Human leucocyte antigen G (HLA-G) and its murine homologue Qa-2 protect from pregnancy loss

During pregnancy, the maternal immune system has to balance tightly between protection against pathogens and tolerance towards a semi-allogeneic organism. Dysfunction of this immune adaptation can lead to severe complications such as pregnancy loss, preeclampsia or fetal growth restriction. The MHC-Ib molecule HLA-G is well known to mediate immunological tolerance. However, no in-vivo studies have yet demonstrated a benecial role of HLA-G for pregnancy success. Myeloid derived suppressor cells (MDSC) are suppressively acting immune cells accumulating during pregnancy and mediating maternal-fetal tolerance. Here, we analyzed the impact of Qa-2, the murine homologue to HLA-G, on pregnancy outcome in vivo. We demonstrate that lack of Qa-2 led to intrauterine growth restriction and increased abortion rates especially in late pregnancy accompanied by changes in uterine gene expression, altered spiral artery remodeling and protein aggregation in trophoblast cells indicating a preeclampsia-like phenotype. Furthermore, lack of Qa-2 caused decreased accumulation of MDSC and impaired MDSC function. Lastly, we show that application of sHLA-G reduced abortion rates in Qa-2 decient mice by inducing MDSC. Our results highlight the importance of an interaction between HLA-G and MDSC for pregnancy success and the therapeutic potential of HLA-G for the treatment of immunological pregnancy complications. and showing in-vivo that application of sHLA-G improves pregnancy outcome. These results give reason to hope that synthetic sHLA-G may nd a place in the of pregnancy complications like abortions and preeclampsia. cells


Introduction
Premature termination of pregnancy either by abortion or by preterm delivery is the most important pregnancy complication. At least 25%, but probably up to 50% of women suffer miscarriage and about 10% of infants are delivered preterm 1,2 . Besides chromosomal or anatomic anomalies and endocrinological disorders, immunological factors play an important role in abortion pathogenesis and preterm delivery. During pregnancy, there is a close contact between maternal immune cells and fetal cells. Thus, the maternal immune system has to balance tightly between protection against pathogens and tolerance towards the semi-allogeneic fetus. Dysfunction of the immune adaptation to pregnancy can lead to severe complications such as pregnancy loss, preeclampsia, preterm birth or fetal growth restriction. The mechanisms facilitating maternal-fetal tolerance are only incompletely understood and therapeutic options are limited.
The major histocompatibility class Ib (MHC Ib) molecule human leucocyte antigen G (HLA-G) is a nonclassical MHC I molecule with low allelic variation and a restricted peptide repertoire 3 . Under physiological conditions, HLA-G is mainly expressed by trophoblast cells at the maternal-fetal interface and can be secreted in a soluble form (soluble HLA-G, sHLAG) to the circulation 3 . In vitro studies show that HLA-G shapes the maternal immune system towards tolerance by induction of a tolerogenic phenotype of antigen-presenting cells (APCs) 4 as well as inhibition of T-cell activity 5 and natural killer cell (NK-cell) cytotoxicity 6 . During an uncomplicated pregnancy, levels of sHLA-G rst increase and then decrease until the third trimester. Undetectable sHLA-G levels or variation in the course of sHLA-G levels seem to be related with gestational complications such as spontaneous abortion and preeclampsia 7 . As yet, however, no in-vivo studies exist that demonstrate a bene cial role of HLA-G for pregnancy success.
Qa-2 has been described as the murine homolog to HLA-G 8 . Mice not expressing the Qa-2 antigen (Qa2 -) have smaller litters and a shorter duration of gestation than Qa-2 expressing animals 9 . On the fetal side, presence of Qa-2 seems to protect from rejection 9 . Mechanisms mediating the protective role of Qa-2 during reproduction are only incompletely understood.
Myeloid derived suppressor cells (MDSC) are myeloid cells with various immune suppressive properties.
