Single cell transcriptional signatures of the human placenta in term and preterm parturition.

More than 135 million births occur each year; yet, the molecular underpinnings of human parturition in gestational tissues, and in particular the placenta, are still poorly understood. The placenta is a complex heterogeneous organ including cells of both maternal and fetal origin, and insults that disrupt the maternal-fetal dialogue could result in adverse pregnancy outcomes such as preterm birth. There is limited knowledge of the cell type composition and transcriptional activity of the placenta and its compartments during physiologic and pathologic parturition. To fill this knowledge gap, we used scRNA-seq to profile the placental villous tree, basal plate, and chorioamniotic membranes of women with or without labor at term and those with preterm labor. Significant differences in cell type composition and transcriptional profiles were found among placental compartments and across study groups. For the first time, two cell types were identified: 1) lymphatic endothelial decidual cells in the chorioamniotic membranes, and 2) non-proliferative interstitial cytotrophoblasts in the placental villi. Maternal macrophages from the chorioamniotic membranes displayed the largest differences in gene expression (e.g. NFKB1) in both processes of labor; yet, specific gene expression changes were also detected in preterm labor. Importantly, several placental scRNA-seq transcriptional signatures were modulated with advancing gestation in the maternal circulation, and specific immune cell type signatures were increased with labor at term (NK-cell and activated T-cell signatures) and with preterm labor (macrophage, monocyte, and activated T-cell signatures). Herein, we provide a catalogue of cell types and transcriptional profiles in the human placenta, shedding light on the molecular underpinnings and non-invasive prediction of the physiologic and pathologic parturition.

Introduction

Parturition is essential for the reproductive success of viviparous species (Romero et al., 2006a); yet, the mechanisms responsible for the onset of labor remain to be elucidated (Norwitz et al., 1999; Norwitz et al., 2015). Understanding human parturition is essential to tackle the challenge of prematurity, which affects 15 million neonates every year (Muglia and Katz, 2010; Blencowe et al., 2012; Romero et al., 2014a). Bulk transcriptomic studies of the cervix (Hassan et al., 2006; Hassan et al., 2007; Hassan et al., 2009; Bollopragada et al., 2009; Dobyns et al., 2015), myometrium (Charpigny et al., 2003; Romero et al., 2014b; Mittal et al., 2010; Mittal et al., 2011; Chan et al., 2014; Stanfield et al., 2019), and chorioamniotic membranes (Haddad et al., 2006; Mittal et al., 2009; Nhan-Chang et al., 2010) revealed that labor is a state of physiological inflammation; however, finding specific pathways implicated in preterm labor still remains an elusive goal. A possible explanation is that gestational tissues, and especially the placenta, are heterogeneous composites of multiple cell types, and elucidating perturbations in the maternal-fetal dialogue requires dissection of the transcriptional activity at the cell type level, which is not possible using bulk analyses. Recent microfluidic and droplet-based technological advances have enabled characterization of gene expression at single-cell resolution (scRNA-seq) (Klein et al., 2015; Macosko et al., 2015). Previous work in humans (Tsang et al., 2017; Pavličev et al., 2017; Vento-Tormo et al., 2018) and mice (Nelson et al., 2016) demonstrated that scRNA-seq can capture the multiple cell types that constitute the placenta and identify their maternal or fetal origin. Such studies showed that single-cell technology can be used to infer communication networks across the different cell types at the maternal-fetal interface (Vento-Tormo et al., 2018). Further, the single-cell-derived placental signatures were detected in the cell-free RNA present in maternal circulation (Tsang et al., 2017), suggesting that non-invasive identification of women with early-onset preeclampsia is feasible. However, these studies included a limited number of samples and did not account for the fact that different pathologies can arise from dysfunction in different placental compartments. In addition, the physiologic and pathologic processes of labor have never been studied at single-cell resolution.

Results and discussion

In this study, a total of 25 scRNA-seq libraries were prepared from three placental compartments: basal plate (BP), placental villous (PV), and chorioamniotic membranes (CAM) (Figure 1A). These were collected from nine women in the following study groups: term no labor (TNL), term in labor (TIL), and preterm labor (PTL). scRNA-seq libraries were prepared with the 10X Chromium system and were processed using the 10X Cell Ranger software, resulting in 79,906 cells being captured and profiled across all samples (Supplementary file 1). We used Seurat (Butler et al., 2018) to normalize expression profiles and identified 19 distinct clusters, which were assigned to cell types based on the expression of previously reported marker genes (Tsang et al., 2017; Pavličev et al., 2017; Vento-Tormo et al., 2018) (see Materials and methods, Figure 1—figure supplement 1 and Supplementary file 2–3). The uniform manifold approximation and projection (UMAP Becht et al., 2019) was used to display these clusters in two dimensions (Figure 1B). With this approach, the local and global topological structure of the clusters is preserved, with subtypes of the major cell lineages (trophoblast, lymphoid, myeloid, stromal, and endothelial sub-clusters) being displayed proximal to each other. The trophoblast lineage reconstruction displayed in Figure 1—figure supplement 2 shows the progression from cytotrophoblasts to either extravillous trophoblasts or syncytiotrophoblasts, which recapitulates the differentiation structure previously reported (Tsang et al., 2017; Vento-Tormo et al., 2018).

Figure 1.

Transcriptional map of the placenta in human parturition.

(A) Study design illustrating the placental compartments and study groups. (B) Uniform Manifold Approximation Plot (UMAP), where dots represent single cells and are colored by cell type. (C) Distribution of single-cell clusters by placental compartments. (D) Average proportions of cell types by placental compartments and study groups. (E) Distribution of single cells by maternal or fetal origin. STB, Syncytiotrophoblast; EVT, Extravillous trophoblast; CTB, cytotrophoblast; HSC, hematopoietic stem cell; npiCTB, non proliferative interstitial cytotrophoblast; LED, lymphoid endothelial decidual cell.

Each row represents a gene marker, and color represents normalized and scaled gene expression values derived from Seurat. Cell-type colors are consistent across all figures in this paper unless otherwise indicated.

Lines reconstruct the most likely differentiation path of the different cell-types starting for cells that may be in a trophoblast progenitor state. The first branch seems to separate EVT from CTB; then CTB splits in STB and npiCTB. STB, Syncytiotrophoblast; EVT, Extravillous trophoblast; CTB, cytotrophoblast; npiCTB, non proliferative interstitial cytotrophoblast.

npiCTB cell-type is highlighted inside the green circle. CTB, cytotrophoblast; npiCTB, non proliferative interstitial cytotrophoblast.

For each cell, an index is derived that represents the total number of reads mapping to the Y chromosome genes divided by the total number of reads mapping to autosomal chromosomes. (A) Density plot across all cells of the Y index. (B) UMAP plot where each facet represents a different placental compartment and each cell color is scaled proportionally to the Y index.

