Metabolic dynamics and prediction of sFGR and adverse fetal outcomes: a prospective longitudinal cohort study

This was a prospective longitudinal cohort study of sFGR. Pregnant women with MCDA twin pregnancies who underwent cesarean section between September 2017 and December 2021 were recruited for this study. The inclusion criteria were MCDA twin pregnancies, cesarean delivery, and agreement to undergo follow-up after delivery. The exclusion criteria were maternal factors, including chronic diseases; obstetric and delivery complications; fetal factors, including major congenital anomalies; and other complicated MCDA twin pregnancies undergoing selective reduction and other delivery complications (Fig. 1A). In our study, chorionicity was determined using ultrasonography by two different obstetric sonographers at 6–14 weeks when twin pregnancies were independently detected. We followed the guideline recommendations which include the following criteria: (1) the number of gestational sacs at 6–9 weeks [28] and (2) the identification of the “T” sign of the amniotic-placental junction between the twins at 10–14 weeks [29‐31]. After the twins were determined as MCDA twin, a board-certified twin-specialized obstetric sonographer was assigned to perform the ultrasonography examination and record fetal biometry/Doppler indices. sFGR was identified based on the diagnosis of the International Society of Ultrasound in Obstetrics and Gynecology (ISUOG) in 2016 that one fetus estimated fetal weight (EFW) < 10th centile and the intertwin weight discordance > 25% [32]. Finally, among 820 recruited twin pregnancies, 34 pairs of sFGR twins and 45 pairs of uncomplicated MCDA twin pregnancies were included in this study.

Maternal and fetal clinical characteristics, including maternal age, body mass index (BMI) before pregnancy and delivery, gestational week, fetal birth weight and height, sex, 1- and 5-min Apgar scores, and umbilical blood pH, were measured and recorded within 24 h after delivery (Table 1). Before delivery, fetal growth indicators, including the head circumference (HC), abdominal circumference (AC), biparietal diameter (BPD), and femur length (FL), were determined using ultrasound measurements (84 cases in the MCDA group and 48 in the sFGR group) (Additional file 1: Table S1). Additionally, within 5 days after delivery, data from neonatal cerebral ultrasonography (88 cases in the MCDA group, 68 in the sFGR group), including abnormal signal in brain, choroid plexus, and white matter, were collected to identify signs of neonatal brain injury (Additional file 1: Table S2). Two fixed obstetric sonographers conducted ultrasonography and determined the conditions of fetal growth and neonatal brain injury.

The neurocognitive behavioral development of infants aged 2–3 years was assessed using the Ages and Stages Questionnaire-third edition subscale (ASQ-3), which included the first speaking and walking times and five developmental domains: communication, gross motor, fine motor, problem solving, and personal social, with a total of 60 points available in the combined evaluation for each infant [33]. According to the suggestion provided by the World Health Organization (WHO) for anthropometric analysis of children younger than 5 years of age, z-scores were calculated using the “anthro_zscores” function in R package “anthro” (version 1.0.0): (1) “Height for age” is used to evaluate whether a child’s height is within the normal range for their age group; (2) “Weight for age” is used to assess whether a child’s weight falls within the expected range for their age; (3) “Height for weight” is used to assess the current nutritional or developmental status of a child; (4) “BMI for age” is used to assess whether the proportion of a child’s weight to their height is within the appropriate range. All follow-up surveys were conducted via telephone or online questionnaires, with oral responses.

Peripheral blood samples were collected and centrifuged at 1650 rcf (g) for 10 min at 4 °C, and plasma was collected and stored at – 80 °C. The peripheral blood of pregnant women in the first trimester was collected at 8–10 weeks (41 cases in the MCDA group, 13 in the sFGR group), that in second trimester was collected at 20–22 weeks (43 cases in the MCDA group, 23 in the sFGR group), and that in third trimester was collected at 30–32 weeks (43 cases in the MCDA group, 24 in the sFGR group). Cord blood was drawn quickly after the fetus was delivered (85 cases in the MCDA group, 63 in the sFGR group). All samples underwent only one freeze–thaw cycle prior to detection.

