Single cell profiling at the maternal–fetal interface reveals a deficiency of PD-L1+ non-immune cells in human spontaneous preterm labor – Scientific Reports

Overview

The main goal of this study was to use mass cytometry to characterize maternal and fetal cell types at the MFI by focusing on decidua tissues isolated from human term and preterm placentas. To understand the biological roles of cells at the MFI and discover distinct cell-type features as potential biomarkers of labor at term and preterm, we compared cell profiles between: (1) spontaneous term laboring (TL) and spontaneous preterm laboring (PL), (2) spontaneous TL and term non-laboring (TNL), (3) spontaneous PL and preterm non-laboring (PNL), and (4) TNL and PNL. An overview of our protocol for profiling single cells at the MFI is shown in Fig. 1A.

We included singleton pregnancies from women who were 18 years or older and who self-identified as Black. Membranes from 55 placentas were processed; 52 passed quality control after mass cytometry and were included in subsequent analyses. The final sample included placentas from 27 term and 25 preterm pregnancies; the 14 PNL placentas were from pregnancies with preeclampsia or eclampsia. The characteristics of the 55 women are shown in Table 1.

To enhance differences between the two groups, we selected term pregnancies from among women who delivered ≥ 38 + 0 weeks gestation and preterm pregnancies from among women who delivered between 22 + 0 to 34 + 6 weeks gestation We defined spontaneous labor as ≥ 6 regular uterine contractions in 60 min and either: (i) cervix ≥ 2 cm dilated or (ii) cervix ≥ 75% effaced. We excluded pregnancies with a fetal anomaly or a positive test for HIV, ZIKA, Hepatitis B, Hepatitis C or herpes. Based on pathological examinations of all preterm placentas, one was diagnosed with chorioamnionitis. However, the tissue sample from that placenta did not have increased neutrophils (a marker of infection) and did not appear as an outlier for any of the cells examined. Therefore, we retained this sample in our analyses.

A human decidual cell atlas

To address the main goal of our study, we first identified the cell types present in decidual tissues from the four study groups using a mass cytometry panel of 32-metal isotope-tagged antibodies (Supplementary Table S1). This panel contained validated commercial antibodies that assess polarization of known immune populations that are also present in the periphery¹⁷ as well as antibodies that identify maternal and fetal non-immune cell populations at the MFI. We used Hierarchical Stochastic Neighbor Embedding (HSNE) for analyses of the hyperspectral satellite imaging data, as implemented in the Cytosplore platform, to cluster the data with a stepwise increase to the single-cell level (Fig. 1B–D)^18,19. Abundances of cells in each subset for each of the four study groups and summary statistics for the analyses reported below are shown in Supplementary Tables 3–5.

We first divided the data into the immune (CD45⁺) and non-immune (CD45^–) compartments (Figs. 1B and 2), and then applied additional filters for deeper clustering. As the first filter, we used 14 cell-surface markers to define six major immune cell types and 11 cell-surface markers to define six non-immune cell populations (Fig. 1C; Supplementary Fig. 1).

Twelve distinct cell lineages were identified following the first filter. Cells were clustered as CD45⁺ (immune) cells (A) and CD45^– (non-immune) cells (B). Six major immune populations (A) and six non-immune cell populations (B) were color coded by Cytosplore^+HSNE. hSNE visualization of the CD45 + (C) and CD45- cells (D) as a density plot for each of the four groups is shown. Each point is an individual subject. The cells were grouped based on the Euclidian distances of marker expression and populations were assigned based on the markers. The representative cells were found by a weighted k-nearest neighbor (kNN) graph and served as landmarks. The size of each point indicates the area of influence (AoI) of the landmarks. Scatterplots of marker expression are shown in Supplementary Fig. S1. The ratios of CD45pos to CD45neg cells by study group (E).

