Skip to main content
NCBI home page
mSystems logoLink to mSystems
. 2026 Jan 9;11(2):e01405-25. doi: 10.1128/msystems.01405-25

Machine learning and causal inference applied to the gut metagenome-metabolome axis reveals a link between neonatal jaundice and autism spectrum disorder

Xianhong Chen 1,2,3,#, Cheng Chen 2,#, Xiucai Lan 4,#, Xueli Zhang 5, Tingting Li 2, Peng Zhang 6, Guoqiang Cheng 6, Wei Zhou 7, Zhangxing Wang 5, Yingmei Xie 2, Shujuan Zeng 3,✉, Wenhao Zhou 7,✉, Mingbang Wang 2,✉
Editor: Hongwei Zhou8
Reviewed by: Xingyin Liu9
PMCID: PMC12911356  PMID: 41511111

ABSTRACT

Neonatal jaundice (NJ) might increase the risk of autism spectrum disorder (ASD) in children. This study examined whether alterations in the gut microbiota could explain the link between NJ and ASD. We analyzed three cohorts: NJ cohort 1 comprised 68 neonates with NJ and 68 healthy controls (HCs); NJ cohort 2 included 56 infants with NJ and 14 HCs; and the ASD cohort consisted of 43 children with ASD and 31 typically developing children. Fecal samples were collected aseptically. We performed 16S rRNA sequencing (NJ cohort 1), liquid chromatography with tandem mass spectrometry metabolomics (NJ cohort 1 and ASD cohort), and shotgun metagenomics (NJ cohort 2 and ASD cohort). We characterized the gut DNA virome, quantified bile acid metabolism genes, and integrated multi-omics data using causal mediation and machine learning causal inference. Both NJ and ASD were associated with increased diversity of bile acid metabolism genes, suggesting biomarker potential. The gut DNA virome was also identified as a potential biomarker. Causal mediation analysis showed that the gut DNA virome influences bile acid metabolism genes in both conditions. Using machine learning-based causal modeling, we further found that gut human betaherpesviruses and human mastadenoviruses contribute to NJ and ASD, respectively, mediated by gut bile acid-metabolizing bacteria. These findings suggest that perturbations in the virome and bile acid-metabolizing bacteria may explain the link between NJ and ASD. Our results indicate that NJ and ASD are associated with bile acid metabolism alterations, which are also influenced by the gut DNA virome. Dysbiosis of the gut DNA virome and bile acid-metabolizing bacteria may mechanistically link NJ and ASD.

IMPORTANCE

Human epidemiological studies have established an association between perinatal pathogenic infections and autism spectrum disorder (ASD), and the gut microbiota plays an extremely important role in this relationship. Neonatal jaundice (NJ) may increase the risk of ASD in children. However, it remains unclear whether alterations in the gut microbiota affect the association between NJ and ASD. Both NJ and ASD are linked to altered gut bile acid metabolism and significantly elevated gene diversity among bile acid metabolism enzymes, and these relationships are influenced by the gut virome. Gut human betaherpesviruses and human mastadenoviruses influence the development of NJ and ASD, respectively, by influencing the abundance of gut bile acid-metabolizing microbes. Alterations of the gut virome and bile acid-metabolizing bacteria appear to explain the link between NJ and ASD. There is a lack of effective treatment options for ASD. We found that both NJ and ASD are linked to altered bile acid metabolism. Gaining a comprehensive understanding of the role of the bile acid-gut microbiota axis in the pathogenesis of NJ and ASD, as well as regulating this axis, may be crucial for developing novel preventive and therapeutic strategies for ASD.

KEYWORDS: gut microbiota, virome, bile acid metabolism, autism spectrum disorder, neonatal jaundice

INTRODUCTION

Human epidemiological evidence suggests that prenatal maternal infection by rubella, herpes simplex virus type 2, cytomegalovirus, or other pathogens and the resulting immune activation might influence the pathogenesis of neuropsychiatric disorders such as autism spectrum disorder (ASD) (1–6). The intraperitoneal administration of lipopolysaccharide (mimicking bacterial infection) or poly I:C (mimicking viral infection) to mice at day 12.5 of conception induces maternal immune activation (MIA); offspring born to mothers with MIA have been used as a classical animal model to understand the pathogenesis of ASD because they often exhibit ASD-like behavioral abnormalities (7–9). In support of the epidemiological data, studies based on animal models of MIA indicated that the maternal gut microbiota promotes neurodevelopmental abnormalities in mouse offspring (10), and human clinical studies also revealed that fecal microbiota transplantation (FMT) alters the gut microbiota composition of children with ASD while improving their behavioral symptoms (11). In conclusion, human epidemiological studies have established a link between perinatal pathogenic infections and ASD, and studies have indicated that alterations of the gut microbiota play an extremely important role in this association.

Neonatal jaundice (NJ) can increase the risk of ASD in children (12–14). NJ is a common neonatal complication, with its prevalence reaching 80% in preterm infants (15, 16), and preterm birth is associated with an increased risk of ASD (17, 18). NJ can increase the risk of ASD in children, and alterations of the gut microbiota link perinatal/infant risk factors with behavioral abnormalities. Disturbances of the gut microbiota early in life can affect brain development and behavior in offspring. Using a cohort of 213 mothers and 215 children, Dawson et al. found that alpha diversity in the maternal fecal microbiota during late gestation predicted internalizing behavior in children at 2 years of age, and taxa from the butyrate-producing families Trichophyceae and Coccidae were more abundant in the mothers of children with normal behavior (19). A healthy prenatal diet indirectly reduces internalizing behavior in children by increasing the alpha diversity of the maternal gut microbiota (19). There is a strong association between the composition of the gut microbiota in infancy and subsequent behavioral problems. By analyzing the gut microbiota at 1, 6, and 12 months of age and measuring behavioral outcomes at 2 years of age in 201 children, Loughman et al. found a significant association between the decreased abundance of Prevotella spp. in fecal samples collected at 12 months of age and increased behavioral problems at 2 years of age (20).

The gut microbiota is involved in the metabolism of bile acids, and the composition of the gut microbiota is influenced by bile acids themselves (21). A comprehensive understanding of the gut bacteria involved in the metabolism of bile acids is necessary. Bile acids are cholesterol-derived steroid molecules that play important roles in energy homeostasis, host metabolism, and the maintenance of innate immunity through G protein-coupled receptors and/or nuclear receptors (22). There is a bidirectional interaction between the gut microbiota and bile acids. On the one hand, bile acids reshape the gut microbiota by promoting the growth of bile acid-metabolizing bacteria and inhibiting the growth of other bile-sensitive bacteria, and on the other hand, the gut microbiota can influence bile acid metabolism and the composition of the bile acid pool by producing a variety of enzymes, such as bile salt hydrolases (BSHs) and hydroxysteroid dehydrogenases (HSDHs), which modify primary bile acids to secondary bile acids and further degrade them via other enzymatic mechanisms, thus helping to maintain cholesterol homeostasis (23).

The mechanisms by which the human gut DNA virome is established early in life and by which it interacts with the microbiota and metabolome, thereby playing a role in human disease, are unclear. Multiple enterovirus communities that infect both microbial and animal hosts collectively constitute the human gut DNA virome, and recent advances in human gut metagenomic sequencing and analysis revealed the complexity of the human gut DNA virome and facilitated the discovery of new viruses (24, 25). We previously used multi-omic tools such as metagenome–metabolome bioinformatic analysis to capture key bacteria involved in critical glutamate metabolism in the gut microbiota of patients with ASD and revealed an increase in the abundance of the bile acid-metabolizing bacterium Eggerthella lenta and its association with glutamate metabolism (26). Newly developed causal inference methods have also accelerated the use of multi-omic techniques to unravel the underlying pathogenesis of complex diseases (27–29). In the present study, we explored the landscape of the gut DNA virome, bile acid-metabolizing bacteria, and the metabolome in patients with NJ and ASD through a multi-omics approach and used machine learning causal inference methods to obtain a comprehensive understanding of the composition of the gut microbiota and metabolites that influence the clinical symptoms of NJ and ASD.

MATERIALS AND METHODS

Participant recruitment and sample collection

In total, three cohorts were included in this study. NJ cohort 1 included 68 patients with NJ and 68 matched healthy controls (HCs). NJ cohort 2 included 56 infants with NJ and 14 HCs. The ASD cohort included 43 children with ASD and 31 typically developing (TD) children. The inclusion and exclusion criteria for the three cohorts are detailed in the Supplemental methods. Researchers wearing medical sterile masks and sterile gloves collected fresh fecal samples within 1 h of participants’ defecation using disinfected disposable fecal collection tubes to prevent contamination. All samples were assigned unique identifiers to ensure anonymity. Samples were processed within 2 h of collection, aliquoted into 3–5 g portions per tube, and stored at −80°C. Written informed consent was obtained from the parents of each participant, and the study was approved by the Longgang District Maternity and Child Healthcare Hospital (approval no. LGFYKYXMLL-2024-100). Details of the cohorts are provided in Tables S1 to S3. The overall design of this project is presented in Fig. 1.

