AI-2 combined with Lactobacillus rhamnosus GG remodels gut microbiota structure to alleviate intestinal oxidative stress injury in a mouse necrotizing enterocolitis model

Abstract

Background

Necrotizing enterocolitis (NEC) is a devastating intestinal disease primarily affecting preterm infants. This study aimed to explore the efficacy of Lactobacillus rhamnosus GG (LGG) combined with quorum-sensing molecule autoinducer-2 (AI-2) in a neonatal mouse model of NEC.

Methods

NEC was induced in neonatal mice, which were then randomly assigned to the NEC or treatment groups (NEC + AI-2, NEC + LGG, and NEC + LGG + AI-2), with uninduced mice as control group. Disease severity, intestinal barrier integrity, inflammatory responses, scanning electron microscopy (SEM), gut microbiota structure, transcriptomic profiling, and oxidative stress markers were comprehensively evaluated.

Results

The LGG + AI-2 co-treatment demonstrated the most effective protection against NEC. It markedly alleviated clinical symptoms and intestinal histopathological damage, with additive benefits compared with LGG or AI-2 monotherapy. Mechanistically, the combination enhanced the intestinal barrier by upregulating the tight junction protein zona occludens-1 (ZO-1), reducing the levels of tumor necrosis factor-α (TNF-α) and interleukin-6 (IL-6), and increasing the levels of interleukin-10 (IL-10) by inhibiting nuclear factor-kappa B (NF-κB) activation. SEM results indicated that LGG combined with AI-2 restored intestinal biofilm formation and colonization of beneficial commensal bacteria. Gut microbiota analysis revealed that LGG + AI-2 ameliorated microbial balance, increasing diversity and selectively enriching beneficial bacteria such as Clostridium butyricum while suppressing the growth of pathogens such as Escherichia coli. Transcriptomic analysis identified 131 core differentially expressed genes, predominantly enriched in glutathione metabolism and oxidative stress pathways. Accordingly, the combination treatment rescued redox homeostasis, evidenced by increased reduced glutathione (GSH) levels, decreased malondialdehyde (MDA) and oxidized glutathione (GSSG) contents, as well as restored expression levels of the key antioxidant regulators glutathione peroxidase 4 (GPX4) and nuclear factor erythroid 2-related factor 2 (NRF2).

Conclusion

The combination of LGG and AI-2 confers a potent protective effect against NEC by simultaneously improving intestinal integrity, modulating inflammation and microbiota, and alleviating oxidative stress, highlighting a promising novel combined therapeutic strategy for NEC.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12967-026-08646-5.

Introduction

Necrotizing enterocolitis (NEC) remains a devastating gastrointestinal emergency in preterm infants, defined by high mortality and limited treatment options [1, 2]. Despite advances in neonatal care, current therapeutic strategies, including antibiotics, bowel rest, and surgical intervention, frequently result in suboptimal outcomes and high recurrence rates [3, 4]. This crucial therapeutic gap emphasizes the urgent need to develop novel treatment approaches [5].

Probiotic supplementation has emerged as a promising preventive strategy for NEC [6]. Lactobacillus rhamnosus GG (LGG) improves intestinal barrier integrity, modulates immune responses, and fosters a balanced microbial environment [7, 8]. However, the protective efficacy of LGG monotherapy against NEC remains inconsistent across clinical and preclinical studies [9], underscoring the need for combined therapeutic strategies to augment and stabilize probiotic benefits [10].

Concurrently, molecules related to bacterial communication have garnered attention as potential microbiota-targeting therapeutics [11]. Autoinducer-2 (AI-2), a key quorum-sensing molecule, regulates intestinal microbial community structure and interspecies communication [12, 13]. Accumulating evidence indicates that NEC pathogenesis is closely linked to intestinal oxidative stress, characterized by excessive reactive oxygen species (ROS) production, glutathione (GSH) depletion, and dysregulation of the nuclear factor erythroid 2-related factor 2 (NRF2)/glutathione peroxidase 4 (GPX4) antioxidant axis [14, 15]. Our previous studies demonstrated that fecal AI-2 levels correlated with NEC severity in infants and exogenous AI-2 ameliorated intestinal inflammation and dysbiosis in neonatal mouse models [16–18]. Furthermore, we reported that AI-2 and LGG co-administration enhanced probiotic colonization, restored microbial architecture, and strengthened barrier integrity [19]. However, the combined mechanisms of LGG + AI-2 in NEC—especially whether they converge to affect intestinal barrier repair, inflammatory signaling, microbial reshaping, and oxidative stress alleviation—remain unclear. Moreover, the incomplete mechanistic understanding of this combination limits its clinical translation.

Based on these considerations, we hypothesized that LGG and AI-2 act in combination to protect against NEC via complementary and potentially additive mechanisms targeting intestinal barrier integrity, inflammatory signaling cascades, microbial ecology, and oxidative stress homeostasis. To test this hypothesis, we established a neonatal mouse NEC model to systematically assess the combined protective effects of LGG + AI-2 and elucidate the underlying mechanisms via multi-dimensional analyses of histopathology, inflammation, scanning electron microscopy, gut microbiota, transcriptomics, and oxidative stress.

Materials and methods

Preparation of LGG

LGG was cultivated as described previously [7, 19]. Briefly, LGG (BNCC136673, China) was cultured anaerobically at 37 ℃ in De Man, Rogosa, and Sharpe medium (Hopebio, Qingdao, China) for 24 h. The OD600 was adjusted to 0.5 (~ 10⁸ CFU/mL). The suspension was centrifuged at 5000 rpm for 5 min at 4 °C, washed, and resuspended twice in sterile saline before use.

