The CD94/NKG2A inhibitory receptor educates uterine NK cells to optimize pregnancy outcomes in humans and mice

NKG2A educates mouse uNK cells

Most murine and human peripheral NK cells display a bi-modal expression pattern for NKG2A. We show here that both splenic and uterine mouse NK cells expressed NKG2A but neither expressed NKG2C or NKG2E at embryonic day (E) 9.5, because cells from NKG2A-deficient Klrc1^−/− mice on a C57BL/6 (B6) background did not stain with an antibody that reacts with NKG2A, NKG2C, and NKG2E (Figure S1A), consistent with published data (Rapaport et al., 2015). NKG2A can also be expressed by activated CD8⁺ T cells, but at E9.5, there was no expression of NKG2A on T cells in the uterus or the spleen of WT B6 mice (Figure S1B). Therefore, in B6 mice, NKG2A is solely expressed on NK cells and not accompanied by other NKG2 receptors. In the early-gestation mouse uterus, there are three subsets of innate lymphoid cells (ILCs): conventional NK cells (cNKs), uterine ILC1 (uILC1), and tissue-resident uNK (trNK) cells (Doisne et al., 2015; Filipovic et al., 2018). We confirmed that all three subsets express NKG2A in early gestation, from ∼40% in cNK and trNK to ∼60% in uILC1 (Figure S1C). For simplicity, in this paper, we refer to these three subsets collectively as uNK cells. Both the NKG2A⁺ and NKG2A⁻ uNK subsets expressed the cell-surface markers of education (DNAM-1) and maturation (KLRG1) (Figure S1D). The NKG2A ligand Qa-1 (the mouse ortholog of HLA-E) is not found on mouse trophoblast (Madeja et al., 2011), and we show here that it was expressed by uterine CD45⁺ leucocytes, including T lymphocytes, B lymphocytes, and NK cells, in early pregnancy (Figures S1E−S1F). In mice, interferon gamma (IFN-γ) produced by NK cells is the key cytokine involved in remodeling the uterine spiral arteries (Ashkar et al., 2000), a necessary step for optimal placentation. Production of IFN-γ through stimulation of activating receptors is also a standard functional readout of both human and mouse NK cell education. To quantify the contribution of NKG2A to uNK cell education in mice, we gated on all uNK cells and compared the percentage of IFN-γ⁺ cells upon NK1.1 stimulation within NKG2A⁺ and NKG2A⁻ subsets in B6 dams (Figures 1A and 1B). This was greater within NKG2A⁺ uNK cells than within NKG2A⁻ uNK cells (Figure 1C), although mean IFN-γ expression per cell was similar (data not shown). Because NKG2A synergizes with specific inhibitory Ly49 receptors to educate mouse peripheral NK cells (Zhang et al., 2019), Ly49 might contribute to the superior functional competence of NKG2A⁺ uNK cells. To test this, we measured the percentage of cells expressing the self-specific educating Ly49I receptor in B6 mice (Fernandez et al., 2005). Because we found no significant difference in Ly49I expression between NKG2A⁺ and NKG2A⁻ subsets, we could rule out the potentially confounding effect of Ly49I education in NKG2A⁺ uNK cells (Figure 1C). DNAM-1 expression correlates with peripheral NK cell education (Wagner et al., 2017), and here we find that it was expressed in greater percentages of NKG2A⁺ compared with NKG2A⁻ uNK cells (Figure 1C). The greater functional competence of NKG2A⁺ uNK cells could be a secondary effect of advanced maturity of NKG2A⁺ uNK cells. KLRG1 and CD11b correlate with peripheral NK cell maturation. However, KLRG1 expression was found in similar percentages of NKG2A⁺ and NKG2A⁻ uNK cells (Figure 1C) and CD11b was higher for NKG2A⁻ NK cells (Figure 1C). This suggests that NKG2A⁺ uNK cells are not more mature than NKG2A⁻ uNK cells, as shown previously in human NK cells (Björkström et al., 2010); therefore, the superior functional response of NKG2A⁺ cells more likely results from education than from differences in maturation. This dataset shows that NKG2A-educated uNK cells are more functionally competent than NKG2A⁻ uNK cells in response to NK1.1 crosslinking.