(A) SingleR method using the human primary cell atlas (HPCA) reference panel Aran et al. (2019), (B) SingleR method using the human placenta in the first trimester (HPFT) reference Vento-Tormo et al. (2018), (C) Seurat label transfer method using the HPFT reference Stuart et al. (2019).

(A) SingleR method using the HPCA reference panel Aran et al. (2019), (B) SingleR method using the HPFT reference Vento-Tormo et al. (2018), (C) Seurat label transfer method using the HPFT reference Stuart et al. (2019).

To determine the cell-type we used Seurat label transfer function and the HPFT panel as a reference Vento-Tormo et al. (2018). Only cells with a label transfer score >0.001 are shown.

(A) UMAP plot showing cells identified as potential doublets (shown on top to avoid occlusion by the more common singlets) by DoubletFinder McGinnis et al. (2019) (B) % of cells for each cell-type that are predicted as potential doublets (overall average = 0.898%).

Figure 1—figure supplement 1.

Heatmap of the top gene expression markers defining each cell-type.

Each row represents a gene marker, and color represents normalized and scaled gene expression values derived from Seurat. Cell-type colors are consistent across all figures in this paper unless otherwise indicated.

Figure 1—figure supplement 2.

UMAP plot highlighting the trophoblast cell-types and their inferred differentiation path using slingshot R package.

Lines reconstruct the most likely differentiation path of the different cell-types starting for cells that may be in a trophoblast progenitor state. The first branch seems to separate EVT from CTB; then CTB splits in STB and npiCTB. STB, Syncytiotrophoblast; EVT, Extravillous trophoblast; CTB, cytotrophoblast; npiCTB, non proliferative interstitial cytotrophoblast.

Heatmap showing the correspondence between our final curated cluster labels and automated cell-labeling methods.

The cell type composition differed both among placental compartments (Figure 1C) and due to the presence of physiologic and pathologic processes of labor (i.e. term in labor and preterm labor) (Figure 1D). While extravillous trophoblasts (EVT) were present in all three compartments, cytotrophoblasts (CTB) were especially pervasive in the placental villi, which is explained by the fact that CTBs are abundant in the parenchyma of the placentas. CTBs were also present in the basal plate since this placental compartment is adjacent to the placental villi (Figure 1A). The phenotypic similarities between trophoblasts in proximity to the decidua parietalis (layer attached to the chorioamniotic membranes) and those found in the basal plate have been previously documented (Genbačev et al., 2015; Garrido-Gomez et al., 2017). Importantly, non-proliferative interstitial cytotrophoblasts (npiCTB) were identified for the first time in the placental villi as forming a distinct cluster. This new cluster was also observed in the basal plate, but not in the chorioamniotic membranes, suggesting that this type of trophoblast has specific functions in the placental tree. Lineage reconstruction by slingshot (Street et al., 2018) revealed that npiCTBs are likely derived from conventional CTBs (Figure 1—figure supplement 2). The non-proliferative nature of npiCTBs was evidenced by the reduced expression of genes involved in cell proliferation such as XIST, DDX3X, and EIF1AX (Figure 1—figure supplement 3). npiCTBs displayed an increased expression of PAGE4 (Figure 1—figure supplement 3), a gene expressed by CTBs isolated from pregnancy terminations (Genbacev et al., 2011), suggesting that this type of trophoblast cell is present earlier in gestation. As expected, trophoblast cell types were of fetal origin, and decidual cells present in the basal plate (including the decidua basalis) and chorioamniotic membranes (including the decidua parietalis) were of maternal origin (Figure 1E and Figure 1—figure supplement 4).

Figure 1—figure supplement 3.

Single marker gene expression UMAP plot for genes differentially expressed between CTB and npiCTB.

npiCTB cell-type is highlighted inside the green circle. CTB, cytotrophoblast; npiCTB, non proliferative interstitial cytotrophoblast.

Figure 1—figure supplement 4.

Analysis of the fetal/maternal origin of the cell-types based on data from three pregnancies with a male fetus.

For each cell, an index is derived that represents the total number of reads mapping to the Y chromosome genes divided by the total number of reads mapping to autosomal chromosomes. (A) Density plot across all cells of the Y index. (B) UMAP plot where each facet represents a different placental compartment and each cell color is scaled proportionally to the Y index.

In terms of immune cell types, the chorioamniotic membranes largely contained lymphoid and myeloid cells of maternal origin, including T cells (mostly in a resting state), NK cells, and macrophages (Figure 1C and E and Figure 1—figure supplement 4). In contrast, the basal plate included immune cells of both maternal and fetal origin, such as T cells (mostly in an active state), NK cells, and macrophages. The placental villi contained more fetal than maternal immune cells, namely monocytes, macrophages, T cells, and NK cells. Two macrophage subsets were found in placenta compartments: macrophage 1 of maternal origin that was predominant in the chorioamniotic membranes, and macrophage 2 of fetal origin that was mainly present in the basal plate and placental villi. Together with previous single cell studies of early pregnancy (Vento-Tormo et al., 2018), these results highlight the complexity and dynamics of the immune cellular composition of the placental tissues, including the maternal-fetal interface (i.e. decidua), from early gestation to term or preterm delivery.

Importantly, a new lymphatic endothelial decidual (LED) cell type of maternal origin was identified in the chorioamniotic membranes, forming a distinct transcriptional cluster that was separate from other endothelial cell-types (Figure 1C and E). LED cells were rarely observed in the basal plate and were completely absent in the placental villous. Similar to other endothelial cell types, LED cells highly expressed CD34, CDH5, EDNRB, PDPN, and TIE1 (Figure 2—figure supplement 1). The signature genes of this novel cell type were enriched for pathways involving cell-cell and cell-surface interactions at the vascular wall, extracellular matrix organization (Figure 2—figure supplement 2), tight junction, and focal adhesion (Figure 2—figure supplement 3), indicating that LEDs possess the machinery required to mediate the influx of immune cells into the chorioamniotic membranes. Immunostaining confirmed the co-expression of LYVE1 (lymphatic marker) and CD31 (endothelial molecule marker) in the vessels of the decidua parietalis of the chorioamniotic membranes, but not in the basal plate or placenta (Figure 2A). The co-localization of LYVE1 and CD31 proteins (i.e. LED cells) in the chorioamniotic membranes is shown in Figure 2B and Figure 2—video 1. LED cells also expressed the common endothelial cell marker CD34 (Figure 2C, green arrow). LYVE1 was also expressed by the fetal macrophages present in the placental villi and basal plate (Figure 2C, red arrow), yet the protein was only visualized by immunostaining in immune cells located in the villous tree (Figure 2A, red arrows). This finding conclusively shows the presence of lymphatic vessels in the decidua parietalis of the chorioamniotic membranes, providing a major route for maternal lymphocytes (e.g. T cells) infiltrating the maternal-fetal interface (Arenas-Hernandez et al., 2019).

