Proteogenomic and V(D)J Analysis of Human Decidual T cells Highlights Unique Transcriptional Programming and Clonal Distribution

Abstract

Immunological tolerance towards the semi-allogeneic fetus is one of many maternal adaptations required for a successful pregnancy. T cells are major players of the adaptive immune system and balance tolerance and protection at the maternal-fetal interface, however, their repertoire and subset programming is still poorly understood. Using emerging scRNA-seq technologies, we simultaneously obtained transcript, limited protein, and receptor repertoire at the single-cell level, from decidual and matched maternal peripheral human T cells. The decidua maintains a tissue- specific distribution of T cell subsets compared to the periphery. We find that decidual T cells maintain a unique transcriptome programming, characterized by restraint of inflammatory pathways by overexpression of negative regulators (DUSP, TNFAIP3, ZFP36) and expression of PD-1, CTLA-4, TIGIT and LAG3 in some CD8 clusters. Finally, analyzing T cell receptor clonotypes demonstrated decreased diversity in specific decidual T cell populations. Overall, our data demonstrate the power of multi-omic analysis in revealing regulation of fetal-maternal immune co-existence.

Introduction

T cells are major players of the adaptive immune system, having the capacity to undergo recombination of their antigen-specific receptor. T cell receptor (TCR) recombination, or V(D)J rearrangement, is the basis behind the specific antigenic response of T cells (1) and involves the combination of random segments of variable (V), diversity (D), and junctional (J) segments on both the α and β chains (2). After activation with their cognate antigen, T cells undergo clonal expansion, leading to a specific response to the immunological challenge and subsequently immunological memory. T cell repertoire in tissues is the front line of adaptive immune recognition, yet it is still poorly understood, especially in pregnancy.

Pregnancy requires many maternal adaptations, including the ability to maintain immunological tolerance towards the semi-allogeneic fetus yet remain vigilant to pathogens. Tolerance towards the fetus is accomplished with a combination of mechanisms: 1) lack of classical MHC I molecules, HLA-A and -B, by human fetal trophoblasts(3); 2) limiting traffic of immune cells to/from the mouse, but not human, decidua (4–6); and 3) induction of tolerogenic regulatory T cells (T_regs) in humans and mice (7–10). In fact, many have shown that the decidua maintains a significant proportion of T_regs (11–13), with maternal-fetal HLA-C mismatch leading to expansion and activation of decidua T_regs (14), presumably to maintain fetal tolerance. Moreover, dysregulation of decidual T_regs has been linked to pregnancy pathologies- preeclampsia (15, 16) and preterm birth (17), indicating their importance in maintaining fetal tolerance and a healthy pregnancy.

TCR clonal diversity is the central mechanism of immunological responses against a diverse host of antigens. Differences in clonal repertoire have been reported between peripheral T cells and different tissues, both in human health and disease states (18–20). Similar clonal restriction has been observed in human decidua (15, 21–23), although these studies have been limited by looking at TCR β chain usage only. Furthermore, no conclusive link has been made between clonal diversity and pregnancy pathologies (22, 24, 25), leaving this important question unanswered.

Utilizing single-cell proteogenomic and V(D)J-sequencing we set out to confirm previous observations regarding decidual T cell composition and address the hypothesis that there are transcriptional and clonal differences between decidual and peripheral T cells. We analyzed single-cell transcription, surface proteins, and CDR3 sequences in matched samples from healthy term pregnancies. Our results show that the decidua maintains a unique T cell population, composed primarily of CD8 T cells with an exhausted/effector memory phenotype and very few naive T cells, confirming previous studies (26, 27). Gene expression analysis further shows that, globally, decidual T cells maintain a transcriptome geared towards immune tolerance and attenuation. Lastly, clonal analysis of decidual T cells illustrate that the decidua maintains unique T cell clones compared to the periphery. Overall, our data show that decidual T cells are programmed by the local decidual environment and are clonally distinct from peripheral T cells.