Mouse model of NEC

C57BL/6 mice were acquired from Chongqing Medical University and maintained in a specific-pathogen-free environment. To breed pregnant mice, female and male mice were co-housed in a ratio of 1:2. The NEC model was developed using previously published methods with modifications [20]. The 7-day-old C57BL/6 mice were randomly allocated to five groups using a computer-generated randomization sequence. The control (CON) mice were reared by their mothers. AI-2 (S-4,5-dihydroxypentane-2,3-dione, DPD) was purchased from Omm Scientific, Inc. (Dallas, TX, USA) (http://www.ommscientific.com/products.html), as previously synthesized and characterized [21]. For NEC and intervention groups, formula milk was prepared as follows: the mice in the NEC group received 2 g Similac Advance in 10 mL of 33% Esbilac puppy milk replacer; the mice in the AI-2 group received 2 g Similac Advance with 1 mL of 5 mM AI-2 solution in 9 mL of 33% Esbilac (final concentration 500 nM); the mice in the LGG group received 2 g Similac Advance in 9 mL of 33% Esbilac supplemented with LGG (10⁸ CFU/mL); and the mice in the LGG + AI-2 group received 2 g Similac Advance with 1 mL of 5 mM AI-2 solution in 9 mL of 33% Esbilac supplemented with LGG (10⁸ CFU/mL). All groups of mice were fed every 4 h via a silicone tube (1.9 Fr) at 30 mL/kg of body weight for 3 days. Following feeding, the mice in the NEC and intervention groups were subjected to asphyxiation (100% nitrogen, 90 s) and cold stress (4 °C,10 min) twice daily for 3 days. The CON mice were not subjected to these procedures. After the 3-day modeling period, the mice were rested for 12 h prior to tissue and content harvest. Investigators were blinded to group allocation for all downstream analyses (histopathology, SEM, transcriptome sequencing, and 16S rRNA sequencing), with randomization and blinding performed by independent researchers to minimize bias. The Animal Care and Use Committee of Chongqing Medical University approved this study (Approval No. 20231013001).

Histopathological examination

Three days after NEC induction, the mice were euthanized and a 1-cm segment of the distal ileum was excised and fixed in 4% paraformaldehyde. The tissues were dehydrated, embedded in paraffin, and cut into 4-µm sections. The sections were stained with hematoxylin and eosin and observed under an optical microscope (Nikon, Japan). The grading criteria were as follows: normal (0); epithelial cell lifting or separation (1); necrosis to the villus level (2); complete villous necrosis (3); and full-layer necrosis (4). For each sample, one randomly selected field was evaluated by a blinded assessor in a double-blinded manner, with the highest injury score taken as the final histopathological score [22, 23]; a score of ≥ 2 indicated NEC positivity [24].

Tissue protein extraction

For biochemical analyses, the intestinal tissues were homogenized in RIPA lysis buffer (Keygen BioTECH, China) containing 1% PMSF. The homogenate was centrifuged at 12,000 rpm for 10 min at 4 °C, and the supernatant was collected. For Western blotting, the protein concentration was determined using the bicinchoninic acid assay (Keygen BioTECH, China). For the enzyme-linked immunosorbent assay (ELISA), the supernatant was used directly.

Enzyme-linked immunosorbent assay

ELISA was performed following the manufacturer’s protocol (ABclonal, China). Briefly, the frozen samples and kit reagents were equilibrated to room temperature for 30 min. Standards and samples (100 µL/well) were added to the corresponding wells and incubated at 37 °C for 90 min, followed by four washes. Biotinylated antibody working solution was added and incubated at 37 °C for 60 min, followed by four washes. Enzyme conjugate was added and incubated at 37 °C for 30 min, followed by four washes. Chromogenic reagent was introduced, and the reaction was terminated with the stop solution in the dark. The optical density was measured at 450 nm using a microplate reader (BioTek, USA) and blank-corrected. The sample concentrations (pg/mL) were determined from the standard curve for IL-6, TNF-α, and IL-10.

Western blotting

The protein samples were combined with SDS-PAGE sample buffer in a 4:1 ratio and denatured at 100 °C for 10 min. Proteins were resolved by 10% SDS-PAGE and transferred to methanol-activated polyvinylidene difluoride membranes (Millipore, USA) at 100 V for 90 min at 4 °C. The membranes were blocked with 5% BSA in Tris-buffered saline with Tween-20 for 1 h at room temperature and incubated overnight at 4 °C with primary antibodies against zona occludens-1 (ZO-1; Proteintech, China), phosphorylated nuclear factor-kappa B (p-NF-κB; CST, USA), NF-κB (Proteintech, China), GPX4 (Proteintech, China), NRF2 (Proteintech, China), and β-actin (ABclonal, China). After incubation with HRP-conjugated secondary antibodies for 1 h at room temperature, the bands were visualized using an enhanced chemiluminescence (ECL) detection kit (Millipore, USA) and imaged with the Bio-Rad ChemiDoc Touch system (Bio-Rad, Hercules, USA). The intensities were quantified using Image Lab and ImageJ software, normalized to β-actin.

SEM observation of mouse intestinal tissue

The experiment employed 10-day-old mice. The mouse intestinal tissue was excised, rinsed with phosphate-buffered saline, and fixed in 3% glutaraldehyde. After three 10-min washes with ultrapure water, the tissue was post-fixed in 1% osmium tetroxide for 1–2 h, followed by three additional 10-min washes. Dehydration was performed using graded ethanol (30%, 50%, 70%, 90%, and three rounds of 100%, 15 min each). The tissue was subjected to critical point drying, mounted with a conductive adhesive, and gold-sputtered. Imaging was performed using a JSM-IT700HR scanning electron microscope (JEOL Ltd., Tokyo, Japan). Low-magnification fields were examined first, followed by high-magnification imaging of selected regions to analyze mucosal structure, biofilm presence, and bacterial colonization [25]. Quantitative image analysis was performed using ImageJ software following established protocols [26, 27]. For each sample, two random fields were selected at consistent magnification (n = 3 biological replicates per group). The biofilm coverage area (% of mucosal surface covered by the biofilm matrix) was measured by scale calibration and threshold-based area calculation. Adherent bacterial density (bacteria per 100 μm²) was determined in two random high-power fields per sample. The CON group was excluded from bacterial density quantification due to its dense, multilayered biofilm.

RNA isolation and cDNA synthesis

Total RNA was extracted from intestinal tissues using QIAzol lysis reagent (Qiagen, Germany) for transcriptome analysis or an RNA extraction kit (Promega, China) for quantitative polymerase chain reaction (qPCR) following the manufacturers’ protocols. For qPCR, RNA was reverse transcribed into cDNA.

