Reactivation of rheumatoid arthritis after pregnancy: Increased phagocyte and recurring lymphocyte gene activity

Authors


Abstract

Objective

Pregnancy is associated with reduced disease activity in rheumatoid arthritis (RA) and frequently with disease exacerbation after delivery. This study was undertaken to generate a systematic overview of the molecular mechanisms related to disease remission and postpartum reactivation.

Methods

Transcriptomes of peripheral blood mononuclear cells (PBMCs) were generated from RA patients and healthy women by transcription profiling during the third trimester and 24 weeks after delivery. For functional interpretation, signatures of highly purified immune cells as well as Kyoto Encyclopedia of Genes and Genomes pathway annotations were used as a reference.

Results

Only minor differences in gene expression in PBMCs during pregnancy were found between RA patients and controls. In contrast, RA postpartum profiles presented the most dominant changes. Systematic comparison with expression signatures of monocytes, T cells, and B cells in healthy donors revealed reduced lymphocyte and elevated monocyte gene activity during pregnancy in patients with RA and in controls. Monocyte activity decreased after delivery in controls but persisted in RA patients. Furthermore, analysis of 32 immunologically relevant cellular pathways demonstrated a significant additional activation of genes related to adhesion, migration, defense of pathogens, and cell activation, including Notch, phosphatidylinositol, mTOR, Wnt, and MAPK signaling, in RA patients postpartum.

Conclusion

Our findings indicate that innate immune functions play an important role in postpartum reactivation of arthritis. However, this may depend not only on the monocyte itself, but also on the recurrence of lymphocyte functions postpartum and thus on a critical interaction between both arms of the immune system.

Pregnancy induces amelioration or remission of disease in the majority of patients with rheumatoid arthritis (RA) (1). Within 3 months after delivery, the disease relapses or flares (2). It appears reasonable to postulate that disease remission is related to the physiologically occurring mechanisms of tolerance, which are not restricted to immunologic changes at the maternal–fetal interface but become visible at the systemic level. Pregnancy-induced remission offers a unique opportunity to study pathogenetic mechanisms of relapse.

Different mechanisms contribute to maternal tolerance of the fetus (3). Several studies have demonstrated a decrease in the number of T cells during pregnancy (4–6). In addition, in healthy women, cytokine production of T cells changes from a Th1-mediated to a Th2-mediated response during pregnancy (7, 8). Our previous investigations of RA during pregnancy did not reveal differences regarding levels of circulating cytokines (interleukin-1β [IL-1β], IL-10, and interferon-γ) (9); however, a predominance of Th2 and a suppression of Th1 cytokine secretion by peripheral blood mononuclear cells (PBMCs) during pregnancy was demonstrated. Collectively, the available data do not fully explain the profound disease-remitting effect of pregnancy on RA. Therefore, in the present study we used a different methodologic approach.

Expression profiling has been applied with increasing success to describe molecular processes of disease not only in inflamed tissue (10–12), but also in blood from patients with systemic lupus erythematosus (SLE) (13, 14), rheumatoid arthritis (15, 16), or psoriatic arthritis (17). Furthermore, this technology has been used to search for diagnostic patterns associated with therapeutic responsiveness (18, 19). Based on the fact that transcriptome analysis currently provides the most comprehensive overview, we applied this technology to search for regulatory factors/pathways involved in disease remission during pregnancy and in postpartum relapse.

Analysis of RA patients and healthy women in the third trimester and 24 weeks after delivery allowed us to compare gene profiles of RA patients whose disease was in remission with those of healthy controls in the third trimester and to describe genetic activity at postpartum relapse. Importantly, PBMCs could be analyzed in the almost complete absence of medication during pregnancy. Because of possible changes in cellular composition during pregnancy, expression profiles of monocytes, T cells, B cells, and granulocytes were generated. These helped to differentiate between changes in cell numbers and changes in cell activation. Finally, cellular pathways were analyzed.

PATIENTS AND METHODS

Patients and controls.

Six pregnant patients with RA and 8 age-matched healthy pregnant controls were studied at gestational weeks 32–34 and 24 weeks postpartum. The study was performed at the Department of Rheumatology and Clinical Immunology/Allergology of the University Hospital of Bern after approval by the institutional review board of Bern. RA patients fulfilled the American College of Rheumatology (formerly the American Rheumatism Association) criteria (20). The mean age of patients was 31 years (range 21–38 years), and the mean age of healthy controls was 33 years (range 21–40 years). Disease activity was assessed using the RA Disease Activity Index (RADAI) (21), the tender and swollen joint count, and C-reactive protein (CRP) level (Table 1). Nonsteroidal antiinflammatory drugs were allowed until week 32 of gestation, and antimalarials, sulfasalazine, and a maximum of 10 mg/day of prednisone were allowed throughout pregnancy. Flares postpartum were treated according to standard protocols.

