalysis 3.4 (Bruker Daltonics), the spectra were transformed into text format (x, y). Massytools (version 0.1.7.1 beta) was used to perform calibration and integration of the text files (33). Specifically, calibration was performed with a high precision calibration list, requiring at least 5 calibrants to be present at a RMS S/N ≥ 6 (119 spectra were excluded during this step, which included all 40 blanks). The putative glycan structures were extracted covering 95% of the isotopic envelope, summing per isotope the background-corrected area.

Spectra with quality control values below 3 SD of the mean values [i.e., fraction of spectrum in analytes, fraction of analyte area above S/N 9, and highest main peak (H5N4E2) S/N] were excluded from further analysis (13 spectra). Putative glycan signals were excluded from further analysis if failing to be present in 15% or more spectra of any analytical group (healthy, RA, standard) with an S/N ≥ 6 and a ppm error of at most 15 (which led to the exclusion of 21 signals). After curation, we retained 1,841 spectra with 78 analytes (Table S1 in Supplementary Material). Single N-glycan areas were normalized to the sum of all areas. To prevent outlier influence on the data analysis, individual glycosylation values surpassing 5 SD from their mean were removed from further analysis (118 values). Derived traits were calculated from the single glycans (Table S2 in Supplementary Material).

Measurement repeatability was assessed by calculating the mean, SD, and CV for all glycans and derived traits within the plasma controls (Table S3 and Figure S1 in Supplementary Material). Statistical analyses were performed using R 3.1.2 in an environment of RStudio 0.98.1091 (RStudio Team, Boston, MA, USA) (34). Figures were annotated with glycan symbols created in GlycoWorkbench 2.1 following the nomenclature proposed by the Consortium for Functional Glycomics (35, 36). For all statistical analyses, glycosylation values were scaled and centered, making regression effect sizes (B) representative of a 1 SD change in the glycan value.

Association between glycosylation and, respectively, pregnancy, RA, and DAS28(3)-CRP was explored by mixed linear regression. To establish the effect of pregnancy on glycosylation, a binary time point classification was constructed between (A) the first and third trimesters of pregnancy (coded, respectively, as 0 and 1), (B) the third trimester of pregnancy and 6 weeks postpartum (respectively, 0 and 1), and (C) the third trimester of pregnancy and 26+ weeks postpartum (respectively, 0 and 1). With an added random intercept per individual, glycan variables were used as outcome variable and time point classification as predictor variable (model: glycosylation ~ β₁·time point) (Table S4 in Supplementary Material). The effect of RA (healthy = 0; RA = 1) on glycosylation was modeled by applying a random intercept for all time points, using glycosylation as outcome variable and RA as predictor variable (model: glycosylation ~ β₁·RA) (Table S5 in Supplementary Material). The effect of DAS28(3)-CRP on glycosylation was modeled by either having a random intercept per time point (to analyze between individuals) or a random intercept per individual (to analyze within individuals), using glycosylation as outcome variable and DAS28(3)-CRP as predictor variable [model: glycosylation ~ β₁·DAS28(3)-CRP] (Table S6 in Supplementary Material).

To validate associations between glycosylation and disease activity during pregnancy, the difference in DAS28(3)-CRP [ΔDAS28(3)-CRP] and glycosylation (Δglycosylation) was established between the first trimester and third trimester of pregnancy (third trimester − first trimester). The association between these was established by linear regression [Δglycosylation ~ β₁·ΔDAS28(3)-CRP] (Table S7 in Supplementary Material). This process was similarly performed for the determination of the relationship between ΔDAS28(3)-CRP and Δglycosylation during the postpartum period, which was established between the third trimester of pregnancy and 12 weeks postpartum time point (12 weeks postpartum − third trimester). The time point at 12 weeks postpartum was chosen for the calculation because it displayed the highest disease activity after delivery, since most patients had already restarted anti-rheumatic therapy afterward.

Throughout the study, the Benjamini–Hochberg procedure was employed to control for multiple testing, which, under a study-wide false discovery rate of 5%, resulted in a significance threshold of α = 3.03 × 10⁻² (Table S8 in Supplementary Material) (37).