Transcriptomic analysis of intestinal tissue

RNA integrity was evaluated using an Agilent 5300 Bioanalyzer, and the concentration was measured with a NanoDrop ND-2000 spectrophotometer. Only high-quality samples (OD260/280 1.8–2.2, OD260/230 ≥ 2.0, RIN ≥ 6.5, 28S:18S ≥ 1.0, total RNA ≥ 1 µg) were used for library construction. Subsequent RNA purification, reverse transcription, library construction, and sequencing were performed by Majorbio Biotech Co., Ltd. (Shanghai, China). Briefly, mRNA was isolated by poly(A) selection using oligo(dT) beads and fragmented. Double-stranded cDNA was synthesized using a SuperScript kit (Invitrogen, USA) with random hexamer primers (Illumina, USA). After end-repair, phosphorylation, and “A” base addition, the libraries were size-selected (~ 300 bp) and amplified using Phusion DNA polymerase (New England Biolabs, USA) for 15 cycles. Following Qubit 4.0 quantification, paired-end sequencing was performed on the NovaSeq 6000 (Illumina, USA) with 2 × 150-bp reads. Raw reads were trimmed using fastp (v0.20.0) and aligned to the GRCm39/mm39 reference genome using HISAT2 (v2.1.0). Mapped reads were assembled with StringTie (v2.1.4), and gene abundance was quantified using RSEM (v1.3.1) as transcripts per million. Differential expression was analyzed using DESeq2 (|log2FC| ≥ 1, P ≤ 0.05). Functional enrichment [Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG)] analysis was performed using GOATOOLS and KOBAS (v2.0), with significance set at Bonferroni-corrected P ≤ 0.05 [28, 29].

Quantitative real-time reverse transcription PCR

cDNA was amplified using the CFX96 real-time PCR detection system (Bio-Rad, USA) with GAPDH and Actin as the reference genes. The primer sequences were as follows: mouse GAPDH: forward 5’-CACGGCAAATTCAACGGCACAG-3’, reverse 5’-TCGCTCCTGGAAGATGGTGATGG-3’; mouse Nrf2: forward 5’-TGCCACCGCCAGGACTACAG-3’, reverse 5’-GCGTGCTCAGAAACCTCCTTCC-3’; mouse Gstm3: forward 5’-TCTGCTGCAGTCCCGATTTT-3’, reverse 5’-CCTCTTGCCCAGGAACTCAG-3’; mouse Gsta4: forward 5’-TGATGATGATTGCCGTGGCT-3’, reverse 5’-CAACTGAGCTGGTTGCCAAC-3’; mouse Gsta13: forward 5’-CCCTTTCCCTCTGCTGAAGG-3’, reverse 5’-CCATGGGAGGCTTTCTCTGG-3’; mouse Gstm1: forward 5’-TCCCGACTTTGACAGAAGCC-3’, reverse 5’-TCCATCCAGGTGGTGCTTTC-3’; mouse Actin: forward 5’-ACTGCCGCATCCTCTTCCTC-3’, reverse 5’-AACCGCTCGTTGCCAATAGTG-3’. The results were expressed as 2^−ΔΔCt ± standard error of mean.

Fecal microbiota analysis

Fresh feces were collected under sterile conditions immediately after defecation, snap-frozen in liquid nitrogen, and stored at −80°C. Total microbial genomic DNA was extracted from thawed fecal samples (0.2 g per sample) using the FastPure Stool DNA Isolation Kit (MJYH, Shanghai, China) following the manufacturer’s protocol. DNA quality was evaluated by 1% agarose gel electrophoresis, and the concentration and purity were determined using a NanoDrop 2000 UV-Vis spectrophotometer (Thermo Scientific, USA). High-quality DNA (OD260/280 = 1.8–2.0, OD260/230 ≥ 1.8) was used for subsequent analysis. The bacterial 16S rRNA gene was amplified using universal primers 27F (5’-AGRGTTYGATYMTGGCTCAG-3’) and 1492R (5’-RGYTACCTTGTTACGACTT-3’) [30], each tailed with unique PacBio barcodes. The PCR mixture (20 µL) contained 4 µL of 5× FastPfu buffer, 2 µL of 2.5 mM dNTPs, 0.8 µL each of 5 µM forward and reverse primers, 0.4 µL of FastPfu DNA polymerase, 10 ng of template DNA, and DNase-free water. Amplification was performed using a T100 Thermal Cycler (Bio-Rad, USA). PCR was performed under the following conditions: 95 °C for 3 min; 27 cycles of 95 °C for 30 s, 60 °C for 30 s, and 72 °C for 45 s; final extension at 72 °C for 10 min; and hold at 4 °C. Triplicate reactions were performed per sample. PCR amplicons were purified with AMPure PB beads (Pacific Biosciences, USA), quantified using Synergy HTX (BioTek, USA), pooled equimolarly, and used for library construction with the SMRTbell Prep Kit 3.0 (Pacific Biosciences) before sequencing on the PacBio Sequel IIe System (Pacific Biosciences) by Majorbio Bio-Pharm Technology Co. Ltd. (Shanghai, China). PacBio raw reads were processed with SMRT Link v11.0 to obtain HiFi reads (≥ 3 full passes, 99% accuracy), which were barcode-identified and length-filtered (1,000–1,800 bp retained). HiFi reads were denoised via the DADA2 plugin in QIIME2 (v2020.2) [31] to generate amplicon sequence variants (ASVs). The resulting ASVs were rarefied to 6,000 sequences/sample (average Good’s coverage, 97.90%), and taxonomically classified using RDP Classifier v2.13 [32] against the NT_16s database (v20230830) with a confidence threshold of ≥ 0.7. The gut microbiota health status was evaluated by calculating the gut microbiome health index (GMHI) and microbial dysbiosis index (MDI). Metagenomic functions were predicted via PICRUSt2 [33] (HMMER for sequence alignment, EPA-NG/Gappa for reference tree placement, castor for 16S copy normalization, and MinPath for gene pathway mapping). Bioinformatic analyses were performed on the Majorbio Cloud platform (https://cloud.majorbio.com) and in R: alpha diversity (observed ASVs, Chao1, Shannon, and Good’s coverage) via Mothur v1.30.2 [34]; beta diversity via PCoA (Bray–Curtis dissimilarity) and PERMANOVA (Vegan v2.4.3); differentially abundant taxa via LEfSe (LDA > 2, P < 0.05) [35]; group-microbiota associations via db-RDA (Vegan v2.4.3, Monte Carlo permutations = 999); and co-occurrence networks via Spearman’s correlation (|r| > 0.5, P < 0.05) [36].