Table 1. Clinical characteristics of the 6 patients with rheumatoid arthritis*
PatientSwollen joint countCRP, mg/literRADAIMedication
  • *

    CRP = C-reactive protein; RADAI = Rheumatoid Arthritis Disease Activity Index; SSZ = sulfasalazine.

Donor 1    
 Third trimester6120.3SSZ 500 mg/day
 Postpartum5<100.3SSZ 3,000 mg/day
Donor 2    
 Third trimester0<100.0None
 Postpartum2<100.0Chlorambucil, prednisone
Donor 3    
 Third trimester1<100.4None
 Postpartum6<100.7None
Donor 4    
 Third trimester1<100.5None
 Postpartum0<100.5Acetaminophen
Donor 5    
 Third trimester5412.4SSZ 2,000 mg/day
 Postpartum1360.8SSZ, prednisone, etanercept
Donor 6    
 Third trimester0<100.0None
 Postpartum5518.6Methotrexate 15 mg/week

Isolation and cryopreservation of PBMCs.

PBMCs were isolated from 50 ml of citrate blood within 60 minutes of collection by Ficoll-Hypaque gradient centrifugation (Biocoll separating solution; Oxoid, Basel, Switzerland). After washing twice with phosphate buffered saline, cells were suspended at a concentration of 106–107 cells/ml in 10% DMSO and 90% heat-inactivated fetal calf serum, frozen at a rate of 1°C/minute to −80°C, and stored in liquid nitrogen until thawed directly for array analysis.

Reference cell types and spiking experiment.

Individual cell types (monocytes, granulocytes, CD4 and CD8 T cells, natural killer [NK] cells, and B cells) were purified as described previously (22). Briefly, after erythrocyte lysis at 4°C, CD15+ granulocytes were extracted by automated magnetic sorting (autoMACS; Miltenyi Biotec, Bergisch Gladbach, Germany). The CD15-depleted fraction was divided for 2 separate 4-channel high-speed fluorescence-activated cell sorts (DIVA; BD Biosciences, Heidelberg, Germany) using one antibody cocktail with Alexa 405–conjugated anti-CD3, phycoerythrin (PE)–Cy7–conjugated anti-CD56, fluorescein isothiocyanate (FITC)–conjugated anti-CD14, and allophycocyanin-Cy7–conjugated anti-CD19 and a second cocktail with Alexa 405–conjugated anti-CD3, FITC-conjugated anti-CD14, PE-Cy5–conjugated anti-CD4, and PE-Cy7–conjugated anti-CD8. Dead cells were excluded by addition of 4′,6-diamidino-2-phenylindole. Purities of cells were >97%, and viabilities were >99%. After sorting, cells were immediately lysed in RLT buffer containing 1% β-mercaptoethanol, and lysates were stored at −80°C until RNA isolation.

For the spiking experiment, a whole blood leukocyte preparation containing 55% granulocytes, 11.5% monocytes, 20.5% T cells (14% CD4+ T cells, 6.5% CD8+ T cells), 7.5% NK cells, and 4.5% B cells was used. CD3+ T cells were isolated and spiked back into the T cell–depleted leukocyte fraction at defined concentrations (0%, 0.8%, 1.6%, 3.1%, 6.3%, 12.5%, 25%, 50%, and 100%). (Details are available upon request from the corresponding author)

RNA isolation and Affymetrix analysis.

Total RNA was extracted from PBMCs and reference samples using the RNeasy Mini kit, according to the recommendations of the manufacturer (Qiagen, Hilden, Germany). For microarray hybridization on HG-U133A GeneChips (Affymetrix, Santa Clara, CA), complementary DNA was synthesized from 3 μg of total RNA and transcribed in vitro (Enzo Biochem, New York, NY) to generate biotin-labeled complementary RNA (cRNA). Fragmented cRNA (50 μg/ml) was hybridized for 16 hours at 45°C. Arrays were washed and stained under standardized conditions (fluidic station) and scanned on a Hewlett Packard Genearray Scanner (Affymetrix) controlled by MAS 5.0 software.