The association of pregnancy and the postpartum period with serum protein N-glycosylation was established by mixed linear regression. Within individuals, comparison was made between the first and third trimester (representative of pregnancy; respectively, coded 0 and 1; adjusted for individual), third trimester and 6 weeks postpartum, and third trimester and 26+ weeks postpartum (Table S4 in Supplementary Material).

With progression of pregnancy, a marked alteration in overall number of antennae per N-glycan was observed (Figure 3, top row). During pregnancy, within the complex type glycans (C), the relative abundance of tri- and tetraantennary N-glycans (CA3, CA4) showed to increase at the expense of diantennary species (CA2) (β_CA2 = −1.05 SE ± 0.06; β_CA3 = 1.08 ± 0.06; β_CA4 = 0.70 ± 0.06), followed by a slow recovery after delivery (β_CA2 = 0.94 ± 0.07; β_CA3 = −0.98 ± 0.07; β_CA4 = −0.56 ± 0.08; Table S4 in Supplementary Material).

For the serum N-glycosylation characteristics likely originating from the Fc portion of IgG (diantennary N-glycans without sialylation; A2S0) (15, 26), we confirmed with progressing pregnancy a decrease in fucosylation (A2S0F, β_A2S0F = −0.59 ± 0.05) and increase in galactosylation (A2FS0G, β_A2FS0G = 0.59 ± 0.04), both rapidly reversing after delivery (β_A2S0F = 0.67 ± 0.04; β_A2FS0G = −0.81 ± 0.04) (Figure 3, bottom row).

For the serum N-glycosylation characteristics likely originating for a large part from IgG-Fab, IgM, and IgA (fucosylated diantennary N-glycans with sialylation; A2FS) (15, 26), we observed a decrease in bisection with trimester progression (A2FSB, β_A2FSB = −0.56 ± 0.04) and a rapid return to pre-pregnancy levels at 6 weeks postpartum (β_A2FSB = 1.11 ± 0.03).

Notably, whereas α2,6-linked sialylation remained relatively stable throughout pregnancy (Figure 4, bottom row), a drastic increase of α2,3-linked sialylation (e.g., per galactose on di- or triantennary glycans, respectively, A2GL and A3GL) was observed up to the third trimester (β_A2GL = 1.11 ± 0.05; β_A3GL = 1.16 ± 0.05), followed by a rapid return to baseline levels at the first time point after delivery (β_A2GL = −1.19 ± 0.05; β_A3GL = −1.42 ± 0.04) (Figure 4, top row).

By mixed linear regression, glycosylation values were compared between individuals with RA and controls (healthy = 0, RA = 1; adjusted for pregnancy) (Table S5 in Supplementary Material). For immunoglobulin type glycosylation, we observed in RA a lower galactosylation (A2FS0G, β_A2FS0G = −0.44 ± 0.07), higher bisection (A2FS0B, β_A2FS0B = 0.55 ± 0.08), and a higher sialylation per galactose (A2FGS, β_A2FGS = 0.75 ± 0.08). In addition, differences were found in several non-fucosylated diantennary traits (higher A2F0G, A2F0GS, and A2GE, lower A2F0B, A2F), all mainly driven by elevated levels of H5N4E2 (β_H5N4E2 = 0.97 ± 0.07) in RA patient serum. For tri- and tetraantennary N-glycosylation, we observe with RA a higher fucosylation (e.g., A3F, β_A3F = 0.34 ± 0.08; A4F, β_A4F = 0.56 ± 0.08) particularly within species with α2,3-linked sialylation (A3LF, β_A3LF = 0.35 ± 0.08; A4LF, β_A4LF = 0.58 ± 0.08), suggesting an increase in sialyl-Lewis X/A.