Materials and Methods

Sample Collections and Processing

Placentas and whole blood were obtained from five healthy women undergoing normal elective cesarean sections (>37 wks GA) on day of intake in accordance with the UW Obstetrical Tissue Bank IRB protocol (#2014–1223) and UnityPoint Health – Meriter Institutional Review Board protocol (#2018–10). Decidua basalis were dissected and processed as previously described (28–30). Matched peripheral blood mononuclear cells (PBMCs) were isolated from whole blood via density gradient centrifugation. PBMC layer was extracted and washed with PBS, counted, and adjusted to an optimal concentration of 4 ×10⁶ cells/mL. Mononuclear cells (MCs) from decidua basalis and whole blood were frozen in FBS/10% DMSO and stored until needed.

T cell Enrichment and Sorting

MCs from decidua basalis and matched PBMCs were thawed and subsequently enriched for T cells using the Dynabeads^® Untouched^™ Human T Cells Kit (Invitrogen). Enriched T cells were then labeled with Zombie NIR^™ Fixable Viability Kit (BioLegend), treated with TruStain FcX^™ (BioLegend), then labeled with fluorochrome- and TotalSeq^™-C oligonucleotide-conjugated antibodies (Table 1). T cells were sorted into PBS (Ca++/MG++ free, 1% non-acetylated BSA, 1 mL/mL RNase inhibitor) using the BD FACSAria, collecting 100,000 live, CD3+ cells per sample. One set of samples (matched decidua and PBMCs) were removed due to low cell number recovery, leaving four pairs of matched samples for sequencing.

Table 1.

Fluorescent and Oligo conjugated antibodies

Marker	Clone	Conjugation	Supplier
CD3	UCHT1	PE	BioLegend
CD14	M5E2	PE-Cy7	BioLegend
CD19	SJ25C1	PE-Cy7	BD Science
CD34	581	PE-Cy5	BD Science
CD45	2D1	A488	BioLegend
CD56	NCAM16.2	BV421	BD Science
TCR Vα;7.2	3C10	TotalSeq™-C	BioLegend
CD161	HP-3G10	TotalSeq™-C	BioLegend
CD4	RPA-T4	TotalSeq™-C	BioLegend
CD8	SK1	TotalSeq™-C	BioLegend
TCR Vα;24-Jα18	6B11	TotalSeq™-C	BioLegend
TCR αβ	IP26	TotalSeq™-C	BioLegend
CD279	EH12.2H7	TotalSeq™-C	BioLegend
CD196	G034E3	TotalSeq™-C	BioLegend
CD194	L291H4	TotalSeq™-C	BioLegend
CD197	G043H7	TotalSeq™-C	BioLegend
CD45RA	HI100	TotalSeq™-C	BioLegend

Library preparations and sequencing

Cell suspensions were submitted to the University of Wisconsin (UW) Biotechnology Center (UWBC) for processing. Collected cells were concentrated and cell viability was further validated using the Countess^™ II (Invitrogen). A total of 56,000 cells (7000 for each of 8 samples) were targeted using the 10X Genomics V(D)J Single-Cell RNA-Seq pipeline. Briefly, GEMs were prepared using the Single Cell A Chip Kit (10X Genomics). Full-length cDNA cleanup was performed using DynaBeads Myone Silane beads (Invitrogen). Full-length cDNA was then amplified by PCR (13 cycles) and post cDNA amplification cleanup was performed using SPRIselect (Beckman Coulter) and QC was performed using Agilent HS DNA chips. Feature barcode (FB) PCR amplification was done in 9 cycles. Sample indexing was then performed using the Chromium i7 Plate N, Set A indices (10X Genomics). Gene expression (GEX), FB, and V(D)J libraries were then prepared respectively. Libraries meeting all quality control criteria were then sequenced using the Illumina NovaSeq platform at the UW Gene Expression Center in collaboration with the UWBC DNA Sequencing Facility, Madison, Wisconsin. Libraries were sequenced at the following depth (reads/cell): GEX: 85,000; V(D)J: 11,000; FB: 6,400, with a total of 53,000 cells submitted for sequencing.

Sequence data processing with CellRanger.

