From October 2013 to December 2017, we recruited 220 participants, including 110 PE patients and 110 controls, from two institutes in Shanghai, China (Department of Obstetrics, International Peace Maternity and Child Health Hospital, and Department of Obstetrics, Renji Hospital, Shanghai Jiao Tong University School of Medicine). The case-control study flow chart is displayed in Figure 1. Fasting peripheral blood specimens were collected before delivery. The inclusion criteria for the PE group were pregnant women with hypertension (≥ 140 mmHg systolic and ≥ 90 mmHg diastolic blood pressure), proteinuria higher than 0.3 g/d or qualitatively + above. In contrast, healthy age- and gravidity-matched pregnant women with normotensive pre-pregnancy and pregnancy and without obstetric complications as the control group.

Table 1 lists clinical features at the patients’ initial diagnosis of PE. We retrospectively collected clinical characteristics of subjects at their initial routine prenatal examination in early pregnancy (12–14 gestational weeks), which is listed in Table 2. Since the Department of Obstetrics at Renji Hospital is a consultation and resuscitation center for maternal near-miss in Shanghai, some PE patients in this hospital were referred from other institutions, so the baseline data of some subjects in early pregnancy (12–14 weeks of gestation) are missing. The study was conducted in accordance with the Declaration of Helsinki and was approved by the Medical Research Ethics Committee of International Peace Maternity and Child Health Hospital (the Ethics Committee Approval Protocol No. 5 of 28 February 2014). All subjects provided broad informed consent for the research use of their biological samples.

Genomic DNA was extracted from the peripheral blood (n = 10/group) of pregnant women and detected using Affymetrix CytoScan HD Cytogenetics Chips according to the manufacturer’s recommended protocol. Quality control analysis indicated that all microarrays were qualified, except one was judged as male by the Chromosome Analysis Suite software for unclear reasons (Supplementary Table 1). Therefore, we excluded this sample from the analysis of the microarray data.

The raw data was processed by Chromosome Analysis Suite analysis software from Affymetrix to obtain all the CNVs, including duplications and deletions. Given the small sample size for microarray analysis, we opted for CNVs that occurred with a frequency difference of more than 30% between the two groups. These CNVs did not overlap with the high-frequency CNV regions of ordinary individuals in the Database of Genomic Variants¹ as differential CNVs. The differential genes within the region of differential CNVs were given gene annotation by the Gene Ontology (GO) database. The GO function enrichment analysis was employed to obtain the special functions for these genes using Fisher’s exact test and χ2 test with the threshold of P < 0.05 and FDR < 0.25.

Total genomic DNA was extracted from peripheral blood specimens (n = 100/group) of pregnant women using the AxyPrep Multisource Genomic DNA Miniprep Kit (Axygen Scientific) following the manufacturer’s instructions. The genes linked to lipid metabolism were verified by quantitative polymerase chain reaction (qPCR), and GAPDH was used as an endogenous control at the DNA level. Details of the primer sequences are shown in Supplementary Table 2. In this study, correlations between candidate genes and clinical lipid profiles were analyzed by SPSS Statistics 25, using Pearson’s correlation test.

HepG2 (ATCC) cell line was cultured in DMEM medium (Gibco) supplemented with 10% fetal bovine serum (Gibco), 1% Penicillin-Streptomycin (Thermo Fisher Scientific) at 37^°C in a 5% CO2 atmosphere. We infected cells with lentivirus carrying APOBEC3A _Flag or negative control and screened the cells using puromycin (Thermo Fisher Scientific) to obtain a pool of cells overexpressing the APOBEC3A gene (or negative control). DMEM medium with 1 mM oleic acid (Sigma) served as the fat accumulation inducer for the cells. Cell cultures were collected after being starved overnight and treated with the fat accumulation inducer for 24 h. The intracellular lipid content was determined by a biochemical analyzer (Roche). The results were normalized to the total protein concentration.

The coding sequence of human APOBEC3A, the 3 × flag tag, and a green fluorescent protein were simultaneously subcloned into the AAV2/8 plasmid (AAV_APOBEC3A) under the control of the thyroxine-binding globulin promoter, a hepatic-specific promoter. This serotype is known to have a stable expression in the liver after injection (11). A virus containing green fluorescent protein alone under control of the thyroxine-binding globulin promoter was used as a control virus (AAV_Mock).

