Peripheral plasma samples from pregnant females were provided by Guangdong Provincial People's Hospital for high-throughput sequencing. All subjects gave their verbal informed consent before enrollment. The study was conducted in accordance with the Declaration of Helsinki, and the protocol was approved by the Ethics Committee of Guangdong Provincial People's Hospital. Blood samples were obtained between September 2017 and November 2018 from pregnant females with children affected by PS, as well as pregnant females carrying unaffected fetuses (n=6 in each group). All blood samples were drawn 1–2 days before delivery. PS was diagnosed using a postnatal cardiac color ultrasound. Both groups consisted of three male infants and three female infants. The mean age of pregnant females in the PS and control groups was 33.3 and 29 years, respectively.

Plasma EVs were extracted using an exoRNeasy Serum/Plasma Midi kit (Qiagen GmbH) and ~1 ml plasma was centrifuged at 16,000 × g for 10 min at 4°C. Cellular materials and coagulated proteins were removed, and the supernatant was transferred into a new tube by careful aspiration. The re-suspended EV liquid was subsequently used for characterization of EVs.

The size distribution of the isolated EVs was analyzed using NanoSight NS300 (Malvern Panalytical). Particles were automatically tracked and sized based on Brownian motion and the diffusion coefficient. After isolation, EVs were diluted in 1 ml of exosome-free PBS, and the mixture was slowly injected into a clean particle-free sample pool to avoid formation of bubbles. The sample pool was covered and placed into the instrument. Manipulations were performed according to the manufacturers instructions. Three recordings were carried out for each sample and results are presented as an average of these recordings.

For TEM, 3 µl of EV pellet were placed on 200-mesh EM copper grids for 5 min, incubated for 5 min at room temperature, and then subjected to standard uranyl acetate staining at room temperature for 1–2 min. The grids were then washed three times with PBS and allowed to semi-dry at room temperature. Subsequently, the grids were visualized with a transmission electron microscope (H7650; Hitachi, Ltd.) at magnification, ×300, using digital micrograph software (v3.8; Gatan, Inc.).

EV lysate supernatants were prepared, total protein was extracted using Exosome cracking solution (Shanghai Umibio Biotechnology), and protein concentrations were determined using a bicinchoninic acid protein assay kit (Thermo Fisher Scientific, Inc.). Proteins (60 µl) were separated via 10% SDS-PAGE, and subsequently transferred onto PVDF membranes (EMD Millipore). Then, 5% non-fat milk in TBS with 0.05% Tween-20 (TBST) buffer was used to block the membranes for 1 h at room temperature. The membranes were incubated with rabbit anti-CD63 (1:500; cat. no. sc-5275; Santa Cruz Biotechnology, Inc.), anti-CD9 (1:1,000; cat. no. ab263019; Abcam) and anti-calnexin antibodies (1:1,000; cat. no. ab22595; Abcam) at 4°C overnight, washed with TBST, then incubated with horseradish peroxidase-conjugated goat anti-rabbit secondary antibody (1:5,000; cat. no. 7074S; Cell Signaling Technology, Inc.). Protein bands were visualized with an ECL reagent (Merck KGaA) using an automatic imager (GE Healthcare).

Long RNAs from EVs were amplified by Epi™ longRNA Ampli kit (Epibiotek). RNA sequencing libraries were constructed using SMARTer Stranded Total RNA-Seq kit (cat. no. 634413; Clontech Laboratories, Inc.). Adapters and low-quality bases were removed using Trimmomatic software (v0.36) (15). Clean RNA sequencing data were aligned to the GRCh38 human genome (hg38) downloaded from Ensembl (www.ensembl.org) using HISAT2 software (v2.1.0) (16). Genes were annotated using GENCODE annotation (v25; www.gencodegenes.org), and the read count for each gene was obtained using featureCounts (v1.6.3; subread.sourceforge.net). Clean read counts were further normalized using the fragments per kilobase per million mapped reads method. Moreover, underexpressed genes in any sample were filtered out. Cluster analysis was performed using the flashClust R (v1.01-2-2) package (17).

