Intestinal microbiome in very-preterm infants at one month of age and association with neurodevelopmental outcome

Abstract

Background

Preterm birth is the leading cause of death in children under five years of age worldwide. The association between preterm birth and long-term outcomes is vaguely known. In very preterm infants, the gut microbiome is highly variable and impacted by the neonatal intensive care unit environment. Our objective was to better understand the crosstalk between the gut microbiome and the host at one month of age in very preterm infants and its impact on neurological outcomes at two years of age. We performed a multi-omics analysis of fecal samples collected in 2011 from 73 very preterm French infants at one month of age, grouped according to their neurodevelopment assessed at two years of age using the Ages & Stages questionnaire. Multi-omics profiling and integrative analyses were performed between 2022 and 2023, including fecal microbiome, metabolome, and host transcriptome characterization using 16 S rRNA gene sequencing, LC–MS, and mRNA sequencing, respectively.

Results

The gut microbiome of very preterm infants at one month is mostly driven by either Escherichia or Staphylococcus, which are differentially associated with host immune markers (CAMP), metabolomic pathways, notably the energy pathway due to the presence of various nicotinamide adenine dinucleotides (NAD+) and two-year neurodevelopmental outcomes.

Conclusion

The gut microbiome at one month of age could be a noninvasive biomarker of gut immaturity and metabolic defects. Escherichia and Staphylococcus proportions were found to be the best indicators of physiological maturity and immaturity, respectively. Escherichia may help the process of intestinal maturation in preterm infants through specific metabolites production and is associated with a better neurodevelopment.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12866-026-04789-z.

Background

The gut microbiome is established during the first three years of life and evolves throughout the lifetime. In the early days of life, the gut microbiome undergoes expansion, which is influenced by dietary factors, the maturity of the immune system, and microbial competition. [1, 2] Environmental factors also shape the gut microbiome, especially during the first 1,000 days, with potential long-term repercussions on adult health (DOHaD concept; Developmental Origins of Health and Disease) [3].

The rate of premature birth (defined as birth before 37 weeks of gestation) worldwide ranges between 4 and 16% depending on the country and is higher in low-income regions [4]. In France, the prematurity rate increased from 4.5% in 1995 to 5.6% in 2016 and 7% in 2019 [5], affecting approximately 55,000 children each year [4]. Many complications, such as Sudden Infant Death Syndrome, breathing difficulties, and visual and hearing impairments in the short term, but also metabolic disorders, learning difficulties, and behavioral problems in the long term, are associated with prematurity [6]. Organs, particularly within the digestive tract, may be underdeveloped due to the premature birth, ultimately affecting their size and function (including mucus layer production, increased permeability, reduced antimicrobial peptides, and fewer Paneth cells) [7].

The gut microbiome of preterm infants differs from that of full-term infants. In full-term infants, the gut microbiome is dominated by Escherichia coli and Enterococcaceae taxa during the first week, allowing subsequent colonization by strictly anaerobic bacteria [8–10] by consuming ambient oxygen in the gut. In preterm infants, the intestinal microbiome is less abundant, less diverse, and composed of higher proportions of facultative anaerobes commonly found in neonatal intensive care unit (NICUs) [11], such as Enterobacter, Klebsiella, Escherichia, Enterococcus, Staphylococcus, and Clostridium [12, 13]. Since the gut microbiome may affect neurological development and immune system function [13, 14], differences in the gut microbiome structure may consequently influence the long-term health of infants. Moreover, the modulation of host gene expression by the gut microbiome has been well demonstrated in germ-free compared to conventional mouse models. For instance, the tricarboxylic acid (TCA) cycle of colonocytes is dysregulated in germ-free mice due to the absence of the microbial metabolite butyrate [15]. Immunity-related gene expression can also be influenced by the gut microbiome composition. In mouse models, more than half of the immunity genes expressed in the gastrointestinal tract are regulated by the gut microbiome, mostly demonstrated for pathobionts such as Helicobacter, Salmonella, and Listeria. [15] These interactions can also be studied using in vitro models (cell lines, organoids, etc.), biopsies collected from the host (invasive methods), or exfoliated colonic cells from feces [16, 17]. In our study, we partially characterized the microbiome-host crosstalk using the fecal transcriptome, which provides a noninvasive approach [18, 19].

We previously showed that very preterm infants (born in France before 31 weeks of gestation) can be stratified into six enterotypes based on the composition of their gut microbiome at one-month-old [13]. Five enterotypes were driven by one dominant bacterial genus (enterotypes 1–5), whereas one was characterized by a very low bacterial load (enterotype 6). These enterotypes were associated with different cognitive development at two years of age, as assessed by the Ages and Stages Questionnaire (ASQ). Enterotypes 4, 5, and 6 (driven by Enterococcus, Staphylococcus, and a low bacterial load, respectively) grouped the preterm infants with the worst outcome, enterotype 3 (driven by Escherichia) grouped the infants with the best outcome, and enterotypes 1 and 2 (driven by Clostridium and Klebsiella/Enterobacter) grouped the infants with heterogeneous outcomes.