We used the CellRanger v3.1 count pipeline to generate filtered gene count matrices for each sample. This pipeline includes demultiplexing, discriminating between cell and background barcodes, and aligning reads to the human transcriptome (GRCh38 3.0). Next, we applied the CellRanger v3.1 vdj pipeline to generate a list of barcodes with CDR3 sequences for each sample. For each barcode, the pipeline assembles the V(D)J transcripts into contigs, aligns the contigs to the TCR reference sequences (GRCh38 3.0), and determines whether the contigs correspond to a CDR3 sequence by annotating both ends of those sequences with V and J genes.

Single cell quality control.

We next enriched the dataset for high quality cells by filtering on several quality control metrics: the fraction of universal molecular identifiers (UMIs) aligning to mitochondrial transcripts, fraction of a list of housekeeping genes detected (list compiled by Tirosh et al. (31)), number of RNA or surface features detected, and number of RNA (FB) UMIs. We converted each value to median absolute deviations (MADs) within each sample and removed cells for which any value was outside of [−3,3]. Finally, although the cells were selected for CDR3 positivity prior to sequencing, we implemented two additional quality control filters to ensure a high confidence data set. First, we required cells to have full-length CDR3 sequence data (from scV(D)J-seq). Next, we plotted the cells by their CD4 and CD8 protein expression (discriminated with antibody feature barcode) and observed that while most cells almost exclusively expressed one or the other, there was a small cluster of apparent double-positives and another of apparent double-negatives. We defined thresholds on CD4 and CD8 to identify and remove the double-positives and double-negatives, which we suspected to be multiplets or false-positive cells. The cells removed by the CD4/CD8 filter were mostly subsumed within the set of cells without full-length CDR3, providing further evidence that they were likely to be artifacts. At the end of the filtering process, both samples from one of the subjects had lost substantially more cells than the other three subjects. We removed both samples for that subject from further analysis, resulting in three subjects each with PBMC and decidua basalis samples included in final analysis. The remaining cell counts per participant and sample type are shown in Table 2.

Table 2.

Cell counts per participant and sample type

Participant	# Decidua T cells	# PBMC T cells	Total
81	1661	4994	6655
83	3378	5398	8776
85	3186	3628	6814

Count normalization.

For RNA values, we log-transformed the depth-normalized counts. For surface proteins profiled by feature barcoding, we used centered log-ratio (CLR) normalization across all cells for each feature.

Batch correction, visualization, and clustering.

We used the Seurat v3 integration pipeline (32) to align the per-sample RNA datasets to reduce subject- and localization-specific variability, using parameter settings recommended in Seurat vignettes (nfeatures=2000, dims=1:30). This approach employs “anchor” cells, which are cells that have similar nearest neighborhoods in the datasets being reconciled. The corrected expression value for a gene in a cell is a function of its proximity to the anchor cells and a score for each anchor. We specified the order of pairwise alignments as follows. First, we integrated the PBMC T cells across subjects, and separately the decidua basalis T cells across subjects. Then we integrated the PBMCs and decidua basalis cells together. The final batch-corrected RNA matrix was used for uniform manifold approximation and projection (UMAP) plots and clustering, but not for expression visualization or statistical analysis (in order to prevent bias incurred during alignment).

Clustering analysis.

Using Seurat, we applied Leiden clustering (33) on the shared nearest neighbor graph with resolution = 0.5. We manually compared the highly expressed surface proteins and marker genes for each cluster to annotations of T cell subset markers in the literature. Clusters were annotated based on both gene and protein expression and following previously published phenotypic classifications (34–42).

Marker gene and differential expression analysis.

To interpret the clusters, we performed rank-sum tests to identify marker genes that distinguished each cluster from cells in all other clusters. Genes were filtered in advance of testing for expression in at least 10% of the cluster’s cells and for absolute average log fold change >0.25. To account for differences in the number of cells available from each sample, we first tested each gene within each sample separately using rank-sum tests (Wilcoxon) on the log-normalized RNA counts. We combined the p-values from the six samples using the sum-log method and adjusted the combined p-values using Bonferroni correction (n = 33548 genes). Final marker genes were selected using a threshold of adjusted p<0.05.