Figure 2—figure supplement 1.

Single marker gene expression UMAP plot for genes that are more highly expressed in lymphatic endothelial decidual (LED) cells.

Each row of panels represents a gene that is highly expressed in LEDs and each column represents a compartment (Basal Plate = BP, Pacental Villi = PV, and chorioamniotic membranes = CAM). Note that LEDs highlighted inside the circle are almost only found in the CAM.

Figure 2—figure supplement 2.

Clusterprofiler dot plot showing the ReactomeDB Pathways enriched for genes that define each cell-type.

Color is scaled to the Benjamini Hochberg adjusted p-value, and dot size is scaled to the fraction of cell-type (column name) specific genes (number in parentheses) that are found annotated in the category (row name).

Figure 2—figure supplement 3.

Clusterprofiler dot plot showing the Kegg Pathways enriched for genes that define each cell-type.

Color is scaled to the Benjamini Hochberg adjusted p-value, and dot size is scaled to the fraction of cell-type (column name) specific genes (number in parentheses) that are found annotated in the category (row name).

Figure 2.

Identification of LED cells in the chorioamniotic membranes.

(A) Cell segmentation map (built using the DAPI nuclear staining) and immunofluorescence detection of LYVE-1 (red) and CD31 (green) in the basal plate (BP), placental villi (PV), and chorioamniotic membranes (CAM). Red arrows point to fetal macrophages expressing LYVE1 but not CD31, and green arrows indicate lymphatic endothelial decidual cells (LED cells) expressing both LYVE1 and CD31. (B) Co-expression of LYVE1 and CD31 (i.e. LED cells) in the chorioamniotic membranes. (C) Single-cell expression UMAP of LYVE-1 (red) and CD34 (green) in the placental compartments.

Each row of panels represents a gene that is highly expressed in LEDs and each column represents a compartment (Basal Plate = BP, Pacental Villi = PV, and chorioamniotic membranes = CAM). Note that LEDs highlighted inside the circle are almost only found in the CAM.

Color is scaled to the Benjamini Hochberg adjusted p-value, and dot size is scaled to the fraction of cell-type (column name) specific genes (number in parentheses) that are found annotated in the category (row name).

Color is scaled to the Benjamini Hochberg adjusted p-value, and dot size is scaled to the fraction of cell-type (column name) specific genes (number in parentheses) that are found annotated in the category (row name).

Immunofluorescence co-expression of LYVE-1 (red) and CD31 (green) represents LED cells.

Figure 2—video 1.

Video with the 3D reconstruction of the lymphatic endothelium in the decidua present in the CAM compartment.

Immunofluorescence co-expression of LYVE-1 (red) and CD31 (green) represents LED cells.

For cell types that were present in more than one placental compartment, major differences in gene expression were identified across locations, indicative of further specialization of cells depending on the unique physiological functions of each microenvironment (Figure 3—figure supplement 1 and Supplementary file 4). Differences in the transcriptional profiles were particularly large for maternal macrophages as well as EVTs, NK cells, and T cells in the chorioamniotic membranes compared to the other compartments. Genes differentially expressed in the chorioamniotic membranes were enriched for interleukin and Toll-like receptor signaling as well as for the NF-κB and TNF pathways (Figure 3—figure supplements 2–4). These results are consistent with previous reports showing a role for these mediators in the inflammatory process of labor (Romero et al., 1989a; Romero et al., 1990b; Romero et al., 1992a; Romero et al., 1990a; Santhanam et al., 1991; Romero et al., 1993; Romero et al., 1991; Hsu et al., 1998; Keelan et al., 1999; Young et al., 2002; Osman et al., 2003; Kim et al., 2004; Abrahams et al., 2004; Kumazaki et al., 2004; Koga et al., 2009; Belt et al., 1999; Yan et al., 2002; Lindström and Bennett, 2005; Vora et al., 2010; Romero, 1989b; Romero et al., 1992b; Lonergan et al., 2003). Conversely, the placental villous and basal plate were more similar to each other, with most differentially expressed genes (DEG) between these compartments being noted in fibroblasts (335 DEG, q < 0.1 and fold change >2) (Figure 3—figure supplements 1 and 5–10). DEGs in the placental villous fibroblasts showed enrichment in smooth muscle contraction, the apelin and oxytocin signaling pathways (Figure 3—figure supplement 9), while DEGs in CAM fibroblasts were enriched in elastic fiber formation and extracellular matrix pathways (Figure 3—figure supplement 2). The latter finding indicates that the same cell type (e.g. fibroblasts) may have distinct functions in different microenvironments of the placenta.

Figure 3—figure supplement 1.

Stacked bar plot summarizing differentially expressed genes across compartments for a cell types that are present on all three of them.

Number of DEGs at a Benjamini Hochberg adjusted p-value<0.1 and fold change greater than 2: (A) between each pair of compartments, or (B) for each compartment (indicated in color) to the other two compartments.

Figure 3—figure supplement 2.

Clusterprofiler dot plot showing the ReactomeDB Pathways enriched for genes that are significantly more highly expressed in the CAM compartment relative to the other compartments for each cell-type.

Color is scaled to the Benjamini Hochberg adjusted p-value, and dot size is scaled to the fraction of cell-type (column name) specific genes (number in parentheses) for the CAM that are found annotated in the category (row name).

Figure 3—figure supplement 4.

Clusterprofiler dot plot showing gene ontology (GO) terms enriched for genes that are significantly more highly expressed in the CAM compartment relative to the other compartments for each cell-type.

Color is scaled to the Benjamini Hochberg adjusted p-value, and dot size is scaled to the fraction of cell-type (column name) specific genes (number in parentheses) for the CAM that are found annotated in the category (row name).

Figure 3—figure supplement 5.

Clusterprofiler dot plot showing the ReactomeDB Pathways enriched for genes that are significantly more highly expressed in the BP compartment relative to the other compartments for each cell-type.

Color is scaled to the Benjamini Hochberg adjusted p-value, and dot size is scaled to the fraction of cell-type (column name) specific genes (number in parentheses) for the BP that are found annotated in the category (row name).

Figure 3—figure supplement 10.

Clusterprofiler dot plot showing gene ontology (GO) terms enriched for genes that are significantly more highly expressed in the PV compartment relative to the other compartments for each cell-type.

Color is scaled to the Benjamini Hochberg adjusted p-value, and dot size is scaled to the fraction of cell-type (column name) specific genes (number in parentheses) for the PV that are found annotated in the category (row name).

Figure 3—figure supplement 9.

Clusterprofiler dot plot showing the Kegg Pathways enriched for genes that are significantly more highly expressed in the PV compartment relative to the other compartments for each cell-type.