In this study, we aimed to investigate the functionality of these enterotypes and determine how they may be associated with different neurodevelopmental outcomes later in life. We characterized and integrated the gut microbial composition, metabolic functions, and host transcriptomic profiles in relation to neurodevelopment in preterm infants.

Methods

Study population, ethics, and sample collection

We selected 73 very preterm infants from the EPIFLORE cohort, an ancillary study of the French national cohort EPIPAGE2 including infants born at 22–34 weeks’ gestation during 2011 [20]. EPIPAGE2 aimed to assess the medium and long-term impact of preterm birth on children’s health. Within EPIFLORE, stools were collected from diapers with a sterile spatula by trained nurses or other qualified staff members within the NICU, placed into a 2 ml sterile tube and immediately frozen at − 80 °C until microbiome analysis. Families received information and agreed to participate in this cohort by providing informed consent as recommended by French law in case of noninterventional research. The study was approved by the National Data Protection Authority (n8911009), by the Consultative Committee on the Treatment of Information on Personal Health Data for Research Purposes (number 10.626), and by the Committee for the Protection of People Participating in Biomedical Research (CPP SC2873).

Based on the previously identified microbiome–outcome associations, we consolidated the enterotypes into three outcome-based clusters: cluster A, corresponding to favorable cognitive outcomes (infants from enterotype 3); cluster B, representing heterogeneous outcomes (infants from enterotypes 1 and 2); and cluster C, encompassing poor outcomes (infants from enterotypes 4 and 5). Enterotype 6, which grouped infants with the poorest outcomes and a very low bacterial load, was not included in these clusters.

Twenty infants from each of the three clusters A, B and C and 13 from enterotype 6 (low microbial biomass) were selected to represent the overall gut microbiome distribution and the main clinical and demographic characteristics of the study population. All infants were born between 25 and 31 weeks of gestation, and fecal samples were collected at approximately 21 days. For clusters A, B and C, selection criteria were: availability of data on digestive transit type, absence of placenta praevia, presence of a central catheter, absence of major comorbidities or NEC, postnatal antibiotic treatment on day 0, and availability of APGAR scores. This led to the pre-selection of 226 infants (n = 32 Cluster A, n = 144 Cluster B, n = 50 cluster C). To reduce sample size while preserving the distribution of key baseline characteristics across groups, criteria mainly included gestational age, birth weight categories, and premature rupture of membranes. For enterotype 6 (n = 93 infants in EPIPAGE2 cohort), absence of comorbidities and sample availability were the main criteria. Post-selection clinical characteristics are detailed across the four groups in Table 1.

Table 1.

Associations between cluster of microbiota, neonatal characteristics and individual therapies in 73 infants

	Cluster A n = 20	Cluster B n = 20	Cluster C n = 20	Enterotype 6 n = 13
Gestational age (mean, in weeks)	28.75	28.3	27.35	26.38
Weight at birth
Normal	65%	55%	30%	15%
Low	5%	10%	20%	77%
Very low	30%	35%	50%	8%
Female	40%	35%	40%	62%
Context preterm
Induced	30%	30%	25%	23%
Spontaneous accepted	25%	30%	10%	38%
Spontaneous sub	45%	40%	65%	38%
Cesarean delivery	50%	80%	75%	38%
Multiple birth	45%	30%	45%	23%
Necrotizing enterocolitis (NEC cases)	0	0	0	1
Premature rupture of membranes	50%	55%	50%	0%
APGAR Superior D7	80%	85%	85%	77%
Practice of skin-to-skin contact during the first week of life	75%	80%	60%	62%
Gastrointestinal transit considered normal (> 1 stool a day)	75%	45%	35%	54%
Augmentin	0%	0%	10%	0%
D0 postnatal antibiotics	100%	90%	100%	69%
Secondary postnatal antibiotics	30%	45%	50%	77%
Mortality or severe morbidity	0%	0%	0%	23%
Insulin	25%	15%	35%	77%
Never intubated	25%	25%	10%	5%
Surfactant during first days of life	65%	70%	75%	85%
catecholamine	5%	0%	15%	8%
Sup 3 g protein	65%	75%	95%	45%
Breast milk during the first week	45%	10%	20%	31%
Breast milk D28
Other feed	20%	5%	5%	8%
Both breast feeding and formula feeding	70%	95%	95%	85%
Breast feeding (BF)	10%	0%	0%	0%
Breast feeding at hospital discharge	55%	45%	35%	15%
BMI mother (mean)	23.32	24.20	21.29	25.84
ASQ score at 2 years (mean)	217.75	193.35	161.85	200.34
Score communication at 2 years	41.75	42	29.5	43.36
Score personal social at 2 years	44.8	39.25	30	46.36
Score problem solving at 2 years	42.75	33.7	31.65	44.31
Score gross motor at 2 years	50.2	44.9	34	50.18
Score fine motor at 2 years	48.1	39.5	36.7	52.54