Statistical analysis

Data were presented as the mean ± standard error of mean. Statistical analyses were performed using SPSS version 22 (IBM Corp., NY, USA) and GraphPad Prism version 9 (GraphPad Software, CA, USA). One-way analysis of variance (ANOVA), followed by Tukey’s honestly significant difference post hoc test, was used for parametric data, whereas the Kruskal–Wallis H test with Dunn’s post hoc test was applied for nonparametric data. Normality assessments and logarithmic transformations were performed where appropriate [37]. A significance threshold of P < 0.05 was established.

Results

LGG combined with AI-2 alleviated intestinal histopathological damage and enhanced intestinal barrier function in mice with NEC

As depicted in Fig. 1a, the body weight was significantly lower in the NEC group than in the CON group (P < 0.0001), with no significant differences among the three treatment groups. As presented in Fig. 1b, the mortality rate was significantly higher in the NEC group than in the CON group (P < 0.01), with no notable discrepancies detected among the AI-2, LGG, and LGG + AI-2 groups. Morphologically, the distal ileum of mice in the NEC group exhibited pronounced gas accumulation compared with that in the CON group (Fig. 1c). Both AI-2 and LGG monotherapies ameliorated intestinal gas accumulation compared with that in the NEC group; the LGG + AI-2 combination showed a more substantial improvement than that in the NEC, AI-2, and LGG groups.

Fig. 1.

LGG + AI-2 improved intestinal pathology and barrier function in mice with NEC. (a) Changes in body weight of mice in each group (n = 10 per group). (b) Mortality rate of mice in each group (n = 20 per group). (c) Appearance and gas accumulation of intestinal tissue in mice of each group. (d) Hematoxylin–eosin staining results of terminal ileal tissues in mice of each group (n = 4 per group; original magnification, 100×). (e) NEC pathological scores of mice in each group (n = 4 per group). (f) Expression and quantification of ZO-1 protein in the intestines of mice in each group (n = 4 per group). Values are presented as the mean ± standard error of mean. Significance was tested using one-way ANOVA. AI-2, Autoinducer-2; LGG, Lactobacillus rhamnosus GG. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001

As presented in Fig. 1d and e, the histopathological analysis demonstrated that the mice in the NEC group had significantly higher intestinal histopathological scores than those in the CON group (P < 0.001). LGG monotherapy significantly reduced these scores (P < 0.05), with an even more pronounced decrease in the LGG + AI-2 combination group (P < 0.01) compared with the NEC group. As illustrated in Fig. 1f, ZO-1 protein expression was markedly lower in the NEC group than in the CON group (P < 0.05), but significantly elevated in the LGG + AI-2 group compared with the NEC, AI-2, and LGG groups (all P < 0.05). In summary, the LGG and AI-2 combination alleviated NEC-induced intestinal histopathological injury and restored barrier integrity, independent of significant effects on body weight or mortality.

LGG combined with AI-2 attenuated inflammation in mice with NEC and suppressed NF-κB activation to regulate cytokine balance

We next examined the inflammatory response by assessing the p-NF-κB/NF-κB ratio and cytokine levels (IL-6, TNF-α, and IL-10). As shown in Fig. 2a-b, total NF-κB protein levels remained unchanged across all groups whereas the p-NF-κB/NF-κB ratio was significantly higher in the NEC group than in the CON group (P < 0.01). AI-2, LGG, and their combination all significantly reduced this ratio compared with that in the NEC group (P < 0.05, P < 0.01, and P < 0.001, respectively), with the LGG and AI-2 combination group displaying the lowest expression among treatments, without significant divergence from the monotherapy groups. For inflammatory cytokines (Fig. 2c and e), the NEC group displayed markedly elevated levels of IL-6 (P < 0.0001) and TNF-α (P < 0.001), along with significantly reduced levels of IL-10 (P < 0.001), compared with the CON group. All interventions reversed these alterations, with the LGG + AI-2 group showing the most pronounced improvement, superior to that for either monotherapy. In summary, these data suggested that combined LGG and AI-2 exerted a more robust anti-inflammatory effect than individual interventions in the NEC model by inhibiting NF-κB activation and modulating cytokine balance.

Fig. 2.

Effects of LGG combined with AI-2 on intestinal inflammation in mice with NEC. (a-b) Expression and quantification of p-NF-κB/NF-κB ratio in the intestines of mice in each group (n = 4 per group). (c) Levels of IL-6 in the intestines of mice in each group (n = 8 per group). (d) Levels of TNF-α in the intestines of mice in each group (n = 8 per group). (e) Levels of IL-10 in the intestines of mice in each group (n = 8 per group). Values are presented as the mean ± standard error of mean. Significance was tested using one-way ANOVA. AI-2, Autoinducer-2; LGG, Lactobacillus rhamnosus GG. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001

LGG combined with AI-2 restored intestinal mucosal homeostasis in mice with NEC by promoting biofilm formation and beneficial bacterial colonization

SEM analysis revealed distinct mucosal and biofilm phenotypes across groups (Fig. 3a–e). The CON group displayed a continuous, well-organized mucosa with a dense, uniform sheet-like biofilm and abundant adherent commensal bacteria. In contrast, the NEC group exhibited severe villous fragmentation, disruption of the mucosal barrier, sparse biofilm formation, markedly reduced adherent bacteria (predominantly atypical cocci), and increased numbers of free-floating bacteria. Both AI-2 and LGG monotherapies ameliorated these abnormalities. AI-2 partially restored barrier continuity and biofilm density, whereas LGG further attenuated tissue damage and largely reestablished microecological homeostasis. The LGG + AI-2 combination achieved the most prominent repair, with mucosal structure approaching that in the CON group, a uniformly dense sheet-like biofilm, abundant adherent bacilli, and no free bacteria.