We used a similar method to test for differential expression between decidua and peripheral T cells within each cluster. We performed the tests first between the matched samples for each participant, resulting in three p-values for each gene per cluster. We called differentially expressed genes using the threshold of combined, adjusted p-value < 0.05.

TCR analysis and visualization.

We separated the CDR3 annotations for each sample. Then for every unique combination of TRA and TRB sequences, we counted the number of cells annotated with that combination in each cluster and calculated the Shannon diversity index (43). We examined the patterns of the unique combinations of TRA and TRB sequences by using the UpSetR R package (44)(v1.4.1). This visualization summarizes the combinatorial occurrences across a list of binary variables. For that purpose, we constructed matrices having unique TRA and TRB combinations for a subset of cells as rows and binary features as columns.

Other statistical analyses of sequencing data.

To quantify the subject’s variation in the TCR data, we calculated the percentage of unique clonotypes for each subject, tissue and cluster combination. Then, we used the following model:

$$y_{ijk}=\mu +{\beta }_i+{\gamma }_j+u_k+{ϵ}_{ijk}$$

Where $y_{ijk}$ is the percentage of unique clonotypes for subject i=1,2 and tissue j=1 and cluster k; $\mu$ is the average % of unique clonotypes across all the data, this parameter corresponds to subject 81 and Decidua tissue; ${\beta }_i$ is the subject $i$ effect, this corresponds to subject 83 and 85, respectively; ${\gamma }_j$ is the tissue j effect corresponding to the PBMC tissue; ${ϵ}_{ijk}~N(0,{\sigma }^2)$ is the random error and $u_k~N(0,{\tau }^2)$ is a random effect representing cluster variation. We fitted this model using the lme4 (45) (v1.1–31) and report (46) (v0.5.5) R packages.

We quantified the overlap between the # of clonotypes in PBMC and Decidua cells using the Morisita-Horn index (MHI) (47) using the divo (48) (v1.01). The MHI is zero when there is not overlap between two sets and one when the overlap is complete.

${\chi }^2$ and one-sided hypergeometric tests were used to compare the number of unique clonotypes in PBMC and Decidua cells, and whether the number of unique clonotypes in Decidua was enriched in any of clusters 1, 6, 9 or 11.

Visualization software.

Visualizations of single cell expression and surface feature data were generated using the R packages Seurat (v3.1.5, v4.0), ggplot2 (49)(v3.3.2), and viridis (50) (v0.5.1). We constructed the chord diagrams in Figure 4B using the circlize R package (51)(v0.4.12).

Figure 4. TCR analysis.

(A) TCR repertoire Shannon diversity per cluster separated by tissue of origin for each subject. (B) Chord diagrams comparing the tissue specific VJ gene pairing in TRA / TRB sequences in CD4 Th-like, CD4 Central Memory, CD4 Treg and CD8 MAIT T cells. (C) UpSet plot comparison of unique TRA-TRB combinations between tissue and CD4 Th-like, CD4 Central Memory, CD4 Treg and CD8 MAIT T cells.

Results

Our understanding of clonal and phenotypic T cell diversity at the maternal-fetal interface is critical to deciphering how decidual T cells contribute to a normal pregnancy. We previously reported on the unique distribution of T cell subsets specific to the term human decidua by flow cytometry (30). To further investigate the specificity of decidual T cells, we utilized emerging scRNA-seq technologies (CITE-seq and scV(D)J-seq) to garner gene, protein expression, and receptor repertoire at the single-cell level, simultaneously from decidual and matched maternal peripheral T cells (Figure 1A). Unsupervised clustering analysis of the transcriptomic component of the resulting data set identified 15 unique T cell clusters, with overlapping regions between decidual and peripheral T cells and areas of clear segregation of tissues (Figure 1B,C). The majority of T cells in the decidua were CD8-positive and included the previously described human decidual CD8+ effector memory subset (Cluster 4) (27, 52) (Figure 1D, top). Interestingly, we found three clusters (1, 6, and 9) that were nearly equally distributed between the decidua and periphery (Figure 1D, bottom). Taken together, this supports our previous observation that the decidua maintains a tissue-specific distribution of T cells compared to the periphery (30).

