In this multi-center intervention study, 49 women with uRPL were included ( Supplementary Figure 1 ). uRPL was defined as having two or more pregnancy losses before 24 weeks of gestation without an identifiable cause. Women were referred to the department of Obstetrics and Gynecology at either the Radboud University Medical Centre (RUMC) or the Maastricht University Medical Centre+ (MUMC+) and were included when no cause for their miscarriages could be identified according to the guidelines of the European Society of Human Reproduction and Embryology (ESHRE). The clinical work-up included a standardized medical history of the couple, screening for uterine abnormalities, thyroid abnormalities, anti-phospholipid syndrome, paternal screening for sperm DNA fragmentation and parental karyotyping (if indicated) (41). Participants were excluded from the study if they met one or more of the exclusion criteria presented in Supplementary Table 1 .

This study was conducted in accordance with the Declaration of Helsinki, the Medical Research Involving Human Subjects Act, and the guidelines for Good Clinical Practice (GCP). After obtaining informed consent, according to the Medical Ethical Committee of the Radboud University Medical Centre and Maastricht University Medical Centre+ NL77307.091.21, women received a questionnaire for assessing baseline characteristics. PB and MB were collected for immunophenotyping.

The intervention consisted of a 12-week, personalized, moderate-intensity, aerobe exercise program for a stationary bike. The exercise program was personalized to individual heart rate reserves (HRR). The first six weeks the participants were instructed to train two times a week for 50 minutes at 50-60% of their HRR, the subsequent six weeks for three times a week for one 50 minutes at 50-60% HRR. Their heart rate was monitored during exercise using a Polar H10 heartrate monitor chest strap (Polar Electro, Inc, Kempele Finland). Women needed to attend a minimum of 80% of 30 prescribed trainings after completion of the intervention in order to be eligible for data analysis. All women were asked not to change their pre-existing training schedules (if applicable) and not to change their diet during the intervention. Before and after the 12-week exercise intervention, identical immune analyses were conducted.

Of the 49 women who were included in the study, 24 women completed the intervention (data is shown only of these women) ( Table 1 ). Both PB and MB samples were collected. However, due to collection issues, PB was available from 23 participants, and MB was available from 22 participants. Different figures might show variable numbers of women included in analyses if unreliable individual markers were excluded due to technical staining issues. Sufficient remaining cells (n=12) were frozen for later functional analyses.

To study uterine lymphocytes, MB was collected using a menstrual cup (40) at 3 time points of 12 hours right after the start of menses, for a total of 36 hours. After every 12 hours, the cup was emptied in a new collection tube, containing RPMI1640 medium (ThermoFisher Scientific, Waltham, MA, USA) supplemented with 10% human pooled serum (HPS) (Zen-Bio, Durham, NC, USA), 1 mM pyruvate (ThermoFisher Scientific), 2 mM GlutaMAX™ (ThermoFisher Scientific), 100 U/mL penicilin/100 μg/mL streptomycin (ThermoFisher Scientific), and 0.3% sodium citrate (Merck, Darmstadt, Germany). PB (10 mL) was collected in ethylene diamine tetra acetic acid (EDTA) tubes.

PB was diluted 1:1 with phosphate buffered saline (PBS) (Fresenius Kabi, GmbH, Graz, Austria). The collected MB from 3 different timepoints was first pooled, after which it was washed with PBS. After centrifugation, the pellet of the MB was filtered through a 70 μm cell strainer (PluriSelect Life Science, Leipzig, Germany) by creating a vacuum using a connector ring (PluriSelect) and a 30 mL syringe (Terumo Corporation, Tokyo, Japan). The filtered MB was then incubated for 20 minutes at room temperature with RosetteSep™ Human Granulocyte Depletion Cocktail (STEMCELL Technologies, Vancouver, Canada) according to manufacturer’s instructions, after which the sample was diluted with an equal volume of PBS+2%HPS.

Subsequently, MB mononuclear cells (MBMCs) and PB mononuclear cells (PBMCs) were isolated using Lymphoprep (STEMCELL Technologies) according to manufacturer’s instructions. The samples were centrifuged for 15 min at 800g, no brake at 20°C. Isolated MBMCs and PBMCs were washed twice with PBS+2%HPS, and resuspended in RPMI1640 + 10% HPS.