Color is scaled to the Benjamini Hochberg adjusted p-value, and dot size is scaled to the fraction of cell-type (column name) specific genes (number in parentheses) for the PV that are found annotated in the category (row name).

Next, we assessed changes due to term and preterm labor in each cell type (Supplementary file 5). The largest number of DEGs between the term labor and term no labor groups were observed in the maternal macrophages (macrophage 1), followed by the EVT (144 and 37, respectively, q < 0.1; Figure 3A). The largest number of DEGs between the preterm labor and term labor groups were observed in EVT and CTB (37 and 33, respectively, q < 0.1; Figure 3A). Figure 3B displays the gene expression changes between TIL and TNL or PTL and TNL that are shared between the two labor groups, representing the common pathway of parturition (defined as the anatomical, physiological, biochemical, endocrinological, immunological, and clinical events that occur in the mother and/or fetus in both term and preterm labor Romero et al., 2006b). Non-shared differences in gene expression with labor at term and in preterm labor were mostly observed in trophoblast cell types such as CTB and EVT as well as in stromal cells (Figure 3C). Some of these changes may be explained by the unavoidable confounding effect of gestational age since placentas from women without labor in preterm gestation cannot be obtained in the absence of pregnancy complications. Specifically, the expression of NFKB1 by maternal macrophages was higher in women with term labor compared to non-labor controls, and this increase was further accentuated in preterm labor (Figure 3D). Consistent with the induction of the NFκB pathway, the labor-associated DEGs in macrophages involved biological processes such as activation of immune response and regulation of pro-inflammatory cytokine production (Figure 3—figure supplement 11A). These results are in line with previous studies showing that decidual macrophages undergo an M1-like macrophage polarization (i.e. pro-inflammatory phenotype) during term and preterm labor (Xu et al., 2016).

Figure 3.

Cell type specific expression changes in term and preterm labor.

(A) Number of differentially expressed genes (DEGs) among study groups (TNL, term no labor; TIL, term in labor; PTL, preterm labor) by direction of change. Shared (B) and non-shared (C) expression changes in term labor and preterm labor relative to the term no labor group (q < 0.01). The length of each whisker represents the 95% confidence interval. (D) The expression of NFKB1 by maternal macrophages in the placental compartments (BP, basal plate; PV, placental villous; CAM, chorioamniotic membranes) and study groups. The notch represents the 95% confidence interval of the median. (E) Differences and similarities in expression changes with preterm labor and term labor by three major cell types (immune, stromal/endothelial, and trophoblast cells).

Number of DEGs at a Benjamini Hochberg adjusted p-value<0.1 and fold change greater than 2: (A) between each pair of compartments, or (B) for each compartment (indicated in color) to the other two compartments.

Color is scaled to the Benjamini Hochberg adjusted p-value, and dot size is scaled to the fraction of cell-type (column name) specific genes (number in parentheses) for the CAM that are found annotated in the category (row name).

Color is scaled to the Benjamini Hochberg adjusted p-value, and dot size is scaled to the fraction of cell-type (column name) specific genes (number in parentheses) for the CAM that are found annotated in the category (row name).

Color is scaled to the Benjamini Hochberg adjusted p-value, and dot size is scaled to the fraction of cell-type (column name) specific genes (number in parentheses) for the CAM that are found annotated in the category (row name).

Color is scaled to the Benjamini Hochberg adjusted p-value, and dot size is scaled to the fraction of cell-type (column name) specific genes (number in parentheses) for the BP that are found annotated in the category (row name).

Color is scaled to the Benjamini Hochberg adjusted p-value, and dot size is scaled to the fraction of cell-type (column name) specific genes (number in parentheses) for the BP that are found annotated in the category (row name).

Color is scaled to the Benjamini Hochberg adjusted p-value, and dot size is scaled to the fraction of cell-type (column name) specific genes (number in parentheses) for the BP that are found annotated in the category (row name).

Color is scaled to the Benjamini Hochberg adjusted p-value, and dot size is scaled to the fraction of cell-type (column name) specific genes (number in parentheses) for the PV that are found annotated in the category (row name).

Color is scaled to the Benjamini Hochberg adjusted p-value, and dot size is scaled to the fraction of cell-type (column name) specific genes (number in parentheses) for the PV that are found annotated in the category (row name).

Color is scaled to the Benjamini Hochberg adjusted p-value, and dot size is scaled to the fraction of cell-type (column name) specific genes (number in parentheses) for the PV that are found annotated in the category (row name).

The two panels correspond to the following cell-types: (A) maternal macrophages, and (B) extra villous trophoblasts (EVT).

Figure 3—figure supplement 11.

Clusterprofiler dot plot showing ReactomeDB pathways enriched using gene set enrichment analysis (GSEA) for genes differentially expressed in term labor compared to term no labor condition.

The two panels correspond to the following cell-types: (A) maternal macrophages, and (B) extra villous trophoblasts (EVT).

Clusterprofiler dot plot showing the Kegg Pathways enriched for genes that are significantly more highly expressed in the CAM compartment relative to the other compartments for each cell-type.

Clusterprofiler dot plot showing the Kegg Pathways enriched for genes that are significantly more highly expressed in the BP compartment relative to the other compartments for each cell-type.

Clusterprofiler dot plot showing gene ontology (GO) terms enriched for genes that are significantly more highly expressed in the BP compartment relative to the other compartments for each cell-type.

Clusterprofiler dot plot showing the ReactomeDB Pathways enriched for genes that are significantly more highly expressed in the PV compartment relative to the other compartments for each cell-type.

When comparing the effect sizes between the PTL/TNL and TIL/TNL juxtapositions on the same gene and cell type, positive correlations were observed for most of the placental cell types (Figure 3E). Genes displaying differential effects in term and preterm labor are mostly found in trophoblast cell types (see off-diagonal points in the scatter plot), which may be explained by the phenomenon of gene expression decoherence (Lea et al., 2019). This lack of proper correlation between biomarkers to their expected normal relationships is commonly found in pathological conditions. Lastly, in EVT the DEGs with labor were enriched for genes implicated in cellular response to stress, including the WNT and NOTCH pathways, as well as cell cycle checkpoints (Figure 3—figure supplement 11B), further supporting the hypothesis that the cellular senescence pathway (i.e. cell cycle arrest) is implicated in the physiologic (Behnia et al., 2015; Polettini et al., 2015) and pathologic (Hirota et al., 2010; Gomez-Lopez et al., 2017) processes of labor.