Fecal microbiome analysis

Total fecal DNA was extracted from 73 samples (weighted aliquots of approximately 100 mg) using chemical (lauryl-sarcosine and guanidine thiocyanate) and mechanical (bead-beating) disruption followed by isopropanol precipitation as described previously [21]. Concentration and integrity of nucleic acids were determined both visually via gel electrophoresis on 1% agarose gels and spectrophotometrically. Sequencing was performed on the GeT-PlaGe platform of Génopole (Toulouse Midi-Pyrénées, France) in 2016 using Illumina MiSeq technology (V3, 2 × 250 bp) targeting the V3-V4 regions of the 16 S bacterial rRNA gene (primers V3fwd: ACGGRAGGCAGCAG, V4rev: AGGATTAGATACCCTGGTA). Positive and negative controls were added to the libraries. The FROGS tool (“Find, Rapidly, OTUs with Galaxy Solution”) was used for the gut microbiome analysis, performing preprocessing, clustering, chimeras removal, and operational taxonomic unit (OTU) affiliation [22]. 678,511 sequences were retained after raw data quality filtering. Taxonomic affiliations were performed using both the SILVA and RDP databases and 1,534 operational taxonomic units (OTUs) were described.

Fecal untargeted metabolomics

To gain insight into the metabolic functions linked to differential bacterial colonization in the preterm gut, the fecal metabolome was analyzed. This approach allowed us to characterize both host- and microbiome-derived metabolites. Metabolomic analysis was performed in 2022 on 50 mg aliquots of frozen fecal samples using liquid chromatography−mass spectrometry (LC-MS, Metabolon Durham, USA) as previously described [23]. Following data acquisition, quality control, metabolite identification, and quantification, a normalization step was applied using the median batch method [24].

Transcriptomic library construction

To determine the host transcriptomic response to differential bacterial colonization in the preterm gut, host mRNA transcripts potentially detectable in preterm fecal samples (from exfoliated eukaryote cells) were analyzed. RNA was isolated from frozen fecal aliquots using RNeasy mini kit (Qiagen) with a modified protocol as previously published [25], eluted in 40 µl of RNAse-free water and stored at −80 °C. Total RNA was extracted from 73 samples and from an additional 31 samples (n = 25 from Cluster A, n = 27 from Cluster B, n = 39 from Cluster C, and n = 13 from enterotype 6). To establish a preliminary list of transcripts potentially impacted by the different bacterial colonization, three RNAs pools were prepared for both clusters A and B, and 2 pools for enterotype 6, with each pool containing RNA from 6 infants. RNA extracted from Cluster C samples did not meet the quality criteria for transcriptomics (no detection of housekeeping genes, low amounts of RNA, and RIN < 6). RNA was reverse-transcribed using a High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems) with oligoDT16 primers to enrich mRNA. Eight mRNA pools were sent to Eurofins (Eurofins, Germany) for deep RNA sequencing.

RT-qPCR of selected genes

Transcripts for housekeeping genes (tubulin beta, TUBB2A, and Hs00742533_s1) [26], S100 calcium-binding protein A8 and A9 (S100A8 and S100A9), cathelicidin antimicrobial peptide (CAMP), and nicotinamide phosphoribosyltransferase (NAMPT) were quantified by quantitative PCR (primers listed in Additional Table 1) after reverse transcription. RT-qPCR assays were performed with TaqMan Universal PCR Master Mix (Thermo Fisher Scientific) for tubulin and Power SybrGreen PCR Master Mix (Thermo Fisher Scientific) for other genes, using a QuantStudio 3 real-time thermocycler (Thermo Fisher Scientific). The amplification conditions were 50 °C for 2 min, 95 °C for 20 s, followed by 60 cycles of 95 °C for 1 s and 60 °C for 20s. Total RNA from HT29 cells and Escherichia coli (strain K12) was used as positive and negative controls, respectively, for eukaryotic and prokaryotic genes.

Biostatistical analyses

Analysis and visualization of gut microbiome communities were conducted in RStudio v2022.12.0 using the Phyloseq package v1.47.0 [27] to calculate the alpha and beta diversity metrics. For metabolomic analysis, Metaboanalyst software v5.0 [28] was used for Principal Component Analysis (PCA) and Partial least squares-discriminant analysis (PLS-DA) using standard normalization and scaling options and MetOrigin software v2.0 [29] was applied for metabolite profiling. For transcriptomic analysis, Illumina output raw sequencing files were filtered using fastQC v0.11.9 for read quality and fastP v0.23.2 for quality control, trimming of adapters, filtering by quality, and read pruning [30]. FASTQ files were aligned to the human genome using two databases, EnsemblOrg and Pub, using STAR v2.7.10a software [31]. Finally, transcript abundances were quantified with RSEM v1.4.2 (RNA-Seq by Expectation-Maximization) software [32]. The output was a gene count file and a normalized count file in TPM (Transcripts Per Million). Transcript pathway analysis was performed with Cytoscape v3.9.1 [33]. Statistical analyses were performed using the DESeq2 package v1.40.2 [34] on non-normalized data with default statistical settings. The functions mapIds and EnhancedVolcano were used for gene annotation and volcano plots, respectively.