Fig. 3.

SEM analysis of intestinal mucosal structure, biofilm morphology, and bacterial colonization in each group. (a–e) Changes in biofilm and bacterial adhesion in the intestinal mucosa of mice in the CON, NEC, NEC + AI-2, NEC + LGG, and NEC + LGG + AI-2 groups (n = 3 per group). SEM images were captured at original magnifications of 2000× (low magnification; scale bar: 10 μm) and 10,000× (high magnification; scale bar: 1 μm). (f) Quantitative analysis of biofilm coverage area in all five groups. (g) Quantitative analysis of adherent bacterial density in the NEC, NEC + AI-2, NEC + LGG, and NEC + LGG + AI-2 groups. Values are presented as mean ± standard error of mean. Significance was tested using one-way ANOVA. AI-2, Autoinducer-2; LGG, Lactobacillus rhamnosus GG. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001

Quantitative image analysis using ImageJ software confirmed these observations. For biofilm coverage area (Fig. 3f), the NEC group showed significantly decreased coverage compared with the CON group (P < 0.0001). The coverage significantly increased in the AI-2, LGG, and LGG + AI-2 groups compared with the NEC group (all P < 0.0001), with the coverage in the combination group significantly exceeding that in both monotherapy groups (vs. AI-2, P < 0.01; vs. LGG, P < 0.0001). For adherent bacterial density (Fig. 3g), the CON group was excluded due to a dense, multilayered biofilm precluding accurate counting. Among the remaining four groups, the NEC group showed the lowest density. Compared with the NEC group, NEC + LGG and NEC + LGG + AI-2 groups significantly increased bacterial adhesion (P < 0.05 and P < 0.0001, respectively), and the combination further surpassed the NEC + AI-2 group (P < 0.001) and the NEC + LGG group (P < 0.05). These quantitative results confirmed that the combination of LGG and AI-2 restored intestinal mucosal homeostasis in NEC mice by promoting biofilm formation and beneficial bacterial colonization more effectively than either monotherapy.

LGG combined with AI-2 modulated gut microbiota structure, enhanced richness, and improved microbiota health in mice with NEC

First, we compared the CON and NEC groups. PCoA of 16S rRNA sequencing data revealed significant differences in microbial community structure between the CON and NEC groups (Fig. 4a; R = 0.9372, P = 0.001). Alpha diversity analysis showed that the NEC group had a significantly higher Shannon index (P < 0.01) and lower Simpson index (P < 0.01) than the CON group (Fig. 4b and c), whereas no significant difference was observed in the Chao index (Fig. 4d). The GMHI was significantly higher (P < 0.05) and the MDI was significantly lower (P < 0.001) in the CON group than in the NEC group (Fig. 4e and f), indicating a healthier microbiota profile in controls.

Fig. 4.

Gut microbiota structure, diversity, and health indices between CON and NEC groups. (a) PCoA analysis of intestinal flora between the CON and NEC groups. (b–d) Alpha diversity metrics (Shannon, Simpson, and Chao indices) in the CON and NEC groups. (e) Gut microbiome health index (GMHI) in the CON and NEC groups. (f) MDI in the CON and NEC groups (n = 7–10 per group). Significance was tested using the Wilcoxon rank-sum test. *P < 0.05, **P < 0.01, ***P < 0.001

We next compared the four intervention groups. PCoA revealed distinct clustering among these groups (Fig. 5a; R = 0.4763, P = 0.001). The LGG + AI-2 group exhibited significantly higher Chao, ACE, and sobs indices than the NEC (all P < 0.01) and AI-2 groups (P < 0.001, P < 0.01, and P < 0.001, respectively) (Fig. 5b and d). Although the Shannon index did not differ between the NEC and treatment groups, the LGG + AI-2 group displayed a significantly higher Shannon index than the AI-2 group (P < 0.01) (Fig. 5e). GMHI was significantly lower in the NEC group than in all treatment groups (P < 0.01), with the index in the LGG + AI-2 group surpassing that in both monotherapy groups (P < 0.01) (Fig. 5f and k). Consistently, the MDI was the lowest in the LGG + AI-2 group, significantly below that in the NEC, AI-2, and LGG groups and approaching the level in the CON group (all P < 0.01) (Fig. 5l). Collectively, these results suggested that gut microbiota dysbiosis contributed to NEC pathogenesis, and LGG combined with AI-2 improved microbial composition and diversity.

Fig. 5.

Gut microbiota structure, diversity, and health indices in the four intervention groups. (a) PCoA in the NEC, NEC + AI-2, NEC + LGG, and NEC + LGG + AI-2 groups. (b–e) Alpha diversity metrics (Chao, ACE, sobs, and Shannon indices) in the four groups. (f–k) GMHI in the four groups. (l) MDI in the four groups (n = 7–10 per group). Significance was tested using the Kruskal–Wallis H test. AI-2, Autoinducer-2; LGG, Lactobacillus rhamnosus GG. *P < 0.05, **P < 0.01, ***P < 0.001

LGG combined with AI-2 regulated species-level gut microbiota abundance and altered dominant taxa composition in mice with NEC

Species-level analysis revealed significant differences between the CON and NEC groups (Fig. 6a): Ligilactobacillus murinus and Rodentibacter pneumotropicus were significantly enriched in the CON group, whereas Escherichia coli and Klebsiella oxytoca predominated in the NEC group (P = 0.00038, P = 0.013; P = 0.00049, P = 0.00025, respectively), reflecting distinct gut microbiota structures. The species-level analysis revealed distinct compositional patterns across the four intervention groups (Fig. 6b). The abundance of E. coli was markedly elevated in the NEC group compared with the other three groups (P = 0.00039). The LGG + AI-2 group showed the highest abundance of Clostridium butyricum and Lactobacillus taiwanensis (P = 0.0012, P = 0.004, respectively), whereas LGG monotherapy significantly enriched Akkermansia muciniphila and Parabacteroides distasonis (P = 0.0014, P = 0.0046) and AI-2 alone predominantly increased the abundance of Ligilactobacillus murinus (P = 0.0011). These results indicated that different treatments significantly altered species-level gut microbiota composition.

