The cultured amniotic fluid cells [healthy controls (no chromosomal or other disease), n=24; CHD cases (fetus with ventricular septal defects), n=24] were obtained from singleton pregnant patients who underwent amniocentesis during gestational weeks 20.0-26.3. The patients were aged from 22 to 43 years old. Inclusion criteria: Patients were willing to have amniocentesis. In CHD group, the fetus was diagnosed as CHD by ultrasound. The indication for amniocentesis in the control group was either older than 35 years, a high risk in serum Down's screening or a high risk of non-invasive prenatal screening. Exclusion criteria: The fetus was affected with abnormalities chromosomes or abnormal copy number variation. The samples were collected in Shengjing Hospital of China Medical University (Shenyang, China) from April 2022 to December 2022. All participants provided written informed consent, and the study was approved by the Ethics Committee of Shengjing Hospital of China Medical University (approval no. 2022PS307K). The sample collection and experimental procedures conformed to the Declaration of Helsinki. All patients were non-smokers and non-alcoholics. The samples were matched for maternal and gestational age and sex of the fetus (Table I). The amniotic fluid cells were cultured with amniotic fluid medium (FUJIFILM Biosciences; cat. no. 99473210805) for 2 weeks in 5% CO₂ at 37˚C (Fig. 1).

The metabolites were extracted from the cell residue with 1 ml precooled methanol/acetonitrile/water (v/v, 2:2:1) under sonication every 10 min for 1 h in ice baths. The mixture was incubated at -20˚C for 1 h, followed by centrifugation at 14,000 x g at 4˚C for 20 min, and the supernatant was transferred to a sampling vial. To ensure data quality for metabolic profiling, quality control (QC) samples were prepared by pooling aliquots representative of all samples for data normalization. The supernatant was evaporated using a high-speed vacuum-concentration centrifuge, 14,000 x g and 4˚C about for 1 h. Dried extracts were redissolved in 50% acetonitrile. Each sample was centrifuged at 14,000 x g and 4˚C for 20 min, and the supernatant was used for analysis.

Metabolomic profiling was performed on UPLC-MS/MS system (1290 Infinity LC, Agilent Technologies, Inc.) and TripleTOF 5600 (SCIEX). Then, 5 µl Samples were separated with a 2.1x100.0 mm ACQUIY UPLC BEH 1.7 µM column (Waters China Ltd.). The flow rate was 0.5 ml/min and the mobile phase contained A (25 mM each ammonium acetate and ammonium hydroxide in water) and B (acetonitrile). The gradient was 95% B for 0.5 min and was linearly reduced to 65% in 6.5 min, 40% in 2.0 min (maintained for 1.0 min) and increased to 95% for 1.1 min, employing a 5-min re-equilibration period. Both electrospray ionization (ESI) positive and negative modes were applied for MS data acquisition. The ESI source conditions were as follows: Ion source gas 1 and 2, 60; curtain gas, 30; source temperature, 600˚C; ion spray voltage floating, ±5,500 V. In MS acquisition, the instrument was set to acquire data over the m/z range of 60-1,200 Da, and the accumulation time for MS scanning was 0.15 sec/spectra. In the auto MS/MS acquisition, the instrument was set to acquire data over the m/z range of 25-1,200 Da, and the accumulation time for the production scan was 0.03 sec/spectra. Blank (50% acetonitrile in water) and QC samples were injected every 12 samples during acquisition. All the samples were run in one batch for the positive and negative ion mode. Each sample was detected and analyzed using UPLC-MS/MS, and two original files (positive and negative ion mode) were obtained. MS total ion chromatogram (TIC) flow diagram was generated.

Data processing and analysis were as previously described with minor amendments (10). Raw MS data were converted to MzXML files using ProteoWizard MS Convert and processed using XCMS for feature detection, retention time correction and alignment. The metabolites were identified by accurate MS (<25 ppm) and MS/MS data, which were matched with a standard database (BaseDeepBP, Shanghai Bioprofile Technology, Human Metabolome Database (hmdb.ca/, MassBank, https://massbank.eu/MassBank/, MetaboBASE, http://metabase.com/, Global Natural Product Social Molecular Networking, https://gnps.ucsd.edu/ProteoSAFe/static/gnps-splash-old.jsp). In the extracted-ion features, only the variables with >50% of the non-zero measurement values in ≥1 group were retained.