Fig 1.

Flowchart showing research across three cohorts examining gut microbiome, metabolome, and DNA virome in NJ and ASD. The diagram illustrates progression from metabolic profiling through bile acid gene analysis to causal modeling.

Open in a new tab

Study flowchart. This study included three cohorts, NJ cohort 1 (68 patients and 68 HCs), NJ cohort 2 (56 patients and 14 HCs), and the ASD cohort (43 children with ASD and 31 TD children). The overall design of this study was as follows. (i) Identification of metabolome alterations in NJ and ASD: liquid chromatography with tandem mass spectrometry (LC-MS/MS) was used to analyze metabolomes in NJ cohort 1 and the ASD cohort. In NJ cohort 1, 16S rRNA gene sequencing was used to profile the gut microbiome. For the ASD cohort, shotgun metagenomic sequencing was employed for profiling. Microbiome-metabolome association analysis identified bacteria and metabolites linked to NJ and ASD. (ii) Association of bile acid metabolism genes with the gut DNA virome: Shotgun metagenomic data from NJ cohort 2 and the ASD cohort were analyzed. The gut DNA virome was characterized using PathoScope 2, and bile acid metabolism genes (BSH and HSDH) were quantified with Diamond. Associations among viruses, bile acid metabolism genes, NJ, and ASD were examined through virome-metagenome integration. (iii) Linking gut DNA virome and bile acid metabolism to NJ and ASD: Causal mediation analysis evaluated the effects of the gut DNA virome and bile acid metabolism genes on NJ/ASD and clinical indicators. Then, we constructed a causal model and evaluated the robustness of the causal model using a machine learning causal inference method.

16S rRNA gene sequencing

In NJ cohort 1, 16S rRNA gene sequencing was used to detect the composition of the gut microbiome with reference to previously published studies (30–33), as detailed in the Supplemental material.

Gut metabolome analysis

In NJ cohort 1 and the ASD cohort, in reference to our previous studies (26, 34, 35), we analyzed the gut metabolome using liquid chromatography with tandem mass spectrometry (LC-MS/MS). Briefly, approximately 200 mg of fresh fecal sample per subject was used for metabolite extraction. The detailed procedures for sample preparation, LC-MS/MS instrumentation, and data processing are provided in the Supplemental material.

Gut metagenomic analysis

In NJ cohort 2 and the ASD cohort, referring to our previous studies (26, 36–38), we performed shotgun metagenomic sequencing, as detailed in the Supplemental material.

Gut bile acid metabolism genes

In NJ cohort 2 and the ASD cohort, metagenomic data sets were used to systematically identify gut bacteria carrying BSH and HSDH genes, as detailed in the Supplemental material.

Gut DNA virome

In NJ cohort 2 and the ASD cohort, metagenomic data sets, together with PathoScope software (version 2.0) (39, 40), were used to identify viral sequences from metagenomic data, as detailed in the Supplemental material. It is important to note that the shotgun metagenomic sequencing approach employed in this study captures only DNA sequences. Consequently, our analysis of the viral community is confined to the DNA virome, and RNA viruses were not detected.

Procrustes analysis

Procrustes analysis was used to assess the concordance of significant markers of differences in multi-omics data using the R package vegan as described by Zhao et al. (41) and as detailed in the Supplemental material.

Causal mediation analysis

Causal mediation analysis was used to assess the causal relationship of the independent variable with the dependent variable based on the mediating variable using the R package mediation (version 4.5.0), as detailed in the Supplemental material.

Causal inference based on machine learning

Causal inference based on machine learning was performed using Microsoft’s DoWhy library (https://github.com/microsoft/dowhy) and the EconML library (https://github.com/econml/), as detailed to in the software manual, and the protocol was described in previous studies (42–45) and the Supplemental material.

Other statistical analyses

To evaluate whether differences in gut microbiota composition, metabolites, virome, or bile acid metabolism genes could be used to distinguish patients with NJ from HCs or patients with ASD from TD children, we employed partial least squares discriminant analysis (PLS-DA). This analysis was conducted using the plsda() function from the R package mixOmics. To examine the correlations of the gut microbiota composition, metabolites, virome, or bile acid metabolism genes that differed significantly between the groups with clinical performance, logistic regression models were fitted using the lm() function in R. The resulting P-values and coefficients of determination were extracted via the summary() function. Additionally, boxplots and scatterplots were generated using the beeswarm R package to visualize group differences, with statistical significance determined by the Wilcoxon rank-sum test (wilcox.test() function).

RESULTS

Children with NJ or ASD exhibit alterations in gut bile acid metabolism

Through the integration of multi-omics data and machine learning-based causal inference in our previous studies (42), we identified significant contributions of gut bacteria and bile acid metabolites in NJ, allowing for the prediction of NJ risk. To investigate the association between gut metabolome alterations and NJ or ASD, LC-MS/MS was employed to characterize the metabolomic profiles of both NJ cohort 1 and the ASD cohort. In NJ cohort 1, the gut microbiome was profiled using 16S rRNA gene sequencing, while shotgun metagenomic sequencing was applied to analyze the gut microbiome in the ASD cohort.

We found that infants with NJ exhibited significantly higher gut bile acid metabolism (P < 0.001; Fig. S1A) and a marked reduction in the levels of the gut bile acid metabolite tauroursodeoxycholic acid compared to HCs (P < 0.001; Fig. S1B). In addition, tauroursodeoxycholic acid levels were negatively correlated with serum total bilirubin (TBIL) levels (Fig. S1C). Notably, patients with ASD also exhibited altered gut bile acid metabolism, demonstrating significantly elevated levels of the bile acid metabolites chenodeoxycholic acid-3-sulfate (P < 0.05; Fig. S1D) and taurocholic acid (P < 0.05; Fig. S1E) compared to TD children. Furthermore, chenodeoxycholic acid-3-sulfate levels showed a positive correlation with IgA levels (Fig. S1F).

Altered gut bile acid metabolism associated with the gut DNA virome composition

To elucidate the role of gut microbiota in bile acid metabolism in patients with NJ and ASD, we conducted shotgun metagenomic sequencing of fecal samples from both NJ cohort 2 and ASD cohort. The key enzymes involved in bile acid metabolism in intestinal bacteria are BSH and HSDH. BSH hydrolyzes bound bile salts to free bile acids and glycine/taurine, and HSDH further oxidizes free bile acids to form secondary bile acids, thereby regulating bacterial populations and maintaining gut microbiota balance (46, 47). We systematically quantified the abundance of BSH and HSDH genes in the metagenomic data sets to characterize the gut bile acid-metabolizing bacterial communities in these patient groups. Furthermore, given the potential regulatory role of gut viruses in shaping bacterial communities, we performed comprehensive virome profiling to identify bacteriophages and pathogenic viruses that may influence bile acid-metabolizing bacteria.

Through metagenomic association analysis, we identified specific BSH and HSDH genes associated with each disease. The bile acid metabolism genes of NJ include WP_035763572.1, 2BJF_A, WP_011590332.1, etc. Genes associated with ASD include WP_182424780.1, AAO76366.1, WP_057826614.1, RHH16704.1, etc. The complete list of BSH and HSDH genes associated with NJ and ASD is provided in Tables S4 and S5, respectively. Based on PLS-DA, we found that the NJ and HC groups could be distinguished to some extent based on the abundance of gut bile acid metabolism genes (Fig. 2A), and these groups could also be clearly distinguished using the gut DNA virome composition (Fig. 2B). Furthermore, we assessed the consistency of significantly altered gut bile acid metabolism genes and changes in the gut DNA virome composition based on Procrustes analysis, and the results revealed the potential consistency of differentially expressed gut bile acid metabolism genes and differences of the gut DNA virome composition between the NC and HC groups (P < 0.001; Fig. 2C).

Fig 2.

Statistical plots showing group separation based on gut bile acid metabolism genes and DNA viral profiles. Procrustes analyses reveal significant concordance between these microbial features across comparison groups of patients and controls.