Statistical and graphical analyses were performed using RStudio and Python. The analysis of numbers was performed using the package ggplot2 v3.4.0 with the chi-square (Khi2) test or G-test. Correlations and visualizations were assessed using the Hmisc (v4.7-1), Corrr (v0.4.3), and ComplexHeatmap (v2.14.0) packages. Spearman correlation and graphical analyses between genera and metabolites were performed using Python packages Pandas (v2.0.3), Numpy (v1.23.0), and Scikit-learn (v1.2.2). Boxplots were performed using the package ggstatsplot v0.11.8 in RStudio. Statistical significance was determined by non-parametric tests (Kruskal-Wallis followed by Dunn test) unless otherwise stated, and corrected for multiple comparisons with Benjamini-Hochberg adjustment.

Results

For each preterm fecal sample, we sought information on the gut microbial composition and diversity (16 S rRNA gene sequencing), metabolite composition (non-targeted metabolomics), and transcriptome profile (mRNA transcriptomics coupled with RT-qPCR). However, sample quality issues and/or insufficient bacterial biomass prevented the combination of all techniques for all samples (Additional Fig. 1). For microbial composition and metabolomics, results were obtained for Cluster A, B, and C (n = 60), but not for enterotype 6 (n = 13; low biomass). The host transcriptomic profiles were obtained from Cluster A, B, and enterotype 6. Most of the samples sequenced by RNAseq were also analyzed by RT-qPCR, with the addition of samples from Cluster C (n = 9) to confirm RNAseq results.

Very preterm gut Microbiome at one month is associated with cognitive outcome at two years

In a cohort of 577 preterm infants, Roze et al. [13] highlighted the statistical stratification of one-month-old infants into six enterotypes named enterotype 1 to enterotype 6 based on the gut microbiome composition (Additional Fig. 2). In our study, infants selected from enterotype 3 (Escherichia-driven) were referred to as Cluster A (n = 20), infants selected from enterotypes 1 and 2 (driven by Enterobacter and Clostridium sensu stricto) were referred to as Cluster B (n = 20), infants selected from enterotypes 4 and 5 (Enterococcus and Staphylococcus driven) were referred to as Cluster C (n = 20) and 13 infants were selected from enterotype 6 (C6). In our sample selection (n = 73), belonging to Cluster C was also associated with significantly lower Ages and Stages questionnaire scores (ASQ) in infants (P < 0.001) (Fig. 1A).

Fig. 1.

Distribution of enterotypes relative to cognitive development and associated gut microbiome composition. A: Frequency of preterm infants with high ASQ (> 185) in blue and low ASQ (< 185) in red within each cluster; P = 0.008 with G-test. Cluster A (n = 20), Cluster B (n = 20), Cluster C (n = 20) and Enterotype 6 (C6; n = 13). B: Principal coordinate analysis based on Bray-Curtis distance matrix highlighting the bacterial clustering between the 3 clusters (n = 20 infants/cluster). C: Alpha-diversity evaluated by observed Shannon and Simpson indices (n = 20 infants/cluster), Dunn test. D: Relative abundances of the five main genera in each cluster. (n = 20 infants/cluster), Dunn test. * = P < 0.05; ** = P < 0.01; *** = P < 0.001

Analyses of the 60 selected samples from clusters A, B, and C highlighted differences in microbial beta diversity (Fig. 1B). Clusters B and C were closer to each other and farther away from Cluster A on the principal coordinate analysis (PCoA). These differences were also reflected in microbial alpha-diversity, which was significantly lower in infants from Cluster C (Fig. 1C). The gut microbiome of Cluster A was dominated by Escherichia, Cluster B by Klebsiella, Enterobacter, and Clostridium, and Cluster C by Staphylococcus and Enterococcus genera (Fig. 1D).

Correlation between Cluster-driving genus, microbial diversity, gestational age, and weight at birth

Preterm infants were divided into three groups based on gestational age at birth (GAB): GAB1 included infants born before 26 weeks of gestation (WG), GAB2 included those born between 27 and 29 WG, and GAB3 included infants born after 30 WG. ASQ differences were not significantly associated with GAB (Fig. 2A). While the proportions of Staphylococcus at one month of age did not significantly differ between GAB groups, there was a trend toward higher abundance of Staphylococcus in the GAB1 subgroup within clusters B and C (Fig. 2B). In Cluster A (Escherichia-driven), the relative abundance of Staphylococcus was not associated with GAB. In contrast, GAB was slightly correlated with the proportion of Escherichia (r = 0.16, P = 0.03) and negatively correlated with Staphylococcus (r= −0.25, P = 0.001; Fig. 2C). These genera were also negatively correlated with each other (r= −0.47, P = 0.02). Although not statistically significant, a positive correlation was observed between ASQ and Escherichia (r = 0.3, P = 0.09). Finally, alpha-diversity was negatively correlated with Staphylococcus abundance (r = −0.25, P = 0.03).

Fig. 2.