They mainly consist of two subtypes named granulocytic MDSC (GR-MDSC) with phenotypic characteristics of neutrophils and monocytic MDSC (MO-MDSC) with phenotypic similarities to monocytes. In mice, GR-MDSC are de ned as CD11b + /Ly6G + /Ly6C lo and MO-MDSC as CD11b + /Ly6G -/Ly6C hi cells 10 . Primarily, MDSC accumulation has been described under tumor conditions where they suppress immune responses against tumor cells, thereby leading to disease progression 11 . Later, MDSC have been shown to accumulate under various other pathologies like infection, trauma, autoimmune disease, obesity and transplantation 12 . In recent years, however, there is increasing evidence that MDSC also play a physiological role during pregnancy by modulating maternal immune responses and protecting the fetus from rejection [13][14][15][16][17] . Accumulation and activation of MDSC are driven by various factors 10 . Recently, we demonstrated, that HLA-G induced and activated GR-MDSC in-vitro 18 .
In the present study, we investigated the in-vivo role of HLA-G for pregnancy success by using mice not expressing the Qa-2 antigen (Qa-2 negative, Qa2 -, B6.K1) and showed that lack of Qa-2 leads to intrauterine growth restriction and increased abortion rates with profound changes in uterine gene expression pro le and uterine spiral arteries and trophoblasts, indicating a preeclampsia-like phenotype.
Immunological adaptation to pregnancy was imbalanced with missing MDSC accumulation and impaired MDSC function. By application of sHLA-G to Qa-2mice we could reduce their increased abortion rate, highlighting its therapeutic potential for treatment of immunological pregnancy complications.

Results
Qa-2 de ciency in mice leads to adverse pregnancy outcome in late pregnancy To evaluate the impact of Qa-2 on pregnancy outcome, we analyzed mice lacking the Qa-2 antigen (Qa-2 -, B6.K1). Compared to WT mice, we found signi cantly smaller litter sizes in Qa-2animals ( Figure 1A). At mid-pregnancy (E10.5), Qa-2mice had similar numbers of intact fetuses compared to WT mice (Supplementary Figure 1A) and slightly increased abortion rates (Supplementary Figure 1B+C). At E18.5, however, lack of Qa-2 led to an abortion rate of 17% in comparison to 5% in WT animals ( Figure 1B+C). Furthermore Qa-2fetuses weighed signi cantly less than WT fetuses ( Figure 1D+E). Qa-2 de ciency leads to changes in the uterine gene expression pro le Since many adaptation processes to pregnancy take place in the uterus, we next performed transcriptome analyses of whole uterine lysates. Of 610 differentially regulated genes, 380 were upregulated in WT uteri and 230 were upregulated in Qa-2uteri. Network analysis using the STRING database 19 showed enrichment of genes encoding for proteins involved in various speci c biological processes in WT animals (Figure 2A), among which "immune system processes", "nervous system development" and "circulatory system development" were the processes with most genes detected. Among immunological processes upregulated in WT uteri, especially cytokine/chemokine signaling, myeloid cell differentiation, apoptosis regulation, leucocyte migration and lymphocyte activation were involved ( Figure 2B). Genes upregulated in Qa-2animals showed no clustering to any biological process.
Qa-2 de ciency leads to altered spiral artery remodeling and altered trophoblast morphology Since spiral artery remodeling is a crucial step in maternal adaptation to pregnancy, and transcriptome analyses revealed profound differences in uterine expression of genes involved in vasculogenesis between WT and Qa-2animals, we next analyzed spiral arteries in E10.5 pregnant WT and Qa-2animals.
In Qa-2animals, we found large areas within the decidua with unorganized trophoblast distribution, while in WT animals, trophoblasts proper organized around the vessels, pointing to an abnormal trophoblastmigration in Qa-2mice ( Figure 3A). In addition, spiral arteries of Qa-2animals had thicker vessel walls than that of WT mice, while luminal areas did not differ (Figure 3 B-D).