Open in a new tab

Altered gut bile acid metabolism genes associated with the DNA viral composition. (A–C) Results from NJ patients and HCs. (A) PLS-DA revealed that the NJ and HC groups could be distinguished to some extent based on the abundance of gut bile acid metabolism genes. (B) PLS-DA revealed that the NJ and HC groups could be distinguished to some extent based on the DNA viral composition. (C) Procrustes analysis indicating significant concordance between DNA viral composition and bile acid metabolism gene abundance in NJ and HC groups (P < 0.001). Panels D–F present the results of patients with ASD and TD children. (D) PLS-DA plot distinguishing ASD and TD groups based on gut bile acid metabolism gene abundance. (E) PLS-DA plot clearly separating ASD and TD groups based on DNA viral composition. (F) Procrustes analysis demonstrating significant concordance between DNA viral composition and bile acid metabolism gene abundance in ASD and TD groups (P < 0.001).

Similar to the results in NJ cohort 2, based on PLSDA analysis, we found that the abundance of gut bile acid metabolism genes could somewhat distinguish children with ASD from TD children (Fig. 2D), and the gut DNA virome composition could clearly distinguish children with ASD from TD children (Fig. 2E). Furthermore, we also evaluated the consistency of differentially expressed gut bile acid metabolism genes and differences of the gut DNA virome composition based on Procrustes analysis, and the results demonstrated the potential consistency of differentially expressed gut bile acid metabolism genes and differences of the gut DNA virome composition between the ASD and TD groups (P < 0.001; Fig. 2F).

To investigate potential associations between gut viruses and key components of bile acid metabolism, we examined correlations with both bile acid-metabolizing bacteria and related functional genes. Our analysis revealed several specific virus-microbe relationships in the NJ cohort 2 (Fig. S3A and B). Most notably, the abundance of Bifidobacteriales was inversely associated with human betaherpesvirus levels (P < 0.001, R2 = 0.22), and similarly, Bifidobacterium showed a negative correlation with Escherichia Stx1 phage (P < 0.001, R2 = 0.22). In contrast, we found a positive correlation between Mycobacterium phage and the bile acid metabolism gene WP_17534087 (P < 0.001, R2 = 0.27). Within the ASD cohort, human mastadenovirus levels were positively associated with the bile acid metabolism gene WP_002565860_1 (P < 0.001, R2 = 0.28; Fig. S3D). Conversely, this same gene showed negative correlations with both Enterobacterial phage (P < 0.001, R2 = 0.24; Fig. S3eE) and Pandoravirus dulcis (P < 0.001, R2 = 0.32; Fig. S3F). These findings reinforce the potential functional links between specific viral elements and microbial (including gene-level) features in both NJ and ASD.

To further evaluate the clinical significance of bile acid metabolism genes, we examined their correlations with clinical phenotypes. The results indicated that in neonates with NJ, the abundance of gut bile acid metabolism genes (including EEA86306_1, EEG48917_1, and WP_ 082174150_1) was positively correlated with serum direct bilirubin (DBIL) levels (Fig. S2A through C). Similarly, among children diagnosed with ASD, a positive association was observed between the abundance of intestinal bile acid metabolism genes (such as WP_002565680.1, WP_002587930.1, and WP_007662998.1) and serum IgA levels (Fig. S2D through F).

Gut DNA virome composition affects NJ and ASD through bile acid metabolism genes

To further investigate whether the gut DNA virome and bile acid metabolism genes share mechanistic features that could partly explain the parallel findings in NJ and ASD, we performed causal mediation analysis to assess the effects of key viral constituents and bile acid metabolism genes on the development and core clinical features of each condition separately. In NJ cohort 1, the analysis revealed that the gut DNA virome can exert indirect physiological effects through specific bile acid metabolism genes, particularly 2HF0 and Q9KK62, while also exerting direct influences via genes such as AAC45411_2, EEA86306_1, and WP (Fig. 3A). Moreover, the virome composition also affected important clinical indicators of NJ, specifically serum DBIL and TBIL, through bile acid metabolism genes (Fig. 3B and C, detailed results in Table S6).

Fig 3.

Alluvial diagrams displaying causal pathways between enteroviruses, bile acid metabolism genes, and clinical outcomes. Flows show viral influence on neonatal jaundice, autism spectrum disorder, and bilirubin levels through metabolic mediation.

Open in a new tab

Causal mediation analysis revealed enteroviruses’ influence on NJ and ASD through bile acid metabolism genes. Results are visualized using alluvial diagrams. In each subplot, the three horizontal columns represent the causal effect of the treatment variable, the mediating variable, and the outcome variable mediated through the intermediate variable. Red indicates an indirect causal effect; blue indicates a direct causal effect. (A–C) Causal mediation analysis revealed that the DNA viral composition affects the development of NJ (A), DBIL levels (B), and TBIL levels (C) through bile acid metabolism genes. (D–F) Causal mediation analysis demonstrated that the DNA viral composition affects the development of ASD (D and E) and gut bilirubin levels (F) via bile acid metabolism genes. Model stability was evaluated via causal mediation analysis with 999 bootstrap replications to generate 95% CIs and P-values for the effects. A significant average causal mediation effect (ACME) (P < 0.05) confirms a stable causal mediation effect. The upward arrows in the figure indicate that these pathways lead to an increased risk of ASD and NJ or an increase in related indicators.

Similarly, in the ASD cohort, causal mediation analysis indicated that the composition of the enterovirus group could affect ASD through bile acid metabolism genes. Specifically, Human mastadenovirus was found to directly affect ASD via genes, including 6UFY_A, AAO76366_1, WP_002587930_1, and WP_007662998_1, while exerting indirect effects through AAB61150_1 (Fig. 3D). Furthermore, enteroviruses were shown to impact gut bacterial composition, particularly affecting species like Acidaminococcus fermentans, Eggerthella, Bifidobacterium animalis, and members of the Erysipelotrichaceae family (Fig. 3E). Additionally, the virome composition was found to influence gut bilirubin levels through its interaction with bile acid metabolism genes (Fig. 3F), as detailed in Table S7.

Finally, we constructed a causal model and evaluated the robustness of the causal model using a machine learning causal inference method. In NJ cohort 2, gut abundance of human betaherpesviruses was significantly elevated in the NJ group compared to HCs (P < 0.01; Fig. 4A) and was positively correlated with serum DBIL levels (P = 0.0002; Fig. 4B). Machine learning causal inference analysis suggested a potential causal effect of gut human betaherpesviruses on NJ, mediated by bacteria harboring bile acid metabolism genes (2HF0_A and 2HF0_B), with breastfeeding and the delivery mode identified as potential confounders (Fig. 4C).

Fig 4.

Data showing betaherpesvirus abundance in jaundice correlates with bilirubin while mastadenovirus in autism links with intestinal IgA. Machine learning identifies bile acid-metabolizing bacteria as mediators in both pathways with specific confounders.

Open in a new tab

Machine learning-based causal inference suggests gut DNA virome influences ASD and NJ through bile acid-metabolizing bacteria. (A–C) Gut human betaherpesviruses contribute to NJ development via bile acid-metabolizing microbes. (A) Human betaherpesviruses were more abundant in the NJ group than in the HC group (P < 0.01). (B) The abundance of human betaherpesviruses was significantly and positively correlated with serum DBIL levels. (C) Human betaherpesviruses have a potential causal effect on the development of NJ, and bile acid-metabolizing microbes (i.e., bacteria rich in the 2HF0_A and 2HF0_B genes as mediators), breastfeeding, and the delivery mode were identified as potential confounders. (D–F) Gut human mastadenoviruses affect the development of ASD by influencing gut microbes involved in bile acid metabolism. (D) Human mastadenoviruses were more abundant in ASD children than in TD controls (P < 0.001). (E) Significant positive correlation between the intestinal abundance of human mastadenoviruses and intestinal IgA content. (F) Machine learning causal inference methods identified a potential causal effect of human mastadenoviruses on the development of ASD, and bile acid-metabolizing microbes (i.e., bacteria rich in the RHH16704_1 and 6UFY_A genes) and Eggerthella spp. as mediators, gastrointestinal problems, and gender were identified as potential confounders. *P < 0.05, **P < 0.01, and ***P < 0.001.

Similar to the results in NJ cohort 2, analyses in the ASD cohort revealed significantly higher gut abundance of human mastadenoviruses in ASD children compared to TD controls (P < 0.001; Fig. 4D), which correlated positively with intestinal IgA content (P < 0.001; Fig. 4E). The machine learning-based causal model indicated a potential effect of human mastadenoviruses on the development of ASD, mediated by bile acid-metabolizing bacteria (carrying RHH16704_1 and 6UFY_A genes) and Eggerthella spp., while gastrointestinal problems and gender were identified as potential confounders (Fig. 4F). Through the summary and analysis of indicators such as gut bacteria, DNA viruses, bile acid-metabolizing bacteria, differentially expressed genes, group B Streptococcus (GBS) genes, and metabolites, we also found that a total of 70 gut markers were significantly elevated in both the ASD and NJ groups. These included 66 gut bacteria, 2 gut bile acid metabolism genes, 1 GBS gene, and 1 gut metabolite. Meanwhile, 33 gut markers were significantly reduced in the disease groups, comprising 31 bacteria, 1 GBS gene, and 1 gut bile acid metabolism gene, as detailed in Table S8.