Relations and correlations between gut microbiome composition and clinical variables. A: Frequency of preterm infants with high ASQ (> 185) in blue and low ASQ (< 185) in red, depending on gestational age (GAB1, < 26 WA; GAB2, between 27 WA and 29 WA; GAB3, > 30 WA). P = 0.06 with G-test. GAB1 (n = 25), GAB2 (n = 29) and GAB3 (n = 19). B: Relative abundance of Staphylococcus in the 3 clusters according to the gestational age. GAB1 (n = 25), GAB2 (n = 29) and GAB3 (n = 19). C: Spearman’s correlation with the significance of quantitative clinical variables and dominant bacterial genera * P < 0.05, ** P < 0.01

Preterm infants were also divided into three groups based on weight at birth: very low birth weight included infants weighing less than 1,500 g, low birth weight included infants weighing between 1,500 g and 2,499 g, and normal birth weight included infants weighing more than 2,500 g. ASQ differences were not significantly associated with birth weight (Additional Fig. 3). Infant weight at birth was positively correlated with the relative abundance of Escherichia (r = 0.36, P = 0.02) and negatively correlated with the relative abundance of Enterococcus (r = −0.22, P = 0.054). The dominant genera of the enterotypes also correlated with each other: Escherichia was negatively correlated with Staphylococcus and Enterococcus (r = −0.47, P = 0.02, and r = −0.22, P = 0.03, respectively) and Enterococcus was negatively correlated with Enterobacter (r = −0.13, P = 0.04).

The differences in the gut microbiome were reflected in the fecal metabolomic profiles

A total of 1,040 metabolites were detected in the feces of very preterm infants at one month of age, and 856 were annotated. Based on MetOrigin analysis, 49 metabolites were solely attributed to the gut microbiome, 6 to the host, and 150 to co-metabolism of the microbiome and the host. The remaining metabolites were associated with drugs (n = 35), food components (n = 75), and of unknown origin (n = 725) (Fig. 3A). According to principal component analysis (PCA) of the composition of fecal metabolites, the fecal metabolome of preterm infants belonging to Cluster C differed slightly from that of preterm infants in clusters A and B. Notably, while Cluster A was distinct from the other two clusters based on gut microbiome composition, the fecal metabolome was even more divergent in infants from Cluster C than in those from clusters A and B (Fig. 3B). Analysis of the super-pathways by partial least squares-discriminant analysis (PLS-DA) highlighted that the pathways related to energy, carbohydrates, and partially characterized molecules discriminated the three clusters, with a relatively higher variable importance in projection (VIP score) between 1.3 and 1.6 (Fig. 3C). The VIP score summarizes the contribution of a variable to the model and is calculated as the weighted sum of the squared correlations between the PLS-DA components and the original variable [35]. Energy super-pathway metabolites were more abundant in Cluster A (Escherichia-driven) and less abundant in Cluster C (Staphylococcus- and Enterococcus-driven). This pathway was positively correlated with Escherichia (r = 0.4, P < 0.001) and Enterobacter (r = 0.4, P < 0.001), and negatively correlated with Enterococcus (r = −0.2, P < 0.05) and Staphylococcus (r = −0.4, P < 0.001) (Fig. 3D). The lipid, peptide, and xenobiotic pathways were the least discriminant with VIP scores close to 0. These pathways were not correlated with bacterial members (except for the xenobiotic pathway in Clostridium, r = 0.2, P < 0.05). The amino acid super-pathway was positively correlated with Klebsiella (r = 0.2, P < 0.05) and negatively correlated with Staphylococcus (r = −0.3, P < 0.05). Nucleotide and cofactor pathways were not correlated with bacterial members with VIP scores lower than 1.

Fig. 3.

Composition of metabolome fecal and importance of super-pathways for each cluster and genera. A: Origin of metabolites detected in the feces. B: Principal Component Analysis of the fecal metabolome. C: Partial least squares-discriminant analysis (PLS-DA) on super pathways. A VIP score is a measure of a variable’s importance in the PLS-DA model, it summarizes the contribution a variable makes to the model. The VIP score of a variable was calculated as the weighted sum of the squared correlations between the PLS-DA components and the original variable. Partially Char = Partially Characterized Molecules. Color under clusters shows the relative ranking of cluster-wise mean normalized abundances for each super pathway. For each pathway, clusters A, B, and C are ordered from lowest to highest abundance and displayed using a discrete three-level color scale (low, intermediate, high). D: Spearman correlation with significance between super pathway and bacterial taxa of interest. * = P < 0.05; *** = P < 0.001

