Pregnancy-induced gene expression changes in vivo among women with rheumatoid arthritis: a pilot study

Women with RA and healthy women of Danish descent were recruited and enrolled in a pregnancy cohort in Denmark, as previously described [13]. A subset of 20 women with RA and 5 healthy women from this cohort were included in the pilot dataset. Subjects with RA fulfilled the 1987 revised American College of Rheumatology criteria for RA [14]. The study was approved by the ethics committee for Region Hovedstaden (Denmark), the Danish Data Protection Agency, and the Children’s Hospital Oakland Research Institute Institutional Review Board. All subjects provided written informed consent prior to enrollment.

Data collected before pregnancy (T0) and at the third trimester (T3) were included in the present study. Therefore, women who had missing data at the T0 baseline or at T3 (eight women with RA) were excluded from further analyses. At both time points, the women with RA were examined by trained study nurses. Disease activity measures including tender and swollen joint counts based on a total of 28 joints (TJC28 and SJC28, respectively) and patient global health (GH) scores were recorded, and a serum sample was collected for measurement of C-reactive protein (CRP) levels. Data on medication use during the 3 months prior to both visits were also recorded. At each visit, blood samples from the women with RA and healthy women were collected in PAXgene RNA tubes (PreAnalytiX, Hombrechtikon, Switzerland).

RA disease activity was assessed using the Disease Activity Score based on 28 joints and 4 variables (DAS28-CRP4, abbreviated henceforth as DAS28), as follows [15]:

Change in DAS28 from T0 to T3 was calculated as: ΔDAS28 = DAS28_T3 − DAS28_T0.

The women with RA were categorized into two subsets: those with negative ΔDAS28 values were considered to show an improvement in DAS28 score at T3 compared to the T0 baseline (referred to as the pregDAS_improved subset), whereas those with positive ΔDAS28 values were included in the “worsened” subset, referred to as pregDAS_worse. Any improvement in DAS28 scores (ΔDAS28 < 0) was used as a surrogate for improvement in disease activity. This was done to capture even small changes in disease activity that may correlate with biological changes observed at the gene expression level. Disease activity categories (remission, low, moderate, or high) were assessed using previously defined criteria [16].

Total RNA was isolated from frozen blood samples, and barcoded cDNA libraries were prepared as previously described [13]. Pseudo-alignment of the de-multiplexed raw sequence reads (FASTQ format) to the Ensembl reference human GRCh38 transcriptome assembly, and quantification of transcript abundances was performed using kallisto (version 0.42.4) [17]. BioMart annotations were used to combine transcript-level counts into gene-level estimates. Pseudogenes, genes with no annotations, and genes with very low read counts (<1 count per million) in at least 25% of all samples were filtered out. Any globin genes and ribosomal RNA (rRNA) transcripts still present were also filtered out. To adjust for variable sequencing depths across samples, the gene-level counts were normalized using the Trimmed Mean of M values (TMM) algorithm as implemented in the edgeR package (version 3.10.5) [18, 19]. To assess batch effects, normalized gene counts from each pair of technical replicates were plotted, and correlations were assessed.

Differential gene expression analysis was performed using edgeR (version 3.10.5) [18] to compare normalized gene-level counts between the T3 and T0 time points within each group of women (i.e., pregDAS_improved, pregDAS_worse and healthy women). A model with a paired-sample design was used for these within-group comparisons. Between-group comparisons (pregDAS_improved vs pregDAS_worse, and pregDAS_improved vs healthy) at each time point were also performed using differential expression analyses. A negative binomial distribution was used to handle the over-dispersion in RNA-seq gene counts. Differential expression was tested using generalized linear model (GLM) likelihood ratio tests. To correct for any batch effects, batch was included as a covariate in the model. A q value threshold of 0.05 was used to assess significance. Because sample sizes were small, FCs in expression were also used in the interpretation of results, focusing on genes with at least a two-fold change in expression from T0 to T3.

Differentially expressed genes were analyzed for over-representation of Gene Ontology (GO) categories using a hypergeometric test implemented in the Web-based Gene Set Analysis Toolkit (WebGestalt) with a threshold of at least five genes per category [20]. A significance threshold of q<0.05 was used to define enrichment. Functional enrichment of gene sets was examined using the STRING database of known and predicted interactions among proteins [21, 22].

Of the 12 women with RA who had data at both T0 and T3, 8 were in the pregDAS_improved group and 3 were in the pregDAS_worse group. One woman was excluded because, although she had an increase in DAS28 at T3, she was in remission at both time points and hence did not fit into the pregDAS_worse group. The DAS28 scores of the 11 women included in the analyses are shown in Fig. 1. The average disease duration among the women with RA was (mean ± SD) 5.9±4.4 years for pregDAS_improved and 8.7±1.0 years for pregDAS_worse. The average age at conception was 30.3±5.7 years for pregDAS_improved, 33.2±1.9 years for pregDAS_worse, and 31.2±5.7 years for the healthy women.