Fig. 6.

Species-level disparities in gut microbiota and distribution patterns of dominant bacterial taxa. (a) Wilcoxon rank-sum test of gut microbiota at the species level between the control and NEC groups. (b) Kruskal–Wallis H analysis of gut microbiota at the species level among the NEC, NEC + AI-2, NEC + LGG, and NEC + LGG + AI-2 groups. (c-d) Application of LEfSe was employed to assess the differences in gut microbiota across five distinct groups (n = 7–10 per group). AI-2, Autoinducer-2; LGG, Lactobacillus rhamnosus GG

LEfSe analysis confirmed distinct microbiota composition between the CON and NEC groups (Fig. 6c). The CON group was dominated by Ligilactobacillus, L. murinus, Lactobacillaceae, and R. pneumotropicus, whereas the NEC group was characterized by Enterobacterales, Enterobacteriaceae, E. coli, and Klebsiella. LEfSe analysis also revealed distinct compositional patterns among the four intervention groups (Fig. 6d). The NEC group was characterized by E. coli, Enterobacteriaceae, and bacilli; the AI-2 group by Gammaproteobacteria, Pseudomonadota, and Ligilactobacillus/L. murinus; the LGG group by Akkermansia muciniphila and Verrucomicrobia; and the LGG + AI-2 group by Bacillota, Clostridia, Eubacteriales, C. butyricum, and L. taiwanensis. Distinct dominant taxa characterized each intervention group, sharply differentiating all treatment groups from the untreated NEC group and underscoring the specific modulatory effects of LGG, AI-2, and their combination on species-level microbiota architecture.

LGG combined with AI-2 regulated gut microbiota functional pathways and improved metabolic imbalance in mice with NEC

Pathway analysis revealed significant differences between the CON and NEC groups (Fig. 7a and b). At level 2, the CON group showed relative enrichment of pathways related to xenobiotic biodegradation and metabolism, lipid metabolism, cellular community–prokaryotes, metabolism of cofactors and vitamins, nucleotide metabolism, replication and repair, translation, and energy metabolism, indicating a healthier and more metabolically active microecology. At level 3, the CON group was enriched in purine metabolism, quorum sensing, amino sugar and nucleotide sugar metabolism, two-component system, ribosome, carbon metabolism, biosynthesis of amino acids, ABC transporters, microbial metabolism in diverse environments, and biosynthesis of secondary metabolites (Fig. 7b), suggesting more comprehensive metabolic and adaptive functions in the CON group. In contrast, the NEC group exhibited deficiencies in these functional pathways, contributing to NEC pathogenesis and microecological imbalance.

Fig. 7.

Differences in gut microbiota functional pathways. (a-b) PICRUSt predictive analysis of gut microbiota between the control and NEC groups. (c-d) PICRUSt predictive analysis of gut microbiota among the NEC, NEC + AI-2, NEC + LGG, and NEC + LGG + AI-2 groups

Pathway analysis also revealed significant differences among the four intervention groups (Fig. 7c-d). At level 2, the LGG + AI-2 group exhibited relative enrichment of lipid metabolism, cellular community–prokaryotes, replication and repair, nucleotide metabolism, translation, signal transduction, metabolism of cofactors and vitamins, and energy metabolism; both the LGG + AI-2 and AI-2 groups were enriched in xenobiotics biodegradation and metabolism, metabolism of other amino acids, and lipid metabolism compared with the NEC and LGG groups. At level 3, the LGG + AI-2 group was enriched in quorum sensing, two-component system, carbon metabolism, biosynthesis of amino acids, and ABC transporters (Fig. 7d), indicating enhanced functional coordination in microbial signaling, energy metabolism, and nutrient biosynthesis. These results suggested that the combined LGG and AI-2 intervention restored the predicted functional potential of the gut microbiota toward a healthier profile.

LGG combined with AI-2 regulated key differentially expressed genes and related pathways in mice with NEC

Transcriptomic analysis identified differentially expressed genes (DEGs) among the five groups using screening criteria of P < 0.05 and |log2FC| ≥ 1 (Fig. 8a). Venn diagram analysis of DEGs from the three pairwise comparisons (AI-2 vs. NEC, LGG vs. NEC, and LGG + AI-2 vs. NEC) revealed 803 DEGs associated with the combined intervention (Fig. 8b). Subsequently, Venn diagram analysis was carried out between these combined-action DEGs and the DEGs from the NEC group versus CON group comparison (Fig. 8c), leading to the identification of 131 key core DEGs. Enrichment analyses using GO, KEGG, and Reactome were conducted on the 131 DEGs. GO enrichment results indicated that the DEGs were mainly involved in pathways such as response to bacterium, xenobiotic metabolic process, response to external stimulus, and response to chemicals (Fig. 8d). KEGG enrichment covered pathways including metabolism of xenobiotics by cytochrome P450, drug metabolism – cytochrome P450, GSH metabolism, and fatty acid degradation (Fig. 8e). Reactome enrichment included pathways such as GSH conjugation, biological oxidations, phase II – conjugation of compounds, and digestion of dietary lipids (Fig. 8f). These findings suggested that the LGG + AI-2 combination alleviated NEC by regulating these pathways, thereby providing insights into NEC pathogenesis and potential intervention targets.

Fig. 8.