Expression of immune-related genes differentiated the clusters

Host transcripts from pooled fecal samples of infants belonging to Cluster A (n = 3; 18 infants) and B (n = 3; 18 infants) were analyzed using RNASeq. Transcriptome profiles could not be determined in Cluster C due to the low amount and quality of eukaryotic RNA. Instead, host transcripts were recovered from infants belonging to enterotype 6 (low biomass; n = 2 pools; 12 infants). Overall, the gene expression patterns of the exfoliated cells discriminated the three clusters on the PCA (Fig. 4A). The genes that were overexpressed in Cluster A as compared to Cluster B were involved in pathways related to inflammation (acute inflammatory response, humoral immune response, immune effector process, immune response, immune system process, inflammatory response, inflammatory response to antigenic stimulus), leukocyte activation and aggregation, and chemokine production (Fig. 4B). In contrast, 33 genes were under-expressed in preterm infants from enterotype 6 relative to Cluster A. There were no statistically significant differences between Cluster B and enterotype 6. Based on these results, we further focused on transcripts expressed by immune cells such as antimicrobials (defensins and cathelicidin) and calprotectin. Cathelicidin Antimicrobial peptide (CAMP) has been described as the only cathelicidin in humans [36]. RT-qPCR analysis showed that CAMP/LL37 expression was lower in enterotype 6 than in Clusters A and B (P = 0.004 and P = 0.01, respectively) (Fig. 4C). Considering defensins, no significant differences were detected in RNAseq experiment between all 3 groups neither for the expression of the lysozyme gene nor for alpha-defensins DEF1, DEF3, DEF4, DEFA5 and DEFA6. Transcriptomic results indicate an overexpression of the genes encoding both subunits of Calprotectin (S100A8 and S100A9) in the gut of infants from Cluster A. Subsequent RT-qPCR analyses confirmed that both S100A8 and S100A9 encoding genes were differentially expressed between clusters A and enterotype 6 (P = 0.0053 and P = 0.063, respectively, Fig. 4D). Moreover, S100A8 gene expression was significantly lower in group enterotype 6 than in clusters B (P = 0.05) and C (P = 0.05) while S100A9 was differentially expressed between clusters A and C (P < 0.001) and between clusters B and enterotype 6 (P = 0.007). Finally, Orthogonal Partial Least Squares-Discriminant Analysis (O-PLSDA) was applied to better differentiate between Clusters A and B and highlighted that Nicotinamide phosphoribosyltransferase gene (NAMPT; also known as visfatin) was the most discriminating transcripts with a VIP score over 8 (Fig. 4E).

Fig. 4.

Intestinal cells mRNA expression profiles in preterm infants. An exfoliome approach was applied to use mRNA from feces as a surrogate for tissue mRNA expression. A: Principal Component Analysis of the fecal host transcriptome. Cluster A (n = 3), Cluster B (n = 3) and Enterotype 6 (C6; n = 2). B: Volcano plot highlighting genes with significant differential expression in Cluster A versus Cluster B (top) and enterotype 6 versus Cluster A (bottom). C: Log2 Fold change for the CAMP/LL37 gene. Cluster A (n = 19), Cluster B (n = 17), Cluster C (n = 12) and Enterotype 6 (n = 5), Dunn test. * = P < 0.05; ** = P < 0.01. D: Log2 Fold change for calprotectin subunits S100A8 and S100A9. Cluster A (n = 19), Cluster B (n = 17), Cluster C (n = 12) and Enterotype 6 (n = 5). Dunn test. E: Partial least squares-discriminant analysis between clusters A and B

Levels of final TCA cycle metabolites associated with Staphylococcus abundance and neurodevelopment

The tricarboxylic acid (TCA) cycle is the major pathway involved in energy metabolism. Metabolites that appear early in the TCA cycle, i.e. citrate (P = 0.01) and aconitate (P = 0.004), were more abundant in Cluster C than in Clusters A and B (Fig. 5A, Additional Fig. 4A-B). In contrast, final TCA cycle metabolites i.e. malate (P < 0.001), fumarate (P < 0.001), and succinate (P < 0.001), were more abundant in clusters A and B than in Cluster C (Additional Fig. 4E-G). Supporting this result, the proportions of Staphylococcus were positively correlated with citrate and aconitate (r = 0.44, P = 0.01, and r = 0.38, P < 0.001, respectively), whereas Escherichia was negatively correlated with aconitate (r = −0.27, P = 0.04) but positively correlated with fumarate and succinate (r = 0.56, P = 0.01, and r = 0.48, P < 0.001, respectively). These two metabolites were negatively correlated with Enterococcus (r = −0.56, P = 0.04, and r = −0.41, P = 0.01, respectively) (Fig. 5B). Isocitrate (isocitric lactone), which is positioned between aconitate and alpha-ketoglutarate in the TCA cycle, did not correlate with any bacterial genera and was not differentially detected between clusters (Fig. 5B, Additional Fig. 4C). Alpha-ketoglutarate was less abundant in Cluster C and was negatively correlated with Enterococcus (r = −0.36, P = 0.02) (Fig. 5B, Additional Fig. 4D). A similar global correlation pattern exists between Staphylococcus and Enterococcus and metabolites of the TCA cycle. Statistical confirmation was provided by the K-means method, which separates the correlations into two groups: the first group with Staphylococcus and Enterococcus and the second group with the other bacteria (Fig. 5B). These findings demonstrate a consistent pattern in Cluster C compared with the other two clusters. Moreover, some metabolites of the TCA cycle are associated with ASQ, particularly fumarate which was more abundant in feces from infants with ASQ > 185 (P = 0.03) (Additional Fig. 5). The transition from metabolites abundant in Cluster C (citrate and aconitate) to metabolites abundant in Clusters A and B was separated by a reaction requiring the presence of NAD+. Isocitrate dehydrogenases catalyze the oxidative decarboxylation of isocitrate to alpha-ketoglutarate and reduce NAD + to NADH. NAD + was less abundant in Cluster C than in Clusters A and B (P = 0.02 and P = 0.04, respectively), thus blocking the continuation of the reactions of the TCA cycle (Fig. 5C). NAD + was negatively correlated with Staphylococcus (r = −0.43, P = 0.02) and positively correlated with Escherichia (r = 0.2, P = 0.05). Instead of being generated de novo from tryptophan, most NAD + is recycled from nicotinamide (NAM) through the salvage pathway to maintain cellular NAD+ levels via NAMPT [37]. This gene was more highly expressed in Cluster A than in clusters B, C and enterotype 6 (P = 0.004, P = 0.003, and P = 0.08, respectively) (Fig. 5D).