A total of 337 plasma samples, including maternal plasma (n = 187) and cord plasma (n = 150), were used to quantify the levels of the 25 target metabolites. The number of metabolites detected in each sample is shown in Figure S1 in Additional file 1. Plasma (20 μL) was aliquoted into a 1.5-mL Axygen tube and mixed with 80 μL of internal standards in methanol. The metabolites in the samples were precipitated by vortexing for 1 min at 4–8 °C, the supernatant was recovered by centrifugation at 20,000 g at 4 °C for 10 min, and 1 μL was injected into the system (QTRAP® 6500 plus LC–MS/MS system, SCIEX, Foster City, CA, USA). The reagents and resources used are listed in Table S3 in Additional file 1.

For amino acid (AA) measurements, including l-alanine, l-serine, l-proline, l-valine, l-threonine, l-lysine, l-glutamic acid, l-methionine, l-histidine, l-phenylalanine, l-arginine, l-tyrosine, l-cystine, l-leucine, and l-isoleucine, the supernatants (3 μL) were analyzed by injection onto an Intrada Amino Acid UPLC column (100 × 3 mm i.d., 3 μm; Imtakt, Kyoto, Japan) at a flow rate of 0.5 mL/min using the LC-20AD Shimadzu pump system and SIL-20AXR autosampler interfaced with the API 6500Q-TRAP mass spectrometer (SCIEX, Framingham, MA). The analytes were monitored using electrospray ionization in negative-ion mode with multiple reaction monitoring (MRM) of the precursor and characteristic production transitions of AAs.

For the phenylacetylglutamine (PAGln) pathway measurement, including PAGln and hippuric acid, supernatants (1 μL) were analyzed by injection onto a Kinetex C18 column (50 mm × 2.1 mm, 2.6 μm; Phenomenex, Torrance, CA) at a flow rate of 0.5 mL/min using the LC-20AD Shimadazu pump system and SIL-20AXR autosampler interfaced with the API 6500Q-TRAP mass spectrometer. The analytes were monitored using electrospray ionization in negative ion mode with MRM of the precursor and characteristic production transitions of the PAGln pathway.

For the trimethylamine-N-oxide (TMAO) pathway measurement, including carnitine, TMAO, betaine, choline, butyrobetaine, trimethyllysine, N, N, N-trimethyl-5-aminovaleric acid (TMVAV), and creatinine, the supernatants (3 μL) were analyzed by injection onto a silica column (2.0 × 150 mm; Luna 5u Silica 100A, Phenomenex, Torrance, CA) at a flow rate of 0.5 mL/min using the LC-20AD Shimadzu pump system and SIL-20AXR autosampler interfaced with the API 6500Q-TRAP mass spectrometer. The analytes were monitored using electrospray ionization in negative-ion mode with MRM of the precursor and characteristic production transitions of the TAMO pathway.

Student’s t-test was used to determine the statistical significance in the intergroup comparison of normally distributed data (if P > 0.05 in the Shapiro–Wilk normality test), whereas the Mann–Whitney U test was used for nonparametric comparisons of data between two different groups (if P < 0.05 in the Shapiro–Wilk normality test). The paired t-test and Wilcoxon’s-sign-rank-test were used for comparisons between large and small fetuses among twins. After correcting for multiple testing, a 2-sided P value of less than 0.002 (0.05/25) was used to indicate evidence of association. The chi-square test was used to compare the proportion of categories between the sFGR and MCDA groups in Table 1. We conducted mixed-effects logistic regression models using the function the function ‘glmer’ in the R package “lme4” (version 1.1–27) [34] to evaluate the statistical correlation between sFGR and the parameters of fetal brain injury estimated by ultrasonography (Additional file 1: Table S2). These models all include a fixed component for sFGR status and a random component for twin pairs to comprehensively assess the effect of sFGR and the relatedness of the individuals within twin pairs. Spearman’s-based correlation test was used to determine the correlation coefficient and P value between any two objects, using the functions “cor.test” and “cor” in the R package “stats” (version 3.6.0). Partial least squares discriminant analysis (PLS-DA) was performed to evaluate the plasma metabolome profiles in maternal plasma samples during pregnancy and cord plasma samples via the function “plsda” in the R package “mixOmics” (version 6.20.0) [35]. Binomial logistic regression analysis was conducted based on the Log2-transformed absolute concentration(s) of individual or multiple selected intergroup significantly differential metabolites using the function “glm” in the R package “stats” (version 3.6.0). The optimal combination of metabolites with the lowest Akaike information criterion (AIC) value in fitting new regression models was further determined using the function “stepAIC” in the R package “MASS” (version 7.3–58.1) [36], with a backward selection strategy. The predictions of a specific model were obtained using the function “Predict” in the R package “car” (version 3.0–10) [37]. Then, the function “roc” and “ggroc” in the R package “pROC” (version 1.17.0.1) [38] were used to build the receiver operating characteristic (ROC) object to evaluate the capacity of metabolites to identify sFGR or fetal brain injury. The 95% confidence interval (CI) of area under the curve (AUC) was calculated using the function “ci.auc” with the default “delong” method in the pROC package [38]. The ROC curves between Model A and B were compared using the function "roc.test" in the pROC package (version 1.17.0.1) with the bootstrap method and the number of bootstrap replicates was set as 10,000 [38]. The three ROC curves built based on the significantly differential metabolites identified in each trimester were compared among each other using function "roc.test" in the pROC package (version 1.17.0.1) with the default parameter “delong” method [38]. The linear mixed-effects model (LMM) were applied to assess the impact of trimesters and disease status (sFGR vs. MCDA or fetal brain injury (abnormal) vs. normal), and their interaction on peripheral metabolite levels employing the ‘lmer’ function in the R package ‘lme4’ (version 1.1–27) [34]. A P value < 0.05 was considered statistically significant (ns: P ≥ 0.05; ∗ P < 0.05; ∗  ∗ P < 0.01; ∗  ∗  ∗ P < 0.001).

