L. paragasseri HM018 is derived from healthy breast milk and is presently conserved at the China General Microbiological Culture Collection Center (CGMCC No. 19749). L. paragasseri HM018, isolated from breast milk, was preserved at the National Engineering Research Center for Dairy Health (Beijing, China). The strain was inoculated in De Man–Rogosa–Sharpe broth (Luqiao Co., Ltd., Beijing, China) at 37°C for 18 h and subcultured twice at a 1% inoculation rate to ensure bacterial activity. Cellular biomass was harvested via refrigerated centrifugation (4°C, 6,000 rpm, 15 min). The supernatant was removed, and the pellet was washed twice with sterile 0.01 M phosphate-buffered saline (PBS; 0.01 M concentration, sterile filtered, Solarbio). Finally, the bacterial suspension was adjusted to low-dose (2.5 × 10⁸ colony-forming units [CFU] mL-1), medium-dose (5 × 10⁸ CFU mL-1), and high-dose (1.5 × 10⁹ CFU mL-1) concentrations for oral gavage.

All animal procedures adhered to the Guidelines for the Care and Use of Laboratory Animals established by Beijing Union University andreceived approval from the Animal Ethics Committee of Beijing Union University (JCZX11-2306-1, Beijing, China).

Fifty adult male Sprague–Dawley rats (body weight 200 ± 20 g; Vital River Laboratories, Beijing, China), raised under specific pathogen-free (SPF) conditions, were acclimatized for one week under standard conditions with a 12-h light/dark cycle. Throughout the acclimatization period, food and water were freely available. Subsequently, the rats were randomly assigned to five experimental groups as follows: (1) a normal diet (ND) control group receiving standard chow and 2 mL of sterile PBS; (2) an HFD model group receiving an HFD and 2 mL of sterile PBS; and three treatment groups (3–5) receiving the HFD supplemented with varying concentrations of L. paragasseri HM018: (3) the low-dose group (LPL) received 2 mL of a 2.5 × 10⁸ CFU/mL suspension; (4) the medium-dose group received 2 mL of a 5 × 10⁸ CFU/mL suspension; and (5) the high-dose group (LPH) received 2 mL of a 1.5 × 10⁹ CFU/mL suspension, all administered via gavage. All experimental diets were provided by Ke’ao Xieli Feed Co., Ltd. (Beijing, China).

After six weeks of intervention, following a 12-h fast after the final gavage, fecal samples were collected and incubated at 25 ± 1°C for 60 min. Subsequently, the rats were euthanized, and blood samples were collected, and Serum was obtained through centrifugation of blood samples at 3,000 × g for 15 min at 4°C. Tissue specimens (liver, epididymal adipose, ileal segments) were rapidly cryopreserved in liquid nitrogen vapor phase and maintained at −80°C.

Serum biomarkers, including TG, TC, LDL-C, HDL-C, glucose, and glycated hemoglobin (GHb), were quantified using commercial assay kits (Nanjing Jiancheng Bioengineering Institute, China) following manufacturer protocols.

Liver tissue samples from rats were fixed in 4% paraformaldehyde at room temperature for 24 h. Subsequently, the samples were dehydrated, embedded in paraffin, and sectioned into 5 μm slices. After dewaxing, the tissue sections were stained with hematoxylin and eosin (H&E) for histological observation.

Using the OMEGA Mag-bind DNA Kit, genomic DNA was isolated from samples. The V3-V4 region of the 16S rRNA gene was then amplified with specific barcoded primers 341F (CCTAYGGGRBGCASCAG) and 806R (GGACTACNNGGGTATCTAAT). Polymerase chain reaction (PCR) products were analyzed using 2% agarose gel electrophoresis and subsequently purified using a Quant-iT PicoGreen dsDNA Assay Kit. Based on the preliminary quantification results from electrophoresis, the recovered PCR products were quantified using a fluorescence quantification system on a microplate reader (BioTek, FLx800). Library construction was performed using an Illumina TruSeq Nano DNA LT Library Prep Kit. The quality of the constructed libraries was assessed using an Agilent Bioanalyzer 2,100 and Promega QuantiFluor, followed by sequencing. Raw sequencing data were processed to remove primer sequences using Cutadapt, and quality control and annotation were performed using QIIME 2 with the SILVA database (version 138). Alpha and beta diversity indices were calculated using QIIME 2.

