Baseline characteristics of the participants
Forty-five full-term neonates diagnosed with CCHD and 50 HCs matched by age and gender were enrolled in this study. The enrollment flowchart and cohort characteristics are shown in Additional file 1: Figure S1 and Table S1, respectively. In brief, no significant differences were observed in birth weight, delivery mode, and breastfeeding status between the two groups.
Differences in gut bacteria between CCHD patients and HCs
We first investigated the gut bacterial composition and diversity between the two groups (CCHD and HD groups) and observed a significant increase in α-diversity in the CCHD group (Fig. 1A). In addition, principal coordinates analysis (PCoA) of bacterial composition also suggested a significant separation between the two groups, which was largely driven by a subset of bacterial species, including Bifidobacterium ramosum, Bifidobacterium biavatii, Lactobacillus delbrueckii, Bacteroides caccae, Enterococcus asini, Pseudomonas oleovorans, and Lachnoclostridium sp.An14 (Fig. 1B). To further delineate the differences in bacterial configurations between groups, we performed enterotype analysis using unsupervised clustering at the genus level. Intriguingly, two clusters driven by a relatively high abundance of the genera Enterococcus (enterotype 1) and Bifidobacterium (enterotype 2) were identified (Fig. 1C–E). The Bifidobacterium enterotype was predominant in the HC group and displayed a convergent microbial community enriched with multiple Bifidobacterium species, whereas the CCHD group was dominated with the Enterococcus enterotype and showed a relatively discrete microbial community enriched with Enterococcus, Klebsiella, and Streptococcus. Next, we identified 58 discriminative bacterial species between the CCHD and HC groups (Fig. 1F). Compared to that in HCs, CCHD patients were characterized by 34 enriched species mainly belonging to the genera Enterococcus (6 species), Enterobacter (5 species), and Clostridium (four species), and by 24 depleted species mainly belonging to the genera Bifidobacterium (9 species), Lactobacillus (4 species), and Veillonella (4 species). Then, a species co-abundance network was constructed to provide an overview of the interplay among these discriminative bacterial species. As shown in Fig. 1G, bacterial species belonging to the same genus clade were closely correlated with each other. For instance, several Bifidobacterium species enriched in HC were closely correlated to generate a covarying cluster. Notably, we observed that species belonging to the genera Bifidobacterium, Lactobacillus, and Veillonella were positively correlated and constituted a symbiotic bacterial community in HC, whereas the species corresponding to the genera Enterococcus, Enterobacter, and Clostridium were discretely enriched in the CCHD group.
Temperate core virome is implicated in early life bacterial perturbations in CCHD
The gut virome is an essential component of the human gut microbiome; however, the ecological interaction between gut viral and bacterial communities remains poorly understood. Here, we characterized the gut virome in both CCHD and HC groups, and investigated its interaction with gut bacteria. We first explored the richness of the gut virome and found a relative increase in α-diversity in the CCHD group (Additional file 1: Figure S2A). As expected, the PCoA of viral composition at the family level also revealed a distinct separation between the two groups, mainly driven by Siphoviridae, Myoviridae, and Herelleviridae (Additional file 1: Figure S2B). After filtering out the low-abundance taxa, 37 viral species responsible for discriminating the two groups were identified, of which 25 species were enriched in CCHD relative to that in HC (Additional file 1: Figure S2C). Notably, Enterococcus and Escherichia phages are the predominant bacteriophages among CCHD-enriched viruses, which is corresponding to the overabundance of Enterococcus and Enterobacter in CCHD group. To initially assess the ecological interaction, we compared bacterial richness with viral richness and found a strong positive correlation in the CCHD group (Additional file 1: Figure S2D). Additionally, positive correlations in community richness between Enterococcus phages and their bacterial host (i.e., genus Enterococcus) were also confirmed in both groups, with the relative abundance consistently higher in CCHD (Additional file 1: Figure S2EF).
