Blood was collected using Vacutainer SST for serum, and CPT tubes (BD Biosciences, Erembodegem, Belgium) containing sodium heparin for cell-based analysis. Serum and peripheral blood mononuclear cells (PBMCs) were isolated according to manufacturer's instructions. PBMCs were taken up in RPMI 1640 (Lonza, Verviers, Belgium) containing 20% fetal calf serum (Thermo Fisher Scientific, Landsmeer, The Netherlands) and 10% dimethyl sulfoxide (Sigma-Aldrich, St Louis, MO, USA) and stored in liquid nitrogen until further use. Serum samples were stored at −80°C.

The fluorescently labeled anti-human monoclonal antibodies used for flow cytometry are shown in Supplementary Table 1. In all experiments, viable cells were analyzed using Fixable Viability Stain 700 (BD Biosciences) or Fixable Viability Dye eFluor 520 (Thermo Fisher Scientific) by performing staining for 15 min at 4°C in the dark. Surface markers were stained for 30 min. Cells were measured using the LSRII-Fortessa (BD Biosciences) and analyzed using both Cytosplore and BD FACSDiva (version 8.0.1) software. For Hierarchical Stochastic Neighbor Embedding (HSNE) analysis in Cytosplore (13), events were downscaled to a maximum of 80,000 per sample. PBMCs of non-relapsing MS patients and healthy controls were measured in the same experiments. PBMCs of relapsing MS patients were measured a year later and thus excluded from HSNE analysis due to the effect of fluorescent intensity shifts on this type of analysis. We excluded Tregs (CD25^highCD127^low) and defined naive, central memory and effector memory populations based on CCR7 and CD45RA expression. Th1, Th2, Th17, Th17.1 and Th17 “double-positive” (DP) subsets were discriminated based on differential expression of CCR6, CCR4, and CXCR3 (14, 15).

CD4⁺ cells were isolated from the blood of healthy non-pregnant females (Sanquin, Amsterdam, The Netherlands) using CD4 microbeads and the autoMACS Pro Separator (both Miltenyi Biotec, Bergisch Gladbach, Germany), frozen as described above and stored in liquid nitrogen until further use. Thawed cells were labeled with CFSE (1:20,000; CFDA-SE, Molecular Probes via Thermo Fisher Scientific) for 10 min at 37°C. After washing twice with RPMI 1640 containing 100 U/ml penicillin and 100 μg/ml Streptomycin (now termed as “Pen/Strep”; Lonza) and 5% FCS (Thermo Fisher Scientific), cells were plated at a concentration of 1 × 10E⁶/ml in RPMI1640 containing Pen/Strep. Anti-CD3/-CD28 dynabeads (Thermo Fisher Scientific) were added (1:5) together with paired third trimester or postpartum serum samples from MS patients or healthy controls till a concentration of 5%. After 72 h of culture, dynabeads were removed and cells were assessed by flow cytometry.

Memory Th cells (CD3⁺CD4⁺CD8⁻CD25^−/dimCD45RA⁻) were isolated using a FACSAria-III machine (BD Biosciences) and plated at a concentration of 0.5 × 10⁶/ml in RPMI 1640 containing 5% inactivated human AB serum (Sanquin) and Pen/Strep. Cells were rested overnight (37°C) and were stimulated with phorbol 12-myristate 13-acetate (PMA; 1:2,000) and ionomycin (1:500; both Sigma-Aldrich) for 5 h. Subsequently, supernatants were harvested and stored at −80°C until further use. Supernatants were diluted 2-fold and analyzed for GM-CSF, IFN-γ, IL-2, IL-6, IL-17A and TNF-α using a custom Luminex multiplex bead immunoassay (R&D Systems, Abingdon, UK). Measurements were performed on a Bio-Plex MAGPIX machine and data were analyzed using Bio-Plex Manager MP software (both Bio-Rad, Hercules, California, USA).