To demonstrate the translational value of single-cell RNA signatures derived from the placenta, we conducted an in silico analysis in public datasets (Tarca et al., 2019; Paquette et al., 2018) to test whether the single-cell signatures could be non-invasively monitored in the maternal circulation throughout gestation (Figure 4A). Previous studies have correlated bulk mRNA expression in the maternal circulation with gestational age at blood draw (Tarca et al., 2019; Al-Garawi et al., 2016), risk for preterm birth (Paquette et al., 2018; Heng et al., 2014; Sirota et al., 2018; Knijnenburg et al., 2019), or both (Heng et al., 2016; Ngo et al., 2018). First, using whole blood bulk RNAseq data, we quantified the expression of single-cell signatures in the maternal circulation. We found that the expression of the single-cell signatures of macrophages, monocytes, NK cells, T cells, npiCTB, and fibroblasts is modulated with advancing gestational age (Figure 4B–C, Figure 4—figure supplement 1A). These results validate the T-cell and monocyte signature changes with gestational age that were previously reported (Tsang et al., 2017; Tarca et al., 2019); yet, here we show that novel placental single-cell signatures (e.g., npiCTB and fibroblast) can also be non-invasively monitored in maternal circulation (Figure 4—figure supplement 1A). In addition, for the first time, we report that the expression of the single-cell signatures of NK-cells and activated T-cells were upregulated in women with spontaneous labor at term compared to gestational-age matched controls without labor (Figure 4D). Furthermore, we found that the average expression of the single-cell signatures of macrophages, monocytes, activated T cells, and fibroblasts were increased in the circulation of women with preterm labor and delivery compared to gestational age-matched controls (24–34 weeks of gestation) (Figure 4E and Figure 4—figure supplement 1B). These findings are in line with previous reports indicating a role for these immune cell types in the pathophysiology of preterm labor (Arenas-Hernandez et al., 2019; Hamilton et al., 2012; Shynlova et al., 2013; Gomez-Lopez et al., 2016).

Figure 4.

In silico analysis to quantify scRNA-seq signatures in the maternal circulation.

(A) Diagram of the longitudinal study used to generate bulk RNAseq data (GSE114037) (Tarca et al., 2019) to evaluate changes in scRNA-seq signatures with advancing gestation. Whole blood samples were collected throughout gestation from women who delivered at term. (B and C) Variation of scRNA-seq signature expression in the maternal circulation with advancing gestation. (D) Diagram of the cross-sectional study used to generate bulk RNAseq data (GSE114037) to evaluate changes in scRNA-seq signatures with labor at term (Tarca et al., 2019). Differences in the expression of scRNA-seq signatures between women with spontaneous labor at term (TIL) and term no labor controls (TNL). (E) Diagram of the cross-sectional study used to generate bulk RNAseq data (GSE96083) to evaluate changes in scRNA-seq signatures in preterm labor (Paquette et al., 2018). Differences in the expression of scRNA-seq signatures between women with spontaneous preterm labor (PTL) and gestational-age matched controls (GA control).

(A) Expression of scRNA-seq signatures in the maternal circulation changing with advancing gestation; (B) perturbations in scRNA-seq signatures with preterm labor.

Figure 4—figure supplement 1.

Quantification of scRNA-seq signatures in maternal circulation (continued from main Figure 4).

(A) Expression of scRNA-seq signatures in the maternal circulation changing with advancing gestation; (B) perturbations in scRNA-seq signatures with preterm labor.

Conclusion

In summary, this study provides evidence of differences in cell type composition and transcriptional profiles among the basal plate, placental villi, and chorioamniotic membranes, as well as between the pathologic and physiologic processes of labor at single-cell resolution. Using scRNAseq technology, two novel cell types were identified in the chorioamniotic membranes and placental villi. In addition, we showed that maternal macrophages and extravillous trophoblasts are the cell types with the most transcriptional changes during the process of labor. Importantly, many of the genes differentially expressed in these cell-types replicate for both conditions of labor. This result shows that we have enough statistical power to detect the changes in gene expression with a large effect size that are general or a common molecular pathway in parturition; yet, additional studies are needed to characterize the different etiologies of the preterm labor syndrome. Lastly, we report that maternal and fetal transcriptional signatures derived from placental scRNA-seq are modulated with advancing gestation and are markedly perturbed with term and preterm labor in the maternal circulation. These results highlight the potential of single-cell signatures as biomarkers to non-invasively monitor the cellular dynamics during pregnancy and to predict obstetrical disease. The current study represents the most comprehensive single-cell analysis of the human placental transcriptome in physiologic and pathologic parturition.

Materials and methods

Sample collection and processing, single-cell preparation, library preparation, and sequencing

Human subjects

Immediately after delivery, placental samples [the villi, basal plate (including the decidua basalis) and chorioamniotic membranes (including the decidua parietalis)] were collected from women with or without labor at term or preterm labor at the Detroit Medical Center, Wayne State University School of Medicine (Detroit, MI). Labor was defined by the presence of regular uterine contractions at a frequency of at least two contractions every 10 min with cervical changes resulting in delivery. Women with preterm labor delivered between 33–35 weeks of gestation whereas those with term labor delivered between 38–40 weeks of gestation (Supplementary file 6). The collection and use of human materials for research purposes were approved by the Institutional Review Boards of the Wayne State University School of Medicine. All participating women provided written informed consent prior to sample collection.

Single-cell preparation

Cells from placental villi, basal plate, and chorioamniotic membranes were isolated by enzymatic digestion, using previously described protocols with modifications (Tsang et al., 2017; Xu et al., 2015). Briefly, placental tissues were homogenized using a gentleMACS Dissociator (Miltenyi Biotec, San Diego, CA) either in an enzyme cocktail from the Umbilical Cord Dissociation Kit (Miltenyi Biotec) or in collagenase A (Sigma Aldrich, St. Louis, MO). After digestion, homogenized tissues were washed with ice-cold 1X phosphate-buffered saline (PBS) and filtered through a cell strainer (Fisher Scientific, Durham, NC). Cell suspensions were then collected and centrifuged at 300 x g for 5 min. at 4°C. Red blood cells were lysed using a lysing buffer (Life Technologies, Grand Island, NY). Next, cells were washed with ice-cold 1X PBS and resuspended in 1X PBS for cell counting, which was performed using an automatic cell counter (Cellometer Auto 2000; Nexcelom Bioscience, Lawrence, MA). Lastly, dead cells were removed from the cell suspensions using the Dead Cell Removal Kit (Miltenyi Biotec) and cells were counted again using an automatic cell counter.

Single-cell preparation using the 10x genomics platform

Viable cells were used for single-cell RNAseq library construction using the Chromium Controller and Chromium Single Cell 3' version two kit (10x Genomics, Pleasanton, CA), following the manufacturer’s instructions. Briefly, viable cell suspensions were loaded into the Chromium Controller to generate gel beads in emulsion (GEM) with each GEM containing a single cell as well as barcoded oligonucleotides. Next, the GEMs were placed in the Veriti 96-well Thermal Cycler (Thermo Fisher Scentific, Wilmington, DE) and reverse transcription was performed in each GEM (GEM-RT). After the reaction, the complementary cDNA was cleaned using Silane DynaBeads (Thermo Fisher Scentific) and the SPRIselect Reagent kit (Beckman Coulter, Indianapolis, IN). Next, the cDNAs were amplified using the Veriti 96-well Thermal Cycler and cleaned using the SPRIselect Reagent kit. Indexed sequencing libraries were then constructed using the Chromium Single Cell 3' version two kit, following the manufacturer’s instructions.