A total of 45 cases from the MCDA group and 34 cases from the sFGR group were included in our study cohort. Maternal plasma samples were collected during different trimesters, whereas cord plasma samples were collected during delivery (Fig. 1A, Table 1). Based on the important role in sFGR emphasized in our previous research and literature reports on untargeted metabolomics, 25 key metabolites were absolutely quantified by targeted metabolomics that were closely related to fetal growth and development [26, 39], vascular system [17, 40‐42], and nervous system damage [16, 27, 43, 44]. To explore the association between maternal and fetal metabolite levels and the long-term neurocognitive behavioral development of infants, we also followed up with infants at the age of 2–3 years using the ASQ3 (Fig. 1B).

There were no significant differences between the MCDA and sFGR groups in maternal age (MCDA: 31.53 ± 4.56 years; sFGR: 31.59 ± 4.07 years), maternal BMI before pregnancy (MCDA: 21.95 ± 2.85 kg/m²; sFGR: 21.96 ± 3.36 kg/m²), maternal BMI before delivery (MCDA: 28.20 ± 3.12 kg/m²; sFGR: 26.11 ± 5.65 kg/m²), in vitro fertilization and embryo transfer pregnancy rate (MCDA: 11.1%; sFGR: 5.9%), and fetal sex (MCDA: 48.9%; sFGR: 58.8%). Moreover, the gestational weeks decreased significantly (MCDA: 35.42 ± 1.50 weeks; sFGR: 32.49 ± 2.22 weeks, P < 0.001), whereas the weight inconsistency rate was increased notably in sFGR (MCDA: 0.07 ± 0.06; sFGR: 0.31 ± 0.12, P < 0.001). Fetal birth weight (MCDA-small [S]: 2244.04 ± 313.09 g; sFGR-S: 1339.12 ± 371.89 g) and height (MCDA-S: 44.78 ± 2.16 cm; sFGR-S: 38.65 ± 3.31 cm) were significantly decreased in the smaller fetuses compared with the larger fetuses (birth weight: MCDA-large [L]: 2425.33 ± 325.99 g, P < 0.001; sFGR-L: 1950.88 ± 464.55 g, P < 0.001; birth height: MCDA-L: 45.31 ± 2.73 cm, P = 0.001; sFGR-L: 42.54 ± 3.54 cm, P < 0.001). Additionally, there was no significant difference in 1- and 5-min Apgar scores in either the MCDA or sFGR group between the larger and the smaller fetuses. However, the cord blood pH in sFGR was significantly different between the larger and the smaller fetuses (sFGR-L: 7.39 ± 0.04; sFGR-S: 7.36 ± 0.05, P = 0.002) (Table 1).