Both cortisol and progesterone levels were measured by ultraperformance liquid chromatography-tandem mass spectrometry (UPLC-MS/MS; Waters TQS, Waters, Etten-Leur, The Netherlands). Both steroids were calibrated using commercially available calibrators (Cerilliant Corporation, Round Rock, TX, USA); article number P-069 (progesterone) and C-106 (cortisol). Trueness was verified for progesterone using reference materials ERM-DA347 and ERM-DA348. For cortisol, trueness was verified using the ERM-DA451 IFCC cortisol reference serum panel (34 levels). Both analytes were measured without bias when compared to the reference materials. Measurement uncertainty was assessed for both analytes by measurement of tri-level matrix matched controls (UTAK, Valencia, CA, USA). For progesterone, the coefficients of variation (CVs) over a period of 1 year were 5.2, 6.6, and 10.4% at 1.4, 17.9, and 65.3 nM, respectively. For cortisol, CVs were 3.6, 3.7, and 4.0% at 40.0, 98.6, and 573.0 nM, respectively. Estradiol levels were measured by automated immunoassay on (Fujirebio Lumipulse G1200, Fujirebio, Tokyo, Japan). CVs over 19 runs were 4.4 and 3.9% at 345 and 1,099 pM, respectively.

Generalized linear mixed models (GLMM) with a gamma distribution were used to assess longitudinal effects and the false discovery rate (FDR) method of Benjamini and Hochberg (BH) was used to correct for multiple testing (R statistical software package version 4.0). Data between clinical groups were compared using Kruskal-Wallis tests incorporating Dunn's multiple comparisons or 2-way analysis of variance (ANOVA) with Bonferroni's multiple comparison tests (GraphPad Prism software version 8, San Diego, CA, USA). Longitudinal results were displayed as connected data points for each individual. For the comparison of data between clinical groups, Box and Whisker plots were used containing individual data points and displaying the median with the interquartile range (IQR). For all tests, a p-value of < 0.05 (^*) after multiple testing correction was considered significant. Categorical data were analyzed using Chi-squared, Kruskal-Wallis or Wilcoxon rank-sum tests.

In this exploratory study, pregnant relapsing-remitting MS patients visiting MS center ErasMS were included and retrospectively validated to match the most recent McDonald 2017 criteria (16, 17). Exclusion criteria were history of recurrent abortion, hypertension and diabetes mellitus. Patients did not receive disease modifying treatment at least 3 months prior to pregnancy and during this study. Third trimester expanded disability status scale (EDSS) scores were similar between patients that did or did not experience a clinically-defined postpartum relapse, as described previously (17) (Table 1). Patients did not experience relapses during pregnancy. Postpartum EDSS scores did differ significantly between non-relapsing and relapsing MS patients (Table 1). Unfortunately, radiological data were not available. Healthy pregnant woman were recruited from the outpatient obstetric clinic at the Erasmus MC. Patient groups and healthy controls were matched for age and selected based on the availability of frozen PBMCs at 28–30 weeks of pregnancy (third trimester) and 4–8 weeks after delivery (postpartum). Other clinical characteristics are summarized in Table 1. The only difference between the healthy controls and the MS groups were the amount of females giving birth for their first time (nullipara).

To assess shifts in phenotypic effector profiles, we first analyzed ex vivo flow cytometry data of CD4⁺ Th cells from paired third trimester and early postpartum blood of healthy controls (n = 12; HC) and MS patients without a postpartum relapse (n = 13; MS-NR) in an unbiased manner using HSNE (Figure 1A and Supplementary Figure 1). For both the HC and MS-NR group, cluster P2 (CCR7^lowCD45RA^low) was decreased, while cluster P1 was increased (CCR7^highCD45RA^high) in third trimester vs. postpartum blood (Figures 1A,B). Cluster P3 (CCR7^dimCD45RA^low) was proportionally decreased during pregnancy relative to the presence of other clusters. For the MS-NR group, this was confirmed by manual gating, showing significantly reduced effector memory (CCR7^lowCD45RA^low) and increased naive (Figures 1C,D; p = 0.002 and p = 0.012) Th cell frequencies. For the HC group, these frequencies did not differ (Figure 1D) and were unrelated to nullipara count (Table 1 and data not shown). In 6 MS patients who did experience a postpartum relapse (MS-R), no significant differences were found in memory Th cell phenotype between third trimester and postpartum samples (Figure 1D). To determine whether pregnancy-relevant factors present in the blood (e.g., hormones) contributed to these phenotypes, we added corresponding serum samples to activated healthy female CD4⁺ Th cells in vitro (Figure 1E). Th cells exposed to paired MS-NR and HC sera showed similar changes in phenotype skewing as found ex vivo (Figure 1F; p = 0.026 and p < 0.001). In contrast to our ex vivo results (Figure 1D), reduced frequencies of effector memory Th cells were also seen after exposure to third trimester compared to postpartum MS-R sera (Figure 1F; p < 0.001). Interestingly, these frequencies appeared higher than in the MS-NR and HC group. Overall, these data suggest that effector memory Th cells are kept in check during pregnancy by serum-related factors, which is lost and may contribute to an MS relapse early after delivery.

