Rheumatoid Arthritis is an autoimmune disease predominantly affecting post-menopausal women, but that can also affect women of childbearing age ^[1]. As a consequence, clinicians are faced with difficult choices regarding the selection of an optimal therapeutic regime that deals with the symptoms of the disease without negatively affecting the pregnancy, as some of the therapeutic regimes for RA are unsafe for use during pregnancy ^[2]. Further, women with RA have an increased risk of adverse pregnancy outcomes ^[3–5]. A better understanding of the mechanism driving the pregnancy-associated changes in RA will provide us with valuable information to help resolve this problem. In addition, it will provide useful clues regarding the pathogenesis of RA in general, thus opening the way to the development of novel treatments ^[6,7].

Frequently, disease activity in patients with RA decreases spontaneously during pregnancy ranging from an improvement in clinical symptoms to complete remission. However, this effect is transient and the disease relapses shortly after delivery ^[8]. This pregnancy-induced amelioration of RA symptoms was first described by Hench in 1938 ^[9] and was a crucial hint towards the identification of corticosteroids as immunosuppressive drugs for use in treating autoimmunity (Table 1) ^[6]. Since then, a large number of retrospective and prospective studies on RA patients have confirmed that an improvement in disease activity occurs during pregnancy in half to three quarters of patients ^[8,10–12]. It is noteworthy that the higher efficacy of more recently developed therapeutic regimes is thought to lead to lower levels of RA activity in patients prior to pregnancy, thus partially “masking” the beneficial effect of pregnancy in more recent reports ^[7]. Despite decades of study, a mechanistic explanation for the pregnancy-induced remission and post-partum relapse of RA remains elusive ^[12].

Here, we examine the role of CD25⁺ regulatory T (T_R) cells in this beneficial effect. T_R cells are a naturally occurring subpopulation of T cells that are essential for many aspects of immune tolerance, including the suppression of autoimmune responses ^[13]. During pregnancy they protect the fetus from rejection by the maternal immune system in both mice ^[14] and humans ^[15–17] (Table 2). The accumulation of antigen-experienced T_R cells in the uterus ^[18] suggests that the suppression of the anti-fetal immune response occurs in a localized and antigen-specific fashion. Further support for an antigen-specific action of T_R cells comes from studies examining the immune response to the minor transplantation antigen H-Y in the context of maternal–fetal tolerance ^[19]. A hint regarding an involvement of T_R cells in the amelioration of RA comes from the observation that their number inversely correlates with disease activity during pregnancy ^[20] (Table 1). However, experimental proof for a mechanistic involvement of T_R cells remained outstanding.

To examine whether T_R cells mediate the pregnancy-associated remission of arthritis, we studied the phenomenon in Collagen-Induced Arthritis (CIA), a mouse model of the disease. We found that pregnancy protects the mice from developing arthritis. Transfer of CD25⁺ cells from these pregnant-protected mice into non-pregnant recipients protected them from CIA. The fact that transfer of CD25⁺ cells from pregnant mice that were not exposed to CIA induction did not confer protection to the recipients suggests that the T_R cells act in an antigen-specific fashion.

CIA in mice resembles the pathology of RA both in terms of histopathology and serological biomarkers ^[22,23]. To induce arthritis in C57BL/6 mice we injected them with chicken Collagen Type II in Complete Freund’s Adjuvant intra-dermally on day 0 and day 21. Some of the mice were mated allogeneically with BALB/c males on days 31–35 (Fig. 1). We compared the course of CIA in non-pregnant (n = 111) and pregnant (n = 44) mice and found that pregnancy protected the mice from the disease (incidence of 32% vs. 11%; Table 3). This is reflected in both the average clinical score over time (P = 0.0002, two-tailed Wilcoxon signed rank test; Fig. 2A) and the maximum clinical score reached (Fig. 2B).

To verify that this is not due to a pregnancy-induced systemic immunosuppression, we compared the response to intra-nasal influenza HKx/31(H3N2) infection in pregnant (n = 5) and non-pregnant mice (n = 9). We found that pregnancy had no effect on the expansion of CD8⁺ cells specific for the H–2D^b/nucleoprotein (NP) peptide complex (non-pregnant vs. pregnant; 7.44 ± 0.65 vs. 7.48 ± 0.51; P = 1, two-tailed unpaired t-test; Fig. 2C and Table 4). This demonstrates that pregnant mice are capable of launching normal immune responses against this pathogen. Thus, the protection from arthritis cannot be due to a pregnancy-induced systemic immune suppression.