NKG2A is required for peripheral NK cell function

NK cells from NKG2A-deficient Klrc1^−/− mice display normal maturation, cell surface receptor repertoires, and cellular development (Rapaport et al., 2015), and we confirmed here that both uNK cells and CD8⁺ T cells developed in normal numbers in both uterus and spleen of Klrc1^−/− dams (Figures S1G and S1H). To confirm the contribution of NKG2A to peripheral NK cell education, we used a standard in vivo assay based on rejection of MHC-deficient B2m^−/− hematopoietic cells (Höglund and Brodin, 2010). Using this missing-self response assay (Figure 1D), we showed that Klrc1^−/− mice were almost 50% less efficient at rejecting MHC-deficient cells than B6 mice (Figure 1E), confirming the key role of NKG2A in peripheral NK cell education (Zhang et al., 2019). Altogether, these data indicate that NKG2A modulates NK cell education both systemically and in the uterus during pregnancy, demonstrating the suitability of NKG2A-deficient Klrc1^−/− mice to model NK cell education in pregnancy.

NKG2A is required for uterine vascular adaptation to pregnancy

The pattern of expression of both the receptor and its ligand in B6 dams (Figure S1) is ideally suited to test directly the hypothesis that maternal, not fetal, Qa-1 educates uNK cells through NKG2A and that NK cell education affects the outcome of pregnancy. In humans, maternal uterine spiral arteries undergo transformation in early pregnancy through the destruction of smooth muscle media by trophoblast cells to allow optimal fetal nourishment (Pijnenborg et al., 2006). In mice, transformation of the arteries relies primarily on IFN-γ produced by uNK cells (Ashkar et al., 2000). Because we observed reduced IFN-γ in B6 NKG2A⁻ uNK cells, we hypothesized that lack of uterine vascular adaptation would be suboptimal in the absence of NKG2A (Figure 2A). Klrc1^−/− dams displayed increased vascular wall area, normal lumen area, and more informatively, a two-fold greater size of the vascular wall relative to the lumen (Figure 2B; Figures S2A and S2B), with thicker smooth muscle actin (Figure 2C). This demonstrates that NKG2A is required for normal vascular remodeling, because the uterine arteries of Klrc1^−/− dams retain smooth muscle actin and thicker vessel walls relative to their lumen.

NKG2A is required for maternal and fetal hemodynamic changes during pregnancy

We next assessed whether this vascular maladaptation leads to changes in upstream pressure indices in the maternal and fetal circulation. To do this, we compared vascular hemodynamics in dams and fetuses of (Klrc1^−/− ♀ x B6 ♂) and (B6 ♀ x Klrc1^−/− ♂) mating to directly compare genetically matched fetuses developing in either NKG2A-deficient or NKG2A-sufficient uteri. Because the fetuses were heterozygous in both sets of pregnancies, whereas the dams were either Klrc1^−/− or WT, this mating strategy allowed us to isolate the effect of the maternal genotype from that of the fetuses (Figure 2D). Using ultra-high-frequency microultrasound Doppler, we measured blood flow velocity in dams at E14.5 (left panel in Figure 2E)—by which time dilatation of the uterine artery because of spiral artery remodeling is expected (Rennie et al., 2016)—and in the umbilical artery in the same dams at E14.5 (right panel in Figure 2E). Although we observed no difference in baseline uterine artery resistance index (UARI) in a separate group of Klrc1^−/− and B6 virgin females (Figure S2C), end-diastolic blood flow velocity (UA EDV) was significantly reduced in pregnant Klrc1^−/− dams (Figure 2F). This contributed to significantly increased UARI (Figure 2G). Although both velocity and resistance indices appeared only mildly affected in Klrc1^−/− dams, the pregnancy-induced UARI reduction typically displayed by B6 dams did not occur in NKG2A-deficient dams (Figure 2H). Moreover, fetuses from Klrc1^−/− dams displayed abnormal umbilical blood flow (Figure 2I). This was characterized by a reduction in peak systolic velocity, because end-diastolic velocity is absent at this gestational age. These datasets show that NKG2A is required for normal hemodynamics in both maternal and fetal circulation and that these defects are determined by the maternal, not the fetal, genotype.