Separate groups of 6- to 8-week-old wild-type, C57BL/6J female mice (n = 14/group) were injected intraperitoneally with 1.5 × 10¹⁰ vector genomes/mouse of the relevant vectors. Mice were fed conventional chow throughout the study. They were sacrificed to investigate the impact of APOBEC3A on lipid metabolism (n = 6/group) or mated to explore the effect of the APOBEC3A gene on pregnant mice (n = 8/group, a positive vaginal plug was marked as gestation day 0, pregnant mice were sacrificed on gestation day 18) at 5 weeks (35 days) after AAV administration. Mice were fasted for 4 h, anesthetized with isoflurane, and whole blood was collected by cardiac puncture into BD Microtainer EDTA collection tubes. Plasma samples were used for lipid analysis and assayed using a biochemical analyzer (Roche). Liver and placenta specimens were obtained and immediately fast-frozen in nitrogen and stored at −80^°C for subsequent testing. Fetuses were delivered by cesarean section, and their weights were measured. We performed all animal procedures according to protocols approved by the Institutional Animal Care and Use Committee of Shanghai Institute of Materia Medica, Chinese Academy of Sciences.

Total RNA was extracted from HepG2 cells (n = 3/group) and used for RNA-seq analysis. cDNA library construction and sequencing were performed by the Shanghai Majorbio Bio-Pharm Technology Co., Ltd. using the majorbio cloud platform. High-quality reads were aligned to the human reference genome (GRCh38) using Bowtie2. We normalized the expression level of each gene to the fragment of the exon model per million mapped reads (FPKM) based on the expectation-maximization method. NOISeq method was used to screen out differentially expressed genes between the two groups with a fold change ≥ of 1.5. Function and pathway annotation and enrichment analysis were based on the GO database and the Kyoto Encyclopedia of Genes and Genomes database.

We obtained the list of genes connected with cholesterol homeostasis and fatty acid metabolism from the Gene Set Enrichment Analysis website. The FPKM of these genes were logarithmically (fold-change) converted, and we visualized the differences between the two groups through heatmap, which was performed by GraphPad Prism V.8 software.

Cell cultures (n = 3/group) or liver tissues (n = 6/group) were lysed in TRIzol (Invitrogen), and total RNA was extracted according to the manufacturer’s instructions. cDNA was synthesized using PrimeScript™ RT Master Mix (Takara). A reverse transcription-polymerase chain reaction was performed using PowerUp™ SYBR™ Green Master Mix (Thermo Fisher Scientific) on an Applied Biosystems 7500 Fast Real-Time PCR System (Thermo Fisher Scientific). Primer sequences for human cells are listed in Supplementary Table 3 (Supplementary Table 4 is for mouse tissues). All results were normalized to GAPDH expression and calculated using the comparative CT (ΔΔCT) method.

Cell cultures (n = 3/group) were lysed with RIPA lysis buffer (Thermo Fisher Scientific). Protein lysates were separated on 10% SDS-PAGE gels and electrophoretically transferred to polyvinylidene fluoride membranes (Millipore). The membranes were blocked with 5% skim milk and incubated overnight at 4^°C with mouse monoclonal anti-FLAG^® M2 (Sigma) as primary antibodies. Then, membranes were incubated with horseradish peroxidase-conjugated secondary antibody (Goat anti-rabbit, or -mouse IgG, Jackson ImmunoResearch). Protein band chemiluminescence was detected using an ECL Plus immunoblot detection system (Millipore). We performed densitometric analysis on low-exposure images with ImageJ (National Institutes of Health). Rabbit polyclonal anti-β-Tubulin antibody (Proteintech) was used as a loading control to confirm equivalent protein loading on the same membrane.

Fresh liver tissue samples were fixed in 4% paraformaldehyde and embedded in paraffin. We then sectioned them into 4-μm-thick sections for immunohistochemical staining. The primary antibodies used in this assay were rabbit monoclonal IgG [EPR25A]-isotype control (Abcam) and rabbit polyclonal to APOBEC3A (Abcam). Immunostaining was performed according to the standard protocol. Then, labeling was visualized using the diaminobenzidine method. Counterstaining was performed using Mayer’s hematoxylin.