Transcriptomic analysis of intestinal tissues. (a) DEGs of intestinal tissues in mice from the five groups. (b) Venn diagram analysis of DEGs between the NEC + AI-2 group versus the NEC group, the NEC + LGG group versus the NEC group, and the NEC + LGG + AI-2 group versus the NEC group. (c) Venn diagram analysis of unique DEGs in the LGG + AI-2 group and DEGs identified in the NEC versus CON comparison (n = 8–10 per group). (d–f) GO, KEGG, and Reactome enrichment analyses of core DEGs

LGG combined with AI-2 alleviated oxidative stress in mice with NEC linked to GSH metabolism

Based on the transcriptomic identification of GSH metabolism pathways, the oxidative stress indicators were subsequently assessed to validate the findings. The GSH/GSSG ratio, and GSH, GSSG, and MDA levels were measured, whereas NRF2 and GPX4 levels were evaluated by Western blotting and qPCR. As shown in Fig. 9a, significant decreases in GSH levels, increases in GSSG levels, reductions in the GSH/GSSG ratio, and elevations in MDA levels were observed in the NEC group compared with the CON group (all P < 0.0001). These alterations were significantly reversed by all interventions (P < 0.0001), with the most pronounced improvement observed in the LGG + AI-2 group compared with the LGG and AI-2 monotherapy groups. As shown in Fig. 9b, decreased Nrf2 mRNA expression and NRF2 protein levels were detected in the NEC group compared with the CON group, whereas significantly higher NRF2 protein levels were observed in the LGG + AI-2 group compared with the NEC, AI-2, and LGG groups (P < 0.0001, P < 0.01, and P < 0.05, respectively). Similarly, significant reductions in GPX4 protein expression were induced in the NEC group compared with the CON group (P < 0.001). However, significant increases in GPX4 protein expression were observed in the LGG + AI-2 group compared with the NEC, AI-2, and LGG groups (all P < 0.0001). Further, the mRNA expression levels of Glutathione S-transferase family members (Gstm3, Gsta4, Gsta13, and Gstm1) were examined (Fig. 9c). Significantly increased mRNA expression of these four genes was observed in the NEC group compared with the CON group (P < 0.0001, P < 0.0001, P < 0.0001, and P < 0.001, respectively). The expression of these genes was inhibited to varying degrees by AI-2, LGG, or their combination. The lowest expression levels of all four genes were observed in the LGG + AI-2 group. Among these, significantly lower expression levels of Gstm3 and Gsta4 were detected in the LGG + AI-2 group than in the LGG group (P < 0.0001, P < 0.01, respectively), whereas significantly lower Gsta13 expression was observed in the LGG + AI-2 group than in the AI-2 and LGG groups (P < 0.001, P < 0.0001, respectively). In conclusion, these results demonstrated that the LGG + AI-2 combination effectively modulated GSH metabolism and alleviated oxidative stress in NEC, outperforming either monotherapy.

Fig. 9.

LGG combined with AI-2 alleviated oxidative stress in NEC by regulating GSH metabolism and the NRF2/GPX4 antioxidant pathway. (a) Intestinal levels of GSH, GSSG, GSH/GSSG ratio, and MDA in each group (n = 8 per group). (b) Relative mRNA expression of Nrf2 (n = 8 per group), Western blots, and relative protein quantification of NRF2 and GPX4 in intestinal tissues of mice in each group (n = 4 per group). (c) Relative mRNA expression of Gstm3, Gsta4, Gsta13, and Gstm1 in intestinal tissues of mice in each group (n = 8 per group). Values are presented as the mean ± standard error of mean. Significance was tested using one-way ANOVA. CON, Control group; NEC, necrotizing enterocolitis model group; NA, NEC + AI-2 group; NL, NEC + LGG group; NLA, NEC + LGG + AI-2 group; AI-2, Autoinducer-2; LGG, Lactobacillus rhamnosus GG. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001

Discussion

NEC continues to be a severe gastrointestinal condition affecting newborns, with its pathogenesis attributed to intestinal barrier disruption, inflammatory activation, gut microbiota imbalance, and oxidative stress [38, 39]. Our previous studies demonstrated that exogenous AI-2 alone attenuated NEC-associated intestinal inflammation by suppressing Toll-like receptor 4/NF-κB signaling [18]. Although these findings demonstrated that AI-2 alone mitigated inflammatory injury, a key question remained—whether AI-2 could act in concert with probiotics to augment protection beyond that achieved by AI-2 alone. In the present study, combined LGG and AI-2 administration markedly ameliorated NEC-induced intestinal injury. Concurrently, the combined intervention suppressed NF-κB activation and rebalanced the cytokine profile, lowering pro-inflammatory TNF-α and IL-6 levels while elevating anti-inflammatory IL-10 levels. These barrier-protective and anti-inflammatory effects aligned with the trends observed in the LGG and AI-2 monotherapy groups; however, they were consistently more pronounced following co-treatment, underscoring the additional protective benefit against NEC [40].

We integrated 16S rRNA sequencing, predictive functional metagenomics, and intestinal transcriptomics to examine the mechanisms underlying this enhanced protection. Previous studies have established that LGG monotherapy ameliorates NEC severity in neonatal models by attenuating TLR-mediated inflammation and reducing intestinal injury [41]. Concurrently, endogenous AI-2 signaling has been shown to protect against NEC through quorum-sensing-mediated regulation of microbial homeostasis [16]. Beyond LGG, other probiotics have also demonstrated efficacy against NEC. For instance, Limosilactobacillus reuteri exhibits antimicrobial and anti-inflammatory properties in a piglet NEC model [42], whereas multi-strain combinations containing Lactobacillus plantarum have shown protective effects in preterm infants [43]. The LuxS/AI-2 system was found to be crucial for Lactobacillus fermentum biofilm formation and multi-stress resistance, as luxS deletion compromised biofilm production, acid tolerance, bile salt tolerance, and hyperosmotic tolerance [44], highlighting the importance of AI-2 signaling for maintaining probiotic functionality under challenging gastrointestinal conditions. However, the combined mechanisms of LGG + AI-2—especially whether they converge to regulate microbial architecture, metabolic function, and host transcriptional programs—remain unelucidated.