We next used 18, 11, and 5 cell-surface markers as a second filter for clustering of T cells, macrophages/monocytes, and granulocytes subsets, respectively (Fig. 1D, Supplementary Table 2). We identified 31 cell populations in human decidual tissues, with 2 broad subsets of CD45 + and CD45- cells comprised of 6 immune cells and 6 non-immune cells after the first filter (Fig. 2A,B). The proportions of CD45 + and CD45- cells for the four study groups are shown in Fig. 2C,D and Supplementary Fig. 2. Whereas the ratios of CD45 + cells to CD45- cells were similar in TL (median = 0.144), TNL (median = 0.142), and PNL (median = 0.130) (Robust Rank-Order test, P > 0.95 in pairwise comparisons), the ratio was significantly higher in PL (median = 0.230; Robust Rank-Order test, P = 0.0014, 0.0001, 0.028 compared TL, TNL, and PNL, respectively) (Fig. 2E). These combined data suggest that fluctuations in the ratios of CD45 + to CD45 − cells are not a feature of labor vs. non-labor at term or between non-labor at term or preterm. Rather, the increased ratio in PTL compared to the other three groups suggests that spontaneous preterm labor may be characterized by shifts in cell composition that results from increased proportions of immune cells and decreased proportions of non-immune cells at the MFI.

To specifically characterize the immune cell lineages, 3,000 CD45⁺ cells from each sample were randomly extracted and pooled together for cell-type profiling (156,000 CD45⁺ cells in total). Using different combinations of immune-cell surface markers (Supplementary Table S2), we identified six major innate and adaptive immune populations at the MFI: CD3⁺ T cells (pan T cells), macrophage/monocytes, NK cells, dendritic cells (mDCs), granulocytes and B cells (Fig. 2A). Macrophages/monocytes and T cells were the most abundant cell types, followed by NK cells, DCs, granulocytes, and B cells. The NK cell, DCs, and B cell populations were not further filtered. Two granulocyte subsets were identified (eosinophils and neutrophils) but both were sparse and not further investigated (Supplementary Fig. 3). The distributions of the abundances of these six cell types among the four groups are shown in Fig. 2C. None of the major immune cell types defined on the first filter differed among the four study groups after considered the potential effect of fetal sex on the relative abundances, except possibly granulocytes (Fig. 3; Table 1; Supplementary Results and Supplementary Fig. 4). The increased abundances of T cells and NK cells in pregnancies with male infants were seen in both preterm and term, as well as laboring and in non-laboring, pregnancies (Supplementary Table 6).

Six immune cell clusters at the maternal–fetal interface. (A) Violin plots showing cell abundance by infant sex. P-values are shown for pairwise comparisons with P ≤ 0.05. Each point is an individual; red denotes a female infant and blue a male infant. (B) Boxplots showing relative abundances of the six immune cell populations in the four study groups. The horizontal lines show the median percent of total immune cells, and the vertical lines show the interquartile ranges. Pairwise differences between groups at nominal P ≤ 0.05 are shown. Test statistics (Robust Rank-Order test) with P-values and descriptive statistics are shown in Supplementary Tables S3 and S5, respectively.

Using the cell-surface markers defined on the second filter of immune cells (Fig. 1D), we identified five macrophage/monocyte and 10 T cell subsets (Fig. 4; Supplementary Fig. 5; Supplementary Tables 7, 8). Cells in macrophage/monocyte cluster 2 were more abundant in PL compared to PNL (z = 2.63, P = 0.0083) and in TNL compared to PNL (z = 3.03, P = 0.0024), while cells in cluster 5 were more abundant in TL compared to TNL (z = 2.02; P = 0.043) (Supplementary Fig. 5). The macrophage/monocyte clusters 2 and 5 cells both had an immune-suppressive phenotype (Fig. 4A), defined as PD-L1⁺CXCR3^highCCR4^high and PD-L1⁺ CXCR3^lowCCR4^low, respectively, suggesting an increase in these cell types during labor. The observation of increased proportions of myeloid cells in TL compared to TNL is consistent with studies in mice^20,21 and in women²². The T cells are discussed in more detail below.

Deep profiling of T cells and macrophages/monocytes at the maternal–fetal interface. 10 clusters of T cells based on the expression of 19 T-cell surface markers (see Figs. 1, 2) (A) and five clusters of macrophages/monocytes based on the expression of 11 cell surface markers (B). Each point represents a cell. The heat map views (center panels) are ordered by similarity as shown by the dendrogram above the heatmaps. The tables (right panels) describe the cell-subtype assignments and their potential cellular functions. Descriptive statistics of these data are shown in Supplementary Table S5.