Furthermore, we found changes in placenta histology between E18.5 old WT and Qa-2fetuses; placentas from both genotypes showed similar cross-section areas ( Figure 3E); however, while placentas from WT animals showed long and thin villi with proper morphology, placentas from Qa-2animals showed irregular and short villi and abnormal vacuolization of the trophoblast and numerous eosinophilic aggregates. On a scale of 0 (no aggregates) to 2 (prominent aggregates), the phenotype of WT placentas was 0.2 ± 0.4 while that of Qa-2placentas was 1.4 ± 1.0 ( Figure 3F+G and Supplementary Figure 2). These aggregates were still present in PAS-diastase staining, indicating that they were not glycogen ( Figure 3H). Proteome analyses of whole placenta lysates from WT and Qa-2animals showed strong enrichment in proteins involved in protein metabolism processes (GO:0019538), especially in translation Systemic and uterine accumulation of MDSC during pregnancy is abrogated in Qa-2mice Since it is known that HLA-G plays an important role in immune regulation during pregnancy 20 and transcriptome analyses showed upregulation of genes involved in immune system processes, we analyzed immune cell populations in spleen and uterus of Qa-2and WT mice. We observed a strong increase in total splenic MDSC, as well as in splenic GR-MDSC and MO-MDSC between non-pregnant WT animals and WT animals at E18.5. In Qa-2animals however, there was only a marginal increase in total splenic MDSC at E18.5, while neither GR-MDSC nor MO-MDSC numbers increased ( Figure 4A-D). Correspondingly, we found strongly increased numbers of uterine MDSC in WT animals at E18.5 in comparison to non-pregnant controls, but not in Qa-2animals ( Figure 4E+F). Conversely, in Qa-2mice we found an increase in splenic T-cells and a decrease in splenic B-cells upon pregnancy, while splenic T-cell and B-cell numbers in WT mice remained unchanged. In WT mice, but not Qa-2mice, splenic NK-cells and Qa2 de ciency leads to changes in T-cell subpopulations during pregnancy and to a decreased capacity of MDSC to induce T regs We further investigated whether there were any differences in T-cell subpopulations between pregnant WT and Qa-2animals at E18.5. Gating strategy for T-cell subpopulations is depicted in Figure 5A, and phenotyping strategy is depicted in Figure 5B. No differences were found in percentages of T-helper cells and cytotoxic T-cells between WT and Qa-2animals ( Figure 5C+G). Qa-2animals had higher numbers of effector memory CD4 + and CD8 + T-cells and lower numbers of central memory CD4 + and naïve CD8 + Tcells ( Figure 5D-F and H-J). Furthermore, Qa-2animals had signi cantly less T reg -and more Th17-cells, while there were no differences in numbers of Th1-and Th2-cells ( Figure 5K-P). Decreased numbers of T regs in Qa-2animals were con rmed by intracellular staining of FoxP3 ( Figure 3Q+R).
Induction of T regs is a main feature of MDSC 21,22 . We thus analyzed the capacity of Qa-2 -MDSC to induce T regs in comparison to that of WT MDSC in vitro. T regs were induced by both, WT and Qa-2 -MDSC, although to a lower extend in the latter ( Figure 6A+B). Capacity of Qa-2 -MDSC to inhibit T-cell proliferation was also reduced in comparison to WT MDSC, but the difference did not reach signi cance ( Figure 6C+D).

Expression of Qa-2 on MDSC is regulated by estrogen via HIF-1α
As Qa-2 seemed to be relevant for MDSC function, we next asked how Qa-2expression may be regulated. Flow cytometric analyses of Qa-2 expression on MDSC and T-cells revealed that in non-pregnant WT mice between 10% and 60% of MDSC and all T-cells expressed Qa-2; pregnancy induced the expression of Qa-2 both on MDSC and T-cells ( Figure 7A-C). The same effect could be observed for HLA-G-expression on human MDSC ( Figure 7D+E). To get hints on a potential hormonal regulation of Qa-2 expression on immune cells, we analyzed blood of female mice during the menstrual cycle and found increased Qa-2 expression on MDSC during proestrus and estrus, the phases with higher estrogen levels 23 , than during metestrus and diestrus ( Figure 7F+G). To further evaluate the effect of estrogen on Qa-2 expression on MDSC, we next stimulated spleen cells of WT mice with increasing concentrations of estrogen and showed that Qa-2 expression on MDSC increased upon estrogen stimulation in a concentration dependent manner, while Qa-2 expression on T-cells did not change ( Figure 7H-J). Recent data showed that expression of HLA-G on tumor cells can be regulated by the transcription factor hypoxia-inducible factor 1α (HIF-1α) 24,25 and that HIF-1α regulates MDSC function during murine pregnancy 15 . We thus assumed that expression of Qa-2 on MDSC may be regulated by HIF-1α and stimulated spleen cells of WT mice with classic (anoxia) and alternative (Escherichia coli, E. coli) stimuli of HIF-1α. We showed that both anoxia and E. coli stimulation led to an increased expression of Qa-2 on MDSC ( Figure 8A-C), but not on T-cells ( Figure 8D+E). Correspondingly, MDSC isolated from pregnant mice with targeted deletion of HIF-1α in myeloid cells (HIF-KO) expressed lower levels of Qa-2 than MDSC isolated from WT mice ( Figure 8F). Stimulation of myeloid HIF-KO MDSC with estrogen did not result in an upregulation of Qa-2 expression ( Figure 8G). Taken together, our results show that expression of Qa-2 on MDSC is at least partly regulated by estrogen via HIF-1α.