DISCUSSION

Accumulating evidence underscores the critical role of the gut microbiota in human health, particularly through its involvement in bile acid metabolism (48–52). The neonatal gut microbiome is underdeveloped and highly vulnerable to environmental influences, such as medical care settings. Notably, neonates with NJ are often transferred to the neonatal intensive care unit, which has been identified as an environmental risk factor for ASD (53). This underscores the importance of the early-life gut microenvironment in neurodevelopment. Alterations in gut microbiota-mediated bile acid metabolism have been strongly associated with various liver diseases. For instance, Schneider et al. found that the gut microbiota and bile acid composition were altered in patients with primary sclerosing cholangitis and that gut microbiota-mediated negative feedback control of BA synthesis led to increased hepatic bile acid concentrations and disrupted bile duct barrier function (54). New evidence suggests that disturbances of gut microbiota-mediated bile acid synthesis play an important role in neurological disorders. It is known that bile acids can cross the blood-brain barrier and act directly on their receptors in the brain or more indirectly by activating intestinal receptors, leading to the release of signals such as fibroblast growth factor and glucagon-like peptide 1, thereby affecting neuronal activity in multiple brain regions or the vagus nerve (55–57). Proper regulation of bile acid homeostasis is essential for normal brain function; defects in these pathways are linked to neuropathological outcomes in both mice and humans, including demyelination, motor deficits, neuroinflammation, seizures, and cognitive impairment (58). In the present study, we observed that patients with NJ or ASD exhibited altered gut bile acid metabolism along with significantly increased diversity of gut bile acid metabolism genes, suggesting that these genes may serve as potential biomarkers for both conditions (see Fig. 5).

Fig 5.

Diagram of bile acid metabolism from liver to intestine via bacterial actions. Shows enterohepatic circulation and gut microbiota changes in NJ and ASD cases, with increased bile acid-metabolizing bacteria and Eggerthella spp. Includes viral influences.

Open in a new tab

Potential mechanisms by which alterations in the gut DNA virome and bile acid-metabolizing bacteria contribute to the development of NJ and ASD. Cholesterol is metabolized in the liver to synthesize primary bile acids, which are secreted into the intestine via the gallbladder. In the gut, bacterial BSH deconjugates primary bile acids into free bile acids, which are then transformed into secondary bile acids through oxidation catalyzed by bacterial HSDH. The majority (>95%) of intestinal bile acids are reabsorbed and returned to the liver via enterohepatic circulation. Alterations in the gut microbiota can disrupt this process, affecting bile acid homeostasis. Patients with NJ and ASD showed increased diversity of bile acid-metabolizing bacteria. Those with ASD also exhibited significant increases in the abundance of bile acid-metabolizing bacteria (Eggerthella spp.) and IgA levels. Gut bile acid metabolism was influenced by human mastadenoviruses and human betaherpesviruses. We propose that virome-induced changes in bile acid-metabolizing bacteria may contribute to the link between NJ and ASD.

Gut microbiota-mediated Th17/interleukin-17A (IL-17A) abnormalities play extremely important roles in ASD (59, 60). Previous studies in mice demonstrated that the maternal IL-17A signaling can promote autism-like behavior in offspring in a gut microbiota-dependent manner (10, 61). Bile acids act as important immunomodulators that regulate a range of immune cell functions by activating receptors such as TGR5 and FXR (62, 63). Recent studies indicated that gut microbiota-derived bile acid metabolites can act as ligands for the key transcription factor retinoid-related orphan receptor-γt (RORγt) to directly regulate the differentiation of Th17 cells. For example, Huang et al. found that the bile acid metabolite 3-oxoLCA inhibited Th17 cell differentiation by directly binding RORγt, and the administration of 3-oxoLCA to mice reduced Th17 cell differentiation in the gut lamina propria (64). In our study, we identified a positive correlation between the abundance of specific bile acid metabolism genes (WP_002565680.1, WP_002587930.1, and WP_007662998.1) and IgA levels in children with ASD. WP_002587930.1 and WP_002565680.1 encode choloylglycine hydrolase, a putative BSH gene potentially involved in bile acid deconjugation, while WP_007662998.1 encodes an short-chain dehydrogenase/reductase (SDR) family oxidoreductase, likely participating in bile acid oxidation. Additionally, levels of the gut bile acid metabolite chenodeoxycholic acid-3-sulfate were significantly elevated in children with ASD and correlated positively with IgA levels, supporting a potential interaction between bile acid metabolism and gut immune responses in ASD. Similarly, as observed in ASD, patients with NJ also exhibit altered gut bile acid metabolism. In the present study, we found that gut bile acid levels (cholic acid) were significantly higher, and tauroursodeoxycholic acid levels were significantly lower in patients with NJ than in HCs, whereas tauroursodeoxycholic acid levels were negatively correlated with serum TBIL levels. These findings suggest that bile acid metabolic alterations may be a common feature of both NJ and ASD.

The gut DNA virome, comprising both eukaryotic viruses and bacteriophages, is now recognized as a key modulator of human health and disease through its interactions with the gut microbiota and host immunity (65, 66). Bacteriophages are viruses that infect intestinal bacteria and represent one of the most abundant members of the gut microbiome (67). There is an interaction between phages and bacteria. On one hand, phages can lyse bacteria, thereby regulating bacterial populations and maintaining gut microbiota balance (46, 47). On the other hand, symbiotic relationships exist between phages and bacteria, as exemplified by temperate phages integrating into bacterial genomes for co-evolution (47, 68). Although intrauterine infections such as cytomegalovirus and hepatitis B virus are established risk factors for NJ—particularly cholestatic jaundice (69)—the role of the gut DNA virome in NJ remains poorly understood. In this study, we provided an initial characterization of the gut DNA virome in NJ, observing distinct viral community structures between NJ and HCs. Furthermore, both in the NJ and ASD cohorts, we observed a potential functional association between the gut DNA virome and bile acid metabolism genes. Based on these findings and supported by recent literature, we propose several mechanistic hypotheses that may explain how virome composition influences bile acid-metabolizing bacteria. First, shifts in phage abundance could directly modulate populations of bile acid-metabolizing bacteria through predator-prey interactions (70). Second, viral-encoded auxiliary metabolic genes (AMGs) may alter bacterial metabolic capacity (71), although such AMGs have not yet been fully characterized in bile acid pathways. Third, broader virome-mediated reshaping of the bacterial community (72) may have created niche opportunities that indirectly favor or suppress specific bile acid-modulating taxa. Finally, the gut DNA virome may trigger host immune responses that secondarily alter the gut environment and bacterial function (73).

The specific role of viruses in ASD is currently unknown. What is known is that administration of poly I:C, a molecule that mimics viral infection, to pregnant mice at day 12.5 of gestation can directly cause ASD-like behavior in their offspring, whereas the development of ASD-like behavior in offspring is dependent on the gut microbiota (2). A clinical FMT trial by Kang et al. reported concurrent alleviation of ASD symptoms and shifts in gut viral and bacterial communities following treatment (11). As observed for gut bile acid metabolism genes, a clear distinction could also be made between children with ASD and TD children based on the DNA viral composition, and notably, there was a potential concordance between gut bile acid metabolism genes and the ASD-specific DNA viral composition. In short, our results found a potential interaction between the gut DNA viral composition and gut bile acid metabolism genes.

Identifying important gut microbial, metabolite, viral, and phenotypic features from high-throughput multi-omic data, such as metagenomic, metabolomic, and viromic data, as well as host phenotypic features, is a challenging task. In the present study, in addition to gut microbiome–metabolome association analysis, we identified key bile acid metabolism genes and changes of the DNA viral composition affecting NJ and ASD by causal mediation analysis. Our results suggest that the gut DNA virome may influence NJ both directly and through bile acid metabolism genes, affecting core clinical indicators such as DBIL and TBIL. Similarly, in ASD, the virome may contribute to disease phenotype and bilirubin metabolism via bile acid-related pathways. Using machine learning-based causal inference, human betaherpesviruses and mastadenoviruses were implicated in the development of NJ and ASD, respectively, mediated by bile acid metabolic genes. These observations suggest that virome-induced shifts in bile acid composition might partially explain the link between NJ and ASD.