To generate an integrated view of the cross-kingdom interaction between gut virome and bacterial community during the earliest stage of life, we further investigated the replication mode of the predicted phages and conducted prophage- and clustered regularly interspaced short palindromic repeat (CRISPR)-based bacteria–phage association analyses. Notably, temperate phages were the predominant viruses in both the CCHD and HC groups, accounting for more than 70% of the viral sequences (Fig. 2A). Additionally, the α-diversity of temperate phages was higher in the CCHD group compared to HC group (Fig. 2B). Next, we analyzed prophage-based host bacteria–phage associations in samples from both the groups. In total, 935 and 694 prophages were successfully identified in the CCHD and HC groups, respectively (Fig. 2C), corresponding to a higher viral richness in CCHD (Fig. 2B). By classifying the detected prophages according to their host bacteria, we observed a similar taxonomic distribution pattern at the phylum level between the two groups (Fig. 2C). Among the four representative intestinal bacterial phyla—Firmicutes, Actinobacteria, Proteobacteria, and Bacteroidetes—the number of identified prophages was highest in Firmicutes, accounting for more than 58% of all prophages in both groups. Nevertheless, compared to those in HCs, the proportions of prophages integrated into the genomes of Proteobacteria and Bacteroidetes were significantly elevated in CCHD, whereas the proportion of Actinobacteria with prophages declined (P < 0.001, chi-square test, Fig. 2C). We then examined the viral taxonomy of the detected prophages based on their bacterial phyla (Fig. 2D). Generally, phages classified as Siphoviridae show major interactions with bacteria. For instance, 51.47% of the Firmicutes-derived prophage sequences in the CCHD group were identified as belonging to Siphoviridae (Fig. 2D). Notably, Siphoviridae has generally been identified as a temperate phage with an inherent ability to mediate the transfer of genes between bacteria and co-evolve with its host [38]. To determine whether there is a host bacteria–temperate phages co-evolution relationship mediating the overgrowth of Enterococcus in neonates with CCHD, we extracted Enterococcus-derived prophage sequences classified as Siphoviridae in the CCHD group and annotated the gene function (mainly the open reading frames, ORFs) using the Pfam database. Intriguingly, 56% of the annotated ORFs were classified as transposase, phage integrase, and enzymatic genes associated with transcriptional regulation and catabolism (including transcriptional regulators, methylases, hydrolases, peptidases, and glycosidases; Fig. 2E), indicating that temperate phages targeting Enterococcus can extensively affect the genetic makeup and metabolic traits of their bacterial hosts. Furthermore, the same prophage sequences were aligned to virulence factor database (VFDB) and comprehensive antibiotic resistance database (CARD), with the purpose to investigate whether there is a complex genetic repertoire of virulence factors and antibiotic resistance genes (ARGs) derived from temperate phages affecting the bacterial hosts’ behavior and fitness. Notably, 47.56% of the predicted ORFs were classified as virulence factor genes, and the functional genes encoding offensive virulence factors associated with adherence, toxin, and secretion system were the most frequent category (Additional file 1: Figure S3A). In addition, we identified an extensive repertoire of ARGs that may confer resistance to up to 29 types of antibiotics (Additional file 1: Figure S3B, C). The most diverse ARG type was identified as antibiotic efflux pump genes, which account for 58.43% of all annotated ARGs and demonstrate resistance to multiple antibiotics including macrolide, fluoroquinolone, tetracycline, and aminoglycoside. Besides, the dominant ARG types against distinct drug classes included macrolide (relative abundance 15.40%), tetracycline (11.59%), fluoroquinolone (10.26%), peptide antibiotic (7.37%), and penam resistance genes (5.30%). Lastly, we selected a host bacterium–prophage pair to decipher the cross-kingdom relationship (Fig. 2F). The viral sequence classified as Enterococcus phage vB_EfaS_IME197 (39,017-nt-long) was identified as a prophage sequence (99.6% identity) in a bacterial sequence classified as Enterococcus faecalis (83,776-nt-long). In addition to the phage structural protein genes, the presence of functional genes encoding virulence factors associated with adherence and toxin, enzymes associated with catabolism, and antibiotic efflux pump was confirmed in the prophage sequence.