PLS-DA based on the level of 25 metabolites showed that maternal plasma samples in the three trimesters were apparently separated well (Fig. 2A, Table 2, Additional file 1: Figure S2A and Table S4), which is consistent with findings of previous reports that maternal metabolite levels vary in different trimesters [22]. With an increase in gestational weeks, we observed a consistent trend between the MCDA and sFGR groups: nine metabolites were gradually decreased, two metabolites were gradually increased, and nine metabolites showed a notable increasing or decreasing trend in the second trimester (Fig. 2C, Table 2, Additional file 1: Figure S3 and Table S4). Additionally, five metabolites, including l-threonine, l-histidine, l-cystine, l-leucine, and TMVAV, showed a contrasting trend between the MCDA and sFGR groups (Fig. 2C, Table 2, Additional file 1: Figure S3 and Table S4). Specifically, the maternal plasma metabolites in the second trimester showed the most distinct separation between the sFGR and MCDA groups (Fig. 2B). l-leucine was the only significantly different metabolite whose level decreased in the sFGR group in the first trimester, while the levels of l-glutamic acid and l-histidine in the third trimester were notably increased in the sFGR group compared to the MCDA group (Fig. 2C). Up to eight metabolites showed significant intergroup differences between the MCDA and sFGR groups in the second trimester. The levels of l-proline, l-phenylalanine, l-arginine, l-tyrosine, l-leucine, l-isoleucine, and betaine were significantly decreased, while the level of TMVAV was significantly increased in the sFGR group (Fig. 2C, Table 2). Among these, the P values of l-phenylalanine and l-tyrosine were less than 0.002 (0.05/25) (Table 2).

PLS-DA showed an unsatisfactory separation trend in cord plasma (Fig. 3A). However, up to seven differential metabolites, including l-valine, l-threonine, l-isoleucine, l-leucine, l-arginine, choline, and hippuric acid, were notably elevated in the sFGR group (Fig. 3B, Additional file 1: Table S5). Among the seven differential metabolites identified, l-threonine, l-isoleucine, choline, and hippuric acid also presented significantly positive maternal–fetal correlations between cord plasma and maternal plasma in the third trimester (Fig. 3B). Additionally, based on these 25 metabolites, significantly different metabolite was not observed in cord plasma between large and small fetuses in the sFGR group (Additional file 1: Table S5).

In the clinical setting, ultrasonography is used to evaluate intrauterine fetal development before delivery. Fetal development indicators including HC (MCDA: 29.69 ± 4.12 cm; sFGR: 28.34 ± 2.73 cm, P < 0.001), AC (MCDA: 28.34 ± 4.29 cm; sFGR: 25.83 ± 4.03 cm, P < 0.001), BPD (MCDA: 8.21 ± 1.21 cm; sFGR: 7.87 ± 0.87 cm, P < 0.001), FL (MCDA: 6.24 ± 1.01 cm; sFGR: 5.71 ± 0.82 cm, P < 0.001), and FL/BPD (MCDA: 0.76 ± 0.05; sFGR: 0.72 ± 0.05, P < 0.001) were significantly lower in the sFGR group than in the MCDA group (Additional file 1: Table S1). To further evaluate the efficacy of differential metabolites between sFGR and MCDA in predicting sFGR globally, logistic regressions were constructed based on individual metabolites or a combination of the differential metabolites in first trimester, second trimester, and third trimester, separately (Fig. 4A–B, Additional file 1: Figure S4A–C). Specifically, based on the aforementioned differential metabolites (P < 0.05), the predictive efficiencies in first trimester (area under the curve [AUC]: model A [l-leucine], 0.793) and third trimester (AUC: model A [l-glutamic acid and l-histidine], 0.744) were less than 0.8, and the highest predictive efficiency was obtained in second trimester (AUC: model A [l-proline, l-phenylalanine, l-arginine, l-tyrosine, l-leucine, l-isoleucine, betaine, and TMVAV], 0.885; model B [l-phenylalanine, l-leucine, and l-isoleucine], 0.878) (Fig. 4A–B, Additional file 1: Figure S4A–C and Table S6). In addition, we compared the ROC curves based on maternal plasma metabolite levels in different trimesters, and the result showed that the performance of the model based on the second trimester was significantly better than that of the third trimester (P = 0.036) (Additional file 1: Figure S4D). In cord plasma, ROC curves were constructed by differential metabolites between the sFGR and MCDA groups. Remarkably, the predictive efficiency (AUC: model A [l-valine, l-threonine, l-isoleucine, l-leucine, choline, and hippuric acid], 0.715; model B [l-valine, l-isoleucine, and hippuric acid], 0.702) in cord plasma (Additional file 1: Figure S4E–F and Table S6) was lower than that in all trimesters in maternal plasma. Additionally, there was no significant difference between models A and B neither for maternal plasma metabolites in second trimester nor for cord plasma (Fig. 4B, Additional file 1: Figure S4F).