Figure 1. Single-Cell RNA-seq analysis reveals unique T cell clusters in term human decidua.

(A) Experimental approach. (B) UMAP indicating cell origin from decidua basalis and peripheral blood (PBMC). (C) UMAP visualization with cluster annotations. (D) Cell number identified per cluster separated by tissue of origin (top); frequency of tissue representation across clusters identified (bottom). (E) UMAP showing targeted protein expression measured by CITE-seq.

Proteogenomics has emerged as a powerful tool to assess simultaneous protein and gene expression, allowing for corroboration of expression, or lack thereof, of low-copy transcripts. We used CITE-seq (53) to measure protein expression in both peripheral and decidual T cells (Figure 1E). Important features noted were the clear demarcation of CD4 and CD8 protein expression, as well as chemokine receptors CD196 (CCR6), CD194 (CCR4), and CD197 (CCR7) (Figure 1E). Average protein expression levels, in conjunction with gene expression information, were then used to annotate the 15 T cell clusters identified by clustering (Figures 1C & 2; Supplementary Figure 1A). Further gene expression analysis across the 15 clusters identified revealed cluster-specific gene expression patterns (Figure 2). Specifically, we found expression of subset-defining transcription factors GATA3 and RORC in cluster 1 and 7, respectively, confirming their classification as Th-like and Th17 T cells, while FOXP3 expression was entirely restricted to cluster 9 (T_regs) (Figure 2). We found a few cells in cluster 3 expressing the B-cell marker CD79A, which has been reported to be expressed by T cells in some instances (54–56). Two clusters (2 and 10) were classified as Progenitor Exhausted CD8 T cells due to moderate protein expression of PD1 (CD279). However, we found that cluster 10 was mostly confined to decidua basalis and additionally expressed CTLA4, TIGIT and LAG3, while cluster 2 contained a mixture of both decidua basalis and peripheral T cells (Figure 1D).

Figure 2. Gene expression across T cells in the term human decidua and matched peripheral T cells.

Heatmap of the top 10 marker genes for each cluster, filtered to genes expressed in at least 60% of cells in the cluster. Top annotations indicate tissue (decidual vs peripheral) and cluster. Up to 300 cells randomly selected per tissue/cluster.

To illustrate the importance of simultaneous protein and gene expression evaluation, and to make recommendations for other researchers who are beginning to analyze multi-omics sequence data we highlight cluster 6 (Central Memory T Cells). Our first step in annotation was to assess average protein expression of the clusters (Supplementary Figure 1A), which showed high levels of CD4-protein expression, leading us to classify this cluster as CD4+ Central Memory T cells. However, after examination of the data at the level of single cells, we found that cluster 6 was composed of both CD4-positive and CD8-positive cells (Figure 1E). Analysis of RNA expression of CD4 and CD8A/CD8B found that CD4 transcripts were not detected in most cells (Supplementary Figure 1B,C). This prompted us to reassess the reliability of using average protein expression for cluster phenotyping, and found some notable discrepancies. For example, we saw apparently high levels of TCR-Vɑ24-Jɑ18 average protein expression (Supplementary Figure 1A), a marker of NKT cells (57, 58). However, these clusters did not express other classic NKT markers, such as CD161 (58, 59). Visualization of RNA expression confirmed that the transcripts for TRAV24 and TRAJ18 were not expressed (Supplementary Figure 1C) despite the apparent detection of TCR-Vɑ24-Jɑ18 protein when z-scaled (Supplementary Figure 1D). Notably, this detection was likely an artifact of scaled data visualization of background level of staining, which we confirmed by histogram analysis of absolute (not z-scaled) Va24Ja18 expression showing little to no transcripts in any cluster (Supplementary Figure 1E). We also confirmed the division (non-overlap) of CD4 and CD8, both at the transcript and protein level, in cluster 6 (Supplementary Figure 1C&D), indicating that despite extremely similar transcriptional programming, this cluster was a mix of CD4 and CD8 cells. These results show how assessing protein expression simultaneously with gene expression, and performing multiple complementary visualizations of the data at both single-cell level and in aggregate, is important in the context of assigning phenotypic properties to immune cells.

