This study was approved by the University of Toyama Institutional Review Board (ID: R2016144 and R20161442). Informed consent was obtained from all participants. Four patients who underwent artificial abortion in the first trimester were enrolled at the Ladies’ Clinic We! Toyama. Three healthy pregnant women and three patients with PE were enrolled at the University of Toyama. One healthy pregnant woman provided peripheral blood and decidua. All abortions were performed by manual vacuum aspiration. All pregnancies in healthy women and preeclamptic patients were delivered via Caesarean section. PE was defined as gestational hypertension accompanied by proteinuria, maternal organ dysfunction, or uteroplacental dysfunction, at or after 20 weeks of gestation (8). Detailed characteristics of the participants are provided in Supplementary Table 1 .

First-trimester decidual tissues were isolated from the uterine content obtained by manual vacuum aspiration. The decidua basalis from healthy late gestation and PE deliveries was macroscopically identified and dissected from the maternal surface of the delivered placenta. The samples were processed within 24 h of collection. Decidua was rinsed with phosphate buffered saline (PBS) until the blood was removed, minced with a pair of scissors to produce 1–2 mm pieces, and filtered through 41-μm nylon net filter. The filtered cell suspension was layered with Ficoll Hypaque and centrifuged at 1500 rpm for 30 min with the brakes off. The immune cells at the gradient interface were collected, washed with PBS, and cryopreserved. Peripheral blood mononuclear cells (PBMC), similar to decidual immune cells, were isolated using density gradient centrifugation.

The cryopreserved mononuclear cells were thawed and incubated with the following antibodies for 20 min: PE anti-CD3, APC-Cy7 anti-CD4, APC anti-CD14, PE-Cy7 anti-CD25, FITC anti-CD37, and BD human Sample Tags ( Supplementary Table 2 ). CD3⁺CD4⁺CD14^-CD37^-cells were sorted into CD4⁺ T cells using a FACSAria II flow cytometer (BD Bioscience, Franklin Lakes, NJ, USA). scTCR/Targeted RNA-seq libraries of CD4⁺ T cells were constructed using a BD Rhapsody system with BD Rhapsody Targeted mRNA and AbSeq Reagent kit (BD Bioscience, Franklin Lakes, NJ, USA), and a BD Rhapsody T Cell Expression Panel Hs (BD Bioscience, Franklin Lakes, NJ, USA) with human IL-10 and NRP1 primer supplements (BD Bioscience, Franklin Lakes, NJ, USA) according to the manufacturer’s instructions, except that cDNA denaturation was performed using 0.1M NaOH for 5 min at room temperature. Sequencing was performed using Illumina NovaSeq 6000 and NovaSeq S4 flow cells (200 cycles kit, read 1 67 bp, read 2 155 bp) (Illumina, San Diego, CA, USA) by ImmunoGeneTeqs, Inc. (Chiba, Japan). Details of the key resources are provided in Supplementary Table 2 .

Fastq data preprocessing was conducted using ImmunoGeneTeqs, Inc. (Chiba, Japan), as follows. After adapter trimming and quality filtering, the base composition of the sequencing data was analyzed using FastQC-v0.11.9. Pair-ended FASTQ files (R1: cell barcode reads; R2: RNA reads) were processed, and the filtered cell barcode reads were annotated using the Python script provided by BD, with minor modifications for compatibility with Python3.8. The associated cDNA reads were mapped to reference RNA (build GRCh38 release-101). Then, the cell barcode information of each read was added to the bowtie2-mapped BAM files, and the read counts of each gene in each cell barcode were determined using mawk. The cells with a total read count above the inflection point were considered valid. To assign each tag to a cell barcode, the read counts of each tag in each valid cell barcode, defined by the cDNA matrix, were extracted from the tag/cell barcode expression matrix. Unassigned cell barcodes were labeled as “not-detected” cells. Then, the sum of the total read counts of each tag was normalized to 10 M reads and the log2 fold change between the first most counted tags and second most counted tags within each cell barcode. Doublet cells (double-positive cells of any pair of Tags (identified by the flowDensity package)) and log2 fold change between the first and second most counted tags < 0.2415 were determined. Finally, the remaining cell barcodes were assigned to the first most-counted tag. The tag expression in each cell barcode was log2(x+1)-transformed and z-scaled by each cell barcode.