Statistical analysis.

CEL files were generated with GCOS 1.2 (Affymetrix) and subjected to robust multichip analysis for quantile-normalized signal calculation (23). Fold changes were calculated and classified as “increased” or “decreased” if differences ≥1.5 fold were observed in pairwise comparisons. For group comparisons, log-transformed signal ratios of all pairwise comparisons were averaged and converted to the mean fold change. “Increased” or “decreased” changes were summarized as percentage of changes in all pairwise comparisons. Cluster analysis was performed with Genesis (developed by Alexander Sturn, Graz Institute for Genomics and Bioinformatics [Graz, Austria]) using log-transformed and z-normalized signal values. Gene expression and clinical data were correlated using Pearson's correlation coefficient and Spearman's correlation coefficient. A relational database was built with FileMaker Pro, version 8.0 (FileMaker, Unterschleissheim, Germany) to enable complex queries. Effects of differential composition were determined using the standard profiles of the highly purified CD14+ peripheral blood monocytes, CD4+ and CD8+ T cells, NK cells, CD19+ B cells, and CD15+ granulocytes.

For each gene (G), the average of the log-transformed signal ratios (meanSLRG) was calculated by comparing a given cell type “A” (CTA) with all other cell types (CTBi):

equation image

To generate the cell type association score for each gene and cell type, all meanSLRmath image were scaled relative to the maximum of these values in this comparison:

equation image

This score revealed positive values for higher and negative values for lower expression in a particular cell compared with all other cell types. With 1 as the maximum score for the most typical gene of a cell type, the scores 0.5 and 0.25 were found at an average rank of 83 and 559, respectively. Since differences between CD4+ and CD8+ T cells were limited, CD4+ cells were used to compare T cells with non–T cells, and each T cell subset was compared only with non–T cells. Differentially expressed genes were ranked, and the cell type association scores were cumulatively plotted to present an unbiased overview of cell type associations. To estimate the quantity of a cell type in a sample, the relative signal intensities between the sample and the purified reference cell type were determined for the 20 most specific genes of this cell type and averaged.

Pathway analysis.

A systematic analysis was performed with 1,791 genes in 32 different Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways classified into the following groups: cell communication, signal transduction, signaling molecules and interaction, immune system, infectious diseases, and cell growth and death. Pathway-related gene transcription was scored by calculating the mean value over all genes of a pathway using the cell type association scores for profiles of reference cell types and using the log-transformed signal ratios for group comparisons. To estimate significance of differences between groups, signal values of all pathway-related genes were z-normalized by gene, averaged for each PBMC sample and pathway, and subjected to the Welch test.

RESULTS

Patient characteristics and clinical outcomes during pregnancy.

All patients and healthy controls had uncomplicated pregnancies with delivery of healthy children between 36 and 40 weeks of gestation. Disease activity scores and CRP values are shown in Table 1. Disease activity during pregnancy was low in all but 1 patient. Accordingly, 4 patients received no drugs during pregnancy. Due to disease activity, 4 patients were receiving disease-modifying antirheumatic drugs (DMARDs) 24 weeks after delivery.

Comparison of patients and controls during and after pregnancy.

For group comparisons between postpartum data and third-trimester data within each donor group (postpartum RA patients versus third-trimester RA patients and postpartum controls versus third-trimester controls), as well as for group comparisons at the 2 time points between the 2 groups (postpartum RA patients versus postpartum controls and third-trimester RA patients versus third-trimester controls), genes that had ≥1.5-fold differential expression in ≥50% of all pairwise comparisons were selected. The fewest differences were found in samples from postpartum healthy women versus samples from pregnant healthy women (10 genes increased and 50 decreased). Notably, few differences were also found in pregnant RA patients versus pregnant healthy subjects (39 genes increased and 48 decreased).

Although individual genes were differentially expressed in a maximum of 83% of all pairwise comparisons, hierarchical cluster analysis separated the samples correctly into the different expression patterns of the groups, except for 1 sample in each comparison (Figures 1A and B). It was possible to separate almost all pregnancy profiles of healthy women from those of RA patients; however, differences between patients whose RA was in complete remission and controls and differences between patients not taking DMARDs and controls were inhomogeneous. Of note, the patient in whom disease remained active in the third trimester (patient 5) presented a distinct pattern (Figure 1B).

Figure 1.