Fecal samples were collected from 10 rats per group (ND, HFD, and LPH groups; n = 30 total). Samples underwent liquid nitrogen flash-freezing and were maintained at −80°C Untargeted metabolomic analysis was performed at Shanghai Applied Protein Technology Co., Ltd. (Shanghai, China) using an ultra-high-performance liquid chromatography (UHPLC) system (1,290 Infinity II, Agilent Technologies) coupled with a quadrupole time-of-flight mass spectrometer (AB Sciex TripleTOF 6,600).

Fecal samples (50 mg) were homogenized with 400 μL methanol/acetonitrile (1:1, v/v) via vortex mixing, followed by sequential processing: 30-min sonication at 5°C (40 kHz), 30-min incubation at-20°C, and centrifugation (1,400 × g, 20 min, 4°C). The resulting supernatant was vacuum-dried and reconstituted in 120 μL acetonitrile/water (1:1, v/v). For quality assurance, 10 μL QC sample was introduced prior to final 5-min sonication/centrifugation, with processed supernatant ultimately transferred to LC–MS vials for analysis.

Chromatographic separation was performed on a Waters Acquity UPLC BEH Amide column (100 mm × 2.1 mm, 1.7 μm) employing hydrophilic interaction liquid chromatography (HILIC). The mobile phase consisted of (A) 25 mM ammonium acetate/ammonium hydroxide aqueous solution and (B) acetonitrile, with the following gradient profile: initial 95% B (0–0.5 min), linear gradient to 65% B (0.5–7.5 min), stepwise reduction to 40% B (7.5–7.6 min), isocratic hold (7.6–8.6 min), immediate return to 95% B (8.6–8.7 min), and 3-min column re-equilibration (8.7–11.7 min). The chromatographic system was maintained at 0.3 mL/min flow rate and 40°C.

Instrument parameters were configured as follows: ion source gases (60 psi), curtain gas (30 psi), 600°C source temperature, ±5,500 V ion spray voltage. The method employed 35 V (±15 eV) collision energy, ±60 V declustering potential, 4 Da isotope exclusion, and 10-cycle candidate ion-monitoring acquisition. Raw data were converted into mzXML format using ProteoWizard MSConvert. Peak detection, alignment, and annotation were performed using the XCMS workflow, with metabolites identified by matching accurate mass (m/z error <10 ppm) and MS/MS spectra against an in-house reference database.

Total RNA was extracted from the liver tissues of rats in the three groups (ND, HFD, and LPH) using the TRIzol reagent. RNA purity was evaluated by measuring the A260/A280 ratio using a Nanodrop ND-2000 spectrophotometer (Thermo Scientific, USA), and RNA integrity was assessed using an Agilent Bioanalyzer 4,150 system (Agilent Technologies, CA, USA). Subsequently, paired-end libraries were prepared using the ABclonal mRNA-seq Lib Prep Kit (ABclonal, China) following manufacturer protocols. Sequencing was conducted across dual platforms: Illumina Novaseq 6,000 and MGISEQ-T7 systems.

Sequencing reads were preprocessed using Trimmomatic (adapter removal/quality trimming). Genome alignment was performed using HISAT2. DESeq2 (v1.40.2) implemented in R was used to determine differentially expressed genes (criteria: |log2FC| > 1, FDR < 0.05). Enrichment analysis was performed using GO and KEGG pathways (clusterProfiler v4.10.0). Statistical significance was set at p < 0.05.