We further analyzed the data to probe for potential differences in transcriptional programming between decidua basalis and peripheral T cells. Differential expression analysis revealed a total of 872 genes that were differentially expressed in any of the clusters (Figure 3A). Of these, 152 genes were globally (7 or more clusters) upregulated in decidual T cells (Supplementary Figure 2). These included genes involved in activation (DUSP, CD69) (24) chemotaxis (CCL4, CXCR4) (60, 61), inflammation (TNFAIP3, NR4A2) (62–64), attenuation/suppression (ZFP36) (65), glucose transport (SLC2A3) (66), trafficking (RGS1) (67), and granule trafficking (SGRN) (68), all of which were heavily enriched in decidual basalis T cells. We then asked whether there were cluster-specific genes that were up- or down-regulated in decidual T cells in fewer than seven clusters. Of the genes shared across fewer than seven clusters, we examined the top 5 up-regulated and top 5 down-regulated genes for each cluster. The union of these top genes included 29 up-regulated and 46 down-regulated genes, with 23 genes unique to one cluster. Some of the top genes were shared between clusters in similar states. These included ARRDC2 and SOX4 (upregulated) and TSPO (downregulated) in both CD4+ and CD8+ Naïve T cells. To further illustrate that decidual T cells maintain a distinct transcriptome independent of phenotype, we focused our attention on the top genes identified for T cell clusters that were equally distributed between the decidua and the periphery (clusters 1, 6 and 9, Figure 3B). We found CST7 to be upregulated in decidual cells, while MYC, RASGRP2, AES, and GIMAP4 were downregulated (Figure 3B). Moreover, specific decidual signatures were found for both clusters 1 and 9, with IFI27 upregulated in cluster 1, and BATF, LINC01943, and TNFRSF18 upregulated in cluster 9. Interestingly, none of the top down-regulated genes were exclusive to one cluster (Supplementary Figure 3). We next turned our attention to the decidual cells in the two clusters labeled Progenitor Exhausted CD8+ T Cells (clusters 2 and 10). Comparisons of gene expression between the decidual cluster 2 and 10 cells found that cluster 2 decidual cells differentially expressed genes involved in protein synthesis, while decidual cluster 10 differentially expressed genes involved in cytotoxicity and chemotaxis (Figure 3C), despite both having the same tissue origin and phenotype. Overall, this supports our hypothesis that decidual T cells would maintain a unique transcriptome programming regardless of phenotypic similarities with peripheral T cells.

Figure 3. Summary of differentially expressed genes between decidua and peripheral T cells.

(A) All differentially expressed (DE) genes from each cluster. Color is the average log2 fold change across subjects. (B) Expression in PBMC versus decidua basalis for clusters 1, 6, and 9. Genes that are up-regulated in basalis are colored in red; down-regulated in blue. Labeled genes are the top “global” DE genes (DE in 7+ clusters; darker red/blue) and top non-global-specific genes (>7 clusters; lighter red/blue), based on ranking by negative log p-value. The top 5 up-regulated and top 5 down-regulated genes from the global and non-global groups are labeled. (C) Gene expression comparison between decidua basalis in clusters 10 versus 2. Red and blue points indicate DE genes (adjusted p<0.05); gray have no significant difference. Labeled genes are top 10 up-regulated and top 10 down-regulated genes, as ranked by negative log adjusted p-value.