Library preparation

cDNA was fragmented, end-repaired, and A-tailed using the Chromium Single Cell 3' version two kit, following the manufacturer’s instructions. Next, adaptor ligation was performed using the Chromium Single Cell 3' version two kit followed by post-ligation cleanup using the SPRIselect Reagent kit to obtain the final library constructs, which were then amplified using PCR. After performing a post-sample index double-sided size selection using the SPRIselect Reagent kit, the quality and quantity of the DNA were analyzed using the Agilent Bioanalyzer High Sensitivity chip (Agilent Technologies, Wilmington, DE). The Kapa DNA Quantification Kit for Illumina platforms (Kapa Biosystems, Wilmington, MA) was used to quantify the DNA libraries, following the manufacturer's instructions.

Sequencing

Sequencing of the single-cell libraries was performed by NovoGene (Sacramento, CA) using the Illumina Platform (HiSeq X Ten System).

Immunofluorescence

Samples of the chorioamniotic membranes, placenta villi, and decidua basal plate were embedded in Tissue‐Tek Optimum Cutting Temperature (OCT) compound (Miles, Elkhart, IN) and snap‐frozen in liquid nitrogen. Ten‐µm‐thick sections of each OCT‐embedded tissue were cut using the Leica CM1950 (Leica Biosystems, Buffalo Grove, IL). Frozen slides were thawed to room temperature, fixed with 4% paraformaldehyde (Electron Microscopy Sciences, Hatfield, PA), and washed with 1X PBS. Non-specific background signals were blocked using Image-iT FX Signal Enhancer (Life Technologies) followed by blocking with antibody diluent/blocker (Perkin Elmer, Waltham, MA) for 30 min. at room temperature. Slides were then incubated with the rabbit anti-LYVE-1 antibody (Novus Biologicals, Centennial, CO) and the Flex mouse anti-human CD31 antibody (clone JC70A, Dako North America, Carpinteria, CA) for 90 min. at room temperature. Following washing with 1X PBS and blocking with 10% goat serum (SeraCare, Milford, MA), the slides were incubated with secondary goat anti-rabbit IgG–Alexa Fluor 594 (Life Technologies) and goat anti-mouse IgG–Alexa Fluor 488 (Life Technologies) for 30 min. at room temperature. Finally, the slides were washed and coverslips were mounted using ProLong Gold Antifade Mountant with DAPI (Life Technologies). Immunofluorescence was visualized using a confocal fluorescence microscope (Zeiss LSM 780; Carl Zeiss Microscopy GmbH, Jena, Germany) at the Microscopy, Imaging, and Cytometry Resources Core at the Wayne State University School of Medicine. Tile scans were performed from the chorioamniotic membranes, placental villi, and basal plate and the complete imaging fields were divided into six‐by-six quadrants.

scRNA-seq data analyses

Raw fastq files obtained from Novogene were processed using Cell Ranger version 2.1.1 from 10X Genomics. First, sequence reads for each library (sample) were aligned to the hg19 reference genome using the STAR (Dobin et al., 2013) aligner, and expression of gene transcripts documented in the ENSEMBL database (Build 82) were determined for each cell. Gene expression was determined by the number of unique molecular identifiers (UMI) observed per gene (QC metrics are shown in Supplementary file 7). Second, data were aggregated and down-sampled to take into account differences in sequencing depth across libraries using Cell Ranger Aggregate to obtain gene by cell expression data. Third, Seurat (Butler et al., 2018) was used to further clean and normalize the data. Then, only data from cells with a minimum of 200 detected genes, and from genes expressed in at least 10 cells were retained. Cells expressing mitochondrial genes at a level of >10% of total gene counts were also excluded, resulting in 77,906 cells and 25,803 genes (summary in Supplementary file 1). Gene read counts were normalized with the Seurat ‘NormalizeData’ function (normalization.method = LogNormalize, scale.factor = 10,000). Genes showing significant variation across cells were selected based on ‘LogVMR’ dispersion function and ‘FindVariableGenes’. Ribosomal and mitochondrial genes were next removed, yielding 3147 highly variable genes which were subsequently analyzed using Seurat ‘RunPCA’ function to obtain the first 20 principal components. Clustering was done using Seurat ‘FindClusters’ function based on the 20 PCAs (resolution of 0.7). Visualization of the cells was performed using Uniform Manifold Approximation and Projection for Dimension Reduction (UMAP) algorithm as implemented by the Seurat ‘runUMAP’ function and using the first 20 principal components.

Assigning cell type labels to single-cell clusters (Appendix 1)

Multiple methods were utilized to label the cell clusters identified by Seurat. First, marker genes showing distinct expression in individual cell clusters compared to all others were identified using the Seurat FindAllMarkers function with default parameters (Supplementary file 3). Marker genes with significant specificity to each cluster (Figure 1—figure supplement 1 and Supplementary file 3) were compared to those reported elsewhere (Tsang et al., 2017; Pavličev et al., 2017). We also used previous known markers used by our group and https://www.proteinatlas.org/ to manually curate the labels. Further, the xCell (http://xcell.ucsf.edu/#) (Aran et al., 2017) tool was utilized to compare the pseudo-bulk expression signatures of the initial clusters to those of known cell types.

Additionally, we compared our manually curated cluster cell type labels to those derived from two automated cell labeling methods: SingleR (Aran et al., 2019) and Seurat (Stuart et al., 2019), using a human cell atlas reference and the placenta single-cell data in early pregnancy (Vento-Tormo et al., 2018) (see Appendix 1 for more details, Figure 1—figure supplements 5–7). Finally, we used the R package DoubletFinder (McGinnis et al., 2019) (https://github.com/chris-mcginnis-ucsf/DoubletFinder) to identify potential doublets. None of our clusters were impacted by doublets (Figure 1—figure supplement 8).

Figure 1—figure supplement 5.

Alluvial diagram showing the correspondence between our final curated cluster labels and automated cell-labeling methods.

(A) SingleR method using the human primary cell atlas (HPCA) reference panel Aran et al. (2019), (B) SingleR method using the human placenta in the first trimester (HPFT) reference Vento-Tormo et al. (2018), (C) Seurat label transfer method using the HPFT reference Stuart et al. (2019).

Figure 1—figure supplement 7.

Uniform Manifold Approximation Plot (UMAP), where dots representing single cells and color represents Seurat predicted cell type labels.

To determine the cell-type we used Seurat label transfer function and the HPFT panel as a reference Vento-Tormo et al. (2018). Only cells with a label transfer score >0.001 are shown.