TCR data matrices containing VDJ read counts, nucleotide sequences, amino acid sequences of CDR3, and gene usage were provided by ImmunoGeneTeqs Inc. (Chiba, Japan). Cells with paired alpha/beta chains were used for TCR analysis.

Distribution-based error correction (DBEC), which included the BD Rhapsody targeted scRNA-seq workflow to reduce background read counts of each gene that were possibly derived from RNA diffusion during the cell lysis step in the BD Rhapsody cartridge, and reverse transcription were conducted by ImmunoGeneTeqs, Inc. (Chiba, Japan).

We created a Seurat object, in which two batches were combined. We calculated metrics (percent mito, percent ribo) and filtered out mitochondrial gene-high cells (> 0.25), doublets and non-hashtag-assigned cells. Because BD Rhapsody-targeted data are non-UMI data (raw count data), we performed global normalization to 1 M tags and scaling using the Seurat ScaleData function by regressing out the library size (total raw read count of each cell) as a confounding factor. We split the combined Seurat objects into a list, with each batch considered an element. We identified variable features individually for each using the FindVariableFeatures function with default parameters. Next, we identified the anchors for each batch and integrated them using the FindIntegrationAnchors function with 50 dimensions. These anchors were then passed to the integrated data function. We used this integrated matrix for the downstream analysis. Dimension reduction was performed using the RunPCA function on an integrated object with the first 50 PCA dimensions. Clusters were visualized using FindNeighbors, FindClusters, and RunUMAP functions with the top 15 dimensions (resolution set to 0.8). Differential markers between the clusters were detected using the FindAllMarker function. Cell types were determined according to previously reported scRNA markers for T cells from peripheral blood, decidua, and cancers (33, 34, 38–40). Module scores of Treg exhaustion-related genes (TNFRSF18, TNFRSF4, PDCD1, IFNG, LAG3, CCR5, CXCR3, CCR6, HAVCR2, CXCR6, NKG7, SLAMF1, KLRB1, CD2, and CXCR1) were calculated as previously reported (39, 41) using the AddModuleScore function. Differential expression analysis was conducted using the FindMarkers function with default settings. Software and algorithms are listed in Supplementary Table 2 . Genes with log fold change > |0.5| and adjusted p-values ≤ 0.05 were regarded statistically significant.

We uploaded the list of differentially expressed genes (DEGs) identified using the FindMarkers function in Seurat to the Matascape (https://metascape.org/gp/index.html#/main/step1) main page and chose “Express Analysis.”

To assess decidual CD4⁺ T cell diversity, we conducted a single-cell transcriptome analysis of CD4⁺ T cells isolated from the decidua basalis of healthy early gestation (6–7 weeks, n = 4), healthy late gestation (37–40 weeks, n = 3), and gestation with early onset PE diagnosed before 34 weeks who delivered between 34 and 37 weeks of gestation (n = 3) ( Supplementary Table 1 ). Women with autoimmune diseases were excluded from this study. CD4⁺ T cells from PBMC from one healthy late gestation patient were also included. CD4⁺ T cells were sorted and subjected to a BD Rhapsody Single-Cell Analysis ( Supplementary Figure 1A ). Dead cells and doublets were removed from the scRNA-seq data. Dimension reduction and clustering were performed on 24,002 cells that passed quality control (QC) tests.

Thirteen CD4⁺ T cell clusters were identified based on cluster marker feature genes and previously reported scRNA-seq data for CD4⁺ T cells (33, 38, 39) ( Figures 1A, B , Supplementary Figure 1B , Supplementary Table 3 ). CD4⁺ T cells included i) three clusters of CCR7- and SELL-expressing naïve (0–2-Tn), ii) one CXCR5⁺ follicular T cell like cluster (3-Tfh like), iii) three clusters of memory CD4⁺ T cells (4–6-Tm), iv) one cluster of Th1/Th2 intermediate cells expressing GZMK and TBX21, which suggests Th1 type, as well as GATA3, which suggests Th2 type (7-Th1/Th2 int); v) one cluster of Th1-type cells expressing high levels of GZMK, HLA-DR, and TBX21 (8-Th1), vi) one cluster of activated CD4⁺ T cells with high PRF1 expressing activated T cells (9-PRF⁺ Tct); vii) one cluster of FOXP3^- CD4⁺ T cells expressing PDCD1, LAG3, and IFNG (10-PDCD1⁺FOXP3^- Treg) with a gene expression signature similar to that of FOXP3^-PD-1⁺ Tregs reported previously in decidua (42), viii) one cluster of Th17-type cells with high RORC expression (11-Th17); and viii) one cluster of FOXP3⁺ Treg also expressing TIGIT, IL2RA, TNFRSF18 (also known as GITR) and ENTPD1 (also known as CD39) (12-FOXP3⁺ Treg) in concordance with prior studies (42) ( Figures 1A, B , Supplementary Figure 1B ). CD69 was expressed across all clusters of CD4⁺ T cells, with the exception of 6-Tm cells ( Figure 1B ), suggesting their residence in tissues as described previously (33, 35, 43). Thus, decidual CD4⁺ T cell populations were found to be heterogeneous and include distinct types of activated memory, helper, cytotoxic, and regulatory T cells.