The gut microbiota plays a direct role in bile acid metabolism. As is known, cholesterol is generated in the liver by cholesterol 7-hydroxylase (CYP7A1) and sterol 27A hydroxylase (CYP27A1), discharged into the intestine through the gallbladder, hydrolyzed and desalted by bacterial BSH in the gut, and further transformed into secondary bile acids by oxidation under the action of gut bacterial HSDH (23). The majority (more than 95%) of bile acids in the gut can be reabsorbed by the liver through the enterohepatic circulation, and only a small amount (5%) of bile acids is excreted in feces (23). Structural and functional perturbations in the gut microbiome can disrupt this process. In our study, both NJ and ASD patients showed increased diversity of bile acid metabolism genes. Individuals with ASD also exhibited a higher abundance of bile acid-metabolizing bacteria (e.g., Eggerthella) and elevated IgA levels. Based on causal mediation and machine learning causal inference analyses, we found that bile acid metabolism was influenced by the composition of gut viruses such as human mastadenoviruses and human betaherpesviruses. In summary, we hypothesized that the altered abundance of bile acid-metabolizing bacteria caused by changes in the DNA viral composition might play extremely important roles in the development of NJ and ASD and established a link between NJ and ASD (see Fig. 5). Currently, we have not yet fully identified all gut bacterial species capable of expressing BSH or HSDH. Our next step involves isolating these specific bacterial strains and conducting 16S rRNA sequencing for accurate identification.

A key aspect of this study is the use of a multi-omic framework, incorporating metagenomic, metabolomic, and viromic data to identify functional genes and viral components involved in bile acid metabolism in NJ and ASD. Furthermore, the combination of causal mediation and machine learning approaches helped infer potential virus-bile acid interactions underlying disease mechanisms.

However, this study had several limitations. First, the causal relationships identified here are derived from computational inference and observational data. While these models highlight plausible mechanistic pathways, future experimental validations using animal models or interventional studies are necessary to definitively establish causality. Second, the shotgun metagenomic libraries we analyzed were limited to DNA, meaning our findings provide a comprehensive view of the gut DNA virome but do not encompass the role of RNA viruses. Given the potential significance of RNA viruses in human health and disease, future investigations that incorporate RNA sequencing will be crucial to obtain a complete picture of the entire viral community and its interactions with the host and bacteriome. Furthermore, due to budgetary constraints, 16S rRNA sequencing was employed in NJ cohort 1, limiting the resolution of gene-level metabolic profiling. Additionally, the metabolome was not completed in NJ cohort 2 because of the lack of a sufficient number of completed macrogenomic sequencing samples. Furthermore, there are many variables that affect the gut microbiome, including host genetic factors, but these factors are difficult to control. Finally, our parent questionnaire showed that only 7 of 47 ASD children had a history of NJ. The sample size of the ASD cohort remains relatively small, and the retrospective collection of NJ history from parents is susceptible to recall bias, potentially leading to underreporting of milder cases. Future prospective studies with larger cohorts, systematic tracking of perinatal exposures, and functional experiments are needed to verify and extend these findings.

Currently, there are no universally effective treatments for ASD, although dietary interventions, probiotics, and FMT have shown promise in alleviating symptoms, possibly through microbial and metabolic restoration (74–76). Our analyses suggest that targeting the virome-microbiota-bile acid interface might offer a potential therapeutic strategy for ASD, a hypothesis that merits further investigation.

Conclusion

Patients with NJ and ASD exhibit alterations of gut bile acid metabolism that are influenced by the DNA viral composition, and alterations in the DNA viral composition and bile acid-metabolizing bacteria appear to explain the link between NJ and ASD.

Supplementary Material

Reviewer comments
reviewer-comments.pdf (361.5KB, pdf)

ACKNOWLEDGMENTS

This study was financially supported by National Natural Science Foundation of China (No. 82571963), Natural Science Foundation of Guangdong Province, China (No.2025A1515012162, No.2024A1515010590), Shenzhen Science and Technology Program (No.s JCYJ20250604145739052 and JCYJ20240813144117023), the Research Initiation Fund of Longgang District Maternity and Child Healthcare Hospital of Shenzhen City (No. Y2024001), and by the Shenzhen Municipal Human Resources and Social Security Bureau (Postdoctoral fellow stationed in Shenzhen, second batch in 2022).

We would like to thank all participants for their invaluable contributions to this research.

X.C.: Data curation, investigation, formal analysis, visualization, writing—original draft, and writing—review and editing. C.C.: Data curation, investigation, formal analysis, visualization, and writing—original draft. X.L.: Data curation, investigation, formal analysis, visualization, and writing—original draft. X.Z.: Data curation, investigation, and formal analysis. T.L.: Methodology and investigation. P.Z.: Methodology and investigation. G.C.: Investigation. W.Z.: Investigation. Z.W.: Investigation. Y.X.: Investigation. S.Z.: Supervision, project administration, and writing—review and editing. W.Z.: Conceptualization, supervision, project administration, resources, and writing—review and editing. M.W.: Conceptualization, supervision, funding acquisition, project administration, resources, and writing—review and editing.

Contributor Information

Shujuan Zeng, Email: zengshujuansz@163.com.

Wenhao Zhou, Email: zhouwenhao@fudan.edu.cn.

Mingbang Wang, Email: mingbang.wang.bgi@qq.com.

Hongwei Zhou, Southern Medical University, Guangzhou, Guandong, China.

Xingyin Liu, Nanjing Medical University, NanJing, China.

DATA AVAILABILITY

The data of the NJ cohort 1 project have been uploaded to CNSA with project ID CNP0003066. The data of the ASD cohort and NJ cohort 2 projects have been uploaded to NCBI and can be downloaded according to the BioProject IDs, namely, PRJEB23052 and PRJEB23053.

ETHICS APPROVAL

This study complied with the principles of the Declaration of Helsinki and was approved by the Medical Ethics Committees of Longgang District Maternity & Child Healthcare Hospital (Approval No. LGFYKYXMLL-2024-100). Written informed consent was obtained from the parents of each participant.

SUPPLEMENTAL MATERIAL

The following material is available online at https://doi.org/10.1128/msystems.01405-25.

Supplemental Material. msystems.01405-25-s0001.docx.

Supplemental figures and methods.

msystems.01405-25-s0001.docx (528.3KB, docx)
DOI: 10.1128/msystems.01405-25.SuF1
Supplemental Tables. msystems.01405-25-s0002.xlsx.

Tables S1 to S8.

msystems.01405-25-s0002.xlsx (58.4KB, xlsx)
DOI: 10.1128/msystems.01405-25.SuF2
OPEN PEER REVIEW. reviewer-comments.pdf.

An accounting of the reviewer comments and feedback.

reviewer-comments.pdf (361.5KB, pdf)
DOI: 10.1128/msystems.01405-25.SuF3

ASM does not own the copyrights to Supplemental Material that may be linked to, or accessed through, an article. The authors have granted ASM a non-exclusive, world-wide license to publish the Supplemental Material files. Please contact the corresponding author directly for reuse.