All multivariate data analyses and modeling were conducted using the SIMCAP software (Version 14.0, Umetrics; Sartorius AG). After mean-centering the data using Pareto scaling, models were developed using principal component analysis (PCA), orthogonal partial least squares discriminant analysis (OPLS-DA) and PLS-DA. All models were evaluated for overfitting using permutation tests. Descriptive performance was assessed using cumulative R2X [ideal model: R2X (cumulative)=1 and R2Y (ideal model: R2Y (cumulative)=1] values, while prediction performance was measured using cumulative Q2 [ideal model: Q2 (cumulative)=1)] and a permutation test. In the permuted model, R2 and Q2 values at the Y-axis intercept should be lower than those of the non-permuted model. OPLS-DA was used to identify discriminating metabolites using variable importance on projection (VIP) scores, which indicate a variable contribution to class discrimination. VIP scores were calculated as the weighted sum of squares of the PLS weights, with values >1 considered statistically significant. High VIP scores indicate strong discriminatory ability and help in selecting biomarkers. Discriminating metabolites were obtained using a statistically significant threshold of VIP values from the OPLS-DA model and two-tailed Student's t-test (P-value) on normalized raw data. Metabolites with VIP values >1 and P<0.05 were considered statistically significant. The fold-change was calculated as the ratio of the average mass area of the CHD group to that of the control group. Identified differential metabolites were used for cluster analyses using the R package (R package version 1.0.12, CRAN.R-project.org/package=pheatmap). Hierarchical clustering for each group was performed for heatmap.

To identify perturbed biological pathways, differential metabolite data were subjected to KEGG pathway analysis using the KEGG database (kegg.jp/). KEGG enrichment analyses were performed using the Fisher's exact test and false discovery rate correction for multiple testing was also performed.

A total of 18 Female C57BL/6J mice (age, 8-10 weeks, Shenyang Lan Pudas technology co., ltd) were used (mean weight, 22.2±0.81 g; range, 20.8-23.6 g). The mice were kept in standard cages under a 12/12-h light/dark cycle, ad libitum food and water, temperature was 20-26˚C and humidity was 40-70%. To obtain pregnant mice, female mice mated with male mice (n=6, 8-10 weeks age, 23.2-25.6 g weight, Shenyang Lan Pudas Technology co., ltd) and underwent testing every morning. Females with a vaginal plug were immediately separated from the males and indicated as embryonic day 0.5 (E0.5). Pregnant mice were randomly assigned to treatment (either arginine or lysine, Sigma) and control group (both n=6). The treatment group received a single intraperitoneal injection of 1.5 mg/g body weight/day arginine/lysine from E7.5 to E14.5. The control group received an equivalent volume of saline. Doses of arginine or lysine administration were determined as previously described (11-13). In addition to the routine feeding, mouse health and behavior were checked every 12 h. The humane endpoints were cessation of eating and movement and listlessness or lack of response when people approached. No animals reached these endpoints. Pregnant mice at E15.5 were anaesthetized with 1.5% isoflurane for 2-3 min. The anesthetized mice had even heartbeat and breathing, relaxed muscles and no limb activity or pedal reflex. A total of 100 µl peripheral blood was collected from the inner canthus by capillary tube. Then, mice were euthanized by cervical dislocation. Death was confirmed when heartbeat and respiration stopped and the pupils dilated for >5 min. Maternal plasma arginine concentration was determined by LC-MS/MS. All experiments were approved by the Ethics Committee of Shengjing Hospital of China Medical University (approval no. 2024PS202K).

The embryonic trunk was fixed in 4% paraformaldehyde solution at 4˚C for ≥48 h. The samples were dehydrated in 75, 85, 95 and 100% ethanol for 1 h. Following xylene treatment at room temperature, 100% paraffin was used for tissue embedding. Then, continuous slices of the embryonic heart (4 µm) were cut from the top of the aortic arch to the apex of the heart and stained with hematoxylin and eosin at room temperature for 1 min using automatic stainer (Leica GmbH) and images were captured under a light microscope.

Maternal peripheral blood (96 controls and 69 CHD cases; gestational age, weeks 18.0-30.0, aged from 20 to 44 years old, collected in Shengjing Hospital of China Medical University from April 2022 to December 2022, the same Inclusion/exclusion criteria as amniotic fluid cells collection) were collected to detect the change of amino acids using a non-derivatized amino acids detection kit (Neobase 2, Revvity, Inc.) on LC-MS/MS (QSight 210MD, PerkinElmer). The sample was detected with multiple reaction monitoring mode scanning in positive ion mode. The ionization conditions were as follows: Drying gas, 120 l/min; electrospray voltage, 4,900 V; source temperature, 150˚C; nebulizer gas, 220 l/min; hot surface-induced desolvation temperature, 250˚C. The injection volume was 10 µl. Chromatographic mobile phase A (included in the detection kit) and the chromatographic gradient elution procedure was as follows: 0.00-0.15 min, flow rate, 0.16 ml/min; 0.15-0.90 min, flow rate, 0.03 ml/min; 0.90-1.00 min, flow rate, 0.7 ml/min and 1.0-1.15 min, flow rate, 0.16 ml/min. The parameters of amino acid detection MS are listed in Table SI.