The level of clonal diversity at the maternal-fetal interface remains an active area of investigation. To increase our understanding of clonal representation in term human decidua we sequenced V(D)J chains at the single-cell level of both decidual and peripheral T cells. A unique clonotype in this study was defined as a productive full-length TCR alpha and TCR beta pairing with only a single repetition in the entire dataset. As expected based on previous work (15, 21–23), T cells in the decidua overall displayed a lower level of clonal diversity compared to peripheral T cells across all sampled individuals (Figure 4A). Minor exceptions were found, however, particularly in individuals 83 and 85, where 3 clusters were found to have higher clonal diversity in the decidua, including CD8+ Progenitor Exhausted and CD8+ Effector Memory T cells (Figure 4A). We compared the at least one unique clonotype between each subject, tissue and cluster combination (Supplementary Figure 4A), and fitting a linear mixed model revealed that the tissue is the main source of variation after testing for the difference between PBMC and Decidua (p = 0.009). Furthermore, the subject differences from the reference are not significant (p = 0.982 and p=0.377, respectively). Taken together, this indicates that tissue localization is a more important factor in clonotype distribution, even after accounting for cellular recovery differences than inter-subject variation. Focusing on the equally distributed clusters, 1, 6, and 9, we found that frequency of chain usage varied across the three clusters, with cluster 1 (CD4+ Th-like) being the most restricted in the decidua (Figure 4B). Surprisingly, both cluster 6 (Central Memory) and cluster 9 (T_regs) displayed more diverse ɑ and β pairings in the decidua compared to their peripheral counterparts (Figure 4B). As expected, MAIT cell ɑ-chain was highly restricted, more so in the decidua, with more diversity observed in the β-chain (Figure 4B).

Assessing individual clonotype frequency in decidual and peripheral T cells, we found that the decidua had fewer proportion of cells with unique clonotypes (1840) compared to peripheral (3327) T cells (p=9.99×10⁻⁵) when looking at clusters 1,6, and 9 (Figure 4C, list of individual clonotypes and abundance per cluster in Supplementary Table I). Interestingly, although MAIT cells exhibit restricted Vɑ-segment usage, they maintained a higher number of unique clonotypes in the periphery (182) than in the decidua (78) (Figure 4C). There was a small number of shared clonotypes between decidual and peripheral T cells, with a total of 146 clonotypes within cluster 1 (CD4+ Th-like), cluster 6 (Central Memory), and MAIT cells, and only one shared clonotypes with the T_reg cluster (cluster 9), in agreement with previous T_reg clonotype analysis in pregnancy (15) (Figure 4C), the MHI is 2.67% (99.99% CI 1.35%−4.19%) suggests that there is little overlap between both tissues. Coincidentally, clonotype analysis of both CD8 Progenitor Exhausted T cells (clusters 2 and 10), revealed more shared clonotypes within cluster 2 than cluster 10, suggesting that Cluster 2 T cells are allowed to traffic between the periphery and decidua (Supplementary Figure 4B).

Discussion

The maternal-fetal interface represents an immunological paradox: how to defend against pathogens while maintaining tolerance towards the semi-allogeneic fetus? It is understood that this balance is achieved in part by limiting fetal antigen presentation (4, 69) and limiting T cell traffic to the maternal decidua (5, 70). We previously showed that the term human decidua maintains a unique milieu of T cell subsets (30). Here, we expand on those observations by applying a single cell multi-omic approach to better understand the depth of decidual T cell partitioning. We confirmed that the human decidua maintains a unique distribution of T cell subsets compared to matched peripheral blood. Furthermore, we show that decidual T cells maintain a unique transcriptome even when phenotypically identical to their peripheral counterparts. Clonal analysis further illustrates how decidual T cells are specific to the decidua with little clonal overlap between the decidua and periphery. Overall, our results highlight the uniqueness of decidual T cells, both transcriptionally and clonally.

We first assessed T cell subset diversity in human term decidua and matched PBMCs (Figure 1) using CITE-seq, a proteogenomic approach which allows for simultaneous measurement of protein and gene expression (53). Having these sets of information allowed us to confidently annotate the T cell clusters identified in our data set. Notably, we found that one cluster (cluster 6) was mischaracterized based on protein expression and consisted of CD4- and CD8-positive T cells (Figure 1; Supplementary Figure 1). Simultaneous protein assessment has proven important when confirming expression of low copy transcripts (71), such as those of CD56 (72) and CD4 (71). We determined our mischaracterization of cluster 6 was a result of visualizing protein and transcriptomic expression, with cluster annotation based on median protein expression (Supplementary Figure 1A). However, in the absence of protein expression data, cluster 6 would have been mischaracterized as CD8-positive T cells based solely on mRNA detection, as the CD4-transcript was detected in very few cells (Supplementary Figure 1B,C). This serves as an example of why both protein assessment and appropriate data visualization are important to properly assess T cell phenotype in single cell transcriptomics.