Fig. 5.

Low levels of final TCA cycle metabolites are associated with Staphylococcus-enriched gut microbiome and unfavorable cognitive development. A: Abundance of metabolites constituting the TCA cycle in three 3 different clusters. In the TCA cycle, red: metabolites higher in Cluster C, green: metabolites higher in clusters A and B, yellow: no difference, and green: no difference in LC-MS; B: Spearman correlation between TCA cycle metabolites and bacteria. Separation by k-means in the two groups. C: Abundance of NAD+ in each cluster, Dunn test. D: Relative expression of the NAMPT gene with RT-qPCR analysis. Cluster A (n=20), Cluster B (n=20), Cluster C (n=20) and Enterotype 6 (n=13), Dunn test. * = P <0.05; ** = P<0.01; *** = P <0.001

Discussion

Anticipating neurodevelopmental trajectories in preterm children poses an ongoing challenge, yet it holds the promise of enabling tailored interventions to protect or enhance health in vulnerable populations. While developing novel machine learning-based predictive models to assess neurodevelopmental outcomes in preterm children, Juul et al. highlighted that it remains difficult to accurately predict the results of complex long-term assessments such as neurodevelopmental scores, even when using over 150 clinical, demographical, and general variables [38]. Our study highlights that the proportions of Escherichia and Staphylococcus/Enterococcus colonizing the gut of one-month-old preterm infants were relevant microbial signatures of clinical birth variables and may help anticipate neurodevelopmental outcomes at two years of life. Specifically, in preterm infants, a gut microbiome enriched in Escherichia at one month was associated with better clinical health (e.g. higher birth weight) and higher ASQ scores at two years compared to children with a gut microbiome enriched in Staphylococcus. A previous study indicated a positive association between alpha-diversity in the gut microbiome of one-year-old infants and the Mullen score [39] which evaluates infant and young child cognitive functioning. In our study, a microbiome enriched in Staphylococcus or Enterococcus reflected lower microbial diversity, which may negatively impact infants’ health, particularly during the first 1,000 days of life (DOHaD concept) [40] and cognitive development [41–45]. While Klebsiella overgrowth in preterm infants has been previously highlighted as predictive of brain damage and associated with a pro-inflammatory immunological tone [14], we did not find an association between Klebsiella (nor with Cluster B) and the cognitive outcome at 2 years of age in our study. However, brain physiology was not assessed at 1 month of age in the present work and a transient effect cannot be ruled out. Moreover, the impact of Klebsiella may be species- or strain-specific.

The gut microbiome can influence the host immune system, including for instance the expression of anti-microbial peptides produced by Paneth cells [46]. We observed that infants from enterotype 6 had lower expression of LYZ genes (Lysozyme; Paneth cell markers), suggesting reduced production of this antimicrobial peptide, whereas alpha-defensin genes (DEF1 to DEF6) were not significantly different between clusters. On the other hand, the expression of Cathelicidin (CAMP/LL37) was lower in enterotype 6 compared to other clusters; since CAMP contributes to both adaptive and innate immunity, these findings suggest it may play a role in neonatal gut microbiome establishment [47]. Overall, our results indicate that having detectable levels of bacteria in the gut is associated with higher expression of certain immune-related genes, with expression levels varying by dominant bacterial genus. An increase in the expression of immune system-related genes at the beginning of life is beneficial to long-term health. Previous studies highlighted that an increase in fecal calprotectin levels in infants during the first month of life was associated with health [48–50], and we accordingly observed that children with a depleted gut microbiome and poorer neurodevelopment had low expression of CAMP and calprotectin at one month.