Hierarchical cluster analysis of gene expression in peripheral blood mononuclear cells from healthy controls (normal donors [ND]) in the third trimester of pregnancy (3T), healthy controls 24 weeks postpartum (pp), rheumatoid arthritis (RA) patients in the third trimester of pregnancy, and RA patients 24 weeks postpartum. A, Comparison of genes differentially expressed in ≥50% of all pairwise comparisons in healthy controls in the third trimester versus healthy controls postpartum. B, Comparison of genes differentially expressed in ≥50% of all pairwise comparisons in healthy controls in the third trimester versus RA patients in the third trimester. C, Comparison of genes differentially expressed in ≥60% of all pairwise comparisons in healthy controls postpartum versus RA patients postpartum. D, Comparison of genes differentially expressed in ≥85% of all pairwise comparisons in RA patients in the third trimester versus RA patients postpartum. Increasing thresholds in C and D were necessary to reduce the number of genes to a level which allowed visualization of gene symbols.

In contrast to the subtle differences between controls and RA patients in the third trimester and between controls during pregnancy and controls after pregnancy, many more genes with up to 100% differential expression were identified when comparing RA patients postpartum versus healthy controls postpartum (155 genes increased and 30 decreased) or RA patients postpartum versus RA patients during pregnancy (997 genes increased and 48 decreased). For cluster analysis, subselections of genes differentially expressed in ≥60% (for RA patients postpartum versus controls postpartum) or ≥85% (for RA patients postpartum versus RA patients during pregnancy) of the pairwise comparisons were selected (Figures 1C and D), and these separated correctly into the different groups without any exception. Patient 6 had the highest level of inflammatory activity postpartum, based on CRP level and RADAI (Table 1) and separated from other RA patients, with a strong overexpression of a broad panel of genes, when compared with healthy controls postpartum (Figure 1C).

Gene expression signatures of purified cell populations and effect of differential cellular compositions on gene expression profiles.

Differential blood count may confound interpretation of gene expression analysis in PBMCs. Therefore, CD3+ T cells were depleted from whole blood leukocytes and then spiked back in defined concentrations to produce 10 samples with different cellular compositions. (Details are available upon request from the corresponding author.) Approximately 3,000 probe sets were found to be increased and 2,500 to be decreased in CD3+ T cells compared with T cell–depleted leukocytes. With decreasing concentrations of T cells, relative differences in T cells were 2-fold for each spiking step, but absolute differences in the cellular composition of the samples were decreasing. This was reflected by a decrease in the number of T cell–related genes differentially expressed at low T cell concentrations, which may be explained by a loss of discriminatory power, especially at low concentrations where the quality of detection and signal stability are reduced. On the other hand, with increasing leukocyte concentration, absolute and relative differences of non–T cell populations were decreasing with each titer step. This was reflected by a decreasing number of non–T cell–related genes differentially expressed at high leukocyte concentrations. Nevertheless, signals of non–T cell genes were increasing, demonstrating that cell concentration and signal intensity were related when signals were within the range of stable detection.

Signatures of highly purified leukocyte populations (CD4+ T cells, CD8+ T cells, CD56+ NK cells, CD11+ granulocytes, CD19+ B cells, and CD14+ monocytes) were compared with each other, and a normalized mean of the log-transformed signal ratios was calculated to score each individual gene for its preferential expression in each cell type. This score was applied to genes differentially expressed between spiked samples. Genes with increased expression in the sample with higher T cell concentration were sorted by fold change. The scores for the top 100 genes were cumulatively added for each cell type separately and plotted (Figures 2A and B). This revealed a dominance of CD4+ and CD8+ T cell–related genes not only in comparisons of samples with high T cell concentrations (100% versus 50%), but also in samples with low T cell concentrations (12.5% versus 6.3%). These genes were in part also related to other lymphocytes and either not related or negatively related to phagocytes. This relationship of molecular signatures with cell types was confirmed when comparing the scores of the 20 most characteristic genes for each cell type. (Details are available upon request from the corresponding author.)

Figure 2.

Cell type association scores of genes with increased expression in 2 comparisons of spiked samples and in 4 comparisons between peripheral blood mononuclear cell samples obtained from RA patients and healthy controls during and after pregnancy. After ranking differentially expressed genes by percentage of increase in pairwise comparisons, the 100 leading candidate genes were selected and scores for cellular dominance for monocytes (Mo), CD4+ T cells (CD4), CD8+ T cells (CD8), and B cells (CD19) were cumulatively plotted. A and B, The spiking experiment provided proof of concept that the scores derived from comparisons of the different cell types reflected differences in cellular composition. C–F, After pregnancy, healthy controls showed an increase in T cell–associated but not monocyte-associated gene expression (C), while changes in RA patients were less dominant (D). Comparisons between RA patients and healthy controls during pregnancy (E) and especially after pregnancy (F) revealed explicitly higher monocyte-associated gene expression in RA patients. See Figure 1 for other definitions.