Clinically, MCDA twins with a higher rate of discordance often have more serious adverse outcomes, including intrauterine fetal death [7, 45]. Hence, we performed a correlation analysis between maternal metabolites and the differences (diff) in physical development parameters of pairs of twins. Our findings revealed that l-leucine in the first trimester was negatively correlated with height diff (R =  − 0.32, P = 0.02) (Additional file 1: Figure S5B, Additional file 2: Table S7). l-glutamic acid in the third trimester was positively correlated with weight diff (R = 0.25, P = 0.05) (Additional file 1: Figure S6C, Additional file 2: Table S7). Moreover, in the second trimester, l-leucine was significantly negatively correlated with FL diff (R =  − 0.28, P = 0.04), height diff (R =  − 0.30, P = 0.02), and weight diff (R =  − 0.34, P = 0.007), and l-isoleucine was negatively correlated with HC diff (R =  − 0.34, P = 0.01) (Fig. 4C, Additional file 2: Table S7). l-phenylalanine in the second trimester was significantly positively correlated with birth weight (L: R = 0.28, P = 0.02; S: R = 0.32, P = 0.009), BPD (L: R = 0.34, P = 0.01; S: R = 0.29, P = 0.04), HC (L: R = 0.36, P = 0.009; S: R = 0.38, P = 0.005), and FL (L: R = 0.28, P = 0.04; S: R = 0.32, P = 0.02) in the large and small fetuses in the sFGR group (Additional file 1: Figure S6A, Additional file 2: Table S7). In addition, we found that seven elevated differential metabolites in cord plasma in the sFGR group were significantly negatively correlated with different fetal physical development parameters (Fig. 3B, Additional file 1: Figure S7, Additional file 2: Table S7).

To investigate potential associations between early-life metabolic alterations and later physical development, we analyzed the correlation between the level of maternal–fetal metabolites and evaluation indices of physical development in infants aged 2–3 years after birth. There were no significant differences in the age-standardized weight, height, BMI, and weight for height between the sFGR and MCDA groups (Fig. 5A). However, l-leucine in the second trimester was significantly related to BMI for age (R = 0.48, P = 0.029) and weight for height (R = 0.45, P = 0.039), whereas l-isoleucine in the second trimester was significantly positively correlated with height for age (R = 0.6, P = 0.0037) in the sFGR group (Fig. 5B).

Neonatal brain injury is a serious adverse outcome in MCDA twins, and those with sFGR suffer from higher morbidity and severity [1, 7, 46, 47]. Consistent with this, the incidence of brain injury in the sFGR group was higher than that in the MCDA group according to the cerebral ultrasonography results of all neonates in the cohort, including abnormal brain structure, choroid plexus, and white matter damage (Fig. 6A). The mixed-effects logistic regression analysis further revealed that the occurrence of sFGR were significantly associated with the abnormity of choroid plexus (P = 0.040), while not with abnormal brain structure (P = 0.062) and white matter damage (P = 0.141) (Additional file 1: Table S2). To explore the metabolites associated with the occurrence of fetal brain injury, samples were grouped into fetal brain injury or not. Intergroup comparison analysis revealed that l-serine and l-histidine in the first trimester, creatinine in the second trimester, and cord plasma were significantly decreased, while l-arginine in the first trimester, l-glutamic acid in the third trimester, and l-arginine in cord plasma were significantly increased in the fetal brain injury group (Additional file 1: Figure S8A). Furthermore, logistic regressions to distinguish brain injury were constructed based on the differential metabolites identified in maternal plasma and cord plasma, respectively (Fig. 6B–E). The model for maternal plasma metabolites showed outstanding value in distinguishing fetal brain injury (AUC: model A [l-serine, l-arginine, and l-histidine in first trimester, creatinine in second trimester, and l-glutamic acid in third trimester], 0.94; model B [l-serine, l-arginine, and l-histidine in first trimester, and creatinine in second trimester], 0.94) (Fig. 6B–C, Additional file 1: Table S8). Although l-arginine and creatinine were used in the maternal plasma and cord plasma models to predict fetal brain injury, the model in cord plasma (AUC: model A [l-arginine and creatinine in cord plasma], 0.689) showed low accuracy (Fig. 6B–E, Additional file 1: Table S8). Additionally, increased concentrations of l-glutamic acid in the third trimester were observed in the sFGR and fetal brain injury groups (Fig. 2C, Additional file 1: Figure S8A). We also found that the maternal plasma metabolites in the second trimester between the sFGR and MCDA groups showed suboptimal values in distinguishing fetal brain injury (AUC: model A, 0.762; model B [l-phenylalanine and l-arginine in second trimester], 0.732) (Additional file 1: Figure S8B–C and Table S8).