According to the HSNE analysis, effector memory (CCR7^dim/lowCD45RA^low) Th clusters P2 and P3 (Figure 1B) mainly expressed CXCR3, a hallmark of IFN-γ-producing subsets (18). Cluster P3 also showed upregulation of Th17 lineage marker CCR6 (14). Manual gating revealed that CXCR3 and not CCR6 expression was lower in third trimester compared to postpartum samples (Figures 2A–C). This was significant for both HC (p = 0.002) and MS-NR groups (p < 0.001), but not for the MS-R group. In line with our HSNE analysis (Figure 1B), these differences were seen for all CXCR3-expressing Th subsets including Th1 (CCR6⁻CXCR3⁺CCR4⁻), Th17 DP (CCR6⁺CXCR3⁺CCR4⁺) and Th17.1 (CCR6⁺CXCR3⁺CCR4^−/dim) cells (Supplementary Figure 2). Interestingly, only MS-associated Th17.1 cells (19) were decreased in MS-R postpartum blood and showed a prominent effector memory phenotype in comparison to other Th subsets (see Supplementary Figures 2, 3). In sharp contrast to the HC group, CXCR3 was downregulated on in vitro proliferating memory Th cells exposed to MS-NR and MS-R third trimester vs. postpartum sera (Figure 2D; p < 0.001). This was comparable to the in vitro effects on effector memory phenotypes (Figure 1F). Although no differences were seen on ex vivo Th memory cells from patients and controls, CCR6 expression levels were reduced in vitro after the addition of third trimester compared to postpartum MS-NR sera (Figure 2D; p < 0.001). As expected, mass spectrometry revealed that cortisol levels were strongly elevated in third trimester sera from each clinical group (Figure 2E; p < 0.001). Progesterone and estradiol could not be detected in postpartum samples. Third trimester cortisol, progesterone and estradiol levels did not differ between clinical groups (Figure 2F). These data indicate that CXCR3⁺ memory Th cells are selectively controlled by serum-related factors in pregnant MS patients, which cannot be explained by changes in cortisol, progesterone or estradiol serum levels between clinical groups.

Finally, we explored whether the activation and pro-inflammatory capacity of memory Th cells is differentially regulated in our clinical groups. For this purpose, co-expression of CD38 and HLA-DR was investigated on third trimester and postpartum memory Th cells (Figure 3A). Only in the MS-R group, the proportions of CD38^highHLA-DR^high memory Th cells were increased in postpartum compared to third trimester samples (Figure 3B; p < 0.001). These proportions were higher than those in the HC group (Figure 3C; p = 0.008). To address how this is related to their ability to produce pro-inflammatory cytokines, memory Th cells were purified from paired third trimester and postpartum samples and stimulated with PMA and ionomycin. Although not significant for each cytokine, higher levels of interleukin-6 (IL-6), tumor necrosis factor α (TNF-α), IL-2, IL-17A, interferon gamma (IFN-γ) and granulocyte-macrophage colony-stimulating factor (GM-CSF) were found in supernatants from postpartum memory Th cells of the MS-R group (Figure 3D). Notably, Th1-associated cytokine IFN-γ and not Th17-associated cytokine IL-17A pre-dominated these samples. Importantly, the same was true for cells derived from third trimester blood in this clinical subgroup (Figure 3D). No significant differences in cytokine production were seen between paired third trimester and postpartum samples from each clinical subgroup (see Supplementary Figure 4). These findings reveal that pro-inflammatory memory Th cells of MS patients with a future postpartum relapse are more trigger-happy already during the third trimester of pregnancy.