Human fetoplacental arterial and venous endothelial cells are differentially programmed by gestational diabetes mellitus, resulting in cell-specific barrier function changes

Sample collection

Ethical approval was obtained from the Medical University Graz (25-008ex12/13 and 27-268ex14/15). All women participating in the study provided written informed consent. Control placentas were collected from pregnancies of non-smoking (self-reported) women who had a negative 75 g OGTT performed at 25–28 weeks of gestation and were free from medical disorders or pregnancy complications. GDM was diagnosed according to WHO/IADPSG criteria [16]. The women who had a positive OGTT were either recommended a diet and classified as GDM A1 or were additionally treated with insulin (NovoRapid plus Insulatard; Novo Nordisk Pharma, Vienna, Austria) and classified as GDM A2. Control and GDM samples were matched for ethnicity and fetal sex, but not for maternal BMI, since being overweight is major risk factor for GDM. However, neither BMI nor gestational weight gain differed between control and GDM groups. In addition, the mean and median cell passage of primary cells from control and GDM women was similar. We also performed covariate analysis on multiple factors (gestational age, cord blood insulin, fetal weight and length, fetal ponderal index, placental weight, maternal C-reactive protein [CRP], maternal height, maternal weight and BMI before pregnancy and before birth and maternal gestational weight gain) using Bioconductor package limma (https://bioconductor.org/packages/release/bioc/html/limma.html) and identified an effect on a total of 27 CpG sites at 21 genes (electronic supplementary material [ESM] Table 1), representing only a small subset of total differentially methylated positions and suggesting that observed effects are a result of GDM. We should note, however, that the sample size is too small for such covariate analysis.

Cell culture

Primary AEC and VEC were isolated from third-trimester human placentas after healthy and GDM-complicated pregnancies following a standard protocol [15]. Cells were characterised by immunocytochemical analysis [17] and cultured on 1% (vol./vol.) gelatin-coated flasks (75 cm²) using endothelial basal medium (EBM; Cambrex Clonetics, Baltimore, MD, USA) supplemented with the EGM-MV BulletKit (Cambrex Clonetics). Cell isolations were used up to passage ten as no phenotypical change or altered responses of the cells to culture were observed.

DNA methylation analysis

DNA (1 μg) isolated from fetoplacental AEC (n = 9), VEC (n = 9), AEC from GDM pregnancies (dAEC) (n = 5) and VEC from GDM pregnancies (dVEC) (n = 9), obtained from nine control and nine GDM placentas in total, was bisulphite converted using MethylEasy Bisulphite Modification Kit (Human Genetic Signatures, Sydney, NSW, Australia). Conversion efficiency was assessed by bisulphite-specific PCR (not shown). Hybridisation of bisulphite-treated samples to Illumina Infinium Human Methylation450 (HM450) BeadChips (Illumina, San Diego, CA, USA) was performed according to the manufacturer’s instructions. Based on power calculations for the HM450 array, our sample size would allow us to detect changes of Δβ > 0.2 and p value <0.05 [18]. The BeadChips were scanned using Illumina iScan (Illumina) and raw data were exported as IDAT files. Minfi Bioconductor package (https://bioconductor.org/packages/release/bioc/html/minfi.html) [19] imported data into R (version R 2.15.1), performed quality control, pre-processing and normalisation using the subset-quantile within array normalisation (SWAN) method [20]. The limma package [21] was used to fit a linear model to compare dAEC and dVEC vs control samples, with patient as random effect and allowing for batch effects. False discovery rate (FDR) was calculated by the Benjamini–Hochberg method. M values were calculated after removing probes on the sex chromosomes to eliminate potential sex bias and poor-performing probes. β values were derived from intensities defined by the ratio of methylated (M) to unmethylated (U) probes given by β = M / (U + M + 100). For details of quality control, outlier identification and HM450 platform validation using locus-specific SEQUENOM MassARRAY EpiTYPER (Agena Bioscience, San Diego, CA, USA) [22], see ESM Methods and ESM Figs 1–3. For information on whether CpGs are located in differentially methylated regions (DMRs), see ESM Table 2 and for a discussion of advantages vs disadvantages of the HM450 platform, see ESM Methods. Unadjusted and adjusted p values and information on whether CpGs are located at potential SNPs, within topological domains (TADs) published in HUVECs [23] or part of a DMR, are given in the lists of significantly methylated CpGs available on Gene Expression Omnibus (GEO) database account GSE106099 (www.ncbi.nlm.nih.gov/geo).