Next, CD4⁺ T cells were separated into early and late gestation and PE and PBMC groups ( Figures 1C-E ). Decidual CD4⁺ T cells contained more abundant populations of memory, 7-Th1/Th2 int, 8-Th1, 9-PRF⁺ Tct, 10-PDCD1⁺FOXP3^- Treg, 11-Th17, and 12-FOXP3⁺ Treg than PBMC ( Figures 1C–E ). The frequency of CD4⁺ T cell subsets did not differ significantly among healthy early, healthy late, and PE groups because of their small sample sizes and inter-patient heterogeneity. However, an increased tendency of 3-Tfh like and 11-Th17 cells was observed in early gestation, whereas late gestation had a modest increase in tendency toward enrichment with the 0-Tn group ( Figure 1E ). Interestingly, decidual CD4⁺ T cells from PE samples showed an increased tendency toward 9-PRF⁺ Tct cells compared to cells found in late gestation ( Figure 1E ). The frequency of decidual 12-FOXP3⁺ Treg varied in PE ( Figure 1E ), as a recent meta-analysis also describes (44). Thus, changes in CD4⁺ T cell cluster frequencies are related to gestational age and the presence or absence of PE.

Next, we conducted a TCR analysis to assess clonal expansion within each cluster. Gene expression and TCR repertoire data were combined from 14,020 cells of five subjects (two early gestational decidua, one late gestational decidua, one PE decidua, and one PBMC sample). In total, 13,538 (96.6%) TCR alpha/beta-paired cells were detected. Clonal expansion was primarily observed in the 7-Th1/Th2 int, 8-Th1, 9-PRF⁺ Tct, and 12-FOXP3⁺ Treg clusters ( Figures 1F , 2A ). 10-PDCD1⁺FOXP3^- Treg was clonally expanded in one early gestation subject (early-decidua-2) ( Figures 1F , 2A ). Clonal expansions in 7-Th1/Th2 int and 8-Th1 were observed in early, late gestation and in PE ( Figures 1F , 2A ). Clonal expansion of 12-FOXP3⁺ Treg was observed in late gestation and PE, but not in early gestation ( Figures 1F , 2A ), which is consistent with our previous report (25). Inter-cluster overlap of expanded clonotypes (clone size = 2 or more) showed that the majority of expanded clones belonged to one cluster, whereas a few had multiple phenotypes ( Figures 2B-E ), confirming that TCR usage partly shapes the CD4⁺ T cell phenotype, as previously described (38, 45–47). Fewer shared clones with Tregs and CD4⁺ Th subsets have been reported in term decidua (48). We confirmed that clonally expanded cells in the 12-FOXP3⁺ Treg cluster shared fewer overlapping clones with other clusters ( Figures 2B–E ). In contrast, some clones overlapped among 7-Th1/Th2 int, 8-Th1, 9-PRF⁺ Tct, 11-Th17, and memory T cell subsets ( Figures 2B–E ).

Collectively, these results indicated that decidual CD4⁺ T cells are phenotypically and clonally heterogeneous. Naïve, CXCR5⁺ Tfh like, memory, Th1, PRF⁺ Tct, Th17, FOXP3^-PD-1⁺ Tregs, and FOXP3⁺ Treg clusters observed here were consistent with previously reported clusters confirmed by gene expression and flowcytometry (33, 38, 39, 42); however, 7-Th1/Th2 was not identified in previous reports. Despite inter-subject heterogeneity, clonal expansion was predominantly observed in 7-Th1/Th2 int, 8-Th1, 9-PRF⁺ Tct, and 12-FOXP3⁺ Treg, but not in naïve clusters in the donors analyzed. The 8-Th1, 9-PRF⁺ Tct, 11-Th17, and 7-Th1/Th2 clusters shared some clones, suggesting that they also share some differentiation processes. However, the major expanded clones were unique to each cluster ( Figures 2B–E ), supporting the validity of the presence of Th1/Th2 clusters in the decidua. The distinct TCR repertoire between 12-FOXP3⁺ Treg and CD4⁺ T cell subsets suggests that decidual FOXP3⁺ Tregs have clonal origins different from those of decidual CD4⁺ Th subsets.