Application of sHLA-G improves pregnancy outcome Lastly, we asked whether we could restore pregnancy success in Qa-2animals by application of sHLA-G.
Pregnant Qa-2mice received either 1µg/g bodyweight sHLA-G or PBS at E10.5 and E14.5. Application of sHLA-G led to a pronounced reduction in the abortion rate of Qa-2animals ( Figure 9A), accompanied by a partial restoration of normal trophoblast morphology ( Figure 9B). Furthermore, it marginally increased splenic MDSC, but strongly increased uterine MDSC ( Figure 9C-E). Simultaneous depletion of MDSC with sHLA-G application reversed the pregnancy-protective effect of sHLA-G ( Figure 9F). To con rm the bene cial effect of sHLA-G on pregnancy outcome in another model, we treated abortion-prone DBA/2Jmated CBA/J mice with sHLA-G or PBS at E0.5, E3.5, E6.5 and E9.5. Also in this model, application of sHLA-G signi cantly reduced abortion rates ( Figure 9G+H).

Discussion
Our data show that Qa-2 is relevant for pregnancy success and protects from late pregnancy loss by regulating immune adaptation to pregnancy in terms of promoting MDSC accumulation and modulating T-cell homeostasis. We further show that expression of Qa-2 on MDSC is relevant for their functionality and regulated by estrogen via HIF-1α. Lastly, we show that application of sHLA-G to abortion-prone Qa-2mice decreases abortion rates via induction of MDSC.
It has been shown for a long time that HLA-G is highly expressed during pregnancy especially by trophoblast cells 3 mediating various immune-modulatory effects in vitro 4-6 . Furthermore, alterations of HLA-G expression during pregnancy are associated with adverse pregnancy outcome [26][27][28] . Until now, however, no studies provided in vivo evidence that HLA-G indeed is needed for a successful pregnancy. We used Qa-2 de cient mice to evaluate the in vivo role of Qa-2/HLA-G for pregnancy outcome and found smaller litter sizes and growth restriction in surviving fetuses in Qa-2animals in comparison to WT animals. These results con rm previous studies also describing smaller litters and smaller offspring in Qa-2animals 9,29 . Interestingly, expression of Qa-2 on the fetal side was found to be advantageous for survival, leading to a higher embryonic cleavage rate 9,30 . We now show increased rates of pregnancy loss in Qa-2animals especially during late pregnancy accompanied by profound changes in maternal adaptation to pregnancy in comparison to WT mice, demonstrating that Qa-2 not only plays a local role in fetal tissue but is also systemically needed in the maternal organism to facilitate a successful pregnancy.
In transcriptome analyses of whole uteri, we found prominent differences between Qa-2and WT animals in transcripts regulating angiogenesis. Corresponding to that, we found altered spiral artery morphology in Qa-2in comparison to WT animals. Remodeling of uterine spiral arteries is one of the critical steps in maternal adaptation to pregnancy as it permits normal placental perfusion and fetal growth and development 31 . Inadequate spiral artery remodeling results in placental hypoxia and may lead to development of preeclampsia and fetal growth restriction 32 . In Qa-2placentas, we observed profound changes in trophoblast morphology in comparison to WT placentas with cytoplasmic storage of eosinophilic aggregates and enrichment of proteins involved in protein metabolism. Interestingly, it has been shown that during preeclampsia, misfolded proteins accumulate in urine, serum and placenta similar to the protein accumulation observed in neurodegenerative disorders like Alzheimer´s disease [33][34][35] . Our ndings of dysregulation of uterine vasculogenesis, fetal growth restriction, late abortions and pathological protein storage in placenta may suggest the development of a preeclampsia-like phenotype in Qa-2mice. This assumption is supported by recent data showing that injection of an anti-Qa-2 antibody led to preeclampsia symptoms in mice that could be abrogated by simultaneous injection of recombinant VEGF 36 .