RNA isolation

Total RNA was isolated with RNeasy mini Kit (Qiagen, Hilden, Germany) and quality was assessed using a BioAnalyzer BA2100 (Agilent, Foster City, CA, USA) with the RNA 6000 Nano LabChip Kit (Agilent). Samples with an RNA Integrity Number ≥8.5 were further used.

Microarray gene expression analysis

Total RNA from AEC (n = 8) and VEC (n = 8) isolated from eight placentas from normal pregnancies and dAEC (n = 11) and dVEC (n = 10) isolated from 14 placentas from pregnancies complicated by GDM was labelled using Ambion WT Expression Kit for Affymetrix GeneChip Whole transcript (WT) Expression Arrays (Life Technologies, Carlsbad, CA). The cRNA was hybridised to GeneChip Human 1.0 ST arrays according to the manufacturer’s instructions (Affymetrix, Santa Clara, CA, USA). Washing and staining (GeneChip HT Hybridization, Wash and Stain Kit; Affymetrix) was performed with Affymetrix GeneChip fluidics station 450. Arrays were scanned using Affymetrix GeneChip scanner GCS3000. Labelling and hybridisation controls were evaluated with Affymetrix Expression Console EC 1.1. Data were analysed with RMA (robust multi-chip average), including background correction, quantile normalisation, log₂ transformation and median polish summarisation using Genomic Suite v6.5 (Partek, St Louis, MO, USA) [24]. Statistical analysis used one-way ANOVA with fetal sex and mother as random factors. The p values were adjusted for multiple testing using the Benjamini–Hochberg method (R/Bioconductor package ‘multtest’; https://www.bioconductor.org/packages/release/bioc/html/multtest.html) [25, 26].

qPCR

cDNA was synthesized from 50 ng total RNA of different cell isolations (n = 10 per group) then used for microarray analysis according to protocol (SuperScript II Reverse Transcriptase protocol; Invitrogen, Carlsbad, CA, USA). qPCR was performed with TaqMan gene expression assays (Applied Biosystems, Carlsbad, CA, USA) and ABI Prism 5700 Sequence Detection System (Applied Biosystems, Foster City, CA, USA). The mean hypoxanthine-guanine phosphoribosyltransferase 1 (HPRT1) and ribosomal protein L30 (RPL30) expression was used as internal control as their expression was unaffected by GDM (not shown). Data were analysed using the \( {2}^{-\Delta \Delta {\mathrm{C}}_{\mathrm{t}}} \) method [27]. Statistical analysis used Student’s t test in SigmaPlot (Systat Software, San Jose, CA, USA).

Pathway analysis

Gene lists were analysed with Ingenuity Pathway Analysis (IPA, version 2.3) (Qiagen). Cut-off for methylation differences was p < 0.05 and Δβ ≥ 0.2 and for gene expression p < 0.05 and fold change (FC) ≥1.5. When investigating pathways the cut-off was set to Δβ ≥ 0.1 (methylation) and FC ≥1.3 (expression) in order to have sufficient genes for the analysis.

F-actin immunofluorescence staining

AEC (n = 5), VEC (n = 6), dAEC (n = 5) and dVEC (n = 5), each in quadruplicates, were seeded in gelatin-coated chamber slides (50,000 cells/well). Participants’ characteristics are provided in ESM Table 3. After 24 h, slides were transferred to room temperature, washed with HBSS and fixed with 3.7% (wt./vol.) formaldehyde in PBS for 10 min. After washing with PBS, cells were permeabilised with 0.1% (vol./vol.) Triton X-100 in PBS for 25 min, washed with PBS, blocked with 1% (wt./vol.) BSA in PBS for 25 min and incubated phalloidin-488 FITC (1:20, Thermo Fisher, Eugene, OR, USA) with DL550 (DyLight 550 goat-anti mouse, 1:100, Thermo Fisher) for 1 h in the dark. Stained cells were washed with PBS and slides were mounted with Dako fluorescent mounting medium (Dako, Carpinteria, CA, USA) with DAPI (1:2000). After overnight drying, actin organisation was observed using a Zeiss LSM 510 Meta microscope, objective Plan-Apochromat 63×/1.4 Oil DIC, at 495 nm and 518 nm excitation wavelength (Zeiss, Oberkochen, Germany) using Zeiss LSM Image Browser. F-actin staining was performed and photographed by a blinded observer.