CRISPR spacers are DNA loci that lie in the bacterial genome and act as a defensive system against phages; therefore, they can be used as a fingerprint to investigate infectious associations between gut bacteria and phages [28, 29]. Initially, we screened CRISPR spacers on the bacterial sequence using CRISPRDetect [36]. Overall, 2747 and 3072 CRISPR spacers were detected in the CCHD and HC groups, respectively, with most spacers derived from Bacilli, Actinobacteria, and Gamma-proteobacteria (Fig. 2G). Notably, compared to that in HCs, the proportion of spacers detected in Bacilli increased significantly in the CCHD group, whereas the proportion of spacers derived from Actinobacteria decreased (P < 0.001, chi-square test), suggesting a shift in the infectious relationship between the two groups (Fig. 2G). Next, we examined the distribution of CRISPR spacers in both the groups and the number of viral families targeted by the CRISPR spacers (Fig. 2H). Generally, almost all spacers were aligned to viral sequences classified as Siphoviridae. Therefore, we determined which bacterial phyla carry CRISPR spacers specific for Siphoviridae. CRISPR spacers against Siphoviridae were frequently detected for the genera Streptococcus and Rothia in both the groups (Fig. 2I), implying that Siphoviridae might preferentially infect bacteria belonging to Firmicutes and Actinobacteria. To systematically elucidate the infectious relationships, we constructed a network model of phage–host pairs by integrating the associations inferred from CRISPR spacers (Fig. 2J). Multiple scenarios of infectious relationships were observed, wherein certain bacterial species could be infected by different phages (or vice versa). Notably, positive relationships were observed between Enterococcus faecalis and several Enterococcus phages enriched in CCHD, further implying a co-evolutionary relationship between Enterococcus phages and their bacterial hosts.
Collectively, these findings suggest that a temperate core virome represented by Siphoviridae can act as an intrinsic force mediating the infectious kinetics and co-evolutionary relationship between gut bacteria and phages, and further implicates early life bacterial perturbations in CCHD by modifying microbial adaptation with a complex repertoire of functional genes.
Systemic interactions between gut microbiome and fecal metabolome
Fecal metabolomics analysis can provide a functional readout of the gut microbiome. Here metabolomic profiling of paired fecal samples was conducted to characterize the metabolic signatures of both the groups. A total of 748 metabolites were identified by liquid chromatograph-mass spectrometry (LC–MS)–based untargeted metabolomics analysis. PCoA of metabolite distribution demonstrated that the overall metabolic signatures between the two groups were significantly different (Additional file 1: Figure S4A). By leveraging the orthogonal partial least-squares discriminant analysis (OPLS-DA), 120 discriminative metabolites between CCHD and HC groups were identified (Additional file 1: Figure S4B). Specifically, the CCHD group displayed enrichment of 36 metabolites and depletion of 84 metabolites compared to those in the HC group. Moreover, pathway-based differential abundance analysis highlighted that the metabolic pathways of thiamine metabolism, linoleic acid (LA) metabolism, biosynthesis of unsaturated fatty acids, galactose metabolism, phenylalanine metabolism, and tyrosine metabolism were downregulated in the CCHD group, whereas the metabolic pathways of arachidonic acid (AA) metabolism and primary bile acid biosynthesis were upregulated (Additional file 1: Figure S4C).
To explore the potential relevance between bacterial composition and metabolomic phenotypes, we calculated Spearman’s correlation matrices and constructed a co-occurrence network of differential bacterial species and metabolites (Additional file 1: Figure S5). To visualize strong associations between bacteria and metabolites, only results showing Spearman’s correlation coefficients > + 0.6 or < − 0.6 with a significance (i.e., P < 0.05) were plotted. Consistent with our initial findings, the differential bacterial species mainly generated four covarying clusters corresponding to their genus annotations (Bifidobacterium, Lactobacillus, Enterococcus, and Enterobacter), and the differential metabolites were clustered according to the metabolic pathways they were involved in. Interestingly, most bacterium–metabolite associations converged to the Bifidobacterium, Lactobacillus, and Enterococcus clusters. Strong positive correlations were observed between the genus Bifidobacterium and fecal metabolites belonging to amino acid and carbohydrate metabolism through some node species (Bifidobacterium longum CAG:69, Bifidobacterium bifidum, and Bifidobacterium catenulatum) and metabolites (l-arginine, N-acetyl-l-glutamate, l-fucose, and rhamnose). Notably, fecal levels of aromatic lactic acids (including hydroxyphenyllactic acid and indolelactic acid [IAA]), newly recognized probiotic-associated metabolites derived from phenylalanine and tryptophan metabolism, were positively correlated with the abundance of Lactobacillus paragasseri and most Bifidobacterium species, while consistently displaying negative correlations with Enterococcus gallinarum.