To investigate whether the protection from CIA during pregnancy can be attributed to the action of T_R cells, we ‘substituted’ pregnancy with adoptive transfer of CD25⁺ cells (Fig. 3). Non-pregnant mice, in which CIA had been induced, received CD25⁺ cells sourced from either non-pregnant control mice (non-pregnant; n = 21), untreated pregnant mice (pregnant; n = 19), or mice that were protected from the disease by pregnancy despite CIA induction (pregnant-protected; n = 5). Each recipient mouse received all CD25⁺ cells obtained from a donor mouse in a one-to-one fashion. Whilst none of the mice receiving CD25⁺ cells from pregnant-protected donors developed arthritis, 24% (5 out of 21) of recipients of cells from non-pregnant donors and 32% (6 out of 19) of the recipients of cells from pregnant untreated donors developed arthritis (Fig. 4A; 1:1 transfer).

The number of CD4⁺CD25⁺ cells significantly increases during pregnancy from 0.35 to 0.5 × 10⁶ cells in non-pregnant mice to approx. 1.5 × 10⁶ cells in pregnant mice ^[14]. Therefore, we titrated the number of cells transferred to match the numbers that can be obtained from non-pregnant donors. Transfer of 0.35 × 10⁶ CD25⁺ cells from control mice had no effect on the outcome of CIA in the recipients (no transfer vs. non-pregnant; Table 5 and Fig. 4B). Whilst transfer of the same number of CD25⁺ cells from pregnant mice appeared to cause a slight delay in the onset of clinical signs (pregnant, Fig. 4B), the outcome per se was not affected (no transfer vs pregnant; Table 5). In contrast, none of the recipients of CD25⁺ cells from pregnant-protected mice developed any signs of arthritis (pregnant-protected, Fig. 4B and Table 5).

In summary, we observed a significant protection from CIA (pregnant-protected, P < 0.05, two-tailed Fischer’s exact test; Fig. 4A) irrespective of the number of cells transferred and conclude that T_R cells mediate the pregnancy-associated protection from CIA. The fact that T_R cells from pregnant mice that did not undergo CIA induction did not protect from arthritis (non-pregnant vs. pregnant; Table 6 and Fig. 4A) shows that the pregnancy by itself is insufficient to protect from arthritis. Rather, our data suggest that this protective effect requires prior exposure of the T_R cells to arthritis-related antigens in the context of pregnancy (non-pregnant vs. pregnant-protected; Table 6 and Fig. 4A).

Since the first description of the pregnancy-induced amelioration of RA symptoms, numerous studies have attempted to elucidate the underlying mechanism (Table 1). Pioneering work by Whyte and co-workers used a model of CIA in DBA mice to examine both the amelioration of arthritis during pregnancy and the post-partum relapse of the disease ^[24]. Their results suggested that prolactin ^[25] and oestradiol ^[26] have opposite effects on the post-partum course of the disease. Yet, due to the lack of precise temporal correlation with disease activity, doubts were expressed on the role of hormones in this process ^[10]. A better temporal correlation with disease activity was observed for the percentage of IgG-associated agalactosyl N-linked oligosaccharides, which decreases during the amelioration of arthritis ^[27]. However, this could not be explained by a pregnancy-induced clearance of the agalactosyl IgG by mannose-binding lectin ^[28].

A further line of investigation centered on the observation that allogeneically mated B10.RIII females were more protected from CIA than syngeneically mated females ^[29]. This has been attributed to both changes in the ratio of T cell populations ^[30] and changes in cytokine levels ^[31]. In humans, the extent of disparity in HLA-DP and HLA-DQ MHC Class II molecules between the mother and the fetus was found to correlate with remission from arthritis during pregnancy ^[32–34], though a later study on inflammatory polyarthritis did not find such a correlation ^[35].