The relative quantity of each cell type in the spiked samples was assumed to correlate with the signal level of marker genes and calculated as the ratio between the signals of the spiked sample and the purified reference cell type. The median cell type fraction was calculated from up to 50 of the best marker genes, according to cell type association scores. The top 20 genes were sufficient to provide a stable estimation that was highly correlated with the spiking concentrations in all cell types except NK cells (R = 0.989 for CD4, R = 0.994 for CD8, R = 0.983 for monocytes, R = 0.955 for B cells, and R = 0.556 for NK cells) (Figures 3A and B). (Details regarding estimation are available upon request from the corresponding author.) The NK cell fraction was minor; NK cell scores overlapped with those of CD8+ T cells and thus were influenced by the changing T cell concentration.

Figure 3.

Estimation of the quantity of different cell types in the spiking experiments and each peripheral blood mononuclear cell (PBMC) preparation, using the 20 most cell-specific genes. The mean ± SD or values for individual donors (healthy donors 1 [D] 1 through 8 and donors with RA 1 through 6) are indicated for each cell type: monocytes (Mo), granulocytes (polymorphonuclear cells [PMN]), CD4+ T cells (CD4), CD8+ T cells (CD8), natural killer (NK) cells, and B cells (CD19). A and B, The spiking experiment demonstrated that the 20 most specific genes according to the scoring can provide a basis for molecular quantification of the different cell types; however, there was a lack of distinction between CD8+ and CD4+ T cells. Bars show the mean ± SD. C, The decrease in monocytes in healthy controls postpartum and the persistence of elevated levels in RA patients suggested by scoring of differentially expressed genes was confirmed by using the 20 most cell-specific genes and calculating the signal ratios between PBMCs and purified cell types. See Figure 1 for other definitions.

Contribution of differential cellular composition to the differences in gene expression in healthy women and RA patients during and after pregnancy.

The leukocyte populations of the samples obtained from RA patients and controls at the 2 time points were analyzed as in the spiking experiment, as described above. Granulocyte numbers were negligible after Ficoll preparation, and NK cells constituted a very small fraction, similar to B cells (data not shown). Figure 2 shows the cumulative cell type association score for the leading 100 genes differentially expressed in the different group comparisons (Figures 2C–F). For postpartum healthy women compared with pregnant healthy women, a positive association with T cell–related genes and negative scores for phagocytes were found. This T cell association was not as strong in RA patients after pregnancy.

Direct comparison between RA patients and healthy controls during and after pregnancy revealed that samples from RA patients expressed phagocyte-related genes at a higher level. This was more pronounced in the samples from RA patients postpartum. Interestingly, dominance of genes related to monocytes was also found when comparing samples from healthy women during pregnancy with those from healthy women postpartum.

Using the 20 most characteristic genes of each leukocyte population for the molecular cell count (see Supplemental Table 1, available in the online version of this article at http://www3.interscience.wiley.com/journal/76509746/home), a decrease in monocytes in healthy women after pregnancy and a persistent elevation in RA patients was confirmed (Figure 3). CD4+ and CD8+ cells increased postpartum in both groups, and the numbers appeared to be more variable in RA patients. Fractions of B cells were low but increased to a minor extent in controls and in RA patients when tested in paired samples (P ≤ 0.05).

Results of pathway analysis.

The genes differentially expressed in RA patients postpartum were predominantly related to KEGG pathways in the groups cell communication, immune system, infectious diseases, signal transduction, signaling molecules and interactions, and cell growth and death. Therefore, we analyzed all genes in 32 pathways associated with these groups. Average values derived from all genes of a pathway as a set of functionally related genes generated more stable information than analysis of single genes. Reference profiles were investigated using the cell type association scores to identify cell type–specific preferences for particular pathways (Figure 4A) and to distinguish effects of differential cellular composition from those of pregnancy- or disease-related gene activation.

Figure 4.