Integrated network analysis of microbial abundance, functionality, and genomic structural variations with host metabolism
In addition to microbial abundance, the investigation of microbial functionality and genetic variations can provide an extra layer of information that facilitates mechanistic insights into the role of the gut microbiome in host metabolism. To this end, we first functionally profiled gene families in all metagenomes and identified 6103 Kyoto Encyclopedia of Genes and Genomes (KEGG) orthologs (KOs). Based on the PCoA of KEGG pathways, the overall microbial functionality between the two groups was significantly different (Additional file 1: Figure S6A). In addition, 840 differential KOs were identified, most of which were involved in metabolic pathways, especially amino acid metabolism, carbohydrate metabolism, and the metabolism of cofactors and vitamins (Additional file 1: Figure S6B). Linear discriminant analysis (LDA) showed that 33 metabolic pathways were differentially depleted in the CCHD group, most of which belonged to amino acid metabolism (aromatic amino acids (AAAs), branched chain amino acids, lysine, and arginine biosynthesis) and vitamin metabolism (riboflavin, vitamin B6, and pantothenate metabolism), whereas the other 29 metabolic pathways enriched in the CCHD group mainly belonged to carbohydrate and lipid metabolism, such as glycerolipid metabolism and lipopolysaccharide biosynthesis (Additional file 1: Figure S6C).
Next, we systematically detected microbial structural variations (SVs), which are highly variable segments of bacterial genomes deleted from certain species (deletion SVs, dSVs) or present in a variable number of copies (variable SVs, vSVs) in others. We identified 428 SVs from 15 bacterial species present in at least 5 samples in both the CCHD and HC groups (with an average size of 14 kbp per SV), including 157 vSVs and 271 dSVs, at a rate of 1–81 SVs per species (Additional file 1: Figure S7A, B). We further assessed the SV variability among all samples. Interestingly, we found that the genetic variability of SVs differed substantially across species, with Escherichia coli and Staphylococcus hominis SK119 displaying the greatest inter-group genetic variability, whereas Bifidobacterium animalis, Bifidobacterium bifidum PRL2010, and Bifidobacterium longum showed relatively low inter-group genetic variability (Additional file 1: Figure S7C, D).
Then, we explored the associations between multilevel microbial features (bacterial abundance, functionality, and SVs) and fecal metabolites. These investigation identified 1776 significant associations between 142 microbial features and 56 metabolites, including 1161 associations with species and pathway abundance and 615 associations with microbial SVs (Fig. 3A, B). To simplify the intricate correlation network of bacteria and metabolites, we deconstructed and clustered the associations according to metabolite categories and mainly focused on associations with AA metabolites [39, 40], polyunsaturated fatty acids (PUFAs) [41, 42], human milk oligosaccharides (HMOs) [43, 44], AAA metabolites [45], and short-chain fatty acids (SCFAs) [46], all of which are well-known gut microbiome-related metabolites implicated in host health.
Notably, > 60% of the significant associations between metabolites and bacterial abundance were attributed to the genera Bifidobacterium and Enterococcus. By plotting the associations between the abundance of Bifidobacterium, Enterococcus, and the metabolites of interest, we observed a “mutually exclusive” characteristic of associations with Bifidobacterium and Enterococcus (Fig. 3C). Specifically, most Enterococcus species were consistently positively correlated with AA metabolite levels (20-hydroxy-leukotriene B4, leukotriene F4, lipoxin A4, and lipoxin B4), whereas inverse associations were concomitantly found between Enterococcus and PUFA levels (eicosadienoic acid, docosahexaenoic acid, and LA; Fig. 3C). Notably, all these AA metabolites are active compounds involved in the inflammatory cascades [39], whereas the ω-3 and ω-6 PUFAs are capable of alleviating inflammation and oxidative stress [47]. Intriguingly, a significant positive correlation was also identified between the abundance of Enterococcus faecium and 3-nitrotyrosine (Fig. 3D), which is an oxidative product of tyrosine and a biomarker of inflammation and oxidative stress [48].
