Single cell analysis of uterine artery endothelial cells reveals cytokine induced emergence of specific immunomodulatory subtypes: implications for preeclampsia

Abstract

While pregnancy is known to be an inflammatory condition, preeclampsia (PE) is associated with higher chemokines and pro-inflammatory cytokines, and higher Th1/Th2 and Th17/Treg ratios. Since the uteroplacental space can secrete cytokines, including TNF and IL1B, a common assumption is the proinflammatory immune cell profile of Th1 and Th17 cells dominating over Th2 and Treg cells begins in that space. To date, a possible role for endothelium in this initiation process has not been considered. Nonetheless, recent publications show that endothelium can become immunomodulatory on exposure to TNF and IL1B, and in systemic hypertension, endothelium has been shown to exist as multiple cell subtypes. We have recently shown that uterine artery endothelial cells from late-pregnant sheep (P-UAEC) treated with TNF alone secrete many of the chemokines and cytokines further elevated in PE subjects. Herein we show that P-UAEC also exist in multiple subtypes with distinct chemokine and cytokine secretory and immunomodulatory properties. The 5 subtypes are differentially regulated by TNF-alpha (TNF) and IL1-beta (IL1B) that may favor subtype specific binding and interaction with distinct classes of Th cells, and an altered ability to respond to Th secreted cytokines (such as IL17 and IL10). Thus, our data demonstrates the possibility that certain endothelial cell subtypes can be pushed to express immunomodulatory proteins by early exposure to increases in TNF or IL1B of immune cell, trophoblast and decidual origin. This in turn begs the question if such endothelial changes could contribute to subsequent immune disturbances seen at the time of clinical presentation.

Introduction

The uterine artery is particularly important in supplying blood to the uteroplacental unit in pregnancy and as such must undergo considerable functional adaptation. We have previously shown cells of the uterine artery develop enhanced cell signaling to maintain junctional assemblies between cells, and increase connexin connectivity to enhance and coordinate Ca2+ signaling necessary for sustained vasodilatory function (Yi et al 2010, Bird et al 2000, 2013). Conversely, preeclampsia (PE) is a hypertensive condition in pregnancy associated with failure of this endothelial adaptation process. Roughly 5% of all human pregnancies in the USA are diagnosed with PE (Mills et al, 2016), resulting in significant morbidity and risk of maternal or fetal death. There is also an increased subsequent risk of cardiovascular disorders throughout the lifespan for mother and fetus (Maric-Bilkan et al, 2019). While we know how to diagnose PE, we have a poor understanding of the etiology of the condition, and current treatments are directed at the life-threatening symptoms, and not the root cause. As such, approvals of new and effective therapies remain elusive. The clinical presentation phase of PE is associated with a higher degree of ‘inflammation’ (Redman & Sargent 2003), and an immune profile wherein proinflammatory immune cells are dominant (including but not limited to Th1 cells over Th2 cells, and Th17 cells over Treg cells) (Jafri & Ormiston 2017). It is perhaps no coincidence that a prior history of proinflammatory diseases including hypertension and diabetes are also a risk factor for PE (Redman & Sargent 2003). Nonetheless, PE can develop without such a prior history, and many studies on placental tissue explants and purified cells suggest TNF and IL1B can be secreted en mass directly from trophoblasts under excessively hypoxic conditions (Benyo et al., 1997, Hung et al., 2004, Rong et al., 2011). Clinical symptoms including excessive hypertension and edema only present after many weeks. In our companion paper (Clemente et al, 2025), we report that uterine artery endothelial cells from pregnant sheep (P-UAECs) treated with TNF show overall changes in gene expression that are largely associated with changes in chemokine and cytokine secretion in women with PE, and the chemokines include those likely to attract Th1 and Th17 cells known to dominate in PE subjects (Jafri & Ormiston 2017). The recent findings in HUVEC that cytokines can also drive changes in immune modulatory function (Mai et al 2013) would further suggest that early changes in uterine artery endothelium itself could precipitate subsequent adverse endothelial cell immune cell interactions in PE subjects seen in the clinical presentation phase. Our purpose here is to ask the question if, regardless of origin, early rises in TNF or IL1B could promote UAEC to adopt immunomodulatory behavior that could in turn help precipitate or accelerate the immune dysfunction observed in the subsequent ‘clinical presentation’ phase.

Another advance that drives our study is the recent development of the cell atlas. Together with the advent of single cell technologies, there is an increasing recognition that even within a ‘pure’ cell population there may be many cell subtypes with distinct functional variance ready to respond alone or together to multiple insults, as opposed to all cells being identical and adjusting together. The increasing use of spectral flow cytometry to observe more antibody-based surface markers simultaneously, combined with more recent single cell RNA sequencing methodology allows us to ask critical questions about possible subtypes within an endothelial cell population that were not possible ten years ago. To that end, in this study we also looked at the impact of TNF and IL1B (above) to ask if (i) they alter uterine artery endothelium at the level of cell subpopulations, and (ii) any changes in number and/or functional state of any of the cell subpopulations that may match clinical observations of PE associated changes in circulating cytokines, chemokines and/or cell subtypes with the potential for ongoing direct and/or paracrine interactions with proinflammatory immune cells. The combined application of multiple single cell techniques do indeed confirm multiple P-UAEC subtypes exist with distinct paracrine and direct immune cell binding capabilities. Our findings suggest a more direct role for cytokine induced changes in P-UAEC subtypes in promoting the subsequent clinical observations reported in PE than was previously suspected. In particular, early exposure of P-UAEC to TNF and IL1B in particular may lay the foundation for emergence of cell subpopulations more geared to direct Th17 cell binding and adverse bidirectional paracrine interaction. If true, this could have a profound impact on our understanding of the early etiology of PE, and open doorways to new approaches to treatment.

Methods

General Reagents 1:

All cell culture reagents, unless otherwise stated, were obtained from Life Technologies (Carlsbad, CA), and all other basic chemicals and reagents were from Sigma (St. Louis, MO) unless otherwise stated. Gentamycin, penicillin/streptomycin were from Gibco (Grand Island, NY). TNF and IL1B were from R&D Systems, Inc. (Minneapolis, MN). For cell imaging, MatTek 35mm ‘glass dishes’ were from MatTek, Asland MA. Fura-2 AM and Pluronic F127 were purchased from Life Technologies (Carlsbad, CA). For initial preparation of P-UAEC stocks, growth media was prepared from Minimum Essential Medium (MEM), Fetal Bovine Serum, Penicillin/Streptomycin, and Gentamycin. Otherwise, all other ‘treatment’ reagents were made up per manufacturer recommendations in sterile H₂O, sterile PBS, or DMSO. NOTE: Final concentration DMSO in cell culture was below 0.1%, which results in no measurable effects on protein expression/phosphorylation, calcium signaling, or barrier function of P-UAEC (unpublished data).

P-UAEC Cell isolation and pooling prior to experimental use:

Pregnant sheep at >120 days gestation were terminated and uterine arteries collected in accordance with Institutional Animal Care and Use Committees (IACUC) protocol. Uterine artery endothelial cells (P-UAEC) were isolated and grown in culture as described (Bird et al 2000) to passage 3 before freezing and storage in liquid nitrogen dewers. During this time, cells were grown using MEM, 20% fetal bovine serum (FBS), 1% penicillin and 4 ug/ml gentamycin to >80% confluence each passage, and split 1:4. After passaged 3, individual vials of cells from 4 sheep were thawed and grown to near confluence, pooled, split 1:4 and refrozen at passage 4 for experimental use.

Analysis of P-UAEC surface proteins by spectral flow cytometry:

Antibody panel.

All antibodies (Suppl Table 1), Flow Cytometry Cell Staining Buffer, Zombie NIR, and other related flow cytometry related materials were from BioLegend (San Francisco, CA) except TL1A antibody from Thermo Fisher (Waltman, MA), Fractalkine from BD Biosciences (San Jose, CA) and VEGFR2 from NOVUS (St. Louis, Missouri). All antibodies were anti-human. Antibodies were added at half the dosage of manufacturer recommendations at final dilutions (after the initial titration pilot), sufficient to achieve saturation binding to target proteins while minimizing nonselective binding. Other reagents (TNF, IL1B, IFNg) were formulated per manufacturer recommendations (in DMSO, Sterile PBS, or Sterile water) and were diluted 1:1000 before addition as above, so the vehicle was below 0.1%. Flow cytometry tubes (Falcon, with caps) were from Corning Life Sciences (Tewksbury, MA).