CD8+ T cells are the most abundant T cells found in the decidua, with the majority being effector memory cells (52, 73–75). Although we did identify CD8+ Effector Memory (EM) T cells they were not the majority in the decidua, with the majority being CD8+ Progenitor Exhausted (Figure 1C, top). Interestingly, when assessing cytotoxic potential of CD8+ EM T cells, Tilburgs et al. found lower expression of both perforin and granzyme B in decidua (both labored and unlabored) compared to matched peripheral CD8 EM T cells (27). Although seemingly contradictory to our observations of higher granzyme B and perforin expression in decidual CD8+ EM T cells (Figure 2; Supplementary Figure 3), this discrepancy can be explained by translational regulation of cytotoxic granules, such as granzyme B and perforin, which has been well documented (76–78), thus accounting for the differences in protein versus transcript detection. It has been noted that exhausted EM T cells decrease upon labor at term (75). Although we did not include matched samples from labored deliveries, we hypothesize that labor would not affect the composition of TCR repertoire in the decidua, as effector memory T cells are considered tissue localized (27, 79) and based on previous observations that show no changes in shared clonotypes between peripheral and decidual CD8+ T cells under different pregnancy conditions (21).

We focused our attention on clusters 1, 6, and 9 because they were the most evenly distributed clusters between the decidua and periphery (Figure 1) and wanted to illustrate that despite a shared phenotype, these cells are transcriptionally and potentially functionally distinct. Our results, indeed, support our hypothesis that decidual T cells maintain a distinct transcriptional profile compared to their peripheral counterparts (Figure 3) and are in agreement with transcriptional analysis of CD8+ EM T cells, from both early and term decidua (26), and T_regs (80, 81). Furthermore, our clonal analysis of T_regs supports the observation that decidual Tregs are fetal specific (7), and are likely acquiring tissue resident properties once they enter the maternal fetal interface (80). We acknowledge, however, that gene expression differences in some clusters (Supplementary Figure 3) could potentially be influenced by imbalanced tissue representation. Nonetheless, comparison of more-or-less balanced clusters suggest that the decidual environment programs local T cells.

Overall, our multi-omic approach confirms a specific decidual T cell signature in human term decidua. Deeper analysis showed how phenotypic similarities between decidual and peripheral T cells are only superficial, as matched T cells were transcriptionally different. This extends to clonal distribution, with the decidua maintaining a unique set of clones. This raises the possibility that decidual T cells are tissue-resident, or enter the decidua as precursors which differentiate locally. It should be noted however, that this study did not comprehensively evaluate the totality of the repertoire in circulating or decidual sites, so low proportion of shared clones cannot be directly interpreted to mean that there is no trafficking between the two sites. That said, it has already been suggested that MAIT cells are transiently tissue-resident in the endometrium (82); and it is possible that other T cell subsets in the decidua could be similar. Lastly, we show that phenotypic similarities between decidual and peripheral T cells is not an accurate assessment of T cells during pregnancy. This is highlighted not just by distinct clonal diversity but by differential gene expression between decidual and peripheral T cells, illustrating that, regarding decidual T cells, you cannot judge a book by its cover.

Key Points.

• scRNA/CITE-seq results show that decidual T cells maintain a unique transcriptome.

• Decidual T cells overexpress negative regulators of inflammatory pathways.

• Specific decidual T cell clusters display lower clonotype diversity

Data Availability Statement

The single cell sequencing data generated for this publication have been deposited in NBCI’s Gene Expression Omnibus and are accessible through GEO Series accession number GSE192558 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE192558).