Differential pathway-related gene activity in A, purified monocytes (Mo), granulocytes (polymorphonuclear cells [PMN]), CD4, CD8, natural killer (NK), and CD19 cells and B, peripheral blood mononuclear cell samples from healthy controls in the third trimester of pregnancy, healthy controls postpartum, RA patients in the third trimester of pregnancy, and RA patients postpartum. Activity in each cell type was calculated using the gene-specific cell type association scores in A and the z-normalized signal values in B, averaged over all genes assigned to a pathway. See Figure 1 for other definitions.

A distinct pattern was seen for normal phagocytes compared with lymphocytes. Genes related to complement and coagulation cascades, migration, adhesion, signal transduction processes, and pathogen responses were active in monocytes. No preferential activity of genes related to signal transduction pathways was found in these cells. Transcripts related to antigen processing were found in higher levels in monocytes than in any other cell type and were found at a lower level in B cells. As expected, genes related to typical processes of T or B cells were dominant in the corresponding lymphocyte populations. Cell cycle–related gene activity was higher in lymphocytes than in phagocytes.

Pathway scores were calculated for each PBMC sample using z-normalized signal values (Figure 4B and Table 2). According to the shift in lymphocyte and monocyte fractions in samples from pregnant versus postpartum healthy women, scores for T cell and B cell signaling as well as cell cycle were increased, and those for complement and coagulation cascade processes were decreased. Additional changes, which may indicate a low level of gene activation, were only minor, with increased scores postpartum for phosphatidylinositol and vascular endothelial growth factor signaling, and nonsignificant changes in scores for adhesion, migration, pathogen response, and a few other signaling-related gene activities.

Table 2. Mean scores for gene activity in individual pathways and significance of differential expression during and after pregnancy in RA patients and controls*
 Third-trimester controlsPostpartum controlsThird-trimester RA patientsPostpartum RA patientsPPP§PP#
  • *

    There were no significant differences between rheumatoid arthritis (RA) patients in the third trimester and controls in the third trimester. NS = not significant; TLR = Toll-like receptor; FcεRI = Fcε receptor I; NK = natural killer; VEGF = vascular endothelial growth factor; TGFβ = transforming growth factor β; CAMs = cell adhesion molecules; ECM = extracellular matrix.

  • Controls postpartum versus controls in the third trimester.

  • RA patients postpartum versus RA patients in the third trimester.

  • §

    RA patients postpartum versus controls postpartum.

  • RA patients postpartum versus controls in the third trimester.

  • #

    RA patients in the third trimester versus controls postpartum.

Cell communication         
 Adherens junction−0.13−0.05−0.100.34NS0.050NS0.043NS
 Tight junction−0.13−0.05−0.080.33NSNSNS0.046NS
 Focal adhesion−0.140.01−0.110.28NSNSNSNSNS
 Gap junction−0.18−0.07−0.070.41NS0.0340.0320.017NS
Immune system         
 Leukocyte transendothelial migration−0.10−0.06−0.080.30NS0.0160.0210.016NS
 TLR signaling pathway−0.10−0.13−0.040.35NSNS0.0300.037NS
 Complement and coagulation  cascades−0.06−0.240.140.24NSNS0.046NS0.006
 FcεRI signaling pathway−0.170.04−0.190.380.0480.011NS0.0120.045
 Hematopoietic cell lineage−0.10−0.01−0.040.19NS0.0190.0320.014NS
 Antigen processing and presentation0.03−0.04−0.060.09NSNSNSNSNS
 B cell receptor signaling pathway−0.230.11−0.290.470.0170.0020.0470.0030.011
 NK cell–mediated cytotoxicity−0.120.02−0.180.33NS0.019NS0.0300.029
 T cell receptor signaling pathway−0.220.11−0.380.560.0060.0030.0370.0050.000
Infectious diseases         
 Cholera−0.25−0.18−0.020.58NS0.0110.0040.002NS
 Pathogenic Escherichia coli infection−0.090.03−0.040.12NSNSNS0.039NS
 Epithelial cell signaling in Helicobacter pylori infection−0.19−0.06−0.100.42NS0.0260.0340.015NS
Signal transduction         
 Notch signaling pathway−0.130.01−0.340.54NS0.0000.0060.0020.025
 Phosphatidylinositol signaling system−0.270.07−0.290.570.0100.0020.0190.0020.005
 mTOR signaling pathway−0.220.02−0.200.47NS0.0110.0410.0100.028
 Calcium signaling pathway−0.15−0.060.000.26NSNSNSNSNS
 Jak-STAT signaling pathway−0.04−0.11−0.100.31NSNSNSNSNS
 VEGF signaling pathway−0.190.06−0.160.330.0390.017NS0.0140.036
 Wnt signaling pathway−0.16−0.01−0.200.44NS0.0140.0440.0180.028
 TGFβ signaling pathway−0.10−0.04−0.160.37NS0.037NSNSNS
 MAPK signaling pathway−0.160.00−0.150.37NS0.023NS0.022NS
 Hedgehog signaling pathway−0.07−0.07−0.070.27NSNSNSNSNS
Signaling molecules and interaction         
 CAMs−0.06−0.100.000.20NSNSNSNSNS
 Cytokine–cytokine receptor interaction0.00−0.140.020.18NSNSNSNSNS
 ECM–receptor interaction−0.08−0.070.020.17NSNSNSNSNS
 Neuroactive ligand–receptor interaction−0.02−0.120.080.10NSNSNSNSNS
Cell growth and death         
 Apoptosis−0.210.00−0.230.530.0360.0080.0250.0080.015
 Cell cycle−0.200.07−0.240.430.0110.030NS0.031NS