Spectral Flow Cytometry

Three independent spectral flow cytometry experiments were conducted using passage 4 P-UAEC, prepared as described above. The purpose of the experiments described here was to identify changes in the expression levels of known endothelial cell markers and antigen presentation markers following treatment with selected pro-inflammatory cytokines TNF, IL1B, or IFNg. We created an antibody panel based on the observations of Pober et al (2017) that targeted 4 endothelial cell markers and 14 antigen presentation/immunomodulatory markers (Suppl Table 1). All data were acquired between August and October 2022 at the University of Wisconsin Carbone Cancer Center (UWCCC) Flow Cytometry Laboratory on a Cytek Aurora Spectral Flow Cytometer equipped with five lasers that provide 64 channels (16 UV, 350 nm; 16 violet, 405 nm; 14 blue, 488 nm; 10 yellow/green, 561 nm; 8 red, 640 nm). Because we wished to investigate a broad range of protein targets for this pilot study, and commercially available fluorochrome conjugates for several antibodies in our panel were limited, a moderate to high degree of spectral interference between some channels was unavoidable but within acceptable limits (Suppl Fig 1).

Cell Treatments and Data Processing.

Previously frozen P-UAEC were plated in eight T-75 flasks (with initial attachment at 5–10% confluence) in 10ml Endothelial Cell Medium (ECM; 500 ml of basal medium, 25 ml of fetal bovine serum, 5 ml of endothelial cell growth supplement, 5 ml pen/strep and 200 µl gentamycin; ScienCell, Carlsbad, CA) and incubated at 37^oC. Media was replaced on day 2 (48hrs). On day 4, when cells were at ~75% confluence, medium was aspirated from each flask and replaced with 10ml fresh serum-free basal ECM. Following a 6 h incubation in serum-free medium at 37^oC, the eight T-75 flasks were separated into four pairs, and treated with TNF, IL1B, or IFNg (each at a final concentration of 10ng/ml; R&D Systems, Minneapolis, MN) or vehicle, respectively. The flasks were then incubated for an additional 20 hrs. Also on Day 4, a set of 41 labeled Eppendorf tubes was prepared for each of the four treatment groups comprising 19 single stained controls (18 antibodies + 1 viability dye; labeled A through S), 19 FMO controls (labeled -A through -S), two full stained samples (labeled Ta & Tb; NOTE: Technically, only one is needed, but a second tube was added for redundancy), and one unstained sample (labeled U). Based on titrations of several antibodies, a working suspension of each antibody was created by dilution with Cell Staining Buffer (BioLegend, San Diego, CA) such that 10µL of working stock dispensed the amount of antibody listed in Suppl Table 1 to all appropriate tubes. After pipetting the desired amount of each antibody to the appropriate tubes, all 164 sample tubes (4 treatment groups x 41 tubes) were capped and kept refrigerated at 4^oC in the dark overnight. Day 5: Following 20 h incubation, the medium was removed from all cell culture flasks. The confluent monolayers of cells from each treatment pair were then harvested using 5ml of 0.25% trypsin/EDTA (Life Technologies, Grand Island, NY) incubated for 5 min at 37^oC.The trypsinization reaction was stopped by the addition of 5ml ECM. Each treatment pair was then combined into one of four labeled 50ml conical tubes. The cell suspensions were centrifuged at 300g for 4 min, supernatant was discarded, the cell pellet in each tube was resuspended in 2ml PBS, and the four tubes were placed on ice. (NOTE: All further cell preparation/incubation steps were performed on ice at 4^oC.) Cells were counted by hemocytometer to determine yield and concentration for each treatment group. To identify dead cells, 1ml (half of the cells) of each of the four cell suspensions was pipetted into a separate 15ml conical tube and incubated with 0.5µL (1:2000) Zombie NIR Viability Dye (BioLegend, San Diego, CA) for 15 min. Following dye incubation, the Zombie NIR-stained and unstained cell samples were separately centrifuged at 300g for 4 min, supernatant was discarded, and the cell pellet in each tube was resuspended in 1ml flow cytometry buffer (PBS, 2% FBS, 500µM EDTA). For each treatment group, 100,000 cells incubated with Zombie NIR were aliquoted into tubes -A through -R, S, Ta, and Tb. Tubes A through R, -S, and U each received 100,000 cells NOT stained with Zombie NIR. Additional flow cytometry buffer was then added as needed to bring the volume of each tube to 200µL. All 164 sample tubes were then incubated on ice in the dark for 45 min. Following this ‘antibody incubation’, all tubes were centrifuged at 300g for 4 min, supernatant aspirated, and the cell pellet in each tube was resuspended in 200µL flow cytometry buffer. Contents of each Eppendorf tube were then transferred to a corresponding labeled 5ml clear flow cytometry tube on ice for analysis by the Cytek Aurora. Machine QC was performed, and channel voltage settings were set by the UWCCC Flow Cytometry Laboratory staff. Forward scatter was set to 30; side scatter was set to 100. Data from a minimum 20,000 up to 50,000 cells were obtained from each tube. Raw flow cytometry data were unmixed using SpectroFlo software, and further processed using FlowJo v10.8.1. The data from tubes Ta and Tb were concatenated before quantifying the median fluorescence intensity (MFI) of each protein target. The net change in median fluorescence intensity (ΔMFI) was defined as the unmixed MFI of a target fluorochrome in the full stained sample minus the MFI of its corresponding FMO control sample. To normalize the ΔMFI values in each experiment, the FMO-corrected values from each treatment group were divided by the mean ΔMFI value obtained from all four treatment groups (ΔMFI/mean ΔMFI).

Further tSNE Analysis of Flow Cytometry Data:

Following the analysis of bulk data, a preliminary single-cell analysis was then undertaken by Dimensionality Reduction via the t-SNE algorithm. tSNE images of spectral flow cytometry data from independent flow experiment 2 were generated using the tSNE function of FlowJo 10.9 (BD Life Sciences) set to the following parameters: Auto (opt-SNE) Learning Configuration, 1000 iterations, Perplexity – 30, Learning Rate (eta) ~4000–7000, Exact (vantage point tree) KNN algorithm and Barnes-Hut gradient algorithm. Clusters within treatment groups were then identified from tSNE data using “FlowSOM (3.0.18 PP)” (Van Gassen et al 2015) and “Cluster Explorer” plug-ins for FlowJo 10.9. For each of the four treatment groups, FlowSOM was set to generate 4, 5, 6, 7, or 8 clusters. The number of clusters selected to represent each group was the lowest number that preserved relevant differences. Heatmaps and cluster percentage charts were subsequently generated by Cluster Explorer.

Analysis of individual P-UAEC by CITE-Seq Analysis:

The cell treatment protocol for spectral flow cytometry (above) and Cellular Indexing of Transcriptomes and Epitopes by Sequencing (CITE-Seq) were identical for P-UAEC until Day 5, except for the modifications described here. Two ‘biological’ experiments were performed: the first comparing IFNg-treated P-UAEC with vehicle-treated cells, and a second comparing TNF or IL1B (each at a final concentration of 10ng/ml) with vehicle treated cells. The first experiment required fewer cells and fewer T-75 flasks (half the flasks required for the other experiments) since the only treatment groups were vehicle and IFNg. For the second experiment, cells were cultured in eight T-75 flasks: two for each treatment group and two backup flasks for further redundancy. On Day 5, following 20 h incubation, the confluent monolayers of cells from each treatment pair were harvested using 5 ml of 0.25% trypsin/EDTA (Life Technologies, Grand Island, NY) for 5 min in the BSC at room temperature, followed by 5 ml ECM to stop trypsinization. Each treatment pair was then combined into one of the labeled 50 ml conical tubes (one conical tube per treatment). The cell suspensions were centrifuged at 300g for 4 min, supernatant was discarded, the cell pellet in each tube was resuspended in 2 ml PBS, and the tubes were placed on ice. (NOTE: All further cell preparation/incubation steps were performed on ice at 4^oC). Recovered cells were counted by hemocytometer to determine yield and concentration for each treatment group. Working in 15ml conical tubes, cells were diluted to 100,000 cells per treatment in ~50 ul of Cell Staining Buffer (BioLegend, San Diego, CA), and treated with 0.5 µL of Human TruStain FcX™ Fc Blocking reagent (BioLegend, San Diego, CA) for 10 min at 4°C. A panel of 14 TotalSeq™ antibodies (BioLegend, San Diego, CA) was prepared (see Suppl Table 2) by adding 1 μL (of a 1/5 dilution (equivalent to 0.2 μL of antibody) of each of the 14 titrated antibodies to 236 μL of Cell Staining Buffer, bringing the total volume to 250 μL for each treatment group. The TotalSeq™ antibody cocktail was added to the 50 μL cell suspension, incubated for 30 minutes at 4°C, diluted with 3 ml of Cell Staining Buffer, centrifuged at 400g, washed with buffer again, and centrifuged a second time. The supernatant was manually poured off, and MEM with 5% FBS (200 μL) was added to the cells for an approximate final volume of ~250µL. Finally, cells were filtered through a 40 µm Flowmi™ Cell Strainer (Sigma, St. Louis, MO). Antibody-tagged cell suspensions were then delivered to the UW-Madison Biotechnology Center for CITE-Seq quantification of mRNA and surface protein expression.