Furthermore, we found signi cant differences in immunological adaptation to pregnancy between WT and Qa-2animals. Both, systemically and locally in the uterus, we observed a strong increase in MDSC in WT but not in Qa-2mice. The accumulation of MDSC and its relevance for successful pregnancy has been described previously in different mouse models (reviewed in 37 ). However, all these studies focused on early-to mid-gestation, showing that depletion of MDSC in early gestation or genetically determined decreased MDSC accumulation at E10.5 led to complete gestation failure 16 or increased abortion rates 15,17 . Our present results now illustrate that MDSC accumulation also seems to be relevant for pregnancy success in later stages of pregnancy. Contrarily to our results, Ostrand-Rosenberg et al.
showed that depletion of MDSC at E8.5 did not lead to adverse pregnancy outcome 16 . However, since antibody-mediated MDSC depletion has to be repeated every three days 38 , one injection at E8.5 may be insu cient to examine the role of MDSC until the second half of pregnancy. In correspondence to the increased accumulation of MDSC in uteri of WT mice we observed an upregulation of genes involved in myeloid differentiation and leucocyte migration and chemotaxis in transcriptome analyses of WT whole uterine lysates in comparison to Qa-2animals. Furthermore, we previously showed that sHLA-G in-vitro led to a quantitative and functional induction of MDSC 18 . We thus assume that a combination of direct and indirect effects of Qa-2 attract MDSC to the pregnant uterus in-vivo.
Decreased MDSC accumulation in pregnant Qa-2mice was accompanied by an increase in splenic Tcells -an effect that may be associated with miscarriage 39 . Differences in T-cell subpopulations between pregnant WT and Qa-2animals mainly concerned effector memory T-cells, T regs and Th17 cells. The CD44 + /CD62Leffector memory CD4 + and CD8 + T-cell subsets were increased in pregnant Qa-2mice, while naïve CD8 + and central memory CD4 + T-cells were decreased. This points towards a higher activation status of T-cells in Qa-2mice. A recent study showed that effector/activated T-cells led to adverse pregnancy outcome, i.e. preterm birth 40 . Furthermore, patients with preeclampsia downregulate CD62L on T-cells 41 . Since the main effector function of MDSC is an inhibition of T-cell activation, one could hypothesize that increased activation of T-cells in Qa-2animals results from a decreased MDSC in uence. However, contrary to our results, MDSC downregulate CD62L on T-cells 16,42 . As Qa-2 is also highly expressed on T-cells, the lack of Qa-2 itself may lead to differences in T-cell activation between WT and Qa-2mice overlapping the effect of MDSC.
Balance between Th1 and Th2 cells did not differ between WT and Qa-2animals, while T regs decreased and Th17 cells increased in pregnant Qa-2in comparison to pregnant WT mice, which is similar to ndings from normal pregnancies, but the opposite to what is found in patients with recurrent pregnancy loss and preeclampsia 43 . Studies in mice showed that expansion of T regs was relevant for healthy pregnancy and that adoptive transfer of T regs protected from abortions 44,45 . A crosstalk between T regs and MDSC has been described extensively under tumor conditions (reviewed in 46  Due to the observed functional differences between WT and Qa-2 -MDSC, we asked whether Qa-2 expression on MDSC may be upregulated during pregnancy. We found that MDSC isolated from pregnant individuals (mice and women) expressed higher levels of Qa-2, respectively HLA-G, than MDSC from nonpregnant individuals and that Qa-2 expression on MDSC, but not on T-cells, could be stimulated by estrogen. Immunomodulatory effects of estrogens have been repeatedly described, e.g. an expansion of T regs and a modulation of Th-cell cytokine expression 50,51 . Furthermore, it could be shown that estrogen mediates expansion and functional activation of MDSC 52,53 . However, upregulation of Qa-2 expression on MDSC by estrogen is a yet unknown mechanism. We further show that the effect of estrogen on Qa-2 expression on MDSC was mediated through HIF-1α. This is in line with results from other groups showing that estrogen can activate HIF-1α 54 and that activation of HIF-1α stimulated HLA-G-expression in cancer cells 24,25 . We found that expression of HIF-1α was relevant for MDSC accumulation and function during pregnancy and that targeted deletion of HIF-1α in myeloid cells (myeloid HIF-KO) led to pregnancy failure in terms of abortions 15 . Our new results suggest that an impaired expression of Qa-2 on MDSC may at least partially be responsible for the adverse pregnancy outcome in myeloid HIF-KO mice.