Figure 1—figure supplement 8.

Doublet analysis by DoubletFinder.

(A) UMAP plot showing cells identified as potential doublets (shown on top to avoid occlusion by the more common singlets) by DoubletFinder McGinnis et al. (2019) (B) % of cells for each cell-type that are predicted as potential doublets (overall average = 0.898%).

Identification of cell-type maternal/fetal origin

We used two complementary approaches to determine the maternal/fetal origin of each cell-type. First, we used the samples derived from pregnancies where the neonate was male (3/9 cases, 8/25 samples) and we derived a fetal index based on the sum of all the reads mapping to genes on the Y chromosome relative to the total number of reads mapping to genes on the autosomes (Figure 1—figure supplement 4). The second method was based on genotype information derived from the scRNA-seq reads that overlap to known genetic variants from the 1000 Genomes reference panel using the freemuxlet approach implemented in popscle (Figure 1E). The freemuxlet approach extends the demuxlet (Kang et al., 2018) method, which can be useful for cases in which separate genotype information for each individual is not available. The software available at https://github.com/statgen/popscle/ was used with the ‘--nsample 2’ option to map each cell barcode to one of the two possible genomes: fetal or maternal. The trophoblast cells are of fetal origin; therefore, we used this information to determine the fetal genome.

Trophoblast trajectory analysis

We used the slingshot R package (Street et al., 2018) to reconstruct the trophoblast cell lineages from our single-cell gene expression data. This method works by building a minimum spanning tree across clusters of cells and has been reviewed as one of the most accurate tools for this task (Saelens et al., 2019). This analysis focused on the trophoblast cell-types (STB, CTB, EVT, and npiCTB), in which we used as input the unmerged cluster labels (i.e., four sub-clusters for CTB, and two for EVT) and the matrix of cell embedding in UMAP (see Figure 1—figure supplement 2).

Differential gene expression

To identify genes differentially expressed among locations (independent of study group), we created a pseudo-bulk aggregate of all the cells of the same cell-type. Only cell types with a minimum of 100 cell in each location were considered in this analysis. Differences in cell type specific expression were estimated using negative binomial models implemented in DESeq2 (Love et al., 2014), including a fixed effect for each individual and location. The distribution of p-values for DEGs between pairs of compartments was assessed using a qq-plot to ensure the statistical models were well calibrated (Supplementary file 3). To detect DEGs across study groups we aggregated read counts across locations for each cell-type cluster, excluding cell-types with less than 100 cells in each study group (15 clusters). Differences in cell-type specific expression among study groups were estimated using negative binomial models implemented in Deseq2. Differential gene expression was inferred based on FDR adjusted p-value (q-value <0.1) and fold change >2.0.

Gene ontology and pathway enrichment analyses

The clusterProfiler (Yu et al., 2012) package in R was utilized for the identification and visualization of enriched pathways among differentially expressed genes identified as described above. The functions ‘enrichGO’, ‘enrichKEGG’, and ‘enrichPathway’ were used to identify over-represented pathways based on the Gene Ontology (GO), KEGG, and Reactome databases, respectively. Similar enrichment analyses were also conducted using Gene Set Enrichment Analysis (GSEA) (Subramanian et al., 2005) which does not require selection of differentially expressed genes as a first step. Significance in all enrichment analyses were based on q < 0.05.

In silico quantification of single-cell signatures in maternal whole blood mRNA

Analysis of transcriptional signatures with advancing gestation and with labor at term

Whole-blood samples collected longitudinally (12 to 40 weeks of gestation) from women with a normal pregnancy who delivered at term with (TIL) (n = 8) or without (TNL) (n = 8) spontaneous labor, were profiled using DriverMap and RNA-Seq, as previously described (Tarca et al., 2019) and data were available as GSE114037 dataset in the Gene Expression Omnibus. The log2 normalized read counts were averaged over the top genes (up to 20, ranked by decreasing fold change) distinguishing each cluster from all others as described above (single-cell signature). Whole blood single-cell signature expression in patients with three longitudinal samples was modeled using linear mixed-effects models with quadratic splines in order to assess the significance of changes with gestational age. Differences in single-cell signature expression associated with labor at term (TIL vs. TNL) were assessed using two-tailed equal variance t-tests. In both analyses, adjustment for multiple signature testing was performed using the false discovery rate method, with q < 0.1 being considered significant.

Analysis of transcriptional signatures in preterm labor

Whole blood RNAseq gene expression profiles from samples collected at 24–34 weeks of gestation were previously described (Paquette et al., 2018) and data were available as GSE96083 dataset in the Gene Expression Omnibus. The study included samples from 15 women with preterm labor who delivered preterm, and 23 gestational age matched controls. Log2 transformed pseudo read count data were next transformed into Z-scores based on mean and standard deviation estimated in the control group. Single cell signatures were quantified as the average of Z-scores of member genes and compared between groups using a two-tailed Wilcoxon test. Adjustment for multiple signature testing was performed using the false discovery rate method, with q < 0.1 being considered a significant result.

Data and materials availability

The scRNA-seq data reported in this study has been submitted to NIH dbGAP repository (accession number phs001886.v1.p1). All other data used in this study are already available through Gene Expression Omnibus (accession identifiers GSE114037 and GSE96083) and through ArrayExpress (E-MTAB-6701). All software and R packages used herein are detailed in the Materials and methods. Scripts detailing the analyses are also available at https://github.com/piquelab/sclabor. To enable further exploration of the results we have also provided a Shiny App in Rstudio available at: http://placenta.grid.wayne.edu/.

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Acceptance summary:

The overall goal of this paper is to characterize the cell types in the placenta during different stages of labor, namely term-labor, pre-term-labor and term-no-labor by using single-cell RNA-sequencing. The novelty of this study is that they are isolating cells from different placental compartments and from women with different types of labor and stages of gestation, providing thus far the most comprehensive scRNA-seq profiling of these tissues. Overall, this study has several novel components:

a) an exceptionally novel scRNA-seq dataset, b) a set of findings that have implications in detecting pre-term labor, namely i) cell type composition varies in different placental compartments, ii) immune cell types and pathways play an important role in the timing of labor, iii) possible non-invasive monitoring of transcriptional signatures of different gestation signatures all based on maternal circulating RNA.

Decision letter after peer review:

Thank you for submitting your article "Single cell transcriptional signatures of the human placenta in term and preterm parturition" for consideration by eLife. Your article has been reviewed by three peer reviewers, one of whom is a member of our Board of Reviewing Editors, and the evaluation has been overseen by Marianne Bronner as the Senior Editor. The reviewers have opted to remain anonymous.

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

We believe this is solid and exciting work, but have a few concerns which can be addressed with writing and analyses.