During pregnancy, pathway activity was similar in patients with RA and in healthy controls. Scores for Toll-like receptor (TLR) signaling and complement and coagulation cascades were slightly higher, and the score for lymphocyte processes was slightly decreased, in RA patients compared with healthy controls, but these differences did not reach significance (Table 2).

A very different picture was observed with samples obtained from RA patients postpartum. Monocyte-associated pathways were strongly increased. This elevation by far exceeded the level to be expected by the increase in monocyte cell numbers. This clearly indicated that in addition to the persistence of elevated monocyte fractions, there was gene activation in these pathways. Similar observations were made for lymphocyte-associated pathways when comparing patients with RA with healthy controls postpartum. This activation appeared to involve processes of adhesion and migration, TLR signaling, responses to pathogens, and Notch, phosphatidylinositol, mTOR, Wnt, and MAPK signal transduction pathways, as well as cell cycle processes.

DISCUSSION

Regarding remission of RA in late pregnancy and relapse postpartum, this pilot study suggests two most interesting characteristics. First, comparative gene expression analysis between RA patients and healthy controls in the third trimester demonstrated that clinical remission of RA during pregnancy was associated with a gene expression profile similar to that seen in healthy pregnant women. Second, postpartum, RA patients presented with a broad panel of differentially expressed genes compared with RA patients during pregnancy. The pregnancy-associated dominance of monocytic profiles persisted in RA. An additional activation of genes functionally related to processes of the innate and, interestingly, also of the adaptive immune response was observed.

With this study, we had the opportunity to investigate an exceptional group of patients. Although statistical power was limited due to the small sample size, this analysis generated several attractive hypotheses. First, comparisons between healthy controls and RA patients in the third trimester of pregnancy demonstrated only minor differences despite minimal or no immunosuppressive treatment. This suggests that molecular pathologic mechanisms of inflammation are reduced, corresponding to the clinical remission in late pregnancy. Postpartum profiles of RA patients presented a strong contrast to those of RA patients during pregnancy and of healthy controls postpartum. This was exceptional, as clinical parameters of disease activity did not reflect the same consistent change. We speculate that the molecular markers may reflect subclinical disease activity related to the postpartum flares which occurred and had to be treated prior to blood sampling 24 weeks after delivery.

Dissociation of clinical signs and underlying pathogenetic events has recently been discussed with regard to tumor necrosis factor (TNF) blocking therapies (24). The protective effects of TNF inhibition on joint destruction may be poorly correlated with its effect on the disease activity score, which is mainly defined by clinical parameters (25). Consistent with this, our group previously found persistent elevation of immunoinflammatory markers despite clinical remission in a cohort of patients with SLE (26). Thus, our data strongly suggest that molecular mechanisms involved in remission during pregnancy are very broad and extremely potent. The changes observed in RA after pregnancy may relate to a molecular pattern of postpartum susceptibility to autoimmune diseases. Postpartum onset or reactivation of inflammatory diseases is not limited to RA, but has also been demonstrated for multiple sclerosis (27). Sampling at different times before and during clinically detectable relapse may help to clarify this and shed light on maternal immunologic characteristics induced during the puerperal and early breast-feeding period.