NKG2A is required for optimal fetal growth

Low birthweight affects the health of human offspring even in the range of 5%–10% reduction (Juárez and Merlo, 2013), and dams with absent or hypofunctional uNK cells generate pups with weight reduction in the same range (Kieckbusch et al., 2014). To test whether the observed aberrations in both maternal and fetal hemodynamics caused by the absence of maternal NKG2A result in suboptimal fetal growth by the end of pregnancy (E18.5), we weighed 98 fetuses from 13 NKG2A-deficient Klrc1^−/− dams and 70 fetuses from 10 NKG2A-sufficient B6 dams. Both types of dams were mated with males of the reciprocal genotype. Fetuses from these two sets of dams were compared with a population control of 131 fetuses from 17 B6 dams mated with B6 males. Fetal weight was significantly reduced in NKG2A-deficient dams. The mean weight of fetuses from Klrc1^−/− dams displayed significant reduction compared with both the genotype control (1.10 ± 0.08 versus 1.15 ± 0.08, 4.3% reduction, p = 0.036) and the population control (1.16 ± 0.07, 5.2% reduction, p = 0.004) (Figure 3A; Table 1), suggesting that reduced fetal growth was driven by maternal NKG2A deficiency. In line with the results of other mouse strains with absent or hypofunctional NK cells, litter size was not affected in Klrc1^−/− dams (Figure S3A). The observed FGR in Klrc1^−/− dams was not affected by the size of their placentae, which showed normal weight (Figure S3B). Despite the reduction at term, fetal weight was normal in Klrc1^−/− dams at E15.5, consistent with the insufficient remodeling of the uterine arteries making the placenta of Klrc1^−/− unfit to meet the exponential fetal-growth demands typical of late pregnancy (Figure S3C). Fetuses developing in NKG2A-deficient dams were not only smaller overall but were also twice as likely to be classified as small for gestational age (SGA, defined as <10^th percentile of (B6 ♀ x B6 ♂) pregnancies) (Figure 3B). As many as 27/98 fetuses (28%) were SGA in NKG2A-deficient dams, whereas only 10/70 (14%) were SGA in the genotype control dams, which did not significantly differ from the 13/131 SGA fetuses (10%) of the population control (Figure 3B). The results show that maternal NKG2A is required for optimal fetal growth, and the extent of the growth reduction in dams lacking NKG2A is comparable to that of other strains of mice with absent or hypofunctional NK cells (Barber and Pollard, 2003; Ashkar et al., 2003; Kieckbusch et al., 2014; Boulenouar et al., 2016).

NKG2A antibody blocking in WT dams does not phenocopy genetic NKG2A ablation

To distinguish between the effects of constitutive and those of acute absence of NKG2A signaling, we blocked NKG2A in B6 dams by administration of one dose of the blocking monoclonal antibody 20d5 (Vance et al., 1999; André et al., 2018) at E6.5 and assessed remodeling of spiral arteries at E9.5, as in Figures 2A and 2B. In another group of 19 B6 dams, we blocked NKG2A with two doses at E6.5 and E9.5 in 11 dams and assessed litter sizes and fetal and placental weights at E18.5, as in Figure 3A, comparing them with those of 8 isotype-control treated dams. We chose these two time points because the E6.5–E9.5 window is when uNK cells may exert their maximal function and when both uterine vasculature remodeling and placenta formation begin. Immune checkpoint blockade can result in fetal loss, as shown in dams treated with anti-PDL1, anti-CTLA-4, and anti-Tim-3 (Guleria et al., 2005; Xu et al., 2017). Although 2/11 dams treated with anti-NKG2A produced smaller litter sizes, the remaining 9 dams produced litters of normal size (Table 2; Figure S3D), indicating that the anti-NKG2A antibody did not obviously lead to embryonic lethality in this context. Fluorescence-activated cell sorting (FACS) analysis after the injections indicated the NKG2A epitope was 100% saturated with the administered antibody in both uNK and splenic NK cells up to three days after injection (Figure S3E; data not shown), suggesting NKG2A blockade for an appropriate duration, that is, at least up to E12.5, when remodeling of uterine arteries should be well under way. We then compared spiral artery remodeling in these NKG2A-blocked B6 dams with remodeling in NKG2A-deficient dams. Although vessel wall thickness was higher in the anti-NKG2A-treated group, the luminal area of spiral arteries was normal and the ratio between the areas of the wall and the lumen was not significantly different (Figures S3F–S3H). Thus, NKG2A blockade does not phenocopy the defective vascular remodeling of NKG2A-deficient dams (Figure 3C). Furthermore, the 71 fetuses from the 8 isotype-treated B6 dams and the 78 fetuses from the 11 anti-NKG2A-treated B6 dams displayed equivalent fetal weights at E18.5. We noted a non-significant 2.7% increase in mean fetal weight from the anti-NKG2A-treated dams (Figure 3D; Table 2). These results suggest that acute blocking of NKG2A does not phenocopy the effect of genetic NKG2A ablation on fetal weight and spiral arteries. In WT mice, before the blockade, NKG2A signaling had been functional and able to educate uNK cells in steady state, thus suggesting that NKG2A education of uNK cells before pregnancy may be sufficient to provide adequate uNK cell function during gestation, even when NKG2A signaling is acutely blocked during pregnancy.