Electrical cell-substrate impedance sensing

Impedance measurements were performed using an electrical cell-substrate impedance sensing (ECIS) system (Applied Biophysics, Troy, NY, USA) [28]. AEC (n = 10) and dAEC (n = 6) (80,000 cells/well), VEC (n = 8) and dVEC (n = 4) (110,000 cells/well) were seeded in 400 μl EBM on gelatin-coated gold electrodes (8W10E+ arrays; Applied Biophysics), in duplicates. Participants’ characteristics are provided in ESM Table 3. VEC are smaller in diameter and were seeded in higher density. Thus, both AEC and VEC reached confluency after 12 h. Impedance was then recorded at 4 kHz for 24 h. Linear mixed-effect model using R package nlme (https://CRAN.R-project.org/package=nlme) was fitted for AEC and VEC separately. GDM, time and the interaction between them were used as fixed effects. Time was also used as random effect.

GDM alters the DNA methylation profile of fetoplacental AEC and VEC

To analyse changes in DNA methylation profile associated with exposure of fetoplacental AEC and VEC to GDM in utero (dAEC and dVEC), we measured the methylation of >450,000 CpG sites. Principal component analysis (PCA) of the entire methylation dataset demonstrated cell type (AEC vs VEC) as a main source of variation but also showed clear separation according to GDM status (Fig. 1a). Hierarchical clustering identified cell type as the greatest contributor to methylation variation (ESM Fig. 1) while high correlation between technical replicates indicated low inter-array variability (ESM Fig. 2).

Open image in new window

Linear regression indicated that exposure to GDM influenced methylation levels (p < 0.05 and Δβ ≥ 0.2) at 2617 CpGs that annotated to 2063 genes in dAEC and 1568 CpGs that annotated to 1360 genes in dVEC (Fig. 1b). The lists of differentially methylated CpGs are deposited as Excel files at GEO database, accession number GSE106099. Not surprisingly given the sample size, these sites did not remain significant following FDR adjustment for multiple testing applied to control for false discoveries. Of the total differential methylation associated with GDM, 351 genes and 51 CpGs were common to both cell types. In contrast to dVEC, where variation was equally distributed between hyper- and hypomethylation, dAEC showed hypermethylation at only 33% of sites while 67% were hypomethylated (Fig. 1b). This was also reflected by a marginal decrease of overall DNA methylation (average methylation levels detected across all probes) with GDM, particularly in AEC, as illustrated by kernel density plot and methylation index (ESM Fig. 4a, b). The box plots and heat maps shown in ESM Fig. 5 illustrate global and gene-specific methylation differences in GDM-exposed cells in more detail.

Great distance analysis showed that most of the GDM-associated variation in CpG methylation occurs in the proximal region upstream of genes (−5 to 0 kb from the transcriptional start) (Fig. 1e, f) in both cell types. In contrast to this, cell-type-specific methylation (AEC vs VEC) is enriched downstream of the transcription start site (0 to 50 kb; Fig. 1g).

Locus-specific DNA methylation profiling of selected genes significantly correlated with genome-wide measurements (r² = 0.83 for arterial and r² = 0.74 for venous group) and thus validated the HM450 DNA methylation platform (ESM Fig. 3).

Pathways and functions of genes associated with altered methylation pattern in GDM

GDM alters gene expression profile of fetoplacental AEC and VEC

PCA of the global transcriptome levels assessed at >30,000 transcripts in AEC and VEC clearly separated groups according to both cell type and GDM status (Fig. 1c). Unsupervised hierarchical clustering also discriminated AEC from VEC, again highlighting the greatest influence of cell type. Furthermore, dAEC vs control AEC clustered, while there was no clear separation in the venous group (ESM Fig. 6). Analysis of genes significantly influenced by GDM (p < 0.05) with a ≥ 1.5-fold expression change, identified 812 genes in dAEC, with 36% being upregulated and 64% downregulated (Fig. 1d). In dVEC only 211 genes had changed expression, with 57% being upregulated and 43% downregulated (Fig. 1d). Interestingly, only 92 genes were commonly affected by GDM in both cell types. A list of the top 100 differentially expressed genes in dAEC and dVEC vs controls is available at the GEO database, accession number GSE103552. A subset of differentially expressed genes identified by microarray was validated by qPCR, confirming differential expression of six out of nine genes (ESM Table 4).