Approximately 25 mg of the liver (n = 6/group) or placental tissue (n = 8/group) was weighed and made into a 10% saline tissue homogenate, centrifuged at 4^°C for 10 min at 3,000 g to obtain the supernatant for subsequent assays. TG or total cholesterol (TC) levels in the liver were measured by a biochemical analyzer (Roche). Placental growth factor (PLGF) or soluble fms-like tyrosine kinase-1 (sFlt-1) concentrations in plasma or placenta were measured by mouse PLGF or sFlt-1 Enzyme-linked immunosorbent assay kits (R&D Systems). The results were normalized by tissue mass.

Lipids from liver specimens (n = 6/group) were extracted according to the methyl tert-butyl ether method (12). Samples were separated and analyzed using an ultra-performance liquid chromatography system (Waters) equipped with a Q-Exactive hybrid quadrupole-Orbitrap mass spectrometer (Thermo Fisher Scientific). Lipid lists were obtained by annotation using Lipid Search software, and differential lipids were identified by t-test and orthogonal partial least squares discrimination analysis test. The Beijing Genomics Institute performed this assay.

Data were expressed as mean (SD) or median (IQR) when appropriate. Comparisons of continuous variables between two groups were made by unpaired two-tailed Student’s t-test or Mann-Whitney U-test when variables were normally or non-normally distributed, respectively. Statistical tests were performed by GraphPad Prism V.8. Correlation analysis was conducted by SPSS Statistics 25. P ≤ 0.05 (*) was considered statistically significant. Other statistical analysis methods used have been described separately in the respective methods section.

Graphical abstract was created using graphics from www.Biorender.com.

Compared with the control group, patients with PE presented with elevated blood pressure, increased proteinuria, decreased gestational age at delivery, and fetal growth restriction (Table 1), consistent with previous reports (13). The clinical characteristics of 68 PE patients and 107 controls in early pregnancy (12–14 gestational weeks) were retrospectively obtained, as shown in Figure 1 and Table 2. The findings showed that those pregnant women who subsequently developed PE exhibited distinctly elevated plasma TC (P = 0.037) and TG (P < 0.001) levels before the disease onset and had significantly increased body mass index (P = 0.006). However, their blood pressure (systolic blood pressure, P = 0.102; diastolic blood pressure, P = 0.240) did not differ substantially in early pregnancy compared to controls. These findings indicate that PE patients tend to have abnormal lipid metabolism before the onset of the disease, but the causative mechanisms have not been elucidated.

From the genomic analysis of 10 PE patients and 10 controls (clinical characteristics are shown in Supplementary Table 5), we obtained 27 differential CNVs, including 22 deletions and 5 duplications (Figure 2A). The identified variants ranged in size from 1,409 base pairs to nearly 151.7 kb (median = 10.8 kb), and these CNV regions contained 27 differential genes (Figure 2B and Supplementary Table 6). GO functional enrichment analysis revealed that the biological functions of these differential genes were enriched in response to nutrient levels, protein hydrolysis, developmentally programmed cell death, and regulation of triglyceride/cholesterol/lipoprotein metabolism (Figure 2C).

We identified six genes in the differential CNV regions that may be connected with lipid metabolism. The LPA gene located in the 6q26 region from 161.032 to 161.047 Mb was reported in previous research on PE (14–17), and we detected this duplication in one case (10%) and four controls (44.4%). APOBEC3A, APOBEC3A_B, and APOBEC3B in 22q13.1 from 39.350 to 39.390 Mb were detected in one case (10%) and four controls (44.4%). The BTNL3 gene in the 5q35.3 region from 180.379 to 180.431 Mb, with this deletion detected in one case (10%) and four controls (44. 4%). The LMF1 gene in the 16p13.3 region from 1.002 to 1.003 Mb, with this duplication detected in no one case and three controls (33.3%). These four regions of differential CNVs were confirmed by qPCR with significantly increased copy numbers of APOBEC3A, APOBEC3A_B, BTNL3, and LMF1 in PE patients compared with controls (Figures 3A–D). In contrast, the decrease in copies of LPA and the increase in copies of APOBEC3B were not statistically significant (Figures 3E,F).