Distinct decidual T-cell frequencies and transcriptomic signatures in early compared to late gestation have been reported previously (33, 35). However, the gestational age-dependent transcriptomic differences in CD4⁺ T cells are not well understood. An analysis of DEGs between early and late gestation was performed for each CD4⁺ T cell cluster. Enriched GOs from the DEGs were analyzed using Metascape (49). Most CD4⁺ T cell clusters had more downregulated genes in late than early gestation, and DEGs were particularly prevalent in the 5-Tm, 7-Th1/Th2 int, and 12-FOXP3⁺ Treg clusters ( Figure 3A , Supplementary Figures 2A–C ). Focusing on the GOs of downregulated genes in healthy late gestation compared to early gestation in these three clusters, similar GOs were shared across the clusters ( Figure 3B ). In particular, GOs related to migration, such as leukocyte migration and regulation of leukocyte migration, were shared among the 5-Tm, 7-Th1/Th2 int, and 12-FOXP3⁺ Treg clusters ( Figure 3B ). In addition, GOs related to activation, such as leukocyte activation, T cell activation, and adaptive immune response, were downregulated in 7-Th1/Th2 int and 12-FOXP3⁺ Treg in late gestation ( Figure 3B ). Genes involved in leukocyte activation, such as NKG7 and PRF1, and those involved in adaptive immune responses, such as TRAC and CD3G, were downregulated in 7-Th1/Th2 int in late gestation ( Figure 3B , Supplementary Figure 2B , Supplementary Table 4 ). In 12-FOXP3⁺ Treg, LGALS1 and CTLA4, which are related to Treg function, were upregulated in late gestation, whereas more genes related to T cell activation and adaptive immune response, such as HLA-class II molecules, were downregulated in late gestation ( Figure 3B , Supplementary Figure 2C , Supplementary Table 4 ).

In summary, this study revealed significant differences in gene expression of CD4⁺ T cells between healthy early and late decidua. Specifically, 5-Tm, 7-Th1/Th2 int, and 12-FOXP3⁺ Treg in healthy late gestation had substantially downregulated gene sets associated with leukocyte activation and migration compared to those in early gestation. Interestingly, 12-FOXP3⁺ Treg showed downregulation of genes involved in cell activation and immune responses; however, they showed upregulation of genes related to Treg function in late gestation.

Next, we compared DEGs between the healthy late gestation and PE groups to determine transcriptomic changes in T cell clusters. DEG analysis showed that genes upregulated in PE were mainly observed in clusters 5-Tm, 7-Th1/Th2 int, and 12-FOXP3⁺ Treg, which were consistent with the clusters that had the most DEGs in the early and late comparisons ( Figures 3C, E , Supplementary Figures 2D, E ). Next, GO analysis of upregulated genes with PE in the 5-Tm, 7-Th1/Th2 int, and 12-FOXP3⁺ Treg clusters revealed that activation-related GOs (lymphocyte activation, leukocyte activation, and T cell activation) were shared among the three clusters ( Figure 3D ). The cytokine signaling and regulation of lymphocyte activation pathways were upregulated in 5-Tm and 12-FOXP3⁺ Treg in PE ( Figure 3D ). In 7-Th1/Th2 int, leukocyte activation pathways, including NKG2 and PRF1, were activated in PE ( Figure 3D , Supplementary Figure 2E ). In 5-Tm, GOs including Th17 differentiation pathway, inflammatory bowel disease, and allograft rejection, represented by IL12RB1, IRF4, LAT, and IL22, were observed in PE ( Figure 3D , Supplementary Figure 2D , Supplementary Table 4 ). Previous studies have shown excessive Th17 cells in PE (26–29); however, in our study, the Th17 cluster did not show upregulated genes with PE. In addition, neither 5-Tm nor 11-Th17 cells showed clonal expansion in PE ( Figure 1F ), suggesting that a pro-inflammatory environment, and not antigen stimulation, may contribute to increased Th17 signatures in decidual CD4⁺ T cells.