CITE-Seq data analysis:

Initially, the UW Biotechnology Center ran 10x genomics Cell Ranger 7.1.0 (Zheng et al 2017) on the sequenced P-UAEC cells using the ovine reference GCF_016772045.1_ARS-UI_Ramb_v2.0 (Davenport et al 2022). Summary data for the RNA (gene expression, GEX) and antibody capture (ADT) was generated from the cell Ranger results. Based on those results it was decided to only look at the RNA gene expression data. Analysis of the cells was undertaken using Seurat v4.3.0 (Hao et al 2021) following previous work in the Ong lab (Chasman et al 2023).

Quality control was performed on the grouped samples using Median Absolute deviation (MAD) factors for the RNA counts and number of features per cell with a threshold of three standard deviations. After QC data batch effects were identified using principal components analysis (PCA) and linear regression between principal components (PC) and the factors of sample, treatments, and day of experiment. The results indicated batch effects of day, so the software Harmony v0.1.1 was utilized to integrate the experiments together and remove batch effects. As Harmony performs the integration on embeddings such as PC, the Seurat workflow was followed. This involved normalization and scaling the data, finding variable features, PCA with number of PC’s set to five as selected based on an elbow plot, Harmony integration, and lastly UMAP using the Harmony embedding rather than PCA. Harmony implementation involved using the Seurat packages wrapper function RunHarmony.

Next, preliminary cell types were indicated at the full data level using SingleR v2.0.0 with the celldex v1.8.0 Human Primary Cell Atlas Data at both the top-level labels (main) and subcellular level (fine labels). Following this initial cell typing Seurat pipeline was again followed to conduct cluster at different resolutions (0.1, 0.25, 0.75, 1.0, 1.5, and 2.0). With per cluster cell typing also being conducted at each resolution using the same software and reference set but only the fine labels being reported. Selection of the final clustering resolution was decided based on silhouette widths and cluster purity using the bluster v.1.8.0 R package. Following this it was decided to remove one small cluster where the cell types were not endothelial.

Finally, differential gene expression analysis was conducted using DESeq2 v1.38.0 with a pseudo bulk implementation first looking at the full data (similar to bulk RNA seq) and at the cluster level. The contrasts compared were: TNF verses the controls, IL1B verses the controls, IFNg verses the controls, TNF verse IL1B, TNF verses IFNg, and IL1B versed IFNg. Due to one sample per treatment condition and needing to include batch in the model it was necessary to create pseudo replicates per treatment per day. This was done by modifying the pseudoReplicates and uniqueFeatures function from Nicholas Mikolajewicz (NMikolajewicz/scMiko: scRNAseq analysis functions. Developed using Seurat framework. version 0.1.0 from GitHub (rdrr.io)). The number of replicates was set to three, and grouped based on the original sample identities.

Performing DE by the fulldata and per cluster utilized the same steps. The first step was generating an aggregated sum matrix of all counts per gene for each pseudo replicated sample. Second was conversion to DESeq2 object and setting the model to be ~batch + treatment, where batch was either batch1 or batch2 for the different experiment day and treatment of control, TNF, IL1B, or IFNg. Implementation of DESeq2’s DE analysis was conducted by first estimating the size factor and dispersion effects on the data and then performing the negative binomial Wald test. The key difference between full and cluster analysis was that the aggregated count matrix was split per cluster and steps two and three were repeated for each cluster.

All methods using R were run using R version 4.2.2 within a CONDA environment. CellRanger was run via command line.

Results

While we have previously studied extensively the actions of TNF on P-UAEC, but not IL1B, we began by verifying the acute physiologic action of 10 ng/ml IL1B on P-UAEC monolayer integrity by ECIS. The results (Suppl Fig 2A) confirmed the previously reported observation (Ampey et al 2021) that TNF at 10 ng/ml can substantially damage monolayer integrity and reduce monolayer resistance, and 10 ng/ml IL1B degrades the P-UAEC monolayer with a smaller but significant effect. We also confirmed that acute IL1B exposure can cause a significant drop in Ca2+ bursts (Suppl Fig 2B), again consistent with prior findings for acute TNF exposure (Ampey et al 2019). Given these deleterious effects of both cytokines on P-UAEC monolayer integrity and Ca2+ signaling, we proceeded into the single cell analysis phase of the study examining the effects of TNF vs IL1B on endocrine and immune modulatory properties of P-UAEC at the single cell level, and we did so using IFNg as a positive control.

Single Cell Analysis reveals distinct endothelial subtypes.

Before we begin this section, it is worth pointing out the terminology used below is tailored to prevent false assumptions of equivalence across methods. For cell clusters identified from cell surface protein targeted antibodies used in spectral flow cytometry analysis, we refer to the cluster of cells as a subgroup. For cell clusters recognized by principal component analysis based on top 50 transcripts from CITE-Seq transcriptome data, we refer to identified cell clusters as subtype. While both approaches yield clusters that may be equivalent, the actual number assigned to the subgroup or subtype will not necessarily be the same ie subgroup 3 and subtype 3 cannot be assumed to be the same cluster.

Cytokine treatment of P-UAEC alters plasma membrane expression of immunomodulatory proteins.

To investigate how exposure to inflammatory cytokines affects the expression of immune cell binding and co-activation ligands and other proteins of interest on the surface of P-UAEC, we conducted three independent spectral flow cytometry experiments. P-UAEC monolayers approaching confluence were incubated with 10 ng/ml TNF, IL1B, IFNg, or vehicle for 20 hours before cells were harvested and labeled with a panel of 18 fluorescently-tagged antibodies selective for 4 endothelial cell markers, one MHC class II molecule, 5 adhesion protein markers, and 8 immune co-stimulatory markers (selected on the basis of our initial findings above and those reported in other endothelial cells – see particularly the review of Pober et al 2017). It should be noted that all antibodies in our panel are selective to human homologs of our target proteins, with an overall level of bound fluorescence averaged 20 times greater in human umbilical vein endothelial cells (HUVEC) than in ovine P-UAEC (data not shown). Despite this limitation, all target proteins were detected. In preliminary analysis, differences between treatment groups in the median fluorescence intensity (MFI) of each fluorochrome in the fully stained samples, corrected by FMO controls, served as a measure of relative protein expression (Fig 1). P values were calculated by one-way ANOVA with Dunnett’s multiple comparison test.

Figure 1: Spectral flow cytometric quantification of protein expression on the plasma membrane of P-UAEC following cytokine treatment.

Cultured P-UAEC were treated for 20 hrs with TNFα, IL1B, IFNg, or vehicle, then labeled with 18 fluorochrome-tagged protein-selective antibodies as described. Results were obtained from n=3 independent spectral flow cytometry experiments, all conducted using a Cytek Aurora Spectral Flow Cytometer. Data sets were unmixed using SpectroFlo software, then gated and quantified using FlowJo software. The net change in median fluorescence intensity (ΔMFI) is defined as the unmixed MFI of a target fluorochrome in the full stained sample minus the MFI of its corresponding FMO control sample. Panel A: Endothelial Markers. No significant changes in the median expression of endothelial cell markers (PECAM1, VE cadherin, TIE2, and VEGFR2) were detected with the notable exception of TIE2 following IL1B treatment (P = 0.007). In all three experiments, the MFI of PECAM1 in the IFN-treated group was no greater than that of the FMO control. This was not found to be significantly different than vehicle by ANOVA (P = 0.307). Panel B: MHC Markers. No difference in the median expression of the MHC class II molecule HLA-DR was observed between treatment groups. Panel C: Adhesion Molecules. Expression of the adhesion molecule LFA3 was significantly increased by TNF treatment (P = 0.042). TNF also appeared to suppress the expression of ICAM1 when compared to vehicle in all three experiments (P = 0.082). Panel D: Costimulators/Immunomodulatory molecules. ICOS-L expression was strongly stimulated by TNF (P = 0.006). Indeed, 4–1BB-L, OX40L, and ICOS-L (three of the major immune cell co-activation ligands) all show a trend toward elevation following cytokine treatment (4–1BB-L by INFg , P = 0.124; OX40L by TNF , P = 0.085; ICOS-L by IL1B, P = 0.060). In contrast, IL1B treatment strongly reduced TL1A expression compared to the vehicle group (P = 0.044). P values were obtained by one-way ANOVA with Dunnett’s multiple comparison test. Green bars indicate P<0.05 while yellow indicate those approaching but not reaching P<0.05.