The relationships between macrophage/monocyte cell clusters and gestational age, fetal sex, and labor are described in Supplementary Results, Supplementary Figs. 6, 7, and Supplementary Tables 3 and 5.

T cell subsets at the MFI differ between term and preterm woman

Ten phenotypically distinct subsets of T cells were present at the MFI (Fig. 4B; Supplementary Table 7). None of the 10 subsets differed by gestational age (Supplementary Fig. 8), but the abundances of T cells in clusters 1, 3, 4, 5, 6, 9 and 10 were higher in pregnancies with male compared to female infants (Supplementary Fig. 9). Cells in six clusters were less abundant in TL compared to TNL at P ≤ 0.05: clusters 1 (P = 0.0041), 2 (P = 0.0098), 3 (P = 0.042), 7 (P = 7.43 × 10^–9), 9 (P = 0.036), 10 (P = 0.0013) (Fig. 5; Supplementary Tables 3 and 5). However, these differences may be due to sex ratio imbalances between TL and TNL (Table 1, Supplementary Results, Supplementary Tables 3 and 5). The phenotype of cluster 7 is one of activated T cells that express the programmed cell death (PD-1) receptor. PD-1 binds to its receptor, PD-L1, to suppress adaptive immune responses and maintain immune balance²³. The lower abundance of these cells in TL, and the relative higher abundance of these cells in PL compared to TL, raised the possibility that cells at the MFI with inhibitory signals from PD-L1 may also differ between term and preterm labor. Thus, we next investigated the abundances of PD-L1⁺ non-immune cells in this niche.

Distributions of 10 T-cell subtypes at the maternal–fetal interface in each study group. In each boxplot, the horizontal lines show the median percent of total immune cells and the vertical lines show the interquartile ranges. Pairwise differences between four groups at nominal P ≤ 0.05 are shown (Robust Rank-Order test). Summary and descriptive statistics of these data are shown in Supplementary Tables S3 and S5, respectively. T cell abundances by sex and study group are shown in Supplementary Fig. 8.

CD45^– non-immune cell subsets at the MFI in term and preterm pregnancies

To explore the non-immune cell compartment, 10,000 CD45^– cells from each sample were randomly extracted and pooled together for cell-type profiling (520,000 CD45^– cells in total). Six major non-immune cell populations were clustered by 11 markers, including CD10, HLA-G, and PD-L1 (Fig. 2B), defining two clusters of decidual stromal cells (DSCs) and two clusters of EVTs. Two additional unknown cell clusters were defined as CD10 negative and HLA-G negative, and may represent the novel endothelial cells described in scRNA-seq studies¹⁵. Of the six major non-immune cell clusters, cells in EVT cluster 1 (CD45^–CD10⁺HLA-G⁺CCR4,6⁺CXCR3⁺CCR5^lowCD66a⁺PD-L1⁺Fas-L⁺) were less abundant in PL compared to TL (P = 0.042) and cells in undefined cluster 2 (CD45^–CD10^–HLA-G^–CXCR3^–CCR4,5,6^–CD66a⁺PD-L1^–Fas-L⁺) were more abundant in PL compared to TL (P = 0.054). (Fig. 6A, Supplementary Tables 4, 5). These results suggested that PD-L1 + non-immune cells at the MFI are deficient in PL pregnancies compared to PL.

PD-L1⁺ non-immune cells at the maternal–fetal interface. (A) Box plots showing the distribution of six clusters of non-immune cells. The horizontal lines show the median percent of total non-immune cells and the vertical lines show the interquartile ranges. Pairwise differences between groups at nominal P ≤ 0.05 are shown (Robust Rank-Order test). (B) hSNE visualization of PD-L1⁺ non-immune cell lineages as a density plot for each of the four groups. Each point is an individual subject. (C) Box plots of PD-L1⁺ non-immune cells by study group. The horizontal lines show the median percent of total non-immune cells, and the vertical lines show the interquartile ranges. Pairwise differences between groups at P ≤ 0.05 are shown (Robust Rank-Order test). Summary and descriptive statistics of these data are shown in Supplementary Tables S4 and S5, respectively.