Interestingly, gene sets associated with Th1 and Th2 differentiation, including GATA3 and STAT1, were upregulated in 12-FOXP3⁺ Treg from the PE group ( Figures 3D, E ). STAT1 overexpression (39) and GATA3 overexpression in FOXP3⁺ Tregs (50, 51) have been reported in autoimmune diseases. Furthermore, PDCD1, CCR5, and TNFRSF4 were upregulated only in 12-FOXP3⁺ Treg in the PE group ( Figure 3E ). Previous reports have shown that these genes are upregulated in exhausted Tregs with reduced suppressive function in autoimmune disease (39, 41). Another report showed that PD-1^high exhausted Tregs were significantly increased in the peripheral blood of patients with PE (52). Notably, the module score of Treg exhaustion-related genes (39, 41) was significantly higher in PE than in healthy late gestation ( Figure 3F ).

Taken together, 7-Th1/Th2 int and 5-Tm were activated and showed a proinflammatory signature, whereas 12-FOXP3⁺ Treg showed an exhausted signature in PE.

Next, we investigated the FOXP3⁺ Treg sub-cluster by further analyzing all 1354 cells from 11 samples ( Figure 4A ). Although CD45RA was not detected, five FOXP3⁺ Treg sub-clusters were identified that showed similarity in marker gene expression with previously identified FOXP3^lowCD45RA⁺ naïve Treg, FOXP3^highCD45RA^- effector Treg, and FOXP3^lowCD45RA^-T cells (21–23, 38–40, 53). CCR7, SELL, and TCF7 were highly expressed in Treg0 Naïve cluster, similar to naïve Tregs, as previously described. Treg1 TIGIT low cluster had low levels of TIGIT and ICOS, similar to FOXP3^lowCD45RA^- T cells. Treg4 effector like cluster had high levels of ICOS, TIGIT, ENTPD1, and CTLA-4, similar to effector Tregs ( Figure 4B , Supplementary Figure 3A ). The Treg4 effector like cluster expressed exhaustion-related genes such as HAVCR2, LAG3, and PDCD1; however, they also showed the highest expression of soluble suppressors such as PRF1 and LGALS1 ( Figure 4B , Supplementary Figure 3A ), supporting Treg4 effector like cluster as an effector and Treg-like properties. The effector Treg like subset, naïve Treg subset, and FOXP3^lowCD45RA^- T cell subset in decidua were previously reported by flowcytometry based studies (24, 25, 35). In addition, expanded clones were predominantly observed in Treg4 effector like cluster in the healthy late (36.4%) and PE (27.4%) ( Figure 4C ), consistent with our previous report (25). Treg2 and Treg3 had intermediate characteristics of Treg1 Naive and Treg4 effector like subsets in terms of Treg-specific marker genes; however, Treg3 showed higher expression of HLA-class II molecules than Treg2 ( Figure 4B , Supplementary Figure 3A ). Some scRNA-seq-based studies have identified naïve Treg-like subsets, effector Treg-like subsets, and various other undefined subsets (38–40, 53). The functions of these Treg subsets should be defined in future studies. The Treg0 Naïve cluster was abundant in PBMC, whereas Treg2, Treg3, and Treg4 effector like clusters were abundant in the decidua ( Supplementary Figures 3B–G ). Cells in the Treg 4 effector like subset were abundant during late gestation ( Supplementary Figure 3G ). Thus, decidual Tregs are heterogeneous, and changes in cluster frequencies depend on gestational age.

Next, we investigated whether a specific FOXP3⁺ Treg cluster drove the Treg exhaustion signature in PE samples ( Figure 3F ) by performing DEG analysis of Tregs in late healthy and PE samples. The number of upregulated genes in PE was highest in the Treg4 effector like subset ( Figure 4D ), and DEG analysis of Treg4 effector like cells showed that Treg exhaustion-related genes (PDCD1, TNFRSF25, TNFRSF4, CCR5, LAG3, and HAVCR2) (39, 41) were more highly expressed in PE than in late healthy pregnancy ( Figure 4E ). Increased expression of STAT1 and GATA3 in PE samples was also observed in Treg4 effector like cluster ( Figure 4E ). Based on the results of our TCR clonality and DEG analyses, the Treg4 effector like clusters may reflect antigen stimulation, driving the Treg exhaustion signature in PE.