We predicted that if P-UAEC were changing from a “classical” endothelial cell phenotype to an immunomodulatory phenotype, we might also observe a decrease in the expression of classic endothelial cell markers as typified by the previously reported loss of VE CAD in (i) P-UAEC treated with TNF, (ii) in HUVEC from PE pregnancy, (iii) on exposure of normal HUVEC to trophoblasts from PE pregnancy (Ampey et al 2021, Wang et al 2002, 2004). However, no significant changes in the median expression of endothelial cell markers (PECAM1 (CD31), VE Cadherin (VE-Cad), and VEGFR2) were detected following IL1B treatment in our preliminary analysis, while TIE2 was significantly different (P = 0.007). In all three experiments, the ΔMFI of PECAM1 in the IFNg-treated group was zero (i.e. signal no greater than that of the FMO control), and this was not found to be significantly different than vehicle by ANOVA (P = 0.307) (Fig 1A).

IFNg has also been reported to induce greater expression of the critical MHC class II molecule HLA-DR in HUVEC after 96 hours (Geppert & Lipsky 1985; Wedgwood et al 1988), but no difference in the median expression of HLA-DR was observed on P-UAEC between cytokine-treated groups and control after 20 hours (Fig 1B). In contrast, expression of the adhesion molecule LFA3 (which binds to CD2 on T lymphocytes), was significantly increased by TNF treatment (P = 0.042). TNF also appeared to suppress the expression of ICAM1 overall when compared to vehicle in all three experiments (P = 0.082 by ANOVA, but P=0.031 by unpaired t test) (Fig 1C). This result contrasts previously published reports of TNF causing a significant increases in ICAM1 expression in other endothelial cells (Doukas & Pober 1990, Karmann et al 1995). TNF and IFNg have also been reported to enhance expression of CD40 on endothelial cells (Karmann et al 1995, Yellin et al 1995, Wagner et al 2002; Valenzuela 2021), but by this analysis, none of the cytokine-treated groups appeared to increase the level of CD40 on P-UAEC above that of the vehicle-treated group. However, expression of the T lymphocyte co-activator ICOSL was strongly stimulated overall by TNF (P = 0.006). Conversely, 4–1BB-L, OX40L, and ICOSL (three of the major immune cell co-activation ligands) appeared to show elevation, but did not achieve significance at P< 0.05 following cytokine treatment (4–1BB-L by IFNg, P = 0.124; OX40L by TNF, P = 0.085; ICOSL by IL1B, P = 0.060). In contrast, IL1B treatment strongly reduced overall TL1A expression compared to the vehicle group (P = 0.044) (Fig. 1D).

TNF treatment of P-UAEC appears to virtually eliminate expression of CD40 and ICAM1 in a majority of, but not all cells.

The analysis above was performed on data from combined cells en masse, so if the expected changes were confined to subgroups of cells rather than all cells, critical changes may have been missed. To that end, for each of the three independent spectral flow cytometry experiments, contour plots (See Suppl Fig 3 for representative plots) were used to further determine the percentage of cells that were positive for each target protein in each treatment group. Proteins listed in the top half of the heat map tended to be expressed in all or most cells, whereas proteins listed near the bottom were sometimes undetected on a large fraction of cells within treatment groups. Despite reports of TNF treatment causing increases in ICAM1 and CD40 expression in a variety of endothelial cell types (as cited before), Fig 2 shows that ICAM1 was only highly expressed in a minority of TNF treated ovine P-UAEC in two experiments, and CD40 was only detected at higher levels in a smaller percentage of cells than in the vehicle-treated controls in all three experiments. This tentatively suggested to us that there may be subgroups of UAEC that are highly specialized in immune modulatory function, and these populations can change on further treatment. To that end, further analysis at the single cell level was undertaken.

Figure 2: Percentage of cells with detectable fluorescence above basal autofluorescence.

For each of three independent spectral flow cytometry experiments, contour plots (see supplemental data for examples) were used to further determine the percentage of cells with detectable protein in each treatment group. Bar colors indicate only the percentage of cells on which a target protein is detected, not protein expression levels. Despite reports of TNF treatment causing increases in ICAM1 and CD40 expression in a variety of endothelial cell types (Karmann et al 1995, Yellin et al 1995, Wagner et al 2002; Valenzuela 2021), it is noteworthy that ICAM1 was only detected in a minority of TNF-treated ovine P-UAEC in two experiments, and CD40 was detected in a smaller percentage than observed in the vehicle-treated group in all three experiments.

Cellular response to cytokine treatment is heterogeneous within treatment groups.

Given the shift in focus to possible cell subpopulations, we next performed tSNE analysis on each of the three experimental data sets. These tSNE images generated by FlowJo from each of the three experiments further suggest different cell populations within treatment groups expressing distinct surface protein profiles. The images from Experiment 2 shown in Fig 3A-D are generally the most representative of the overall patterns of heterogeneity observed in all three experiments. Note that values analyzed by tSNE are not subject to correction by FMO controls. Therefore, the magnitude of color changes do not necessarily depict the magnitude of protein expression, but they do report the heterogeneity of the direction of change in expression per cell. For both the TNF and IL1B-treated P-UAEC, the plots showed no significant difference in expression of the endothelial cell marker PECAM1 compared to vehicle control; however, PECAM1 was expressed quite robustly on some cells and negligibly on others within the TNF- and IL1B-treated groups, suggesting that there are indeed subgroups of P-UAEC that respond uniquely. For TIE2 and VEGFR2, differences in expression levels within treatment groups were far less striking (Fig. 3A). In contrast to the pattern observed for PECAM1, expression of the MHC class II molecule HLA-DR was only moderately variable across all cells in the TNF and IL1B-treated groups, but was more variable within the IFNg-treated group (Fig. 3B). Cells within all treatment groups strongly expressed the adhesion proteins LFA3 and E-Selectin which remained at a relatively consistent level. In contrast, ICAM1 expression was highly variable within the TNF-treated group, displaying color patterns that were curiously opposite to the patterns displayed by PECAM1. The patterns of expression displayed by VCAM1 also tended to be opposite to those of ICAM1 (Fig. 3C). With regard to T lymphocyte co-activation molecules, the variable patterns of CD40 expression within treatment groups appear to be largely opposite those displayed by PECAM1. Interestingly, the expression of CD40 and ICAM1 appear coincident in TNF-treated cells, but not in vehicle-treated cells. Based on the greater polarization in the tSNE image for CD40, it appears there are sub populations of cells in each treatment group in which CD40 expression was higher than that observed in the other cells. This observation raised the possibility that a subset of P-UAEC may be primed to interact with CD40L on attached immune cells. Beyond CD40/CD40L, the three major co-activation molecules 4–1BB-L, OX40L, and ICOSL tended to show less dramatically polarized patterns of surface expression, but still varying degrees of expression by treatment groups. TL1A expression was largely homogeneous within treatment groups but with notable ‘hot spots’ (Fig. 3D).