We then looked at the correlations between microbial SVs and fecal metabolites (Fig. 3A, B). Notably, 29.9% (184 out of 615) of the associations between microbial SVs and metabolites were related to the dSVs of Enterococcus faecium NRRL B.2354, a genetically unstable species identified in our initial findings (Additional file 1: Figure S6C). Although many SVs harbor genes with unknown functions, we observed several intriguing functions in metabolite-associated SVs. For instance, a 2-kbp vSV in Escherichia coli that contains genes encoding invasion proteins was positively correlated with the abundance of leukotriene F4 (Fig. 3E), suggesting that genetic variation in certain functional genes may be implicated in microbial adaptation and the subsequent activation of the host inflammatory response.
Next, we focused on the correlations between differential microbial features and HMOs. Impressively, negative correlations were observed between the abundance of 2-fucosyllactose (2′-FL) and most Bifidobacterium species, as well as a vSV of two segments in Bifidobacterium bifidum PRL2010 (Fig. 3F). Moreover, the abundance of Bifidobacterium longum was strongly positively correlated with l-fucose (Fig. 3D), which is a microbial catabolite of 2′-FL with immunoregulatory activity. Fucosyllactose-utilization genes are suggested to imprint the early life development of the gut microbiome in infants [49]. This prompted us to perform an in-depth analysis of HMO-utilization genes in fecal metagenomes. Intriguingly, a total of 37 HMO-utilization genes were identified and assigned to five clusters, which were all over-represented in the HC group relative to that in the CCHD group (Additional file 1: Figure S8A), and most of them were inversely correlated with 2′-FL abundance (Additional file 1: Figure S8B). By contrast, the overgrowth of Enterococcus, the depletion of HMO-utilization genes in CCHD, and an 8-kbp dSV that contains genes encoding beta-galactosidase family proteins in Enterococcus faecalis ATCC29212 were positively correlated with 2′-FL abundance (Fig. 3C, G, Additional file 1: Figure S8), indicating the incapability of HMO-utilization.
Another interesting category of metabolites is the aromatic lactic acids, and we observed that the abundance of most Bifidobacterium species was positively correlated with these metabolites (Fig. 3C). Furthermore, the metabolic pathway of phenylalanine, tyrosine, and tryptophan biosynthesis also demonstrated a remarkable positive correlation with IAA (Additional file 1: Figure S9). As a noteworthy example, a 21-kbp dSV in Bifidobacterium longum, which contains genes coding for the enzyme AAA transporter, was positively associated with hydroxyphenyllactic acid abundance (the median level was higher for retention, adjusted P < 0.01, n = 48, 40 retaining; Fig. 3G).
Collectively, these findings indicate that the aberrant gut microbial composition in neonates with CCHD, characterized by the overgrowth of Enterococcus and depletion of Bifidobacterium, together with the alterations in microbial functionality and genetic makeup, considerably impact early life immune development and metabolism. Moreover, microbial metabolites are important agents involved in host–microbe interactions. An in-depth analysis of microbial genetics provides mechanistic insights into the metabolomic perturbations.
Overgrowth of Enterococcus was related to inflammatory response and poor prognosis
To further investigate the role of the gut microbiome in clinical outcomes of neonates with CCHD, we first separated our patient cohort into two subgroups based on surgical prognosis and conducted a comparative analysis. Specifically, composite adverse events were used to define the poor prognosis (Additional file 1: Table S2). Nineteen patients who had one or more adverse event/s were classified as having a poor prognosis. The perioperative clinical metadata are summarized in Additional file 1: Table S3. In brief, no significant differences were observed in the demographic data and preoperative and intraoperative variables, including median age at operation, blood oxygen saturation (SpO2), cardiac surgery complexity (RACHS-1 and ABC categories), cardiopulmonary bypass (CPB) time, and intraoperative blood loss.