REFERENCES

  • 1. Steinmetz JD, Seeher KM, Schiess N, Nichols E, Cao B, Servili C, Cavallera V, Cousin E, Hagins H, Moberg ME, et al. 2024. Global, regional, and national burden of disorders affecting the nervous system, 1990–2021: a systematic analysis for the Global Burden of Disease Study 2021. Lancet Neurol 23:344–381. doi: 10.1016/S1474-4422(24)00038-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2. Careaga M, Murai T, Bauman MD. 2017. Maternal immune activation and autism spectrum disorder: from rodents to nonhuman and human primates. Biol Psychiatry Cogn Neurosci Neuroimaging 81:391–401. doi: 10.1016/j.biopsych.2016.10.020 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3. Gumusoglu SB, Stevens HE. 2019. Maternal inflammation and neurodevelopmental programming: a review of preclinical outcomes and implications for translational psychiatry. Biol Psychiatry 85:107–121. doi: 10.1016/j.biopsych.2018.08.008 [DOI] [PubMed] [Google Scholar]
  • 4. Wang M, Doenyas C, Wan J, Zeng S, Cai C, Zhou J, Liu Y, Yin Z, Zhou W. 2021. Virulence factor-related gut microbiota genes and immunoglobulin A levels as novel markers for machine learning-based classification of autism spectrum disorder. Comput Struct Biotechnol J 19:545–554. doi: 10.1016/j.csbj.2020.12.012 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5. Pesch MH, Leung J, Lanzieri TM, Tinker SC, Rose CE, Danielson ML, Yeargin-Allsopp M, Grosse SD. 2024. Autism spectrum disorder diagnoses and congenital cytomegalovirus. Pediatrics 153:e2023064081. doi: 10.1542/peds.2023-064081 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6. Mohebalizadeh M, Babapour G, Maleki Aghdam M, Mohammadi T, Jafari R, Shafiei-Irannejad V. 2024. Role of maternal immune factors in neuroimmunology of brain development. Mol Neurobiol 61:9993–10005. doi: 10.1007/s12035-023-03749-2 [DOI] [PubMed] [Google Scholar]
  • 7. Yu D, Li T, Delpech JC, Zhu B, Kishore P, Koshi T, Luo R, Pratt KJB, Popova G, Nowakowski TJ, Villeda SA, Piao X. 2022. Microglial GPR56 is the molecular target of maternal immune activation-induced parvalbumin-positive interneuron deficits. Sci Adv 8:eabm2545. doi: 10.1126/sciadv.abm2545 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8. Lombardo MV, Moon HM, Su J, Palmer TD, Courchesne E, Pramparo T. 2018. Maternal immune activation dysregulation of the fetal brain transcriptome and relevance to the pathophysiology of autism spectrum disorder. Mol Psychiatry 23:1001–1013. doi: 10.1038/mp.2017.15 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9. Sarieva K, Kagermeier T, Khakipoor S, Atay E, Yentür Z, Becker K, Mayer S. 2023. Human brain organoid model of maternal immune activation identifies radial glia cells as selectively vulnerable. Mol Psychiatry 28:5077–5089. doi: 10.1038/s41380-023-01997-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10. Kim S, Kim H, Yim YS, Ha S, Atarashi K, Tan TG, Longman RS, Honda K, Littman DR, Choi GB, Huh JR. 2017. Maternal gut bacteria promote neurodevelopmental abnormalities in mouse offspring. Nature 549:528–532. doi: 10.1038/nature23910 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11. Kang DW, Adams JB, Gregory AC, Borody T, Chittick L, Fasano A, Khoruts A, Geis E, Maldonado J, McDonough-Means S, Pollard EL, Roux S, Sadowsky MJ, Lipson KS, Sullivan MB, Caporaso JG, Krajmalnik-Brown R. 2017. Microbiota transfer therapy alters gut ecosystem and improves gastrointestinal and autism symptoms: an open-label study. Microbiome 5:10. doi: 10.1186/s40168-016-0225-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12. Jenabi E, Bashirian S, Khazaei S. 2020. Association between neonatal jaundice and autism spectrum disorders among children: a meta-analysis. Clin Exp Pediatr 63:8–13. doi: 10.3345/kjp.2019.00815 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13. Wiggins LD, Overwyk K, Daniels J, Barger B, Crain H, Grzadzinski R, Moody E, Reynolds A, Reyes N, Rosenberg C, Rosenberg S, Pazol K. 2024. Risk factors and clinical correlates of sensory dysfunction in preschool children with and without autism spectrum disorder. Autism Res 17:162–171. doi: 10.1002/aur.3074 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14. Chou HC, Lin HC, Huang KH, Chang YC. 2023. Associations between neonatal jaundice and autism spectrum disorder or attention deficit hyperactivity disorder: nationwide population based cohort study. J Formos Med Assoc 122:1150–1157. doi: 10.1016/j.jfma.2023.05.010 [DOI] [PubMed] [Google Scholar]
  • 15. Maisels MJ, McDonagh AF. 2008. Phototherapy for neonatal jaundice. N Engl J Med 358:920–928. doi: 10.1056/NEJMct0708376 [DOI] [PubMed] [Google Scholar]
  • 16. Chen K, Yuan T. 2020. The role of microbiota in neonatal hyperbilirubinemia. Am J Transl Res 12:7459–7474. [PMC free article] [PubMed] [Google Scholar]
  • 17. Talmi Z, Mankuta D, Raz R. 2020. Birth weight and autism spectrum disorder: a population-based nested case-control study. Autism Res 13:655–665. doi: 10.1002/aur.2260 [DOI] [PubMed] [Google Scholar]
  • 18. Persson M, Opdahl S, Risnes K, Gross R, Kajantie E, Reichenberg A, Gissler M, Sandin S. 2020. Gestational age and the risk of autism spectrum disorder in Sweden, Finland, and Norway: a cohort study. PLoS Med 17:e1003207. doi: 10.1371/journal.pmed.1003207 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19. Dawson SL, O’Hely M, Jacka FN, Ponsonby A-L, Symeonides C, Loughman A, Collier F, Moreno-Betancur M, Sly P, Burgner D, Tang MLK, Saffery R, Ranganathan S, Conlon MA, Harrison LC, Brix S, Kristiansen K, Vuillermin P, BIS Investigator Group . 2021. Maternal prenatal gut microbiota composition predicts child behaviour. EBioMedicine 68:103400. doi: 10.1016/j.ebiom.2021.103400 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20. Loughman A, Ponsonby A-L, O’Hely M, Symeonides C, Collier F, Tang MLK, Carlin J, Ranganathan S, Allen K, Pezic A, Saffery R, Jacka F, Harrison LC, Sly PD, Vuillermin P, BIS Investigator Group . 2020. Gut microbiota composition during infancy and subsequent behavioural outcomes. EBioMedicine 52:102640. doi: 10.1016/j.ebiom.2020.102640 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21. Collins SL, Stine JG, Bisanz JE, Okafor CD, Patterson AD. 2023. Bile acids and the gut microbiota: metabolic interactions and impacts on disease. Nat Rev Microbiol 21:236–247. doi: 10.1038/s41579-022-00805-x [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22. Hu J, Wang C, Huang X, Yi S, Pan S, Zhang Y, Yuan G, Cao Q, Ye X, Li H. 2021. Gut microbiota-mediated secondary bile acids regulate dendritic cells to attenuate autoimmune uveitis through TGR5 signaling. Cell Rep 36:109726. doi: 10.1016/j.celrep.2021.109726 [DOI] [PubMed] [Google Scholar]
  • 23. Wahlström A, Sayin SI, Marschall H-U, Bäckhed F. 2016. Intestinal crosstalk between bile acids and microbiota and its impact on host metabolism. Cell Metab 24:41–50. doi: 10.1016/j.cmet.2016.05.005 [DOI] [PubMed] [Google Scholar]
  • 24. Li Y, Handley SA, Baldridge MT. 2021. The dark side of the gut: Virome-host interactions in intestinal homeostasis and disease. J Exp Med 218:e20201044. doi: 10.1084/jem.20201044 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25. Liang G, Bushman FD. 2021. The human virome: assembly, composition and host interactions. Nat Rev Microbiol 19:514–527. doi: 10.1038/s41579-021-00536-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26. Wang M, Wan J, Rong H, He F, Wang H, Zhou J, Cai C, Wang Y, Xu R, Yin Z, Zhou W. 2019. Alterations in gut glutamate metabolism associated with changes in gut microbiota composition in children with autism spectrum disorder. mSystems 4:e00321-18. doi: 10.1128/mSystems.00321-18 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27. Liu HM, Lin X, Meng XH, Zhao Q, Shen J, Xiao HM, Deng HW. 2021. Integrated metagenome and metabolome analyses of blood pressure studies in early postmenopausal Chinese women. J Hypertens 39:1800–1809. doi: 10.1097/HJH.0000000000002832 [DOI] [PubMed] [Google Scholar]