Lastly, we aimed to investigate the therapeutic effect of sHLA-G on pregnancy outcome and found that application of sHLA-G reduced the abortion rate in Qa-2animals, restored placental morphology and induced uterine MDSC accumulation. Simultaneous antibody-mediated depletion of MDSC nulli ed the protective effect of sHLA-G. Previous studies in mice showed protective effects of HLA-G on transplant rejection 55,56 and collagen-induced arthritis 57 ; furthermore, sHLA-G was shown to allow tumor evasion from immunosurveillance 58,59 , with some of these effects being mediated by an expansion of MDSC [58][59][60] .
These results together with those reported now suggest that a mutual support of HLA-G and MDSC helps to protect allografts from immune rejection and that this interaction is helpful whenever tolerance is needed to survive (pregnancy, organ transplantation), but detrimental in case of tumor growth.
One limitation of our study is that we used a syngeneic mating model. Since allo-antigens play an important role for immunological pregnancy complications and especially for preeclampsia it may be worth to investigate allogeneic pregnancy. However, in our case, allogeneic mating would have led to expression of Qa-2 by the fetuses making it impossible to investigate the effect of a total lack of Qa-2.
Another limitation is that HLA-G is not endogenously expressed in mice. Thus, we used mice lacking Qa-2, the only homologue-candidate for HLA-G yet known, for analyzing its impact on pregnancy outcome.
However, although HLA-G is a human MHC I molecule, it binds to the murine paired immunoglobulin-like inhibitory receptor (PIR-B) and mediates tolerogenic effects in mice 61 making it possible to analyze its effects in-vivo.
In conclusion, we here describe the impact of Qa-2 on immune adaptation during pregnancy, providing evidence that Qa-2 may prevent the development of preeclampsia and showing in-vivo that application of sHLA-G improves pregnancy outcome. These results give reason to hope that synthetic sHLA-G may nd a place in the prevention of immunological pregnancy complications like abortions and preeclampsia.  To test the effect of sHLA-G on pregnancy outcome in Qa2mice, pregnant mice were injected intravenously at E10.5 and E14.5 with 1 µg/g body weight HLA-G1 tetramers in 100 µl PBS or with 100 µl PBS alone (control).

Determination of mouse estrous cycle
For identi cation of the mouse estrous cycle stage, female C57BL/6J mice were anesthetized with 1.5% iso urane (CP-Pharma, Burgdorf, Germany). Blood was obtained by puncture of the retroorbital vein plexus or the tail vein and vaginal swabs were collected according to an established protocol 23 . A cotton tipped swab (Applimed, Châtel-Saint-Denis, Switzerland) wetted with room-temperature physiological saline was inserted vaginally, gently turned, and then removed. The procedure was repeated twice for four consecutive days with 28 days in between.
The vaginal cells were transferred to a glass slide by rolling the swab over the slide. The slide was air dried, stained with hematoxylin-eosin (HE, Merck GmbH, Darmstadt, Germany) and viewed at 10x magni cation under bright eld illumination. The cycle stage was determined based on the presence or absence of leucocytes, corni ed epithelial cells and nucleated epithelial cells according to 23 .