Pathways and functions of differentially expressed genes in GDM

Molecular functions enriched with the differentially expressed genes in GDM were ‘cell cycle’, ‘cellular assembly and organisation’ and ‘DNA replication, recombination and repair’ in both AEC and VEC (Table 2), indicating a strong common effect of diabetes.

Concordant DNA methylation and expression changes in fetoplacental AEC and VEC

To investigate the relationship between DNA methylation and gene expression, methylation and gene expression datasets were combined. For each CpG, differences in β values between control cells and GDM-exposed cells (Δβ) were calculated and compared with corresponding gene expression differences. Setting the cut-off in the significant methylation difference to ≥10% (Δβ ≥ 0.1) and in the gene expression to FC ≥1.3, concordant changes in DNA methylation and gene expression (hypermethylated – downregulated and hypomethylated – upregulated) were detected. In dAEC, 118 genes were hypermethylated and downregulated, while 290 genes were hypomethylated and upregulated (Fig. 2a). In dVEC, 42 genes were hypermethylated and downregulated, while 117 genes were hypomethylated and upregulated (Fig. 2b). Of all genes showing concordant methylation and expression change in GDM, only six were altered in dAEC and dVEC (Table 3).

Open image in new window

Table 3

Genes commonly regulated by GDM in AEC and VEC

Gene symbol	Gene name	Direction of expression and methylation change
EGFR	Epidermal growth factor receptor	Concordant in AEC and VEC Upregulated in AEC and downregulated in VEC
GABBR1	γ-Aminobutyric acid (GABA) B receptor, 1	Concordant in AEC and VEC Upregulated in AEC and VEC
LMF1	Lipase maturation factor 1	Concordant in VEC Upregulated in AEC and VEC
NCKAP5	NCK-associated protein 5	Concordant in AEC and VEC Upregulated in AEC and VEC
SKIL	SKI-like oncogene	Concordant in VEC Upregulated in AEC and VEC
SULF2	Sulfatase 2	Concordant in VEC Upregulated in AEC and VEC

The table shows genes significantly differentially methylated (p < 0.05; Δβ ≥ 0.1) and expressed (p < 0.05; FC ≥1.3) after GDM exposure in both AEC and VEC. Overlap was performed after integration of DNA methylation and gene expression data

Functional effects of GDM on actin organisation and barrier function

To link pathway analysis results to cellular function, we investigated in vitro actin organisation in control and GDM-exposed cells. Phalloidin staining of F-actin (Fig. 3a) demonstrated that exposure to GDM differentially disrupted actin organisation in AEC. While in control AEC the actin cytoskeleton was organised in parallel actin bundles, the cytoskeleton was disorganised in dAEC. In VEC, GDM did not induce similar effects.

Open image in new window

As the actin cytoskeleton is essential for maintaining the endothelial barrier, we further investigated barrier function. Real-time analysis of the electrical impedance of cell monolayers showed an increased average impedance of 16% in dAEC vs AEC (p = 0.01) (Fig. 3b). In contrast, GDM tended to reduce the average impedance of VEC by 30%, but without reaching significance (p = 0.24). These distinct effects of GDM on barrier function in AEC and VEC parallel the aforementioned differences in actin organisation.

Expression profile of genes involved in endothelial cell adhesion mirrors the effect of GDM on barrier function in fetoplacental AEC vs VEC

Intercellular junction proteins join vascular endothelial cells and proteins of adherens junctions are most important for barrier function. Moreover, focal adhesions anchor endothelial cells to the basement membrane. Linker proteins such as vinculin and paxillin interconnect focal adhesions and intercellular junctions through the cytoskeleton. Their complex interplay in regulating actin organisation and vascular permeability is controlled by Rho GTPases [29]. We used expression levels of genes involved in focal adhesions, endothelial junctions and cytoskeleton rearrangement to generate a heatmap illustrating the effects of GDM on their expression in fetoplacental AEC and VEC (Fig. 4). This analysis revealed that GDM had the greatest effect on AEC, particularly on expression of focal adhesion-related genes: GDM upregulated various cellular integrins and extra cellular matrix genes, such as collagens and laminins, reflecting increased barrier function of dAEC potentially as a consequence of enhanced attachment to extracellular matrix. Rho GTPases and their upstream regulators (i.e. Rho guanine nucleotide exchange factors) and downstream targets (i.e. ROCK1) were also altered in dAEC. Fewer of these genes were affected by GDM in VEC.

Open image in new window