Further hypotheses evolve from the comparative and detailed analysis using highly purified immune cells. Although expression profiles of individual cell types of the blood have been investigated previously (28), the effect of differential cellular composition on whole blood or PBMC expression profiling has not yet been studied in a systematic way. In a first approximation in the present study, a spiking experiment showed that signal intensities of marker genes are correlated with the fraction of cells, and that cellular composition can be estimated from gene expression data. In PBMCs of healthy women, we found an increase in phagocyte- and a decrease in lymphocyte-related gene activity in samples obtained in late pregnancy compared with samples obtained postpartum. Both techniques, calculation based on the 20 marker genes for each cell type as well as scores for the top 100 genes increased in healthy controls postpartum compared with healthy controls during pregnancy, yielded the same results. Notably, this is consistent with the results of earlier studies of pregnancy-related changes in monocyte and lymphocyte fractions in peripheral blood (4–6, 29). It confirms data showing that innate immune cells are more active during pregnancy (30, 31) and may thereby compensate for the suppression of the adaptive immune system, which is necessary for tolerance of the semiallogeneic fetus.

Samples from pregnant RA patients exhibited characteristics similar to those of healthy controls for cell type scores and fraction calculation. In contrast, postpartum RA samples presented a sustained elevation of monocyte-related transcripts compared with samples from healthy controls, suggesting an important role of innate immune processes in relapse. However, lymphocyte-related genes were increased when comparing RA patients postpartum with RA patients during pregnancy, pointing to the recurrence of adaptive immune functions as cofactors in the relapse. Additional mechanisms can be proposed from the results of scoring the cell type association in the top 100 genes differentially expressed. Comparison between samples from healthy women postpartum and during late pregnancy revealed genes with a high score of cell type association as indicated by the constant increase to a high cumulative level. This was similar to the comparison of profiles with 100% and 50% T cells, where changes in cellular composition were the only differences introduced by spiking. In contrast, in samples from RA patients postpartum versus RA patients during late pregnancy, the genes differentially expressed had minor scores for cell type association, suggesting that these genes contribute to a different quality of functional processes which become activated in RA postpartum.

Screening of genes increased in RA postpartum for association with known pathways identified 6 major groups of KEGG pathway annotations. In contrast to healthy women, postpartum changes in gene activity in RA patients by far exceeded the pattern that would have been expected if only differences in cellular composition had occurred. Increased activity was predominantly associated with cell adhesion and migration, innate immune response, and infection as well as several signal transduction pathways related to cell activation, cell growth, and differentiation processes. There was additional gene activity in pathways related to phagocytes according to the reference profiles, in particular the processes of cholera and Helicobacter pylori defense, TLR signaling, adhesion, and leukocyte migration. These were also found to be increased when comparing PBMCs of RA patients postpartum with PBMCs of RA patients or controls during pregnancy, which all contained similar fractions of monocytes. Thus, comparison of RA postpartum data with this network of reference profiles suggests an up-regulation of transcripts in phagocyte-associated pathways and, consequently, that phagocytes are especially involved.

The increased expression of genes related to these pathways may indicate a state of alertness by priming for adhesion and infiltration into inflamed tissues and subsequently a contribution to the clinical flare of the disease. This notion is also supported by the particularly high expression of CCR1, CXCR4, and platelet endothelial cell adhesion molecule 1, genes which are associated with monocytes according to our reference profiles and which are related to chemoattraction and transendothelial migration (32). Recent investigations of PBMCs in RA patients have demonstrated similar results (33), with CXCR4 strongly up-regulated in patients with high disease activity as compared with patients with quiescent disease. However, that and other studies (18, 34, 35) did not test for differential cellular composition, which may interfere with differential expression of genes such as CXCR4.

In summary, using a new bioinformatic approach, the results of this pilot study support the hypotheses that molecular processes of inflammation in RA are suppressed during pregnancy but become activated postpartum. We hypothesize that monocytes are important effectors of relapse, and that interaction with recurring lymphocyte activity may be critical. Extended functional studies and detailed analyses at different time points after pregnancy are needed to confirm these hypotheses and to further improve our knowledge about this highly interesting period of immunologic readjustment.

AUTHOR CONTRIBUTIONS

Dr. Häupl had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study design. Häupl, Østensen, Villiger.

Acquisition of data. Häupl, Østensen, Grützkau, Radbruch, Villiger.

Analysis and interpretation of data. Häupl, Østensen, Villiger.

Manuscript preparation. Häupl, Østensen, Burmester, Villiger.

Statistical analysis. Häupl.

Acknowledgements

The authors thank Beate Möwes and Heidi Schliemann for excellent technical assistance.

Ancillary