  • 28. Lv BM, Quan Y, Zhang HY. 2021. Causal inference in microbiome medicine: principles and applications. Trends Microbiol 29:736–746. doi: 10.1016/j.tim.2021.03.015 [DOI] [PubMed] [Google Scholar]
  • 29. Mu X, Jia F, Qiu S, Li Y, Mei N, Zhao X, Han B, Han X, Zhang J, Yao H. 2025. Predicting and interpreting nitrogen removal performance and functional microbial abundance of single-stage partial nitrification and anammox system using machine learning methods. Bioresour Technol 437:133119. doi: 10.1016/j.biortech.2025.133119 [DOI] [PubMed] [Google Scholar]
  • 30. Xu R, Wu B, Liang J, He F, Gu W, Li K, Luo Y, Chen J, Gao Y, Wu Z, Wang Y, Zhou W, Wang M. 2020. Altered gut microbiota and mucosal immunity in patients with schizophrenia. Brain Behav Immun 85:120–127. doi: 10.1016/j.bbi.2019.06.039 [DOI] [PubMed] [Google Scholar]
  • 31. Liu J, Wang M, Chen W, Ma J, Peng Y, Zhang M, Wang C, Yan G, Lu G. 2021. Altered gut microbiota taxonomic compositions of patients with sepsis in a pediatric intensive care unit. Front Pediatr 9:645060. doi: 10.3389/fped.2021.645060 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32. Li Z, Lu G, Luo E, Wu B, Li Z, Guo J, Xia Z, Zheng C, Su Q, Zeng Y, Chan WY, Su X, Qiu X, Zheng X, Cai Q, Xu Y, Chen Y, Fan Y, Chen W, Yu Z, Chen X, Zheng C, Wang M, Poon WS, Luo X. 2022. Oral, nasal, and gut microbiota in parkinson’s disease. Neuroscience 480:65–78. doi: 10.1016/j.neuroscience.2021.10.011 [DOI] [PubMed] [Google Scholar]
  • 33. Su Y, Shadike Q, Wang M, Jiang H, Liu W, Liu J, Tuerdi R, Zhou W, Li L. 2021. A low abundance of genus Bacteroides in gut microbiota is negatively correlated with blood phenylalanine levels in Uygur patients with phenylketonuria. Transl Pediatr 10:2521–2532. doi: 10.21037/tp-21-426 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34. Liu A, Zhou W, Qu L, He F, Wang H, Wang Y, Cai C, Li X, Zhou W, Wang M. 2019. Altered urinary amino acids in children with autism spectrum disorders. Front Cell Neurosci 13:7. doi: 10.3389/fncel.2019.00007 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35. Zhang P, Wang Z, Qiu H, Zhou W, Wang M, Cheng G. 2021. Machine learning applied to serum and cerebrospinal fluid metabolomes revealed altered arginine metabolism in neonatal sepsis with meningoencephalitis. Comput Struct Biotechnol J 19:3284–3292. doi: 10.1016/j.csbj.2021.05.024 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36. Zhou S, Xu R, He F, Zhou J, Wang Y, Zhou J, Wang M, Zhou W. 2016. Diversity of gut microbiota metabolic pathways in 10 pairs of chinese infant twins. PLoS One 11:e0161627. doi: 10.1371/journal.pone.0161627 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37. Wang M, Zhou J, He F, Cai C, Wang H, Wang Y, Lin Y, Rong H, Cheng G, Xu R, Zhou W. 2019. Alteration of gut microbiota-associated epitopes in children with autism spectrum disorders. Brain Behav Immun 75:192–199. doi: 10.1016/j.bbi.2018.10.006 [DOI] [PubMed] [Google Scholar]
  • 38. Yan G, Liu J, Chen W, Chen Y, Cheng Y, Tao J, Cai X, Zhou Y, Wang Y, Wang M, Lu G. 2021. Metagenomic next-generation sequencing of bloodstream microbial cell-free nucleic acid in children with suspected sepsis in pediatric intensive care unit. Front Cell Infect Microbiol 11:665226. doi: 10.3389/fcimb.2021.665226 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39. Francis OE, Bendall M, Manimaran S, Hong C, Clement NL, Castro-Nallar E, Snell Q, Schaalje GB, Clement MJ, Crandall KA, Johnson WE. 2013. Pathoscope: species identification and strain attribution with unassembled sequencing data. Genome Res 23:1721–1729. doi: 10.1101/gr.150151.112 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40. Hong C, Manimaran S, Shen Y, Perez-Rogers JF, Byrd AL, Castro-Nallar E, Crandall KA, Johnson WE. 2014. PathoScope 2.0: a complete computational framework for strain identification in environmental or clinical sequencing samples. Microbiome 2:33. doi: 10.1186/2049-2618-2-33 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41. Zhao R, Feng J, Liu J, Fu W, Li X, Li B. 2019. Deciphering of microbial community and antibiotic resistance genes in activated sludge reactors under high selective pressure of different antibiotics. Water Res 151:388–402. doi: 10.1016/j.watres.2018.12.034 [DOI] [PubMed] [Google Scholar]
  • 42. Chen W, Zhang P, Zhang X, Xiao T, Zeng J, Guo K, Qiu H, Cheng G, Wang Z, Zhou W, Zeng S, Wang M. 2024. Machine learning-causal inference based on multi-omics data reveals the association of altered gut bacteria and bile acid metabolism with neonatal jaundice. Gut Microbes 16:2388805. doi: 10.1080/19490976.2024.2388805 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43. Zeng S, Wang Z, Zhang P, Yin Z, Huang X, Tang X, Shi L, Guo K, Liu T, Wang M, Qiu H. 2022. Machine learning approach identifies meconium metabolites as potential biomarkers of neonatal hyperbilirubinemia. Comput Struct Biotechnol J 20:1778–1784. doi: 10.1016/j.csbj.2022.03.039 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44. Wang T, Wang M, Liu L, Xie F, Wu X, Li L, Ji J, Wu D. 2023. Lower serum branched-chain amino acid catabolic intermediates are predictive signatures specific to patients with diabetic foot. Nutr Res 119:33–42. doi: 10.1016/j.nutres.2023.08.009 [DOI] [PubMed] [Google Scholar]
  • 45. Zheng L, Jiao Y, Zhong H, Tan Y, Yin Y, Liu Y, Liu D, Wu M, Wang G, Huang J, Wang P, Qin M, Wang M, Xiao Y, Lv T, Luo Y, Hu H, Hou ST, Kui L. 2024. Human-derived fecal microbiota transplantation alleviates social deficits of the BTBR mouse model of autism through a potential mechanism involving vitamin B6 metabolism. mSystems 9:e00257-24. doi: 10.1128/msystems.00257-24 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46. Cortés-Martín A, Buttimer C, Maier JL, Tobin CA, Draper LA, Ross RP, Kleiner M, Hill C, Shkoporov AN. 2025. Adaptations in gut Bacteroidales facilitate stable co-existence with their lytic bacteriophages. Gut Microbes 17:2507775. doi: 10.1080/19490976.2025.2507775 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 47. Fernández L, Rodríguez A, García P. 2018. Phage or foe: an insight into the impact of viral predation on microbial communities. ISME J 12:1171–1179. doi: 10.1038/s41396-018-0049-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 48. Lin X, Yu Z, Liu Y, Li C, Hu H, Hu JC, Liu M, Yang Q, Gu P, Li J, Nandakumar KS, Hu G, Zhang Q, Chen X, Ma H, Huang W, Wang G, Wang Y, Huang L, Wu W, Liu NN, Zhang C, Liu X, Zheng L, Chen P. 2025. Gut-X axis. Imeta 4:e270. doi: 10.1002/imt2.270 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 49. Zhao F, Wang J. 2024. Another piece of puzzle for the human microbiome: the gut virome under dietary modulation. J Genet Genomics 51:983–996. doi: 10.1016/j.jgg.2024.04.013 [DOI] [PubMed] [Google Scholar]
  • 50. Tobón-Cornejo S, Sanchez-Tapia M, Guizar-Heredia R, Velázquez Villegas L, Noriega LG, Furuzawa-Carballeda J, Hernández-Pando R, Vázquez-Manjarrez N, Granados-Portillo O, López-Barradas A, Rebollar-Vega R, Maya O, Miller AW, Serralde A, Guevara-Cruz M, Torres N, Tovar AR. 2025. Increased dietary protein stimulates amino acid catabolism via the gut microbiota and secondary bile acid production. Gut Microbes 17:2465896. doi: 10.1080/19490976.2025.2465896 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 51. Liu Y, Zhong W, Li X, Shen F, Ma X, Yang Q, Hong S, Sun Y. 2023. Diets, gut microbiota and metabolites. Phenomics 3:268–284. doi: 10.1007/s43657-023-00095-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 52. Choi J, Choi YR, Jeong MK, Song HH, Yu JS, Song SH, Park JH, Kim MJ, Park H, Ham YL, Han SH, Kim DJ, Lee DY, Suk KT. 2025. Phocaeicola dorei ameliorates progression of steatotic liver disease by regulating bile acid, lipid, inflammation and proliferation. Gut Microbes 17:2539448. doi: 10.1080/19490976.2025.2539448 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 53. Blackard KR, Krahn KN, Andris RT, Lake DE, Fairchild KD. 2021. Autism risk in neonatal intensive care unit patients associated with novel heart rate patterns. Pediatr Res 90:1186–1192. doi: 10.1038/s41390-021-01381-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 54. Schneider KM, Candels LS, Hov JR, Myllys M, Hassan R, Schneider CV, Wahlström A, Mohs A, Zühlke S, Liao L, et al. 2021. Gut microbiota depletion exacerbates cholestatic liver injury via loss of FXR signalling. Nat Metab 3:1228–1241. doi: 10.1038/s42255-021-00452-1 [DOI] [PubMed] [Google Scholar]