Figure 3: tSNE and cluster analyses of spectral flow cytometry experiments suggests heterogeneity of responses within treatment groups.

tSNE images generated by FlowJo from each of the three experiments suggest different levels of protein expression within treatment groups. The tSNE images from Experiment 2 shown here were generally most representative of the three experiments. Since tSNE analysis is not subject to FMO correction, colors do not necessarily depict the magnitude of protein expression accurately but do reflect the heterogeneity of changes. Panel A: Endothelial Markers. For both the TNF- and IL1B-treated P-UAEC, the endothelial cell marker CD31 (PECAM1) was expressed quite robustly on some cells and negligibly on others. Similarly, CD144 (VE cadherin) expression on the IFNg-treated cells was variable. Expression levels of CD202b (TIE2) and CD309 (VEGFR2) remained consistent within treatment groups. Panel B: MHC Markers. Expression of the MHC class II molecule HLA-DR was similar across all cells in the TNF- and IL1B-treated groups, but was more variable within the IFNg-treated group. Panel C: Adhesion Molecules. Cells within all treatment groups expressed the adhesion proteins CD58 (LFA3) and CD62e (E-selectin) at a consistent level. In the TNF-treated group, CD54 (ICAM1) expression appeared to be highly variable, displaying color patterns that were largely opposite to the patterns displayed by PECAM1. Panel D: Costimulator/ Immunomodulatory molecules. The variable patterns of CD40 expression within treatment groups appear to be largely opposite those displayed by PECAM1. The three co-activation molecules—CD137L, CD252, and CD275 (4–1BB-L, OX40L, and ICOS-L, respectively)—tended to be expressed in a variable fashion relative to control. but with less extreme polarization within treatment groups, similarly to TL1A. Panels E & F: Cluster Analysis. Further cluster analysis of the tSNE data was performed for each treatment group using the ‘FlowSOM’ and ‘Cluster Analysis’ plugins for FlowJo 10.9, as described in Methods. Note that the colored labels assigned to the clusters in one treatment group (Pop 0, Pop 1, etc.) have no connection to the same labels that define clusters in another group. The analyses identified at least 4 to 6 distinct subpopulations of P-UAEC with unique protein expression patterns within each treatment group. The heatmaps generated also show in more detail the heterogeneity of responses suggested by the tSNE images. Consistent with tSNE, there was one cluster in each of the TNF, IL1B, and vehicle groups in which CD40 expression was considerably higher than in the other clusters in the group.

To more clearly quantify the heterogeneity of surface protein expression within treatment groups in greater detail, cluster analysis of the tSNE data was performed for each group using the ‘FlowSOM’ and ‘Cluster Analysis’ plugins for FlowJo 10.9, as described in the Methods section. The analyses successfully identified between 4 and 6 distinct subgroups of P-UAEC with unique protein expression patterns within each treatment group. The vehicle, TNF, and IL1B-treated cells were each divided into three large clusters (each containing 20–47% of the cells in the group) and one to three small clusters (each containing 1–7% of the cells in the group). The IFNg-treated cells exhibited the most uniform response, with one subgroup comprising almost 80% of the total number (Fig. 3E). Heatmaps again show in more specific detail the heterogeneity of responses that were suggested earlier by the tSNE images. Despite the identification of four significant subgroups within the vehicle-treated control cells, many proteins in this subgroup exhibited a similar level of expression across clusters. Nonetheless, VCAM1 (CD106) and ICAM1 (CD 54) showed some degree of polarization by cell subgroup per treatment, and the three major co-activation molecules 4–1BB-L, OX40L, and ICOSL tended to show less dramatic polarization but still varying degrees of surface expression by subgroup. Furthermore, CD40 and galectin-9 exhibited a wider range of expression from one subgroup to another, independent of any cytokine action. CD40 was also differentially expressed between subgroups in the TNF treated group, while all the IL1B and over 90% of IFNg-treated subgroups expressed CD40 at low to negligible levels. Consistent with tSNE analysis, there was one subgroup in each of the TNF, IL1B, and vehicle treatment groups in which CD40 expression was considerably higher than in the other subgroups. In the TNF and IL1B-treated groups, some subgroups exhibited a sharp decrease in expression of PECAM1, while others maintained expression levels comparable to control (Fig. 3F). Combined, these data suggested that within each treatment group there are indeed subgroups of P-UAEC with unique patterns of combined protein expression. Some surface proteins appeared to be proportionately co-expressed in a cytokine-specific manner, while the levels of other proteins appeared to be inversely regulated.

The impact of cytokines on single cell transcriptome is agonist and cell subtype specific.

Given such compelling evidence of unique subgroups of P-UAEC with regard to immunomodulatory protein expression, we next sought to identify and characterize members of these cellular subtypes by determining protein and particularly mRNA transcriptome expression profiles at the single cell level. To that end, we utilized single cell CITE-Seq. Cell transcriptomes common to the 5 detected cell subgroups were compared to a cell atlas and showed 99% were endothelial (groups 0–4). Using a standard amount of each library (10%) for antibody detection, we had limited success in that we could show expression of surface proteins by treatment group, but not with sufficient sensitivity to quantitatively determine if there was differential expression as inferred above by spectral flow cytometry/tSNE analysis. Nonetheless, transcriptome analysis successfully obtained data ~5000 or more individual cells in each treatment group. Cell numbers within each cluster changed dramatically following cytokine treatment (Table 1).

Table 1: Cell distributions by subtype as identified by CITE-Seq.

The distribution of cells by subtype is shown according to the treatment group. Note in each row, values for all treatments add up to 100%. Cell type predicted by transcripts is also shown. (See also summary Fig 7 below for alternate expression of this data as % cells in each Cluster by treatment).

	Raw Counts					% of cluster comprised of Treatment				Size of Pseudo Replicates
Cluster	Con	INFg	TNF	IL1B	Total	% Con	% IFNg	% TNF	% IL1B	Con	INFg	TNF	IL1B	Cell Type Predicted by SingleR
0	2,430	414	3,215	3,223	9,282	26.18%	4.46%	34.64%	34.72%	810	138	1,072	1,074	Endothelial: Blood Vessel
1	2,981	1,845	1,690	1,320	7,836	38.04%	23.55%	21.57%	16.85%	994	615	563	440	Endothelial: Lymphatic TNFa_48h
2	2,893	3,203	21	94	6,211	46.58%	51.57%	0.34%	1.51%	964	1,068	7	31	Endothelial: Lympatic
3	2,494	1,106	499	566	4,665	53.46%	23.71%	10.70%	12.13%	831	369	166	189	Endothelial: Lymphatic TNFa_48h
4	44	177	53	29	303	14.52%	58.42%	17.49%	9.57%	15	59	18	10	Endothelial: HUVEC - VEGF

Clusters (hereafter, subtypes) 2 and 3 show a clear decrease in number and % cell number with TNF and IL1B treatment. A smaller decrease was seen for subtype 1, while TNF and IL1B both promoted an increase in cell numbers for subtype 0. The effects of IFNg were distinctly different, with a dramatic negative impact on subtype 0 and smaller effect on subtypes 1 and 3, and a moderate positive effect on subtype 2 cell numbers.

An initial Pseudo bulk seq analysis of the top 50 DE genes for each cell Group regardless of treatment suggested the principal components discriminating between cell clusters included multiple transcripts for chemotaxis, inflammation, and immune modulation markers (Fig 4). Further analysis (Fig 5A, upper panels) showed an overall agonist specific impact across combined cell subtypes on many transcripts for the endocrine factors and immunomodulatory proteins identified above on the cell surface or secreted to media, with ~50% achieving significance (indicated by * inside each bar). Further dissecting the data by agonist and cell subtype (Fig 5A, Lower Panels) showed agonist specific responses were indeed subtype specific. Of note, cell subtype 4 was typically the weakest responder to treatments, and those responses were more often negative and failed to reach significance. In contrast, cell subtypes 1, 2 and 3 most often gave the strongest DE responses to agonists. Of note, TNF most strongly upregulated CCL5 and CCL20 transcripts in cell subtype 1 and CCL20 in subtype 3, while IL1B most strongly increased CCL20 in subtype 1 cells, and to a lesser extent in subtype 3 cells. IL1B did not stimulate a significant increase in CCL5 in any cell subtype. CXCL8 was upregulated by both TNF and IL1B in subtypes 0, 1, 2 and 3. TNF significantly increased transcripts for CCL8 in subtypes 0,1 and 3, while IL1B had no effect. TNF also increased transcripts for IL6 in cell subtypes 0,1 and 3, and IL1B had a similar effect. Of note, CCL17 transcripts were specifically upregulated by TNF alone in cell subtypes 0 and 1, while CXCL6 was specifically upregulated by IL1B alone in cell subtype 0. In contrast to all these findings specific to chemokines, IFNg had no effect on any of these particular transcripts in any cell subtype.

Figure 4: Principal component analysis of scSeq-data (showing the most prominent DE transcripts defining cell subtypes).

The single cell Seq data from CITE-Seq analysis for all treatment groups was combined and processed as described to determine the transcripts most definitive of cell subtype differences. The top 50 such transcripts are shown as a heat map by cell cluster (subtype) (Bottom axis) and relationship of transcript changes by cell subtype to each other by cluster ranking (top to bottom). Note that while this is an unbiased sort, many of the top 50 principal components encode chemokines, cytokines, and immune modulatory proteins.