Among the eight pregDAS_improved women, three did not take any medications at T0 and only one of these three women started taking medications by T3 (prednisolone + sulfasalazine). The remaining five women were taking prednisolone and/or sulfasalazine at both time points; one of them was also on anti-tumor necrosis factor (anti-TNF) therapy at T0, and another was taking methotrexate at T0. The three pregDAS_worse women were all on anti-TNF therapy and taking prednisolone and/or sulfasalazine (except for one) at T0; at T3, one was taking prednisolone + sulfasalazine, one remained on anti-TNF therapy, and one stopped taking medications.

A total of 1296 genes showed significant differential expression between T3 and T0 in the pregDAS_improved subset, of which 161 displayed two-fold or more change in expression (see Additional file 1). These 161 genes were enriched in a number of immune-related pathways, as shown in Table 1. Similar results were obtained when the woman who was receiving anti-TNF therapy at T0 was excluded from the analysis.

A large proportion of the 161 genes that exhibited at least two-fold change in expression between T3 and T0 among the pregDAS_improved women were also differentially expressed among healthy women (q<0.05), mostly (n=108) with two-fold or more change in expression (see Additional files 1 and 2), although some (n=33) showed more modest FCs in expression (1.5≤FC<2.0). In addition, another 77 genes differentially expressed (q<0.05) in both the pregDAS_improved and healthy groups showed lower FCs among the pregDAS_improved women (1.5≤FC<2.0) than among the healthy women (FC≥2). The genes with FC≥2 in both the women with RA and healthy women were enriched in immune-related pathways such as immune system process (q =7.9×10⁻⁹), response to other organism (q =2.7×10⁻⁷), and defense response to bacterium (q =9.3×10⁻⁶).

Of the 161 genes differentially expressed in the pregDAS_improved subset, some (n=31) were also differentially expressed (q<0.05, FC≥2) among pregDAS_worse women, with most of these genes overlapping with those differentially expressed among healthy women (n=30) (see Additional file 2). The FCs in expression from T0 to T3 were correlated between the two RA subsets (Pearson correlation coefficient r=0.86, p<0.00005).

The remaining 130 genes that were differentially expressed (T3 vs T0: FC≥2, q<0.05) in our data among the pregDAS_improved women, but not among the pregDAS_worse women, were significantly enriched in several immune-related GO processes, such as immune system process (q=4.7×10⁻¹⁰), defense response (q=9.6×10⁻⁶), immune response (q=5.0×10⁻⁵), and innate immune response (q=8.0×10⁻⁶), among others (Table 2). Of these 130 genes, 78 were also differentially expressed among healthy women (see Additional files 1 and 2). A comparison of the FCs in expression from T0 to T3 within each of the RA subsets revealed a small cluster of genes that were significantly over-expressed at T3 among pregDAS_improved women and under-expressed (FC≥2, not significant) among pregDAS_worse women (Fig. 2). By T3, mean normalized expression levels of these genes were higher among the pregDAS_improved women than among the pregDAS_worse women. Interestingly, all of the genes in this cluster were type I interferon (IFN)-inducible (IFI44, IFI44L, IFIT1, HERC5, CMPK2, RSAD2, and SIGLEC1). Among these, only the RSAD2 gene was differentially expressed (under-expressed) at T3 compared to T0 among healthy women (q<0.05, FC≥2). A number of other type I IFN-inducible genes (STAT1, IFI16, IFIT2, IFIT5, IFIH1, and MX1) also exhibited similar patterns of expression, being over-expressed at T3 (q<0.05) among pregDAS_improved women, though FCs were more modest (1.5- to 1.9-fold), and under-expressed among pregDAS_worse women. The PLSCR1 gene, which can enhance the type I IFN response, also showed a significant increase in expression (FC≥2) in the pregDAS_improved women by T3, but no change was observed in the pregDAS_worse women. Of interest, a comparison of normalized expression levels of these genes at T0 between pregDAS_improved and healthy women revealed that some of the IFN signature genes (IFI44, IFI44L, HERC5, CMPK2, SIGLEC1, and MX1) were significantly under-expressed in the pregDAS_improved subset compared to healthy women at T0 (q<0.05, FC≥2). However, by T3, there were no significant differences in expression of these genes between the two groups of women. The IFN-inducible genes that were over-expressed at T3 among pregDAS_improved women, but not among in the pregDAS_worse women, belong to a common functional network, as shown by known and predicted protein interactions from the STRING database (Fig. 3). Other than RSAD2, none of the genes shown in this network were differentially expressed (T3 vs T0) among healthy women.