Next, we examined the gut microbial composition between HCs and patients with good and poor surgical prognosis (termed CCHD-G and CCHD-P respectively). Intriguingly, pairwise comparative analyses revealed significant differences in gut microbial configurations between the three groups (Fig. 4A, Figure S10A–D), with LDA indicating that Enterococcus species were enriched in both CCHD-P and CCHD-G when compared to HC (Additional file 1: Figure S10E, F). More notably, a total of 19 discriminative bacterial species were identified between CCHD-P and CCHD-G, with 14 species enriched in CCHD-P mainly belonging to genus Enterococcus (eight species; Fig. 4B), implying that the overgrowth of Enterococcus is a crucial microbial feature that drives the separation between CCHD-P and CCHD-G. To comprehensively characterize the microbial features between the two subgroups, we constructed a co-abundance network based on the differential taxa and observed a synergistic microbial consortium of Enterococcus harbored in CCHD-P (Fig. 4C). Accordingly, the overall microbial functionality between the two subgroups was significantly different (Fig. 4D). By leveraging LDA, 29 metabolic pathways were found to be downregulated in CCHD-P, mainly associated with amino acid metabolism (eight pathways) and metabolism of cofactors and vitamins (five pathways), whereas the other 17 metabolic pathways upregulated in CCHD-P mainly belonged to carbohydrate and lipid metabolism (six pathways, Fig. 4E).
Interestingly, these findings were in accord with the initial evidence we observed from the comparative analysis between CCHD patients and HCs, which further prompted us to hypothesize that gut microbiota dysbiosis in neonates with CCHD, characterized by the overgrowth of Enterococcus, is implicated in worsening surgical outcomes by mediating inflammatory responses and microbial metabolites. To verify this hypothesis, we first examined the inflammatory status of the patients. Blood samples collected on admission were used to quantify the serum levels of inflammatory cytokines (including interleukin-1β [IL-1β], interleukin-6 [IL-6], interleukin-8 [IL-8], tumor necrosis factor-α [TNF-α], and interferon-γ [INF-γ]) and biomarkers of intestinal permeability (including zonulin, D-lactate, intestinal fatty acid binding protein [iFABP], lipopolysaccharide [LPS], and lipopolysaccharide binding protein [LBP]) [13, 50, 51]. Notably, although at low titers, the serum levels of all these biomarkers were significantly increased in CCHD-P relative to those in CCHD-G (Additional file 1: Figure S11, S12), indicating that subclinical systemic inflammation and gut barrier impairment existed prior to cardiac surgery in neonates with poor prognosis. To investigate an aberrant gut microbiota-disrupted gut barrier–systemic inflammation axis, we employed an integrated correlation analysis of differential microbial features and serum biomarkers of inflammation and intestinal permeability. Notably, strong positive associations were observed between Enterococcus abundance and the serum levels of multiple proinflammatory cytokines and biomarkers of intestinal permeability (Fig. 4F). Furthermore, univariable and multivariable logistic regression analyses were performed to investigate whether there is predictive value of Enterococcus species in prognostic stratification for neonates with CCHD (independent of traditional clinical risk factors, e.g., SpO2, cardiac surgery complexity indexes, and CPB time, Additional file 1: Table S4). Intriguingly, a predictive model comprising of Enterococcus faecium abundance, CPB time, and intraoperative infusion volume was constructed and yielded an area under the curve (AUC) of 0.86 (95% CI 0.74–0.97) in the study cohort (Additional file 1: Figure S13), indicating that Enterococcus faecium abundance could be an independent predictor of surgical prognosis for neonates with CCHD.