Figure 5: DE of transcripts of interest by agonist alone, and further by cell subtype.

Panel A: An initial analysis of DE transcript data for chemokines and cytokines, MMPs, and immunomodulatory proteins using Pseudo bulk seq of combined cell subtypes (Upper row) shows multiple overall changes of which ~50% are significant at P<0.05 (Shown by * inside color box). Lower panels show the same data format further delineated by cell subtype. Panel B: Additional data is then shown for further analysis of transcripts encoding receptors to Th cell specific cytokines.

With regard to MMP transcripts (Fig 5A, top center), while TNF had an effect on MMP9 in the overall Pseudo Bulk data set, this did not show as significant in any individual cell subtype. Nonetheless, MMP2 which was also significantly elevated in the Pseudo Bulk data was seen to be significantly increased by TNF in cell types 0, 1, 2 and 3, while IL1B had a smaller effect in cell types 0 and 1 (Fig 5A lower panel). Again, by comparison, IFNg had little to no effect on MMP transcript expression at the pseudo bulk or single cell level.

Regarding transcripts encoding cell surface immunomodulatory proteins (Fig 5A, right panels), TNF clearly and strongly induced VCAM1 over all other transcripts in cell subtypes 1 and 3. It also induced ICOSL most strongly in subtype 2 and ICAM1 in subtype 3. Beyond that, TEK (TIE2) was upregulated in cell subtypes 0 and 1. TNFSF9 (4–1BBL) and SELP were not upregulated significantly in any cell subtype. CD40 was increased solely in subtype 1 cells. CDH5 (VE-Cad) and PECAM1 (CD31) transcripts were most clearly reduced in cell subtype 3. All other transcripts remained at a similar level or showed changes that were not significant.

By comparison, IL1B induced ICAM1 over all other transcripts most effectively in cell subtype 3, but also in cell subtypes 0 and 1. It also induced ICOSL transcripts most strongly in the same cell types. Unlike the response to TNF, VCAM1 was not altered significantly in any cell subtype. Beyond that, there was no effect of IL1B on TEK or TNFSF9 (4–1BBL), while CD40 was significantly increased in cell subtypes 1 and 2. CDH5 and PECAM 1 transcripts were modulated to a small but significant degree by IL1B in cell subtypes 0 and 1. All other transcripts for these surface proteins remained at similar levels. Otherwise, as previously for chemokine and MMP transcripts, IFNg had the weakest effect on these targets. Only CDH5 (VE-Cad) and PECAM (CD31) transcripts were significantly affected by IFNg in subtype 1 cells.

Because of our recent observation by Milliplex assay that factors secreted by Th1/Th17/Treg cells are generally not secreted by P-UAEC in response to TNF (Clemente et al, 2025), we asked if examining transcripts by cell subgroup may indicate that (i) alterations of expression of the corresponding receptors by cell subtype may give further clues to possible fine tuning of cell subgroups to respond to the specific paracrine secretions of an attached immune cell, and (ii) if that could be further altered by cytokine exposure. Our data shown in Fig 5B suggests this may well be the case. Analysis of bulk combined data by treatment (left panel) showed clear differential regulation of receptors to IL17, IL10, IFNg and TNF. By this Pseudo Bulk seq analysis, of the three agonists, TNF had the broadest range of effects, and IFNg the least. Of note, TNF favored IL17RA expression, and downregulated IL10RB, while also increasing TNFRSF1B, IFNGR1 and reducing IFNGR2. IL1B had similar effects to TNF on IL17RA and IL10RB, while IFNg had no significant effect on any receptor transcript.

By cell subtype (Fig 5B, right panel), the only TNF upregulation of IL17RA was observed in cell subtype 3, while the only downregulation of IL10RB was in cell subtype 1. TNFRSF1B was upregulated in subtypes 0 and 1. IFNGR1 was upregulated in cell subtype 1 and to a lesser degree in subtype 0, but IFNGR2 was also downregulated in cell subtype 0. IL1B also had some significant effects by cell subtype. IL17RA was increased in subtype 0 cells, and IL10RB was strongly downregulated in subtype 3 cells. IFNGR1 was also increased in subtype 3 cells, and IFNGR2 was decreased in both subtype 3 and subtype 0 cells. In contrast IFNg had no significant effect on these transcripts in cell subtypes 0–4.

A summary of changes in % of each subpopulation with cytokine treatments are shown in Fig 6A, alongside the total number of DE transcripts in each case (Fig 6B). The most dominant and significant changes in physiologically key transcripts from Fig 5 are summarized in Fig 6C. Note that while the sub table in Fig 6C shows changes in transcript for gain or loss of function in each cell subtype, data for IFNg is not shown because it had no significant effects on these transcripts in any cell type (see Fig 5). Otherwise, bold text with enlarged or reduced font size in Fig 6C denotes more substantial changes in the transcripts of interest, and brackets indicates all changes that were a decrease in transcript. The overall impression is not only are these cell subtypes showing distinctly different changes in chemokine output, MMP output, immune regulatory surface proteins and ‘Th cell cytokine’ receptors in response to TNF and IL1B treatment in an agonist specific manner, there are also times when each treatment may increase or dramatically reduce a cell subtype as % of the cell population (Fig 6A). While IFNg is predominantly an ineffective regulator of these same key transcripts (Fig 6C), it did impact cell subtypes as % of the total population (Fig 6A) and induced substantial numbers of other DE transcripts in cell subtypes 1,2 and 3 (Fig 6B), so the control agonist was indeed effective. With reference to the transcripts of interest here (Fig 6C), TNF impacts strongly on VCAM1, while IL1B favors ICAM1 expression, but these effects are also specific to cell subtype. Of the immune modulatory surface receptors and ligands, there is heterogeneity of presentation of ICOSL, but also CD40 (subtype 1 in particular). While many of these surface immunomodulatory proteins are a ‘match’ for interaction with Th1 cells, our transcript data in Fig 5 suggest this is not always achieved through changes in transcript expression. OX40L and CD40L changes by flow cytometric analysis (above) were also not paralleled by corresponding DE transcripts in our CITE-Seq analysis. Clearly post transcriptional regulation may also be in play as a control mechanism in at least some cases. Of further direct relevance to Th17 cells, only subtype 3 cells show an induction of IL17RA by TNF and a loss of IL10RB by IL1B (Fig 6C), potentially making that cell subtype much more responsive to local increases in Th17 cell-derived IL17 and/or less responsive to local IL10.

Figure 6: Summary of changes in P-UAEC cell groups as % final cell number, and predominant agonist specific effects on DE transcripts of interest.

Panel A shows changes in % of each cell subtype in control or agonist treatment states. Panel B shows the number of transcripts with DE changes in response to each cytokine treatment for each cell subtype. Panel C shows a table summarizing the significant DE transcripts of interest in response to TNF or IL1B treatment for each cell subtype (Data from Fig 5). Note IFNg did not have any significant effects on these transcripts relative to control even though it does regulate many other transcripts (Panel B). Major fold changes are shown as bold text with larger or smaller font, and all negative fold changes are indicated by placing the text in parenthesis.

Discussion.

While pregnancy is known to be an inflammatory condition, preeclampsia is a more extreme state associated with higher chemokines and pro-inflammatory cytokines, and as well as higher Th1/Th2 and Th17/Treg ratios (Redman & Sargent 2003, Jaffri et al 2017). It is known the uteroplacental space can secrete cytokines, including TNF and IL1B, and a common assumption is the proinflammatory immune cell profile of Th1 and Th17 cells dominating over Th2 and Treg cells begins in that space. To date, a possible role for endothelium in this initiation process has not been seriously considered. In our companion paper we have shown that P-UAEC treated with TNF do in turn secrete many of the chemokines and cytokines reported elevated in PE subjects (Clemente et al, 2025). Recent publications have also shown that endothelium can become immunomodulatory on exposure to TNF and IL1B (Mai et al 2013) and, in systemic hypertension, endothelium has been shown to exist as multiple cell subtypes (Jafri & Ormiston 2017). Herein we show using multiple single cell methods that uterine artery endothelial cells from pregnant sheep (P-UAEC) also exist in multiple subtypes, and each has distinct chemokine and cytokine secretory and immunomodulatory properties.

Flow cytometric analysis implicates and CITE-Seq analysis confirms the existence of distinct subgroups of endothelial cells that in turn display specific immunomodulatory properties in response to cytokine treatment.