Patients
The local ethics committee approved this study (682/2016BO1) and all women gave written informed consent. From August to October 2019 peripheral blood from pregnant women (aged 18-43 years) was collected during routine blood sampling. Patients suffering from severe pregnancy complications (severe infection, preterm rupture of membranes, preterm labour, preeclampsia/eclampsia), chronic diseases (autoimmune diseases, malignancies, chronic infections) or receiving immune-suppressive therapy were excluded.

Mouse tissue collection and single cell preparations
Non-pregnant and pregnant mice at gestational age E10.5 or E18.5 were euthanized by CO 2 inhalation.
Blood (0.5-1 ml) was collected immediately after death by intracardial puncture and placed into EDTAtubes. Blood plasma was collected after centrifugation of whole blood at 400 rpm. Red blood cells were removed from whole blood by ammonium chloride lysis. Spleens were removed and tissue was pushed through a 100 mm lter (Greiner bio-one, Frickenhausen, Germany) using a syringe plunger. Red blood cells of the spleen were also removed by ammonium chloride lysis and the resulting cell suspension was again passed through a 40 mm lter (Greiner bio-one, Frickenhausen, Germany). Uterine horns were removed in toto. Fetuses and the fetal part of the placenta were dissected from uteri; blood vessels were removed. Uteri were placed into PBS, cut into ~1 mm pieces and pushed through a 40 mm lter.
Placentas were pushed through a 100 µm lter using a syringe plunger. Red blood cells were removed by ammonium chloride lysis and the resulting cell suspension was then passed again through a 40 µm lter.
All cell suspensions were then adjusted to 1-4x10 6 cells/ml in PBS or medium.

Cell Isolation and Culture
Human peripheral blood mononuclear cells (PBMC) were prepared from EDTA blood samples by Ficoll density gradient centrifugation (lymphocyte separation medium, Biochrom, Berlin, Germany).
To isolate GR-MDSC from murine splenocytes, cells were labeled with Gr-1 Biotin-Antibody and isolated over Streptavidin microbeads followed by a second isolation step using Ly6G Biotin-Antibody and Anti-Biotin microbeads (modi ed protocol of MDSC Isolation Kit mouse, Miltenyi, Bergisch-Gladbach, Germany). Purity of GR-MDSC after separation was >90%, as determined by ow cytometry.

For isolation of CD4 + T-cells from murine splenocytes, cells were labeled with T-cell Biotin-Antibody
Cocktail followed by two consecutive Anti-Biotin magnetic bead separation steps (Miltenyi) according to the manufacturer´s instructions. Purity of CD4 + T-cells after separation was >90%, as determined by ow cytometry.
In-vitro generation of murine MDSC was performed according to previously established protocols 15,62 . For in-vitro generation of MDSC, non-pregnant WT and Qa-2mice were euthanized and femora and tibia removed. Bone marrow was collected by ushing the bones with PBS using a syringe and a 25G needle.
Bone marrow cells were then washed twice, adjusted to 5x10 5  For intracellular staining of Foxp3 cells were extracellular stained with CD4 and CD25 for 30 minutes at 4°C and then incubated in Foxp3 Fixation/Permeabilization working solution (Thermo Fisher Scienti c, Waltham, USA) for 60 minutes at room temperature and protected from light. Cells were washed in 1x Permeabilization buffer (Thermo Fisher Scienti c) and stained with Foxp3 antibody in 1x permeabilization buffer for 30 minutes at room temperature.

RNA isolation and transcriptome analyses
For transcriptome analyses of whole uterine lysates, uteri of pregnant WT and Qa-2mice at E18.5 were collected, snap-frozen in liquid nitrogen and stored at -80°C until RNA isolation. For RNA isolation, frozen tissue (20-30 mg) was shredded using micro pestles (Sigma Aldrich, St.Louis, USA) and liquid nitrogen to obtain powder. RLT buffer was added and the solution was centrifuged at 8000 rcf. RNA isolation was then performed using the RNeasy Mini Kit (Qiagen, Hilden, Germany).