To identify the predisposing genes causing lipid metabolism disturbance in PE patients, we collected clinical lipid profiles from these participants’ initial prenatal examination (12–14 gestational weeks) and analyzed the correlation between lipid profiles and the copy number of these candidate genes. The results indicated significant positive correlations between the copy number of APOBEC3A and both TC (P < 0.001) and LDL (P = 0.048) levels in serum (Figure 3G). In contrast, we did not find significant correlations between other candidate genes and clinical lipid profiles here, except for copies of both BTNL3 (P = 0.012) and LPA (P = 0.006), which were significantly positively correlated with HDL levels (Supplementary Figures 1A–E). For the first time, we report an interesting finding that the APOBEC3A gene may play a role in regulating lipid metabolism and be related to the pathogenesis of PE. Subsequently, we investigated how the novel gene APOBEC3A participates in regulating lipid metabolism and the pathogenesis of PE.

The liver is recognized as an essential metabolic organ in humans (18). To explore the underlying mechanism of APOBEC3A in lipid metabolism regulation, we constructed an in vitro model of HepG2 (HepG2^APOBEC3AOE), a human hepatocellular carcinoma cell line overexpressing APOBEC3A (Figure 4A). Both TC and TG levels were distinctly upregulated in HepG2^APOBEC3AOE cells compared to mock cells treated with or without fat accumulation inducer (Figure 4B). RNA-seq analysis revealed that the differentially expressed genes in HepG2^APOBEC3AOE were enriched in the biological function of response to lipid (Figure 4C) and the pathway of regulation of lipolysis in the adipocytes (Figure 4D). Furthermore, hallmark gene sets related to cholesterol homeostasis and metabolism of fatty acids in HepG2^APOBEC3AOE (Figure 4E) showed distinct differences compared to the mock cells. The genes linked to cholesterol biosynthesis (FDPS, DGAT2) and fatty acid biosynthesis (SCD, FASN), as well as fatty acid absorption (CD36), were upregulated in HepG2^APOBEC3AOE. On the other hand, the genes associated with cholesterol catabolism (GPD1) and fatty acid β-oxidation (CPT1A) were downregulated (Figure 4F). These findings indicate that APOBEC3A may be involved in both cholesterol and fatty acid metabolic processes.

To explore the potential roles of the APOBEC3A gene in regulating lipid metabolisms and participation in the pathogenesis of PE in vivo, we injected AAV_APOBEC3A and AAV_Mock intraperitoneally in adult C57BL/6J female mice (Figures 5A,B and Supplementary Figure 2). The first important finding was that TC and TG levels were elevated in both plasma and liver of APOBEC3A over-expressing mice (Figure 5C). Moreover, the expression of genes related to cholesterol biosynthesis (Hmgcr, Hmgcs2, Fdps) and fatty acid biosynthesis (Srebp1, Fasn, Acc1) were increased. In contrast, the expression of genes linked to fatty acid β-oxidation (Adiporq, Acoxl) was decreased in the liver of APOBEC3A over-expressing mice (Figure 5D). Lipidomics indicated that fatty acids, (O-acyl)-1-hydroxy fatty acids, triglycerides, and diacylglycerol were elevated in mice over-expressing APOBEC3A (Figure 5E and Supplementary Figure 3A). Moreover, pathway enrichment analysis indicated that differential lipids were enriched in glycerophospholipid metabolism, adipocytokine signaling pathway, fat digestion and absorption, regulation of lipolysis in adipocytes, and cholesterol metabolism (Supplementary Figure 3B). These results imply that APOBEC3A may play a significant role in the biosynthesis of fatty acids and triglycerides.

Fetuses in the AAV_APOBEC3A group had decreased fetuses/placenta ratios, which appeared to be driven by a significant reduction in fetal weight (Figure 5F), showing intrauterine growth restriction, a common clinical manifestation of PE. The sFlt-1/PLGF ratio is a well-known marker for the short-term prediction of PE (19, 20). We measured the levels of PLGF and sFlt-1 in mouse placenta homogenates and serum by ELISA. The protein and mRNA levels of sFlt-1 and the sFlt-1/PLGF ratio were significantly increased in the placenta of the AAV_APOBEC3A group compared with the AAV_Mock group (Figure 5G). In contrast, there was no substantial change in both protein and mRNA levels of PLGF in placental tissues (Supplementary Figure 4A). Besides, these changes mentioned above were not observed in serum samples (Supplementary Figure 4B). These findings indicate that APOBEC3A may be involved in the pathogenesis of PE.