While simple initial analysis of bulk cell flow cytometry data suggested some cell surface proteins are clearly regulated on P-UAEC en masse, few changes reached statistical significance. Nonetheless, that changes were occurring in subgroups of cells was confirmed by our application of further ‘single cell’ analytical methods (e.g. tSNE analysis or CITE-Seq) which revealed cell subgroup-specific responses to treatment that were agonist-specific. Of note, additional CITE-Seq analysis clearly showed 5 distinct cell subtypes, and treatment of P-UAEC with TNF vs IL1B resulted in further agonist specific and cell subtype specific changes in chemokine, cytokine and MMP transcripts in each case, so extending the findings of our companion paper (Clemente et al, 2025). Combined analysis confirmed that overnight treatment with TNF or IL1B in particular pushed the cells to subgroup/subtype specific profiles of attachment factors (particularly VCAM1, ICAM1), costimulatory molecules (particularly ICOSL), and a partial loss of classic endothelial markers (particularly CD31/PECAM and CDH5/VE-Cad). There was general agreement between the outcomes of single cell spectral flow cytometry and transcriptome analysis, but in some cases there was disagreement. Clear increases in VCAM1, ICOSL etc were seen in both cases, but ICAM1 showed decreases in surface protein while transcripts were generally increased. This is not necessarily a contradiction given proteins can be modulated at a post transcriptional level, including surface expression and proteolytic degradation, and compensatory mRNA synthesis may be necessary to keep up. Generally it was still clear that the overall responses to each agonist were distinct, and different cell subgroups/subtypes showed distinct responses and even opposing responses within cytokine treatment subgroups/subtypes. These combined data strongly argue for the presence of different endothelial cell subpopulations with differing phenotypic and functional responses to specific stimuli. Regarding loss of classic endothelial markers associated with a vasodilatory/angiogenic functional state, spectral flow cytometry revealed that PECAM (CD31), VE-Cad (CD144) and to some extent VEGFR2 (CD309) showed higher degrees of variability of surface expression across cell subpopulations in response to TNF treatment. Cell attachment molecules and proteins associated with immune cell rolling also showed TNF dependent polarization of expression for VCAM1 (CD106), and to a lesser extent ICAM1 (CD54). With regard to stimulatory and costimulatory molecules, tSNE analysis showed the most clearly impacted in a polarizing fashion was CD40, while OX40L and ICOSL were more uniformly upregulated by TNF treatment. When we examined the effects of IL1B and IFNg, we saw that changes in response to IL1B tended to be seen in the same set of surface proteins, with more overall suppression of surface expression of VCAM1 (CD106) wherever there was an otherwise a stimulatory effect on ICAM1 (CD54). There was also a more uniform suppression of CD40 expression, but with important exceptions. IFNg had a clearly different overall effect compared to TNF or IL1B, much of it suppressive of surface expression of many of these surface molecules. There was also some limited data suggesting antigen presenting function varied on a cell-by-cell basis, with HLA-DR showing some degree of variation in control and in TNF or IL1B treated cells, but further suppression of expression by IFNg. All of these changes strongly infer P-UAEC do exist in subgroups, and they are differentially regulated in a manner that may favor association with and stimulation by specific immune cell classes. If those specific immune cells can in turn stimulate endothelium through costimulatory or bidirectional paracrine signaling, then immune modulation by P-UAEC is indeed possible, and the results of our subsequent CITE-Seq analysis suggest it is likely.

CITE-Seq data also extends our flow cytometry and other multiomics observations to suggest individual endothelial cell subtypes may be geared for direct binding to and specific paracrine interactions with distinct Th cell subclasses.

In general terms the changes detected in corresponding single cell mRNA transcripts largely agreed with flow cytometric analysis of surface proteins. While there were agonist specific changes in transcripts by cell subtype for classic endothelial cell markers, a finding consistent with that above, the most polarizing subtype specific changes in transcripts in response to TNF, IL1B and IFNg were again heavily related to attachment molecules VCAM1, ICAM1, and costimulatory molecules ICOSL and CD40. TNF had the most polarizing effect, IL1B was similar but weaker in effect but with a notable polarizing effect on CD40, and IFNg had a generally suppressive effect on transcript profiles for these molecules. It should also be noted with regard to MMP transcripts, MMP1, MMP2 and MMP9 were among those reported previously as increased by TNF (Ampey et al 2021). In this study MMP9 and MMP1 were detected, but not altered significantly by cell subtype, while TNF>IL1B>>IFNg impacted significantly on MMP2 expression in multiple cell subtypes. It is important to keep this in mind given that several surface molecules such as VE-Cad and CD31 (PECAM) are known to be shed from the cell surface in response to extracellular protease activity (Ampey et al 2021, Eshaq & Harris 2019). Both MMPs and these shed proteins were detected as elevated in media in response to TNF by proteomics (above). As such, we must conclude by surface protein analysis that there are specific subtypes of endothelial cells that may well offer a polarized cassette of surface proteins characterized by VCAM1 domination or ICAM1 domination, variability in CD40 vs CD40L expression, and dominance or not of ICOSL. While we did detect OX40L and CD40L proteins by flow cytometry, we did not detect corresponding transcripts. This is, however, a study based on the far less well-developed ovine genome, so this is not perhaps unexpected. Otherwise, weak but significant changes in transcripts for surface proteins CD31/PECAM, CDH5/VE-Cad are consistent with relatively weak changes in cell surface proteins, but again we know surface protein expression also reflects the additional impact of protease action at the cell surface as the prime regulator of cell surface activity (See Ampey et al 2021). It is important to note that while the roles of classic markers may commonly be considered distinct from those of attachment molecules or co-stimulators, both CD31/PECAM and CDH5/VE-Cad proteolytic degradation may impact on cytokine responses indirectly. Both molecules are extremely large on the cell surface, so stearic inhibition alone may apply to modulating the function of other glycoproteins as well as underlying receptors. VE-Cad has also been shown to cause direct inhibition of GP130 associated responses (Adan et al 2022), and CD40/CD40L interactions have been shown to signal in a manner similar to IL1B (Smola-Hess et al 2001). So it is understandable that to achieve more immune modulatory function, there would be a need to add this protein to the cell surface, but there would also be a need to remove classical endothelial markers to allow surface cytokine receptors to function and actively couple to distinct signaling systems (GP130 coupled and possibly others?). As such, vasodilatory function and immunomodulatory function would by definition be exclusive states, and this could explain why loss of vasodilation is a specific early marker of PE pregnancy (Sladek et al 1997, Bird et al 2000). Such a proposal is entirely consistent with P-UAEC experiencing an imbalance of Jak/STAT vs NFKB signaling resulting in altered chemokine and cytokine and MMP secretions (Clemente et al, 2025, Smola-Hess et al 2001) as they enter the earliest stages of an endothelial mesenchymal transition (Cho et al 2018).

Beyond regulation of immunomodulatory surface molecule transcripts (and particularly VCAM1 > ICAM1), we also found by CITE-Seq that TNF induced subtype 1 cells to favor high CCL5 and CCL20 mRNA transcripts. TNFRSF1B and IFNRG1 receptors were also upregulated, while IL10RB was reduced (Fig 6). Cell subtype 3 showed predominantly CCL20 transcript elevation, and strong VCAM1 and ICAM1 elevation, with strong ICOSL and IL17RA elevation in response to TNF. In contrast, cell subtype 2 showed modest elevation of CXCL8(IL8) and substantial ICOSL elevation in the absence of strong VCAM1 or ICAM1 elevation. Of note, cell subtype 2 was also heavily depleted in numbers by TNF. While treatment with IL1B showed similar changes by cell subtype, they were not identical. Most notable were the absence of any CCL5 response, and the greater significance of changes in CXCL8 and CXCL6 transcripts across cell subtypes. There was also a greater dominance of ICAM1 but not VCAM1 transcript upregulation across subtypes. While ICOSL and CD40 transcripts were upregulated in multiple cell subtypes, ICOSL was most strongly increased by IL1B in subtype 1 and 3 cells and CD40 was increased in subtype 1 and 2 cells. Again IL1B had a clear depleting effect on subtype 2 cell numbers, but not as much as TNF. In contrast, the influence of IFNg was far weaker overall on multiple transcript profiles, and none achieved significance. Nonetheless while IFNg was associated with a clear reduction in subtype 0 cell numbers, it increased all other cell subtypes. Overall, we must conclude that different P-UAEC cell subtypes do indeed exist, and they are controlled differentially by specific cytokines to call out to specific immune cells using distinct chemokines, offer distinct binding molecules, and interact directionally or bidirectionally through costimulators (particularly CD40/CD40L and ICOS/ICOSL). All these effects may then be further co-modulated by associated MMP secretion. At the same time, cell subtypes may show specifically altered transcripts encoding receptors for IL17 vs IL10 (particularly cell subtypes 0, 1 and 3 across TNF vs IL1B stimulation). Together this begs the question if there are (i) cell subtype specific chemokine secretions to attract a specific immune cell that match specific surface profiles for attachment and costimulation via ICOS/ICOSL or CD40/CD40L, and (ii) specific surface receptor compliments to further allow stimulation in a unidirectional or bidirectional manner (IL17 and IL10 receptors in endothelial cell subtypes to respond to correspondingly attached Th17/Treg cells). The fact MMP’s are secreted to the extracellular space means soluble cytokines may also be released from the parent cell to influence nearby cells or immune cells of a different subtype, and so may also impact on unidirectional vs. bidirectional paracrine stimulation between other endothelium subtypes and matching paired immune cells. It is tempting to speculate that while this complex machinery is a fine control mechanism for normal pregnancy, an adverse runaway loop of bidirectional stimulation may also be the trigger for precipitation of clinical symptoms in PE. Further studies will be needed to confirm if this is so.