The relationships between non-immune cell types, infant sex, gestational age, and labor are described in Supplementary Results, Supplementary Figs. 10 and 11, and Supplementary Tables 4, 5.

PD-L1⁺ non-immune cells are less abundant in preterm labor

The significantly lower abundances of HLA-DR⁺PD-1⁺ CD8 T cells (T cell cluster 7) in TL compared to TNL (P = 7.43 × 10^–9) and a trend toward lower abundances of these cells in TL compared to PL (P = 0.069), as well as more abundant PD-L1⁺ EVT cluster 1 cells in TL compared to PL (P = 0.042) suggested a potential role of the PD-1/PD-L1 pathway in preterm birth. To further explore this possibility, we focused on the two clusters of PD-L1⁺ EVTs and one cluster of PD-L1⁺ DSCs among the non-immune cells (Fig. 6). The PD-L1⁺ cells were similar in abundance between TL and TNL (P = 0.80) and between PL and PNL (P = 0.93) (Fig. 6A; Supplementary Tables 4, 5) but were more abundant in TL compared to PL (P = 0.0070), consistent with the observation of fewer PD-1⁺ CD8 T cells in TL compared to PL (Fig. 5).

Cells from all three PD-L1⁺ clusters (DSC cluster 2, EVT cluster 1, and EVT cluster 2) contributed to these differences (Fig. 6B). Further examination of the relationship between PD-L1⁺ non-immune cell abundances and gestational age (Supplementary Fig. 12) suggested that the abundances of PD-L1⁺ non-immune cells at the MFI are not influenced by gestational age per se but rather may be a feature of labor at term (Supplementary Methods). Infant sex was not likely contributing to the differences in abundances of PD-LI⁺ non-immune cells between TL and PL (Supplementary Methods and Supplementary Fig. 12). Together with the previous observations of increased abundance of PD-1⁺ T cells (T cell cluster 7) in PL compared to TL (Supplementary Table 5), these data suggested that preterm labor may be associated with a lack of suppression of activated T cells at the MFI. However, whether this is due to intrinsic differences in the regulation of PD-LI⁺ cells and causally related to preterm birth cannot be determined in this observational study. Therefore, we used a cell culture model of decidua-derived mesenchymal stromal cells from term and preterm placentas to further investigate this possibility.

Decreased transcription of PD-L1 in cultured stromal cells from preterm placentas

We isolated stromal cells from placental membranes and established primary cell lines from three spontaneous TL and three spontaneous PL placentas that were not included in the mass cytometry studies. To assess transcriptional differences in CD274, the gene encoding PD-L1, we purified RNA from cells cultured in three conditions: media alone, media plus cAMP and MPA (medroxyprogesterone acetate) to induce decidualization and mimic pregnancy conditions, and decidualization media plus trophoblast conditioned media (TCM) that contains fetal-derived signaling molecules secreted by trophoblasts (Materials and methods). After standard processing and QC (Materials and methods and Sakabe et al.²⁴), RNA sequencing data from the three replicates of each sample were pooled. Transcript levels of CD274 were strikingly lower in decidua cells from all three preterm placentas compared to decidua cells from three term placentas when cultured in media alone (Wald test P = 3.0 × 10^–4) (Fig. 7). Transcript levels increased in all samples in response to decidualization treatment, reducing differences between preterm and term cells (Wald test P = 0.079), but levels further increased in response to TCM in the term cells only, restoring differences between cells from PTL and TL (Wald test P = 0.020). Other checkpoint genes were either very lowly expressed (TIM3, LAG3; < 1 transcript per 1000 in all samples) or not expressed (CTLA4; 0 transcripts per 1000 in nearly all samples) in all three conditions (data not shown). These results suggest intrinsic differences in the constitutive expression of CD274 in stromal cells and in the regulation of CD274 by TCM signaling molecules in decidualized cells from preterm compared to term placentas.

Transcript levels of the gene encoding PD-L1, CD274, in cultured mesenchymal stromal cells from term and preterm placentas. CD274 transcript levels were measured in stromal cells from three term and three preterm placentas after culturing in media alone, in decidualization media (cAMP/MPA), and in decidualization media with trophoblast conditioned media (cAMP/MPA + TCM) (Materials and methods). P-values from Wald test.