Asymmetric fetal growth and brain sparing in pups of NKG2A-deficient dams

Asymmetric fetal growth occurs in complicated pregnancies with inadequate maternal resources when the brain is preferentially spared from growth restriction at the expense of other fetal organs (Sharma et al., 2016). Fetal brain sparing is linked to cognitive and behavioral abnormalities in children (Tolsa et al., 2004; Eixarch et al., 2008; Oros et al., 2010) and is observed in mouse models (Cahill et al., 2014). Although there is variation in the growth of individual fetuses in normal litters, as shown in Figure 3A, SGA fetuses developing in normal pregnancies are expected to be healthy and grow symmetrically, similar to healthy human SGA babies. We hypothesized that the altered umbilical blood flow found at E14.5 in fetuses from Klrc1^−/− dams mated with B6 males (Figure 2I) was associated with subsequent asymmetric growth pattern in the smallest fetuses measured at the end of pregnancy (E18.5). To do this, we selected the smallest fetuses from each of 4 litters of Klrc1^−/− and 5 litters of B6 dams. These were compared with the average-weight fetuses within the litters of the same types of dams, as shown in Figure 4A. Fetal weights of the selected groups are indicated in Figure S4A. To accurately assess fetal brain development and symmetric body growth, we used micro-computed tomography (micro-CT) and compared ratios between brain volume and femur length (Figure 4B). Small fetuses from Klrc1^−/− dams had higher ratios than their average-weight littermates, whereas no difference was found between average-weight and small fetuses from B6 dams (Figure 4C). In other words, growth symmetry was found across all fetuses in B6 dams, in which both average-weight and small fetuses had brain volumes proportional to their femur length (Figure S4B). In contrast, the brains of small fetuses from Klrc1^−/− dams were disproportionally larger compared with their shorter femurs (Figure S4B). These results show that in a uterus with NKG2A-deficient uNK cells, the smallest fetuses display asymmetric growth and brain sparing and the origin of both was the maternal, not the fetal, genotype.

NKG2A regulates the placental transcriptome

We aimed to determine whether the altered uterine and umbilical blood flow of NKG2A-deficient dams was associated with aberrant placental gene expression. We compared RNA expression in genetically matched placentae of (Klrc1^−/− ♀ x B6 ♂) and (B6 ♀ x Klrc1^−/− ♂) mating at E15.5 (n = 5 litters, 10 fetuses in each group). To isolate differences resulting from the presence or absence of maternal NKG2A, we matched fetal groups not only by gestational age, genotype, and sex composition but also by weight. Thus, we selected the placentae of the two fetuses closest to the mean fetal weight in each litter. The selection of average-weight fetuses was done to remove bias introduced by randomly selecting those who may be suffering from either FGR or fetal overgrowth. Previously reported mouse placenta housekeeping genes Hprt, Gapdh, Actb, Ubc, Pol2ra, and Ywhaz (Solano et al., 2016) were stably expressed by all placentae, across both groups, suggesting that these housekeeping genes did not bias the comparison (Figure S5). Supporting the robustness of our analysis, we found that all samples were only highly enriched in the gene expression signature titled placenta (data not shown). A list of all processed, filtered gene expression scores can be found in Data S1. The volcano plot in Figure 5A highlights the 19 differentially expressed (DE) genes reaching statistical significance with a false discovery rate (FDR) < 0.05, of which 14 were overexpressed and 5 were downregulated in placentae of Klrc1^−/− dams. Table S1 lists these 19 DE genes and highlights the top 10 DE genes with FDR < 0.01. Eight of these 10 genes were overexpressed in the placentae of Klrc1^−/− dams: ribosomal proteins Chchd1, Rpl22, and Rps27; the Sf3b unit of the ribonucleotide complex of spliceosomes; the Pfdn4 chaperone that helps correct folding of nascent polypeptides; mitochondrial protein transporters Timm8b and Tomm5; and the chaperone of mitochondrial cytochrome c oxidase Pet100. The two DE genes with FDR < 0.01 downregulated in placentae of Klrc1^−/− are the tumor suppressor gene and regulator of nuclear factor κB (NF-κB) signaling Ldoc1 and the transferase involved in purine biosynthesis Atic. Using these top 10 DE genes with FDR < 0.01, we performed a gene functional annotation analysis in Metascape to discover enriched pathways (Zhou et al., 2019; Figure 5B). The gene cluster R-MMU-72766 Translation was the most enriched in the analysis (Log10(p) = −3.86) and included Rpl22, Rps27, and Chchd1. Rpl22, Rps27, and Sf3b6 contribute to mRNA processing (Log10(p) = −2.96) and Metabolism of RNA (Log10(p) = −2.71), whereas Rpl22, Chchd1, and Atic contribute to Amide biosynthetic process (Log10(p) = −2.28) (Figure 5B). These results suggest that RNA biology, protein synthesis, and translation are affected in placentae of dams whose uterus had NKG2A-deficient uNK cells.