Which cell subtypes could most specifically attract, and which could bind/interact with Th17 Cells?

As stated above, there are clearly circumstances whereby cell subtypes treated with TNF tend to show high VCAM1 expression while IL1B treatment results in ICAM1 dominance. Three cell subtypes match this pattern, namely subtypes 1, 2 and 3, but type 2 also diminishes in number dramatically in response to TNF and IL1B. In contrast, subtypes 1 and 3 remain well represented and both show strong ICOSL upregulation with subtype 1 also showing CD40 upregulation in response to both TNF and IL1B. The subtype specific upregulation of VCAM1 and/or ICAM1 may well offer increased binding opportunities to Th cells in general, but a study of the effect of ICOSL on immune responses found that dendritic cells obtained from IL10 knockout mice exhibited significantly higher levels of ICOSL, greater Th1 and Th17 responses, and a reduction in Tregs relative to controls following Chlamydia muridarum lung infection. Interestingly, most Th1 cells were ‘ICOS−’, while Th17 cells were mostly ‘ICOS+’. Blockade of ICOS/ICOSL interactions strongly inhibited the expansion of Th17 cells, suggesting that the Th17 response is ICOSL-dependent (Gao et al 2013). Similarly, suppression of ICOSL expression in gastric epithelial cells by H. pylori infection correlated with reduced Th17 responses both in vitro and in vivo, suggesting that H. pylori maintains chronic infection by reducing expression of ICOSL, thereby evading Th17-mediated clearance (Lina et al 2013). This is consistent with a prior study by Paolos et al (2010) , that found ICOS signaling to be “critical for the differentiation and expansion of human Th17 cells. So the elevation of ICOSL in particular on endothelial cell subtypes 1 and 3 suggests essential targets for Th17 binding are present. Further subtype specific changes may also be important. As noted above, the upregulation of CD40 on subtype 1 cells may offer an added way to stimulate any attached Th cells expressing CD40L in a manner that maintains Th cell paracrine signaling. Conversely, the reduction of IL10RB on subtype 1 endothelial cells may invoke reduced resistance to any effects of local IL17 from attached or adjacent Th17 cells. Certainly subtype 1 cells also show strong upregulation of CCL5, CCL20 and CCL17 transcripts, so they are clearly capable of attracting multiple Th cell classes. In many ways, subtype 1 cells are a viable candidate to attract, bind and stimulate Th17 cells at multiple levels. A counter-argument would be that it is subtype 3 cells that show specific increases in IL17RA transcript on TNF stimulation, and a strong reduction in IL10RB in response to IL1B, suggesting subtype 3 cells undergo a further fine tuning to local IL17 secretion. Overall, while subtype 1,2 and 3 cells each have potential to interact with Th17 cells in particular, subtype 1 and particularly subtype 3 cells seem most finely tuned in the presence of both TNF and IL1B to undergo bidirectional paracrine stimulation with Th 17 cells.

But what of types 0 and 4 cells? A hallmark of subtype 0 is they are present in substantial numbers and show an increase in transcripts for CXCL6 in response to IL1B. This has been linked in brain microvascular cells to inflammatory responses following reperfusion injury in a manner associated with HIF1a and the Foxo/PI-3-kinase/Akt pathway activation, the result of which has a negative impact on monolayer permeability and cell survival (Wang et al 2021). As such, they may contribute perhaps to the edema commonly seen in PE. Cell subtype 4 is most remarkable in it is present in very low numbers and it’s non-responsiveness to cytokines, at least in the context of chemotaxis and inflammatory modulation, which suggests they may be specifically tuned for another purpose. They could simply be in a state of cell division/angiogenesis, or about to be released as circulating endothelial cells, or they could be stem cells, giving rise to the other endothelial cell subtypes.

How our data supports an endothelial origin for immune disturbance in PE subjects – The Sequential Activation Hypothesis.

While our findings are intriguing, further studies are needed to more fully understand the processes behind them. What is clear is that a multiomics analysis of the effects of TNF vs IL1B on P-UAEC cell subtypes in particular have shown TNF can initiate profound changes in the P-UAEC population that involve changes in specific subpopulations of cells, and occurs in a way that could selectively attract and bind/interact with specific Th cell subtypes. Furthermore, the response to TNF and IL1B are subtly different, but still potentially complimentary, and involve promoting what appears to be a supportive environment for the attraction of, binding to, and enhanced interaction with Th17 cells in particular. It is important to note a key finding of our study is this ability to attract and interact with Th17 cells will occur in response to cytokines otherwise classically regarded as from the Th1>Th2 cytokine systems. As such, our study data begs the question if the emergence of PE is in fact a ‘sequential activation’ process. Activation step one is the early local elevation of TNF and or IL1B due to either preexisting conditions and/or from a distressed hypoxic placenta creates an altered uterine endothelial population with subtypes moving away from quiescent vasodilation and moving towards binding of and interaction with Th1 cells. The emergence of ‘activated’ subtypes 1, 2 and particularly 3 now allows the attraction and binding of Th17 cells and so begins a complex interaction through the many paracrine and co-stimulatory processes available. We propose that ‘step two’ is occurs when a sufficient number of bound Th17 cells accumulates particularly on subtype 3 cells to reach a point where mutual co-stimulation through direct binding and paracrine signaling via Th cell derived cytokines (IL17 in particular) precipitates an amplified bidirectional inflammatory response beyond the elevated threshold seen in normal pregnancy. With completion of the two activation steps, the process could then propagate out from the initial uterine ‘priming’ site in a similar way. As the proinflammatory and indeed wounding type chemokine/cytokine/immune response begins to dominate in the uterine artery, this could limit any increase in uterine blood flow in the second and third trimester through cytokine mediated blockade of cell-cell coupling via Cx43 that otherwise underpins the sustained increase in vasodilation normally observed at this critical time (Yi et al 2010, Bird et al 2013). Any resulting fetoplacental hypoxia would only serve to reinforce further secretion of TNF and indeed IL1B from non-vascular sources such as hypoxic trophoblasts (Benyo et al 1997). Should bound Th1 and Th17 immune cells then detach from uterine artery endothelium, they could move further into the uteroplacental space, or they could move to the maternal systemic vasculature to promote further endothelial and immune cell ‘activation’ and impair endothelial cell vasodilatory function in those additional locations. At this point, ‘systemic’ hypertension would present clinically as maternal systemic vascular endothelial failure accelerates exponentially.

On a final note, our proposed model does take into account the clinically observed variability of time of onset of PE symptoms, and that Th1/Th17 dominated preexisting conditions are a risk factor for PE. If the first activation step is already occurring due to higher levels of maternal Th1>Th2 inflammatory cytokines prior to conception, then the endothelium would be primed to reach a Th17 binding and responsive state faster. Whether that is due to a strong proinflammatory condition at a young age or a milder but ongoing condition such as obesity at an older age, the effect would be the same. Likewise, in the event there is no prior condition but suboptimal development of the uterine vasculature and subsequent higher levels of placental hypoxia occur early on, the resulting increase in local cytokine secretion would mean the time to reach the second activation step would also be reduced, and the magnitude of the second activation response could be increased. Perhaps now the identification of multiple uterine artery endothelial subgroups/subtypes has been achieved, it will be possible to identify molecular targets specific enough to treat, and with prenatal screening ultimately prevent PE for the first time.