In addition, we compared by mixed linear regression the association of glycosylation with RA disease activity, as assessed by DAS28(3)-CRP, both between and within individuals with RA (respectively, adjusted for time point and individual) (Figure S2 and Table S6 in Supplementary Material). The increase in DAS28(3)-CRP associated with both the decrease in IgG-Fc-type galactosylation (e.g., A2S0G, β_A2S0G = −0.34 ± 0.02) and the decrease in fucosylation of non-α2,3-sialylated triantennary species (A3L0F, β_A3L0F = −0.13 ± 0.03). Increasing with DAS28(3)-CRP were the bisection of IgG-Fc-type glycans (A2F0S0B, β_A2F0S0B = 0.25 ± 0.03) as well as the sialylation and fucosylation of tri- and tetraantennary species (e.g., A3FGS, β_A3FGS = 0.26 ± 0.02; A4F, β_A4F = 0.13 ± 0.03; A4GS, β_A4GS = 0.17 ± 0.02). Glycosylation features associating with DAS28(3)-CRP only within (and not between) individuals throughout pregnancy were the fucosylation of (α2,6-)sialylated diantennaries (A2SF, β_A2SF = −0.12 ± 0.02) and α2,6-sialylation of triantennary non-fucosylated species (A3F0GE, β_A3F0GE = 0.14 ± 0.02).

Finally, between trimesters 1 and 3 (response timeframe) as well as between trimester 3 and 12 weeks postpartum (flare timeframe), we employed linear regression to confirm the within-individual association of DAS28(3)-CRP [ΔDAS28(3)-CRP] with glycosylation (Δglycosylation) (Table S7 in Supplementary Material). In both timeframes, ΔDAS28(3)-CRP showed a negative association with the Δgalactosylation of IgG-Fc type species (ΔA2S0G, responder, β_ΔA2S0G = −0.29 ± 0.07; flare, β_ΔA2S0G = −0.35 ± 0.07; ΔA2FS0G, responder, β_ΔA2FS0G = −0.29 ± 0.07; flare, ΔA2FS0G; β_ΔA2FS0G = −0.33 ± 0.07), whereas the sialylation of triantennary fucosylated species showed in both cases a positive association (ΔA3FGS, responder, β_ΔA3FGS = 0.26 ± 0.07; flare, β_ΔA3FGS = 0.35 ± 0.07) (Figure 5). The flare timeframe additionally showed a positive association of ΔDAS28(3)-CRP with the (α2,6)-sialylation of (fucosylated) diantennary glycans (ΔA2GE, β_ΔA2GE = 0.27 ± 0.07; ΔA2FGS, β_ΔA2FGS = 0.24 ± 0.07).

To interpret the information contained in this study, several aspects of our MALDI-TOF-MS analysis need to be kept in mind. First, MS assesses glycan chemical compositions and not structures, although sialic acid linkage information is provided by the employed derivatization technique (29). Other structural characteristics are presumed based on a wide array of literature on biosynthetic pathways, and on experiments with enzymatic digestion, nuclear magnetic resonance spectroscopy, and MS fragmentation (12, 15, 38–40). Second, the changes observed in the released N-glycan samples could have originated not only from changes in the glycosylation of proteins but also from changes in the abundance of those glycoproteins. Derived traits have been constructed to reflect differences in biosynthesis, and our current day understanding of the relative contribution of specific glycoproteins and tissues to the serum N-glycome (15, 26). Third, the mass spectrometric analysis did not provide quantitative ratios of N-glycosylation, but the direction and magnitude of observed changes is expected to be biologically representative, as suggested by method comparisons (41).

In addition, it has to be noted that the clinical associations found in the study may, in part, originate from treatment differences between individuals during the study period. For example, methotrexate treatment typically only starts after delivery, potentially influencing the time point comparisons. However, this confounding effect is expected to be limited due to the relatively low number of individuals undergoing the treatment. Details on the medication can be found in Table 2.

Previously, we found glycosylation changes throughout healthy pregnancies, and have in RA patients studied the specific glycosylation of IgG-Fc throughout pregnancy (20, 21, 27, 28, 42, 43). In the current study, the first application of a similar MALDI-MS approach with automated sample preparation, we achieved the analysis of the total serum N-glycosylation throughout the pregnancies of both healthy controls and RA patients, in total leading to the analysis of 1,770 clinical samples (30). Since the current cohort also contains the control individuals (without RA) from prior studies, these previous results were confirmed, but we could also for the first time show that comparable TSNG changes can be observed in RA patients.