Microbial metabolites linked Enterococcus to the immune–inflammatory imbalance
By profiling the fecal metabolites of both subgroups, we identified significant differences in the overall metabolic signatures between the two subgroups (Additional file 1: Figure S14A, B). Accordingly, pathway-based differential abundance analysis revealed that most metabolic pathways associated with amino acids, vitamins, and unsaturated fatty acids were downregulated in the CCHD-P group (Additional file 1: Figure S14C). To initially investigate the pivotal bacterial species that were highly associated with metabolomic alterations, we constructed a co-abundance network of differential bacteria and metabolites. Notably, the interplay between bacteria and metabolites was largely attributed to the negative associations between CCHD-P-enriched Enterococcus species and metabolites involved in lipid and vitamin metabolism (Additional file 1: Figure S15), indicating that antagonistic relationships between Enterococcus and lipid and vitamin metabolism could be central to the metabolomic perturbations in neonates with poor prognosis.
To further expand our understanding of the interactions between the aberrant gut microbiome and metabolomic alterations within neonates with different surgical outcomes, we performed an integrated correlation analysis of microbial features and fecal metabolites. Similarly, we focused on the associations between aromatic lactic acids, LA derivatives, SCFAs, B vitamins, and HMOs (Fig. 5A, B). As expected, we found that Enterococcus abundance and five vSVs in Enterococcus faecalis ATCC 29,212 were consistently negatively associated with aromatic lactic acid levels (Fig. 5C). Intriguingly, LA and its derivatives, including 13S-hydroxyoctadecadienoic acid (13-HODE) and α-linolenic acid, were another subset of metabolites that displayed accordant reverse correlations with Enterococcus abundance (Fig. 5C). Another noteworthy category of metabolites, B vitamins, especially pantothenic acid (vitamin B5) and pyridoxal (vitamin B6), were highly correlated with Enterococcus abundance. For instance, pyridoxal was inversely correlated with the abundance of most Enterococcus species and a 9-kbp dSV in Enterococcus faecium NRRL B.2354 (the median level was lower for retention, adjusted P < 0.01, n = 36, 13 retaining; Fig. 5C, D). This SV contains genes encoding the enzymes CoA synthetase, CoA transferase, and acyl-CoA, which might be involved in vitamin B6 metabolism. Pyridoxal is a critical coenzyme involved in the synthesis of amino acids and neurotransmitters (serotonin and norepinephrine), and its depletion is increasingly linked to the inflammatory response [52].
Collectively, these findings indicate that the overgrowth of Enterococcus together with genetic variations is highly associated with the depletion of probiotic-associated metabolites, especially aromatic lactic acids, LA derivatives, and B vitamins, thereby implicating an active inflammatory response.
Microbiome contributed to host inflammatory response and gut barrier impairment through metabolites
To further evaluate whether metabolites can mediate the microbial impact on the inflammatory response and gut barrier impairment in neonates with CCHD, we performed a bi-directional mediation analysis and revealed 23 mediation linkages (Pmediation < 0.05, Pinverse mediation > 0.05, Fig. 6A). Interestingly, most of these linkages were related to Enterococcus faecium (four linkages), Enterococcus columbae (three linkages), and the microbial functionality of glycine, serine, and threonine metabolism (three linkages). Our mediation analysis suggested that Enterococcus faecium might contribute to gut barrier impairment (characterized by increased serum levels of D-lactate and iFABP) by decreasing fecal levels of 13-HODE (26%, Pmediation < 0.05, Fig. 6B) and α-dimorphecolic acid (9(S)-HODE; 17%, Pmediation < 0.05, Fig. 6C), both of which are LA derivatives. Furthermore, the microbial fatty acid degradation pathway may also contribute to decreased intestinal permeability by affecting fecal AA level (14%, Pmediation < 0.05, Fig. 6D).
We also identified several metabolite mediation effects on the microbial impact on systemic inflammation. An interesting example here is Enterococcus columbae, a bacterium enriched in CCHD-P, which may contribute to the systemic inflammatory response by decreasing fecal levels of 9(S)-HODE (23%, Pmediation < 0.05, Fig. 6E) and dodecanoylcarnitine (25%, Pmediation < 0.05, Fig. 6F). By contrast, the microbial functionality of glycine, serine, and threonine metabolism probably contributes to alleviating the systemic inflammatory response through l-targinine (an arginine derivative; 18%, Pmediation < 0.05, Fig. 6G).