At the microbial level, LGG + AI-2 uniquely reshaped the gut microbiota architecture, enriching beneficial commensals such as Clostridium butyricum—a butyrate-producing species implicated in epithelial repair—while suppressing the growth of pathogenic E. coli, as reflected by the most favorable GMHI and the lowest MDI among all intervention groups [45]. Although the role of AI-2 in promoting the growth of C. butyricum and facilitating LGG colonization has not been previously documented, its contribution to the protective effect and direct quorum-sensing regulation warrant future mechanistic investigation. SEM and quantitative image analysis further revealed that AI-2 markedly enhanced biofilm formation and mucosal colonization, suggesting that quorum-sensing facilitation of probiotic engraftment constitutes a key mechanistic distinction of the combined strategy [19, 46]. Metabolic pathway prediction via PICRUSt2 indicated that the co-treatment restored functions related to lipid metabolism, xenobiotics biodegradation, amino acid biosynthesis, and quorum sensing, trending toward the healthy control profile [47, 48]. These microbial and metabolic alterations were paralleled by host transcriptomic signatures: among 131 core DEGs uniquely regulated by the combined intervention, GSH metabolism, xenobiotic metabolism, and oxidative stress response pathways were prominently enriched. Collectively, these multi-omic findings suggested that LGG + AI-2 exerted protective effects through a coordinated axis of microbiota remodeling, metabolic reprogramming, and host transcriptional rewiring [49]—a convergent mechanism not fully achieved by either monotherapy.

Guided by the transcriptomic enrichment of GSH metabolism, we biochemically validated oxidative stress as a key downstream node mediating the combined protective effect. Oxidative stress is a central pathogenic driver in NEC, exacerbating intestinal mucosal injury and disrupting microbial balance [50]. In this context, NRF2 dysfunction and GPX4 depletion compromise epithelial antioxidant capacity and increase susceptibility to injury [51]. The combination of LGG with AI-2 restored the GSH/GSSG ratio, reduced MDA accumulation, and upregulated the antioxidant regulators NRF2 and GPX4 [52]. Additionally, compensatory elevations of Glutathione S-transferase family members (Gstm3, Gsta4, Gsta13, and Gstm1) observed in NEC were suppressed by co-treatment, consistent with attenuated ROS burden [53, 54]. These results indicated that the alleviation of intestinal oxidative stress, mediated through GSH-dependent antioxidant pathways, represented a convergent mechanism by which LGG + AI-2 protected against NEC injury. Although LGG + AI-2 demonstrated potent antioxidant efficacy in our mouse model, translation to humans is challenged by the complex gastrointestinal milieu, where prolonged transit, alkaline bile, and diverse microbiota may compromise AI-2 activity. Green nanotechnology offers promising complementary approaches, as biosynthesized nanoparticles exhibit strong antioxidant and redox-modulating activities. For instance, selenium nanoparticles from potato peels showed antioxidant and antimicrobial effects [55], palladium and platinum nanoparticles from mung bean demonstrated sustainable bioactivity [56], and ZnS nanoparticles from banana peels displayed photocatalytic and redox properties [57]. These eco-friendly platforms could inspire AI-2-encapsulated delivery systems for sustained intestinal release, preserving quorum-sensing activity to augment LGG colonization [58] and antioxidant protection against NEC-induced oxidative injury.

Despite these promising findings, this study had several limitations: (i) long-term outcomes, sex differences, and mouse model constraints remain unaddressed [59, 60]; (ii) PICRUSt2 predictions require metabolomic validation [61]; (iii) AI-2 safety and clinical application ethics of neonates need further evaluation; and (iv) definitive synergism requires larger sample sizes [62]. Future studies should address these limitations to facilitate clinical translation.

Conclusion

LGG combined with AI-2 protected against NEC by improving intestinal barrier function, attenuating inflammation, restoring microbiota homeostasis, and alleviating oxidative stress via GSH metabolism, thereby providing a basis for clinical translation.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

AI-2

Autoinducer-2

CFU

Colony forming units

DEGs

Differentially expressed genes

DPD

S-4,5-dihydroxypentane-2,3-dione

ECL

Enhanced chemiluminescence

ELISA

Enzyme-linked immunosorbent assay

GMHI

Gut microbiome health index

GO

Gene Ontology

GPX4

Glutathione peroxidase 4

GSH

Glutathione

GSSG

Oxidized glutathione

GST

Glutathione S-transferase

HRP

Horseradish peroxidase

IL-6

Interleukin-6

IL-10

Interleukin-10

KEGG

Kyoto Encyclopedia of Genes and Genomes

LEfSe

Linear discriminant analysis Effect Size

LGG

Lactobacillus rhamnosus GG

LuxS

S-ribosylhomocysteine lyase

MDA

Malondialdehyde

MDI

Microbial dysbiosis index

NEC

Necrotizing enterocolitis

NF-κB

Nuclear factor-kappa B

NRF2

Nuclear factor erythroid 2-related factor 2

PCoA

Principal Co-ordinates Analysis

PICRUSt2

Phylogenetic Investigation of Communities by Reconstruction of Unobserved States 2

p-NF-κB

Phosphorylated nuclear factor-kappa B

PMSF

Phenylmethylsulfonyl fluoride

PVDF

Polyvinylidene difluoride

qPCR

Quantitative polymerase chain reaction

RIPA

Radio immunoprecipitation assay lysis buffer

SDS-PAGE

Sodium dodecyl sulfate polyacrylamide gel electrophoresis

SEM

Scanning electron microscopy

TNF-α

Tumor necrosis factor-α

ZO-1

Zona occludens-1

Author contributions

JC and Z-LW secured funding for the study, proposed the research ideas, designed the experiments, and supervised the entire study. R-QH performed the experiments and collected the data. YY, TY, FL, X-WH, and BT provided technical support throughout the experiments and conducted data analyses. R-QH drafted the manuscript. JC reviewed and revised the manuscript. All authors read and approved the final version of the manuscript.

Funding

This study was funded by the Natural Science Foundation of Chongqing Municipality (Grant No. CSTB2023NSCQ-MSX0178), the Jiangxi Provincial Natural Science Foundation (Grant No. 20242BAB23079), and the National Natural Science Foundation of China (Grant No. 82372559).

Data availability

The data used in this study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

This study was conducted according to the Guidelines for the Laboratory Animal Use and Care Committee of the Ministry of Health, China, and the Animal Research Ethics Committee of the Affiliated Children’s Hospital of Chongqing Medical University (No. 20231013001).

Consent for publication

Not applicable.

Conflict of interest

The authors declare no conflicts of interest.