As such, we detected with the progression of pregnancy an increase of galactosylation (A2FS0G) as prior reported for the Fc part of IgG, as well as a decrease in bisection (A2FSB) (18, 20, 21). Additionally, we observed with pregnancy an overall increase in glycan branching (from CA2 to CA3 and CA4) and an increase of α2,3-linked sialylation (L, in part at the expense of α2,6-sialylation, E). In all cases, the postpartum period led to the return to the values before or at the beginning of pregnancy. The increased branching observed in the TSNG could have originated from the abundance and antennarity of acute-phase proteins such as alpha-1-acid glycoprotein, for which an increased serum level and N-glycan branching has been reported with pregnancy (24, 25). On the other hand, we did not observe the decrease in fucosylation reported for the same protein, potentially obscured by the increased fucosylation of other proteins, and have yet to identify the source of the substantial increase of α2,3-linked sialylation up to the third trimester. In literature, it has been reported that IgG and alpha-1-acid glycoprotein glycosylation may be affected by estrogens (44, 45), known to change significantly throughout pregnancy (46), and it is conceivable that other proteins within the TSNG are under similar high-level control.

The total serum N-glycosylation changes with pregnancy showed remarkably comparable between RA patients and healthy controls, but baseline differences could be detected. For example, RA patients displayed a lower degree of IgG-Fc-type galactosylation and higher bisection when compared to controls (A2FS0G, A2FS0B). A decreased IgG-Fc galactosylation and increased bisection are well-known to associate with a variety of inflammatory conditions, including RA, inflammatory bowel disease, and aging (18, 47, 48), and the same glycosylation phenotypes appear detectable in our TSNG study as well (20, 21). The mechanisms by which IgG-Fc glycosylation may affect inflammatory processes remain for a large part to be elucidated, but increased galactosylation and decreased fucosylation have been implicated in increased FcγRIIa and FcγRIIIa binding and antibody-dependent cellular cytotoxicity (49–51).

Furthermore, RA patients showed in our study a higher α2,6-sialylation and lower fucosylation of diantennary glycans (A2GE, A2F) compared to healthy controls, mainly driven by the N-glycan composition H5N4E2, as well as a substantially higher (multi-)fucosylation of tri- and tetraantennary glycans [A3F(a), A4F(a)]. The increased A3/A4 fucosylation strongly suggests the upregulation of sialyl-Lewis X, which has been reported with inflammatory arthritis for several acute-phase proteins, e.g., alpha-1-acid glycoprotein, haptoglobin, alpha-1-antichymotrypsin, and transferrin (52–56). The sialyl-Lewis X on glycans is known to bind to E-selectin, an inducible receptor expressed by endothelial cells (16). This interaction is implicated in the homing of immune cells to a site of inflammation (57), as well as cancer metastasis (58), and may play a role in RA as well. For instance, the alpha-1-acid glycoprotein observed within RA synovial fluid is thought to be of hepatic origin (59), meaning its circulatory variant might make use of a glycan-mediated mechanism for transportation toward the inflamed synovial tissue.

Next to a negative association with the galactosylation of IgG-Fc-type N-glycans, we additionally report a positive association between (Δ)DAS28(3)-CRP and the sialylation of triantennary fucosylated glycans (A3FGS), notably of the α2,6-linked variety. This is in contrast to the TSNG changes throughout pregnancy, which shows marked increase of α2,3-linked sialylation but not an association with disease activity. A recent study, which compared total plasma N-glycosylation with the levels of various metabolic and inflammatory markers, has indicated a link between A3FGS and CRP, suggesting that the CRP component of DAS28(3)-CRP may in part be responsible for the association found within the current study (60). While the protein source of A3FGS remains unknown, future studies will have to reveal its biomarker potential and facilitate our understanding of RA disease severity.

To summarize, by performing MS-based total serum protein N-glycosylation analysis we detected (1) changes in protein glycosylation throughout pregnancy, (2) glycosylation differences between healthy individuals and RA patients, and (3) glycosylation traits coinciding with the pregnancy-associated changes in RA disease activity. While we confirmed in serum the IgG glycosylation phenotypes that were prior reported to associate with RA disease activity, our glycomics approach has additionally allowed the detection of changes that are presumably independent from IgG, namely, the sialylation of fucosylated triantennary N-glycans (A3FGS).