In weighted gene co-expression network analysis (WGCNA), the connectivity between two genes is the β power of their correlation coefficient. The β value determined both the scale-free topology fitting index and the mean connectivity of genes. The β value for scale-free topology fitting index was ≥0.8 by plotting the index against soft thresholds, which indicated that the topology of the network was scale-free, or independent, and simultaneously, the mean connectivity of genes was as high as possible, and it may result in several modules in the subsequent analysis for an insignificant mean connectivity. To understand relationship between them, the softConnectivity function from WGCNA package (v1.63) (18) was used, and the number of randomly selected genes was set to 5,000, while other parameters were set as default. The power was calculated by the pickSoftThreshold function in the WGCNA package. The expression values were compiled using the collapseRows function in the WGCNA package. Cluster analysis was subsequently performed using the flashClust function.

Hub genes are defined as genes with high correlation in candidate modules. In the present study, genes in the same co-expression modules with module membership ≥0.8 were considered as hub genes. Gene Ontology (GO; www.geneontology.org) enrichment analysis was carried out for each module and the corresponding gene data were mapped to the Database for Annotation, Visualization, and Integrated Discovery (DAVID; david.ncifcrf.gov/summary.jsp). A corrected P-value <0.05 was used as threshold. The most favorable module in the current study was visualized by Cytoscape v3.6.6 software (www.cytoscape.org), and the maximum intramodular connectivity of genes was considered as intramodular hub genes.

EVs were isolated and characterized morphologically and phenotypically. Measurements demonstrated that the concentration of the isolated particles was 9.55×10⁹±3.27×10⁷ particles/ml (data not shown), and the mean diameter of the particles was 137.8±30.2 nm (Fig. 1A-C). EVs were round-shaped and membrane-enclosed (Fig. 1D). The exosomal markers CD63 and CD9 were detected in the isolated vesicles, but not in peripheral blood mononuclear cells, whereas the expression of calnexin, the marker of peripheral blood mononuclear cells, showed the opposite results (Fig. 1E).

EVLR-sequencing yielded a median read count of 27.98 million mapped reads/sample. Overall, mRNA constituted 40% of total mapped reads. Other types of RNA included 37% circRNAs and 23% lncRNAs (Fig. 2A). On average, 13,586 mRNAs, 12,984 circRNAs, and 7,858 lncRNAs were detected from the 12 samples (Fig. 2B). In total, 19,391 genes were detected by EVLR-sequencing in the 12 samples. Moreover, 2,997 genes were filtered out as their expression was too low in any sample. Thus, expression levels of 16,394 genes in 12 samples were used to construct the co-expression network by the WGCNA package, of these 40% were associated with mRNA, 37% with circRNA and 23% with lncRNA.

Cluster analysis was conducted on the mRNAs detected in all 12 samples, using the flashClust package (Fig. 3A). Hub genes are displayed in gene dendrograms, in which they positively correlated together in different samples. Different soft-thresholding power values were analyzed. When the power value reached 25, the scale-free topology fitting index was ≥0.08 (Fig. 3B), and as the power value increased, the connectivity declined (Fig. 3C). Therefore, the power value used to construct co-expression module was 25. A total of 26 distinct gene co-expression modules were constructed altogether (Fig. 3D). The number of genes in each module is listed in Table I.

As shown on the network heatmap plot (Fig. 4A), each module showed an independent expression to each other. Hence, we calculated eigengenes of each module, so as to quantify their co-expression similarity. Modules were clustered on the basis of their correlation with each other, which was consistent with the heatmap plot of the adjacencies (Fig. 4B).

Results of GO enrichment analysis are summarized in Table II. Biological process involved in fetal cardiovascular malformations, such as regulation of glucose transport and apoptotic signaling pathway were included in the blue module. Furthermore, the expression of each module in all samples was analyzed. Among all the 26 modules, the blue module markedly differed between the two groups (Fig. 5A). Additionally, a total of 735 genes were involved in the blue module. These genes were enriched in ‘regulation of glucose transport’, ‘glucose transport’, ‘glucocorticoid receptor signaling pathway’, ‘ATP-dependent chromatin remodeling’, ‘histone deacetylation’, ‘histone H3-K4 methylation’, ‘DNA methylation’ and ‘apoptotic signaling pathways’. These biological processes were not enriched in other modules. The top 33 hub genes in the blue module, which have a high degree of correlation, are represented in an interaction network using Cytoscape software (Fig. 5B).