Essential revisions:

1) The paper as written comes across as overly descriptive and could be improved by adding more interpretations of the findings. Many of the results are listings of a large number of cell types or genes expressed or differentially expressed among them. But it is sometimes difficult to see the overall message. For example, "Other genes highly expressed by LED cells were CD34, CDH5, EDNRB, PDPN, and TIE1 (Figure 2C, Figure 2—figure supplement 1)." Before this, it was already stated that two marker genes that define the LED cells were found. What does listing these additional genes add? Similarly, "The magnitude of cell type composition differences across placental compartments was substantial". Then there's a list of cell types and a list of genes that are highly expressed in one of the tissue. In such places it would be more impactful if there could be a more concise description along with the significance of the finding. This happens in some places, but other places could be more concise.

2) The identification of cell types was based on the expression of their clusters to those in http://xcell.ucsf.edu/#. How robust is this analysis and are all clusters annotated with a label? This is important because this is used for input for DEG between matched cell types between compartments. Also, it seems this resource is for bulk and not single cell, so please clarify how the single cell clusters were annotated with a cell label. Specifically, and very importantly, on the xCell website it is clearly stated “xCell is intended for use with bulk gene expression, for single-cell RNA-seq we recommend using SingleR: https://github.com//SingleR”. Since much of the manuscript is based on these cell/clusters annotations, it is difficult to evaluate how solid the authors claims and interpretations are (identification of new cell types, variability if cell composition amongst placental subtypes and also amongst study groups, etc) using potentially incorrectly or poorly annotated clusters.

Essential revisions:

1) The paper as written comes across as overly descriptive and could be improved by adding more interpretations of the findings. Many of the results are listings of a large number of cell types or genes expressed or differentially expressed among them. But it is sometimes difficult to see the overall message. For example, "Other genes highly expressed by LED cells were CD34, CDH5, EDNRB, PDPN, and TIE1 (Figure 2C, Figure 2—figure supplement 1)." Before this, it was already stated that two marker genes that define the LED cells were found. What does listing these additional genes add? Similarly, "The magnitude of cell type composition differences across placental compartments was substantial". Then there's a list of cell types and a list of genes that are highly expressed in one of the tissue. In such places it would be more impactful if there could be a more concise description along with the significance of the finding. This happens in some places, but other places could be more concise.

We thank the reviewers for making this recommendation. We have revised the main text of the manuscript by incorporating new interpretation of our results and discussing our main findings within the context of previous reports. We also provide a more concise description of our findings.

You can find the changes in the Results and Discussion section.

2) The identification of cell types was based on the expression of their clusters to those in http://xcell.ucsf.edu/#. How robust is this analysis and are all clusters annotated with a label? This is important because this is used for input for DEG between matched cell types between compartments. Also, it seems this resource is for bulk and not single cell, so please clarify how the single cell clusters were annotated with a cell label. Specifically, and very importantly, on the xCell website it is clearly stated “xCell is intended for use with bulk gene expression, for single-cell RNA-seq we recommend using SingleR: https://github.com//SingleR”. Since much of the manuscript is based on these cell/clusters annotations, it is difficult to evaluate how solid the authors claims and interpretations are (identification of new cell types, variability if cell composition amongst placental subtypes and also amongst study groups, etc) using potentially incorrectly or poorly annotated clusters.

We thank the reviewers for this comment, which has greatly helped us to expand our Materials and methods section and clarify the tools used in this study. We agree with the reviewers about the importance of detailed description of the methods used to label cell types. Initially, we used xCell on the pseudo-bulk aggregate for each cluster, since there were no tools available for automated cell type labeling. We also used previous known markers used by our group and from https://www.proteinatlas.org/ to curate the labels manually. This was not clearly explained in our initial submission; thus, this section has been expanded in the revised version of the manuscript. Per the reviewers’ recommendation, we have also performed two additional analyses using SingleR and Seurat to further assess the accuracy of our initial cluster labels. Both SingleR and Seurat require a reference single cell panel to train the models to label the cells. To this end, we have used data from the human placenta in the first trimester (HPFT) (Vento-Tormo et al., 2018) as well as the human primary cell atlas (HPCA) as reference panels. We also attempted to use the data from Tsang et al., 2017 for the automated labeling; yet, the labels for each single cell were not available for this purpose. Notwithstanding, the markers from Tsang et al. were used as a guide in our original labeling. Please see Appendix 1 (“Cell type labelling procedures”) and Figure 1—figure supplement 5-7.

Overall, these additional analyses confirmed the quality of the original cluster annotation and will facilitate the comparison of our study with a single-cell RNAseq study of the placenta in early pregnancy (Vento-Tormo et al., 2018). Below, we describe the new information generated in these additional analyses.

Figure 1—figure supplement 5: Alluvial diagrams showing the correspondence between our initial curated cluster labels and automated cell-labelling methods: A) SingleR method using the HPCA reference panel; B) SingleR method using the HPFT reference panel; C) Seurat label transfer method using the HPFT reference panel.

Figure 1—figure supplement 6: Heatmap plots showing the correspondence between our initial curated clusters labels and automated cell-labelling methods: A) SingleR method using the HPCA reference panel; B) SingleR method using the HPFT reference panel; C) Seurat label transfer method using the HPFT reference panel.

These supplementary figures show the consistency among the major cell types as initially identified in our study and those obtained by the automated cell labelling methods. It is worth mentioning that this identification was limited by the cell types present in the reference panel. For example, some cell types identified in the first trimester study (NK cell subsets) were not present as separate clusters in our study (samples collected in preterm and term gestations) since the immune cell composition of the maternal-fetal interface differs between early and late gestation. Another example is the new cell type “npiCTB” identified in our study, which was not identified in the first trimester study.

Figure 1—figure supplement 7: UMAP plot showing placental single-cells. Color represents Seurat-predicted cell type labels using the HPFT reference panel.

This supplementary figure confirms that the cell types identified in our original study are in agreement with those labels derived from the automated cell labelling method using the placental single-cell study in early pregnancy. It is worth mentioning that there may be a small cluster (identified as innate lymphoid cells type 3, ILC3) that was not reported in our study since we clustered them together with T cells. ILC3s were not separately clustered in our study since they are rare and may not have distinct functions compared to T cells, which are more abundant in the placental compartments at the end of pregnancy. Moreover, a fraction of the CTBs identified in our study were labeled as SCTs (label used by Vento-Tormo et al., 2018) using automated cell labelling methods. This finding may be due to differences in the expression profile of the trophoblast cells types between early and late pregnancy. The SCT in the reference panel (first trimester placental scRNA-seq data) may also include the profile of the transient stage between CTB and STB. This is supported by the trajectory analysis shown in Figure 1—figure supplement 2.

You can find the changes in the subsection “scRNA-seq data analyses”, Appendix 1 (“Cell type labelling procedures”) and Figure 1—figure supplement 5-7.