All data are presented as the mean ± standard deviation. At least three independent experimental repeats. The analysis was performed using SPSS 17.0 software (SPSS, Inc.). Student's unpaired two-tailed t test, one-way ANOVA followed by least significant difference post hoc test and χ² test were used for statistical analysis. P<0.05 was considered to indicate a statistically significant difference.

Metabonomics analysis was performed on 48 samples (24 controls and 24 CHD cases). Each sample was detected and analyzed using UPLC-MS/MS, and two original files (positive and negative ion mode) were obtained. Representative images are shown in Fig. 2A. A total of 883 metabolites were identified (data not shown). The system stability was evaluated using two strategies: MS total ion chromatogram (TIC) flow diagram comparison of QC samples and PCA statistical analysis of overall samples. A total of four QC TIC diagrams were overlaid (Fig. 2B). The results suggested that the response intensity and retention time of each color spectrum peak overlapped, indicating that the variation caused by instrument error was small and the data quality was reliable. The ion peaks of metabolites extracted from all experimental and QC samples were subjected to PCA. Samples were closely gathered together, indicating that the experiment had good repeatability (Fig. 2C). Thus, the instrument analysis system had good stability, and the experimental data were reliable. The metabolic spectrum difference reflected the biological differences between the samples.

An OPLS-DA model was established to compare metabolite profiles between the two groups. The model evaluation parameters and score diagram are shown in Fig. 3. The OPLS-DA model distinguished the two sample groups. The parameters (R2Y=0.925 and Q2=0.76) in the OPLS-DA model suggested that the experimental data were stable and reliable. The intercept of Q2 was -0.319, indicating there was no overfitting in the OPLS-DA model established using the experimental data. The OPLS-DA model showed separation in the metabolic profiles between the CHD and control groups. A total of 292 different metabolites (VIP>1 and P<0.05) were identified, of which 172 metabolites were up- and 120 metabolites were downregulated in the CHD compared with the control group (supplementary data). The two most changed metabolites were lysine (L-lysine/lysine/N-methyllysine) and arginine (homoarginine/L-arginine; Table II).

To identify potential marker metabolites, univariate analysis was performed (FC<0.5 or >2.0). Lysine and arginine were the two most changed metabolites (Fig. 4). The fold-changes of L-lysine, lysine, homoarginine and L-arginine were 1848.4, 629.9, 766.5 and 38.2, respectively.

To evaluate the potential differential metabolites, hierarchical clustering for each group was performed (Fig. S1). Generally, when the candidate metabolites screened are reasonable and accurate, the same group of samples appear in the same cluster. Metabolites in the same cluster have similar expression patterns. Although some samples strayed from their groups, most were concentrated in their respective groups.

The differential metabolites were analyzed using KEGG (Fig. 5; supplementary data). The results demonstrated that amino acid metabolism (involved in ‘ABC transporters’, ‘biosynthesis of amino acids’ and ‘D-arginine and D-ornithine metabolism’) may play an important role in CHD development.

To test whether arginine/lysine affect heart development, arginine/lysine was intraperitoneal injected in pregnant mice. Litter size decreased, and the naked deformity rate increased after arginine treatment (Fig. 6A; Table III). The concentration of maternal plasma arginine increased following arginine treatment (182.6±14.4 vs. 92.3±9.6 µmol/l, Fig. 6B). In the arginine-treated group, the hearts of the embryos with no notable gross abnormality demonstrated enlarged heart cavities and thinner heart walls (Fig. 6C), compared with the control. In the lysine-treated group, no obvious deformity was observed.

As there were several amino acids changes in amniotic fluid cells metabonomics, the present study investigated whether the amino acids change in human maternal peripheral blood between CHD and control group. Compared with control group, glycine and glutamine were downregulated by 17.3% (503.8±147.1 vs. 608.7±266.8 µmol/l, Table IV) and upregulated by 6.7% (381.0±70.1 vs. 357.8±77.30 µmol/l), respectively. Therefore, these amino acids cannot be used as non-invasive markers for CHD screening.