Protein isolation and proteome analyses
For proteome analyses, single cell suspensions were prepared from placentas of pregnant WT and Qa2animals at E18.5. Cells were lysed by adding lysis buffer (5% 1M Tris/HCl pH 7,4, 2% 5M NaCl, 1% Triton X 100, 1% PMSF, 4% protease inhibitor cocktail (Sigma-Aldrich, St. Louis, USA)) on ice followed by snapfreezing in liquid nitrogen. Ten micrograms of each sample were digested in solution with trypsin as described in 63 . After desalting using C18 stage tips, extracted peptides were separated on an Easy-nLC 1200 system coupled to a Q Exactive HFX mass spectrometer (Thermo Fisher Scienti c) as described in 64 with slight modi cations: The peptide mixtures were separated using a 90 minutes segmented gradient from to 10-33-50-90% of HPLC solvent B (80% acetonitrile in 0.1% formic acid) in HPLC solvent A (0.1% formic acid) at a ow rate of 200 nl/min. The 12 most intense precursor ions were sequentially fragmented in each scan cycle using higher energy collisional dissociation (HCD) fragmentation.
Acquired MS spectra were processed with MaxQuant software package version 1.6.7.0 with integrated Andromeda search engine. Database search was performed against a target-decoy Mus musculus database obtained from Uniprot, containing 63.686 protein entries and 286 commonly observed contaminants. Peptide, protein and modi cation site identi cations were reported at a false discovery rate (FDR) of 0.01, estimated by the target/decoy approach. The LFQ (Label-Free Quanti cation) algorithm was enabled, as well as match between runs and LFQ protein intensities were used for relative protein quanti cation. Data analysis was performed using the STRING v11 database 19
For analysis of spiral artery remodeling at E10.5, uterine arteries were ligated by dental oss and the uterus was removed, placed on a polystyrene piece and xed 4.5% formaldehyde (Sigma Aldrich) and para n embedded as described in 65 . 3-5 µm thick sections were stained with H&E and the slides were scanned with the Ventana DP200 (Roche, Basel, Switzerland). Placentas from all animals were analyzed at the midsagittal point, given by the presence of the chorioallantoic attachment. The total vessel and luminal areas of the spiral arteries were measured in the central 2/4 of the decidua basalis 65 . The 5 spiral arteries with the largest and roundest lumen in three consecutive sections (50um between sections) were used for analysis and the mean was calculated.

Statistical analysis
Statistical analysis was done using GraphPad Prism 5.0 (GraphPad Software, La Jolla, CA). Data were analyzed for Gaussian distribution using D`Agostino and Pearson omnibus normality test. Unpaired and normally distributed data were analyzed using the unpaired t-test, unpaired and not normally distributed data were evaluated using the Mann-Whitney test. Paired and normally distributed data were analyzed using the paired t-test and paired and not normally distributed data were analyzed using the Wilcoxon matched pairs signed rank test. A p-value <0.05 was considered as statistically signi cant. Increased abortion rate in Qa-2-mice Wildtype (WT) and Qa-2 de cient mice (Qa2-) were term-bred and the day when a vaginal plug was detected was de ned as day E0.5. Mice delivered spontaneously and litter size was determined or mice were euthanized at E18.5 and uteri containing feto-placental units were removed and inspected. Total implantation sides and resorbing units were counted and fetuses were weighed. (A) Litter size of WT (n=23) and Qa2-mice(n=30). (B) Representative uteri containing feto placental units from WT and Qa2-mice at gestational day E18.5. Arrows show resorbing units. (C) Abortion rate (percentage of resorbed fetuses per litter) of WT (n=39) and Qa2-mice (n=28) at E18.5. (D)

Figure 2
Transcriptome analyses of uteri from wildtype and Qa-2 de cient mice Uteri from wildtype (WT, n=2) and Qa-2 de cient mice (Qa-2-, n=2) were collected at E18.5 of pregnancy and total RNA was isolated and sequenced. (A) Selected biological processes enriched in WT uteri in comparison to Qa2-uteri. (B) Tables of single genes upregulated in WT uteri in comparison to Qa2-uteri assigned to the biological processes "immune system processes" (blue) and "circulatory system development" (red).    Expression of Qa-2 on MDSC is regulated by estrogen (A) Representative pseudocolor plots for Gr-1 versus Qa-2 from splenocytes from non-pregnant (np) and pregnant (p) wildtype (WT) animals showing