  • 55. Mertens KL, Kalsbeek A, Soeters MR, Eggink HM. 2017. Bile acid signaling pathways from the enterohepatic circulation to the central nervous system. Front Neurosci 11:617. doi: 10.3389/fnins.2017.00617 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 56. Needham BD, Kaddurah-Daouk R, Mazmanian SK. 2020. Gut microbial molecules in behavioural and neurodegenerative conditions. Nat Rev Neurosci 21:717–731. doi: 10.1038/s41583-020-00381-0 [DOI] [PubMed] [Google Scholar]
  • 57. Mao Z, Hui H, Zhao X, Xu L, Qi Y, Yin L, Qu L, Han L, Peng J. 2023. Protective effects of dioscin against Parkinson’s disease via regulating bile acid metabolism through remodeling gut microbiome/GLP-1 signaling. J Pharm Anal 13:1153–1167. doi: 10.1016/j.jpha.2023.06.007 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 58. Long-Smith C, O’Riordan KJ, Clarke G, Stanton C, Dinan TG, Cryan JF. 2020. Microbiota-gut-brain axis: new therapeutic opportunities. Annu Rev Pharmacol Toxicol 60:477–502. doi: 10.1146/annurev-pharmtox-010919-023628 [DOI] [PubMed] [Google Scholar]
  • 59. Kugelberg E. 2016. Neuroimmunology: IL-17A mediates a path to autism. Nat Rev Immunol 16:205. doi: 10.1038/nri.2016.35 [DOI] [PubMed] [Google Scholar]
  • 60. Ellul P, Rosenzwajg M, Peyre H, Fourcade G, Mariotti-Ferrandiz E, Trebossen V, Klatzmann D, Delorme R. 2021. Regulatory T lymphocytes/Th17 lymphocytes imbalance in autism spectrum disorders: evidence from a meta-analysis. Mol Autism 12:68. doi: 10.1186/s13229-021-00472-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 61. Choi GB, Yim YS, Wong H, Kim S, Kim H, Kim SV, Hoeffer CA, Littman DR, Huh JR. 2016. The maternal interleukin-17a pathway in mice promotes autism-like phenotypes in offspring. Science 351:933–939. doi: 10.1126/science.aad0314 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 62. Biagioli M, Marchianò S, Carino A, Di Giorgio C, Santucci L, Distrutti E, Fiorucci S. 2021. Bile Acids activated receptors in inflammatory bowel disease. Cells 10:1281. doi: 10.3390/cells10061281 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 63. Ma Z, Wang Z, Cao J, Dong Y, Chen Y. 2025. Regulatory roles of intestinal CD4+ T cells in inflammation and their modulation by the intestinal microbiota. Gut Microbes 17:2560019. doi: 10.1080/19490976.2025.2560019 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 64. Hang S, Paik D, Yao L, Kim E, Trinath J, Lu J, Ha S, Nelson BN, Kelly SP, Wu L, Zheng Y, Longman RS, Rastinejad F, Devlin AS, Krout MR, Fischbach MA, Littman DR, Huh JR. 2019. Bile acid metabolites control TH17 and Treg cell differentiation. Nature 576:143–148. doi: 10.1038/s41586-019-1785-z [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 65. Gregory AC, Zablocki O, Zayed AA, Howell A, Bolduc B, Sullivan MB. 2020. The gut virome database reveals age-dependent patterns of virome diversity in the human gut. Cell Host Microbe 28:724–740. doi: 10.1016/j.chom.2020.08.003 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 66. Nayfach S, Páez-Espino D, Call L, Low SJ, Sberro H, Ivanova NN, Proal AD, Fischbach MA, Bhatt AS, Hugenholtz P, Kyrpides NC. 2021. Metagenomic compendium of 189,680 DNA viruses from the human gut microbiome. Nat Microbiol 6:960–970. doi: 10.1038/s41564-021-00928-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 67. Tobin CA, Hill C, Shkoporov AN. 2023. Factors affecting variation of the human gut phageome. Annu Rev Microbiol 77:363–379. doi: 10.1146/annurev-micro-032421-105754 [DOI] [PubMed] [Google Scholar]
  • 68. Shkoporov AN, Turkington CJ, Hill C. 2022. Mutualistic interplay between bacteriophages and bacteria in the human gut. Nat Rev Microbiol 20:737–749. doi: 10.1038/s41579-022-00755-4 [DOI] [PubMed] [Google Scholar]
  • 69. Feldman AG, Sokol RJ. 2019. Neonatal cholestasis: emerging molecular diagnostics and potential novel therapeutics. Nat Rev Gastroenterol Hepatol 16:346–360. doi: 10.1038/s41575-019-0132-z [DOI] [PubMed] [Google Scholar]
  • 70. Sinha A, Maurice CF. 2019. Bacteriophages: uncharacterized and dynamic regulators of the immune system. Mediators Inflamm 2019:3730519. doi: 10.1155/2019/3730519 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 71. Johansen J, Atarashi K, Arai Y, Hirose N, Sørensen SJ, Vatanen T, Knip M, Honda K, Xavier RJ, Rasmussen S, Plichta DR. 2023. Centenarians have a diverse gut virome with the potential to modulate metabolism and promote healthy lifespan. Nat Microbiol 8:1064–1078. doi: 10.1038/s41564-023-01370-6 [DOI] [PubMed] [Google Scholar]
  • 72. Li XL, Megdadi M, Quadri HS. 2025. Interaction between gut virome and microbiota on inflammatory bowel disease. World J Methodol 15:100332. doi: 10.5662/wjm.v15.i3.100332 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 73. Ullah K, Kanwar R, Ali S, Uddin S, Izza I, Ahmad I, Alzahrani KJ, Qadeer A, Chu CT, Chen CC. 2025. Gut virome and aging: phage-driven microbial stability and immune modulation. Aging Dis. doi: 10.14336/AD.2025.0854 [DOI] [PMC free article] [PubMed]
  • 74. Varesio C, Grumi S, Zanaboni MP, Mensi MM, Chiappedi M, Pasca L, Ferraris C, Tagliabue A, Borgatti R, De Giorgis V. 2021. Ketogenic dietary therapies in patients with autism spectrum disorder: facts or fads? A scoping review and a proposal for a shared protocol. Nutrients 13:2057. doi: 10.3390/nu13062057 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 75. Bermudez-Martin P, Becker JAJ, Caramello N, Fernandez SP, Costa-Campos R, Canaguier J, Barbosa S, Martinez-Gili L, Myridakis A, Dumas ME, Bruneau A, Cherbuy C, Langella P, Callebert J, Launay JM, Chabry J, Barik J, Le Merrer J, Glaichenhaus N, Davidovic L. 2021. The microbial metabolite p-Cresol induces autistic-like behaviors in mice by remodeling the gut microbiota. Microbiome 9:157. doi: 10.1186/s40168-021-01103-z [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 76. Kong X-J, Liu J, Liu K, Koh M, Sherman H, Liu S, Tian R, Sukijthamapan P, Wang J, Fong M, Xu L, Clairmont C, Jeong M-S, Li A, Lopes M, Hagan V, Dutton T, Chan S-T, Lee H, Kendall A, Kwong K, Song Y. 2021. Probiotic and oxytocin combination therapy in patients with autism spectrum disorder: a randomized, double-blinded, placebo-controlled pilot trial. Nutrients 13:1552. doi: 10.3390/nu13051552 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Reviewer comments
reviewer-comments.pdf (361.5KB, pdf)
Supplemental Material. msystems.01405-25-s0001.docx.

Supplemental figures and methods.

msystems.01405-25-s0001.docx (528.3KB, docx)
DOI: 10.1128/msystems.01405-25.SuF1
Supplemental Tables. msystems.01405-25-s0002.xlsx.

Tables S1 to S8.

msystems.01405-25-s0002.xlsx (58.4KB, xlsx)
DOI: 10.1128/msystems.01405-25.SuF2
OPEN PEER REVIEW. reviewer-comments.pdf.

An accounting of the reviewer comments and feedback.

reviewer-comments.pdf (361.5KB, pdf)
DOI: 10.1128/msystems.01405-25.SuF3

Data Availability Statement

The data of the NJ cohort 1 project have been uploaded to CNSA with project ID CNP0003066. The data of the ASD cohort and NJ cohort 2 projects have been uploaded to NCBI and can be downloaded according to the BioProject IDs, namely, PRJEB23052 and PRJEB23053.

Articles from mSystems are provided here courtesy of American Society for Microbiology (ASM)

RESOURCES