Immunophenotyping of MBMCs and PBMCs was done using flow cytometry. For the phenotyping, 500.000 MBMCs and 250.000 PBMCs were used and stained in 50 μL and 25 μL respectively. First, the cells were stained for 30 minutes at 4°C with eBioscience™ Fixable Viability Dye eFluor 780 (ThermoFisher Scientific) to stain for dead cells. Subsequently, the cells were stained for 30 minutes at 4°C for specific surface markers stated in Supplementary Table 2 . Because of the use of multiple BD Horizon™ Brilliant Violet Dyes (BD, Franklin Lakes, NJ, USA), BD Horizon™ Brilliant Staining Buffer (BD) was used in combination with regular FACS buffer (PBS+1%BSA+0.02% Sodium Azide (Merck, Rahway, NJ, USA)) according to manufacturer’s recommendations. After surface staining, all cells were fixated and permeabilized using the eBioscience™ Foxp3/Transcription Factor Staining Buffer Set according to the manufacturer’s instructions (ThermoFisher Scientific). After fixation/permeabilization, part of the cells were stained for the transcription factor FoxP3 (ThermoFisher Scientific). All samples were measured with standardized settings on the BD FACSLyric™ (BD). Data was analyzed using the FlowJo™ v10.8 software (Flowjo™, Ashland, OR, USA) as either a percentage or as Median Fluorescent Intensity (MFI). MFIs of immune markers were normalized against MFIs of Fluorescence Minus One (FMO) controls, critical to ensure that the signals observed are truly due to the marker of interest and not artefacts from overlapping fluorescent markers. Certain markers for specific signals, timepoints or participants could not be analyzed due to various reasons such as sample or reagent availability. PBMCs and MBMCs that were not directly used in the flow cytometry staining were frozen in Recovery™ Cell Culture Freezing Medium (ThermoFisher Scientific).

To assess the functional capacity of NK cells in MB and PB, we performed a CD107a degranulation assay and an intracellular staining on frozen PBMCs and MBMCs. Following thawing of a subgroup of available samples (n=12), 200,000 cells were plated in a round-bottom 96-wells plate. Cells that were used to investigate CD107a and interferon-γ (IFN-γ) expression were stimulated with 100 U/mL IL-2 and 10 ng/mL IL-15. An unstimulated condition was taken as a control. The cells that were used to investigate granzyme B and perforin expression were left unstimulated. The cells were incubated at 37°C and 5% CO₂. After 20 hours of incubation, 25 μL of a 1:50 dilution of CD107a-PE (BD Biosciences) was added to the respective wells. One hour later, 20,000 K562 target cells and 5 μg/mL of Brefeldin A and 5 μg/mL of Monensin were added to inhibit the internalization of CD107a molecules and enhance the detection of intracellular proteins. The cells in the wells for investigation of granzyme B and perforin expression were not stimulated with K562. The cells were then incubated for an additional 3 hours at 37°C in 5% CO₂. Following the incubation period, cell viability was assessed, and surface markers were stained with the following antibodies: anti-CD16 Alexa Fluor 700 (BioLegend), anti-CD45 Krome Orange (Beckman Coulter), anti-CD3 BV605 (BD Biosciences), and anti-CD56 BV711 (BD Biosciences). Subsequently, intracellular staining was performed using anti-granzyme B PerCP-Cy5.5 (BioLegend), anti-IFNγ PE-Cy7 (ThermoFisher Scientific), and anti-perforin Pacific Blue (BioLegend) ( Supplementary Table 2 ). All samples were analyzed on a BD FACSLyric™ flow cytometer (BD Biosciences).

Percentages of immune cells or percentages/MFIs of immune markers before and after exercise were analyzed with a Wilcoxon signed-rank test. All analyses were performed using IBM SPSS statistics, version 29.0.0.0 (SPSS Inc, Chicago, IL, USA) and P-values <0.05 were considered statistically significant. The boxplots were prepared in RStudio using ggplot2 package (42, 43).

To determine the effect of exercise on specific immune cell subsets, we analyzed frequencies of T, B, and NK cells using high-dimensional flow cytometry in both PB and MB from women with uRPL ( Figure 1A ). Baseline characteristics, of the included 24 women who completed the intervention with an overall attendance of 95%, are shown in Table 1 .

The role of NK cells in PB and MB are different. While the majority of pNK cells consist of a subset of CD56^dimCD16⁺ NK cells with high cytotoxic capabilities, uNK cells in MB mainly have a CD56^brightCD16^- phenotype and a immunoregulatory functions (12). Results of these two subsets ( Figure 1B ) revealed no significant differences in the frequency of total CD56⁺CD3^-, CD56^brightCD16^- and CD56^dimCD16⁺ NK cells after intervention in either PB or MB ( Figures 1C–E ). Interestingly, we observed significant heterogeneity in the baseline frequencies of uNK cells among women with uRPL, as well as in their response to the intervention. While some women exhibited increased frequencies of uNK cells, others showed a decrease post-intervention.