Several lines of evidence have suggested that T_R cells have a role in the regulation of arthritis ^[36]. T_R cells in RA patients show functional defects ^[37] and depletion of T_R cells in mice leads to increased disease severity ^[38]. Here, we demonstrate that T_R cells from pregnant-protected mice are sufficient to confer protection from CIA when transferred into non-pregnant mice. This strongly suggests that T_R cells are responsible for the pregnancy-induced amelioration of RA.

Prior to pregnancy, exposure of the mother to paternal transplantation antigens induces a rigorous immune response against the graft ^[39]. In the context of pregnancy, this response is suppressed to prevent a rejection of the fetus ^[14]. It appears that some autoimmune responses such as rheumatoid arthritis and multiple sclerosis ^[40] are also re-assessed during pregnancy, resulting in a temporary amelioration of these diseases.

Autoimmune responses could potentially be suppressed in an antigen-specific fashion or by bystander effects. The accumulation of antigen-experienced T_R cells in the gravid uterus ^[18] suggests that the suppression of the anti-fetal immune response occurs in a localized and antigen-specific fashion (see also Table 2). Similarly, in the case of autoimmune diabetes, the data points to a highly antigen-specific involvement of T_R cells ^[41]. Further support for an antigen-specific action of T_R cell comes from our endeavours to find a cell-mediated therapy for arthritis. Genetically engineered inducible Foxp3 (iFoxp3) can be used to confer T_R cell phenotype to T_H cells ^[42]. This can be used to stop CIA using iFoxp3-transduced, polyclonal T cell autografts. We found that this approach only worked if the iFoxp3-transduced T_H cells were exposed to arthritis antigens prior to switching on Foxp3 ^[42]. If iFoxp3 was switched on prior to exposure to arthritis antigens, the course of the disease was not affected. All these findings point towards an antigen-specific suppression by T_R cells. The data presented here provide evidence that the amelioration of arthritis during pregnancy is also antigen-specific. Only T_R cells isolated from ‘pregnant-protected’ mice conferred arthritis protection to non-pregnant mice. T_R cells from pregnant mice that had not been exposed to arthritis-related antigens could not confer protection.

Some mechanistic insight comes from the observation that pregnancy is accompanied by a shift from T_H1 to T_H2 type responses. It has been suggested that this in itself might lead to a diminution of the underlying immune response driving RA ^[43,44]. Pregnant women with RA display a reduction in the capacity of their peripheral blood mononuclear cells to produce the T_H1 cytokines IL-12 and IFNγ ^[45]. This hypothesis could explain why some autoimmune diseases such as SLE can exhibit flares during pregnancy, presumably due to a T_H2 bias of the underlying immune response ^[46]. It remains to be seen whether the T_H1/T_H2 shift during pregnancy acts in parallel to the action of T_R cells or whether the change in bias is actually mediated by the T_R cells. It is noteworthy that in contrast to the essential requirement for T_R cells, a change in the T_H1/T_H2 bias is not fundamental to maternal–fetal tolerance, as mice deficient in T_H2 cytokines can become allogeneically pregnant ^[47].

We propose that the amelioration of arthritis is a collateral consequence of the immune system’s reassessment of all responses coinciding with pregnancy. By making context-dependent decisions, the immune system can suppress immune responses directed against the fetus whilst remaining vigilant towards pathogens, such as influenza, that are recognized to be a danger to the mother. The finding that pathogen-associated molecular patterns (PAMPs) under certain, specific conditions can block T_R-mediated suppression ^[48] offers a hint to as to how the immune system might interpret the context. The absence of exogenous ‘danger’ signals in ongoing autoimmune responses might be sufficient for the immune system during pregnancy to reassess and suppress them. One might speculate that the transient nature of the pregnancy-associated suppression is of evolutionary advantage, as a more permanent induction of tolerance would be prone to be exploited by pathogens. Indeed, certain pathogens, such as Listeria and Salmonella, appear to be able to take advantage of the pregnancy-induced tolerance mechanisms, as these infections are exacerbated by pregnancy ^[49]. The exact mechanism by which immune responses coinciding with pregnancy are re-interpreted by the immune system warrants further investigation.

The authors declare that they have no competing financial interests.

We thank DT Fearon and his group for assistance with the influenza infection assay and kind gift of the relevant reagents. This work was funded by the Medical Research Council and in part supported by Arthritis Research UK (project grant 18297).