Lower risk of pre-eclampsia in women genetically programmed to NK cell education via NKG2A

Our results in mice suggest that NKG2A drives uNK cell education and that lack of maternal NKG2A in dams, but not in fetuses, affects utero-placental arterial remodeling, hemodynamics, and placental gene expression with downstream effects on fetal growth and brain development. These are features of pre-eclampsia, a heterogeneous, polygenic, and multifactorial syndrome that affects up to 1/10 pregnancies in certain human populations. We therefore set out to investigate whether women carrying the −21T allotypes of HLA-B that do not favor NKG2A education in humans are at greater risk of developing pre-eclampsia. To do this, we assessed the association with pre-eclampsia of the SNP rs1050458 on chromosome 6 (6p21), coding the −21M or −21T HLA-B dimorphism in a genome-wide meta-analysis of 9,515 maternal pre-eclampsia cases and 157,719 controls from multiple cohorts (Steinthorsdottir et al., 2020). The analysis revealed that the G allele coding the −21T HLA-B variant was associated with increased pre-eclampsia risk (p = 0.02). Although this association was not apparent in a smaller cohort of Central Asian mothers (p = 0.44), it was present in 7,219 European cases and 155,660 controls (p = 0.005, odds ratio [OR] = 1.07, 95% confidence interval [CI] 1.02–1.12) (Table 3). The data show that the G allele conferred a 7% risk in Europeans (95% confidence interval, 2%–12%). This result suggests that women genetically programmed to educate NK cells through HLA-B leader peptides that allow HLA-E to engage NKG2A are at lower risk of pre-eclampsia.

Limitations of study

Although we thoroughly investigated the impact of NKG2A genetic ablation in dams and their fetuses, we did not directly compare uNK cell function in vitro from B6 and Klrc1^−/− mice. This approach is ongoing. The results of placental gene expression should be validated to further investigate the observed gene pathway dysregulation. We also did not assess the role of the HLA-B −21 dimorphism on human uNK cell function in vitro. This work is ongoing and could lead to a greater understanding of the mechanisms influencing our observations in the reported GWASs. Finally, the genetic linkage of −21M dimorphism with pre-eclampsia was found in European populations, not the Asian cohort. The reason for this is currently unclear, and the linkage should be tested in other populations. Nevertheless, preliminary work that suggests an interaction between NKG2A and KIR genes may influence disease risk.

Key resources table

Resource availability

Experimental model and subject details

Method details

Quantification and statistical analysis

In general, Shapiro-Wilk test for normality (alpha = 0.05) was applied to all datasets except RNA-seq and fetal and placental weight. For comparison of means across two groups a parametric test was used if both groups passed normality testing. If either or both groups failed normality testing, a Mann-Whitney test was used. For multiple group comparisons, a one-way ANOVA was used for normally distributed data with post hoc t tests corrected for multiple comparisons. All p values are two-sided, unless stated. If a p value was < 0.05, ‘ns’ was stated as not significantly different.