Likewise, the frequencies of total CD3⁺ T cells and their subsets (including conventional effector CD4⁺, cytotoxic CD8⁺ T, and CD127^lowCD25⁺FoxP3⁺ regulatory T cells) remained similar in both PB and MB when comparing pre- and post-intervention measurements ( Supplementary Figures 2A–C ). Although there was a trend towards an increased frequency of total CD19⁺CD20⁺ B cells in PB as well as MB, the frequencies of B cell subsets did not change significantly after intervention ( Supplementary Figures 2D–F ).

To study the effect of the intervention on the phenotype of pNK and uNK cells, we performed comprehensive immunophenotyping by flow cytometry to evaluate changes in activating and inhibitory NK cell receptors (NKRs), frequently described to be involved in the regulation of NK cell activity.

We found a significant decrease in median fluorescent intensity (MFI) of CD161 (KLRB1) in total pNK cells (464 [364-623] to 410 [277-483]; p=0.011), in the CD56^brightCD16^- pNK cell subset (353 [247-478] to 296 [193-367]; p=0.033), and in the CD56^dimCD16⁺ pNK cell subset (493 [393-648] to 424 [296-559]; p=0.035), while no significant differences were observed in uNK cells post-intervention ( Figures 2A–D ).

Among the natural cytotoxicity receptors (NCRs) analyzed, the MFI of NKp30 significantly decreased in total pNK cells (432 [317-494] to 376 [252-481]; p=0.018) and in the CD56^dimCD16⁺ pNK subset (468 [290-550] to 369 [228-520]; p=0.004) post-intervention, whereas this was not observed in uNK cells ( Figures 2E–H ). No significant changes were found in the expression intensities of other NCRs (NKp44 and NKp46) after intervention ( Supplementary Figures 3A, B ).

To study the effect of exercise on NK cell markers involved in maturation and memory, we analyzed the frequency and expression intensity of NKG2A and NKG2C before and after intervention. While the frequencies of NKG2A⁺ NK cells did not significantly change after intervention in either MB or PB ( Supplementary Figures 4A–C ), a marked decrease in the MFI of NKG2A was observed in total pNK cells (886 [781-1218] to 732 [555-962]; p=0.039) and CD56^brightCD16⁺ pNK cells (2472 [1583-3116] to 1497 [1113-1790]; p=0.034) ( Figures 3A–C ). In contrast, uNK cells did not show any significant changes in NKG2A expression ( Figures 3A–D ). Furthermore, no significant changes were observed in both frequency and expression intensity of NKG2C in both MB and PB, except for a slight increase the percentage of in NKG2C⁺ CD56^dimCD16⁺ pNK cells (5.4 [3.3-13.7] to 5.9 [2-8.9]; p=0.021) ( Supplementary Figure 4F ).

Next we investigated co-expression of NKRs by analyzing the expression of NKG2A, NKG2C, and KIRs, on CD56^dimCD16⁺ pNK cells and CD56^brightCD16^- uNK cells. No significant changes were found in the CD56^dimCD16⁺ pNK cells or CD56^brightCD16^- uNK cells populations ( Figure 3E ). Additionally, the frequencies of CD57 and KIRs remained unchanged ( Supplementary Figures 3C–F ).

Lastly, we investigated the effect of the intervention on NK effector function by determining the levels of intracellular granzyme B and perforin. In addition, we analyzed NK cell degranulation (CD107a) and IFN-γ production upon stimulation with K562 target cells or cytokines (IL-2 and IL-15) before and after intervention. Our analysis revealed no significant changes in CD107a expression intensity or frequency in either pNK or uNK cells after intervention ( Figures 4B, C , Supplementary Figure 5A ). Notably, stimulation with K562 cells did not enhance CD107a expression in uNK cells compared to the unstimulated conditions ( Figure 4C ). IFN-γ production was significantly reduced by pNK cells (35.8 [3.5-50.4] to 23.1 [3.2-33.9]; p=0.028) after intervention, with no substantial change observed in uNK cells ( Figures 4D, E ). Overall, uNK cells demonstrated lower IFN-γ production compared to their peripheral counterparts upon stimulation and co-culture. Granzyme B and perforin production remained unchanged post-intervention in unstimulated pNK cells ( Supplementary Figures 4B, C ).