Since inadequate energy intake in preterm infants may contribute to neurodevelopmental delay, early nutrition is critical, though efficacy depends on digestive and absorptive capacity. We observed that energy metabolism was linked with gut microbial patterns. Proportions of Staphylococcus were positively correlated with early TCA cycle metabolites and negatively correlated with late metabolites, whereas Escherichia showed the opposite pattern. NAMPT transcript (visfatin) expression was higher in Cluster A (Escherichia-driven) and catalyzes NAD+ synthesis, an essential component at the beginning of the tricarboxylic acid (TCA) cycle. NAD+ levels were positively correlated with Escherichia proportions and negatively with Staphylococcus, suggesting a potential blockade of the TCA cycle in Staphylococcus-driven Cluster C due to lower NAMPT expression. Enhancing NAMPT expression and NAD+ levels may promote proper early energy metabolism and support beneficial microbial growth while inhibiting pathogenic bacteria, including Staphylococcus. [51] Since the administration of NAD+ precursor (NAD3 dietary supplement) in adults improves neurodegenerative pathologies [52], induction of NAMPT gene expression by bacterial members (e.g. Escherichia taxa) and subsequent increase in NAD+ levels may be beneficial in preterm infants to promote proper energy metabolism in early weeks. Moreover, our findings align with recent observations emphasizing that early gut microbiome composition in preterm infants is associated with immune maturation and neurodevelopmental outcomes, highlighting that microbial colonization may shape neurodevelopmental trajectories through interactions with metabolic and immune pathways. These insights reinforce the notion that early-life gut microbiome profiles, including specific genera and their associated metabolites, may contribute to neurodevelopment by modulating key biological systems during a critical window of development [53].

Despite these promising findings, our study has several limitations that warrant further investigation. First, we analyzed the intestinal host-microbiome crosstalk in a small number of participants, and only at 1 month of age which does not provide information on the evolutionary dynamics of the host-microbial interactions in the premature intestine. Moreover, inclusion was guided mainly by perinatal factors, without accounting for variations in feeding strategies or other interventions received between postnatal days 7 and 21. In addition, the cohort includes only French neonates, limiting generalizability to other populations. Moreover, the low yield of data in some analyses allowed only partial characterization of the clusters, which may affect the interpretation of the findings. A multicentric international study may help decipher the impact of geographic location as well as differences in NICU practices, which vary strongly between countries and have been shown to influence preterm infant’s gut microbiome and health. Finally, future larger-scale studies will be needed to robustly apply integrative systems biology analyses offering additional mechanistic insights.

Conclusion

Our study demonstrates that a one-month-old preterm infant’s gut microbiome, particularly the dominance of Escherichia or Staphylococcus, is strongly associated with birth-related variables and may influence neurodevelopmental outcomes at two years of age. A gut microbiome enriched in Escherichia is associated with a higher gestational age and birth weight and may have more beneficial effects on overall neurodevelopment at 2 years of age, bacterial diversity, TCA cycle metabolism, and primo-immunity. In contrast, a Staphylococcus-dominated gut microbiome is linked to poorer health outcomes, including blocked TCA cycle metabolism, weaker immune expression, and worse overall neurodevelopment at 2 years of age. These findings highlight the potential of early intestinal microbial colonization as a predictor of long-term health and neurodevelopment in preterm infants.

Abbreviations

ASQ

Ages and stages questionnaire

CAMP

Cathelicidin antimicrobial peptide

DEF

Alpha-defensins

DOHaD

Developmental origins of health and disease

GAB

Gestational age at birth

LC-MS

Liquid chromatography−mass spectrometry

LYZ

Lysozyme gene

NAD

Nicotinamide adenine dinucleotide

NAM

Nicotinamide

NAMPT

Nicotinamide phosphoribosyl transferase

NEC

Necrotizing enterocolitis

NICU

Neonatal intensive care unit

O-PLSDA

Orthogonal partial least squares-discriminant analysis

OTU

Operational taxonomic unit

PCA

Principal component analysis

PCoA

Principal coordinate analysis

PLS-DA

Partial least squares-discriminant analysis

RT-qPCR

Reverse transcriptase quantitative polymerisation chain reaction

TCA

Tricarboxylic acid

TPM

Transcripts per million

VIP

Variable importance in projection

WG

Weeks of gestation

Authors’ contributions

P.Y.A., J.C.R., M.J.B. and P.L. conceptualized and designed the study, and secured funding. C.M. and C.P. performed the laboratory work. C.P. and L.M.M. performed statistical analyses. C.P. performed the bioinformatical analyses and wrote the initial draft of the manuscript. P.L. supervised data analyses, manuscript preparation and oversaw all aspects of the research. L.M.M., J.C.R and P.Y.A. performed clinical data collection. All authors read and approved the final version of the manuscript.

Funding

This study was funded by the French National Agency for Research (ANR) and Human Nutrition Division of the INRAE (AlimH). The funders played no role in study design, data collection, analysis and interpretation of data, or the writing of this manuscript.

Data availability

The 16S rRNA gene reads are publicly available from the National Center for Biotechnology Information Sequence Read Archive under the Bioproject accession number PRJNA641157. The metadata are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

The study was approved by the French National Data Protection Authority (No. 8911009), by the Consultative Committee on the Treatment of Information on Personal Health Data for Research Purposes (No. 10.626), and by the Committee for the Protection of People Participating in Biomedical Research (CPP SC-2873). Parental informed consent was obtained in all cases.

Consent for publication

Not Applicable.

Competing interests

The authors declare no competing interests.