Furthermore, we compared the time of first talking and walking between the sFGR and MCDA groups for a simple assessment of neural and behavioral development, and found that the difference between the two groups was not statistically significant (Additional file 1: Figure S9A). However, the first walking time of infants with sFGR showed a significantly strong correlation with the levels of l-serine (R = 0.8, P = 0.018) and l-arginine (R =  − 0.8, P = 0.018) in the first trimester, as well as creatinine (R = 0.53, P = 0.013) in the second trimester (Additional file 1: Figure S9B). We evaluated neurocognitive behavioral development in infants by comparing the occurrence of low scores in each domain contained in the ASQ-3 between the sFGR and MCDA groups. The sFGR group showed a higher frequency of lower scores (< 30) in the five developmental domains than the MCDA group (Fig. 7A). For four differential maternal metabolites distinguishing fetal brain injury, the correlations between creatinine in the second trimester and gross motor (R =  − 0.55, P = 0.0097), fine motor (R = 0.73, P = 0.00016), l-arginine in the first trimester with communication (R =  − 0.77, P = 0.024), gross motor function (R = 0.77, P = 0.026), problem solving (R = 0.77, P = 0.024), l-histidine in first trimester with communication (R =  − 0.77, P = 0.024), personal social (R = 0.92, P = 0.001), l-serine in first trimester with communication (R = 0.77, P = 0.024), gross motor (R =  − 0.77, P = 0.026), and problem solving (R =  − 0.77, P = 0.024) in infants were high and significant (Fig. 7B). These findings provide important clues for the early diagnosis of fetal brain injury and long-term neurocognitive behavioral dysplasia in MCDA twins, especially in sFGR.

Since the sample sizes of each trimester in our cohort were not completely consistent (Additional file 1: Figure S10A), to evaluate the potential impact of sample size on the results, we removed those cases with missing values for any metabolite, and retained 40 cases for sensitivity analysis (Additional file 1: Figure S10B; fetal growth restriction group: sFGR: n = 10, MCDA: n = 30; fetal brain injury group: abnormal: n = 12; normal: n = 28). Intergroup differential analysis between the sFGR and MCDA groups revealed that four original significant differential metabolites (l-proline, l-phenylalanine, betaine and TMVAV) were retained, and l-leucine was also close to the detection threshold (P = 0.055) (Additional file 1: Figure S10C, Additional file 3: Table S9). However, no significantly differential metabolites were detected in the third trimester. In the first trimester, we identified one original significantly differential metabolite: l-leucine, as well as three new significantly differential metabolites (l-proline, betaine and l-methionine) compared with the original cohort. Between the fetal brain injury and normal groups, two original differential metabolites (l-serine and l-arginine) in the first trimester were retained. l-valine was an additional differential metabolite added in the third trimester (Additional file 1: Figure S10C, Additional file 3: Table S9).

In addition, we also conducted a comparative evaluation on the prediction efficiency of the prediction model constructed by the original differential metabolites in different trimesters and the remaining 40 cases. Similar to the previous results, there were no significant differences between the first and third trimesters, whereas there were significant differences between the second and third trimesters (P = 0.016) (Additional file 1: Figure S10D). The ROC curve was also drawn on the prediction model constructed by the original four differential factors (l-serine, l-arginine, and l-histidine in the first trimester, and creatinine in second trimester) on the remaining 40 cases, and its AUC was as high as 0.899 (Additional file 1: Figure S10E). Linear mixed-effects models (LMMs) were used to explore the impact of individual heterogeneity and gestational stages on our results (Additional file 3: Table S9). A total of 21 metabolites were significantly different among three trimesters (P < 0.05), with only l-histidine, l-cystine, l-leucine, and TMVAV, showing no significant difference in expression levels with trimesters. Similar to previous observations in the original cohort, LMMs analysis revealed that l-histidine, l-leucine, and TMVAV showed no significant difference with trimesters (P > 0.05), while l-cystine had significant difference with trimesters in the fetal brain injury group (Additional file 3: Table S9).

Taken together, these findings provide important clues for the early diagnosis of fetal brain injury and long-term neurocognitive behavioral dysplasia in MCDA twins, especially in sFGR (Additional file 1: Figure S11).