Rat liver and ileum tissue samples were collected. Total RNA was extracted using RNA-easy Isolation Reagent (R701-01, Vazyme, China), and RNA was reverse-transcribed into cDNA using the HiScript II Q RT SuperMix for qPCR (+gDNA wiper) (R223-01, Vazyme, China) kit. Subsequently, using ChamQ SYBR Master Mix (Q311-02, Vazyme, China) as the reaction system, real-time fluorescence quantitative PCR (RT-PCR) was carried out on an Applied Biosystems 7,500 Real-Time PCR System to detect the expression levels of the target genes (Abcg5, Abcg8, Socs2, and Cish). The primer sequences are as follows: for Abcg5, the forward primer is 5′-GTCCTTCAGCGTCAGCAACC-3′ and the reverse primer is 5′ -ATGGTCTGGCCACTCTCGAT-3′; for Abcg8, the forward primer is 5′-CACCCTAGACTCTAACTCCA-3′ and the reverse primer is 5′-GGAGCACTGGATAGTATTGG-3′ (Jang et al., 2022); for Socs2, the forward primer is 5′-GAACCACGCTGTCAAACT-3′ and the reverse primer is 5′-CTCCCACTCAGACTACCTATT-3′ (Dang et al., 2020); for Cish, the forward primer is 5′- TACCTCCGGGGATCTGGTTG −3′ and the reverse primer is 5′- CACGGGTGGTTTTGACTGAC -3′. Gadph (forward primer: 5′-GAAACCTGCCAAGTATGA-3′, reverse primer: 5′-GCTGTAGCCGTATTCATT-3′) was used as the internal reference gene for normalization analysis.

Graphical representations, including histograms, box plots, and heatmaps, were generated using Origin software (Version 2024, OriginLab, MA, USA). Statistical analyses were conducted using SPSS (Version 27.0.1, IBM, USA), employing one-way analysis of variance followed by Tukey’s post-hoc test for multiple comparisons. Data are presented as mean ± standard deviation (SD), with statistical significance set at p < 0.05. Additionally, bivariate correlations were evaluated using Spearman’s rank-order correlation method.

After 6 weeks on a high-fat diet, serum TG and TC levels were notably increased in HFD group. In the probiotic intervention groups with varying doses, these parameters showed dose-dependent improvements. Additionally, the HFD group demonstrated significantly elevated LDL-cholesterol levels compared to the ND group. But was significantly reduced after probiotic intervention. However, HDL-C level, which is associated with lipid utilization, significantly decreased after HFD, and there was no change in its concentration after the intervention with HM018. Compared to the ND group, body weight, fasting glucose, and GHb levels showed no statistically significant variations across the experimental cohorts (Supplementary Figure S1). Histological analysis showed that HFD led to fat accumulation, and the ratio of the area of liver adipocytes in the HFD group was significantly higher than that in other groups. In summary, the abnormal elevation of key lipid markers (TG, TC, and LDL-C) confirmed the successful establishment of the HFD-induced rat model, while probiotic intervention significantly ameliorated dyslipidemia (Figure 1).

To elucidate the impact of HM018 supplementation on intestinal microbiome composition, 16S rRNA gene sequencing was conducted across thirty fecal specimens. A total of 3,914 amplicon sequence variants (ASVs) were obtained, with 1,189, 801, and 2,168 ASVs identified in the HFD model, experimental LPH, and control ND groups, respectively. The numbers of unique ASVs in the ND, HFD, and LPH groups were 1,423, 359, and 218, respectively (Supplementary Figure S2A).

Principal coordinates analysis revealed a significant separation in the gut microbiota structure between the HFD (green) and ND (red) groups after HFD feeding. Following HM018 intervention, the LPH group (blue) also exhibited a distinct microbiota composition compared to the HFD group. These results demonstrate significant HFD-induced perturbations in gut microbiota composition and that HM018 intervention modulated these changes (Figure 2A).

Alpha diversity analysis revealed that the HFD significantly reduced microbial richness (Chao1 and ACE indices) in the HFD group, and this reduction was further exacerbated after probiotic intervention (Figure 2B; Supplementary Figure S2B). While Shannon and Simpson diversity indices showed no statistically significant variations in the HFD cohort (p > 0.05), both metrics demonstrated consistent decreasing trends. In the dose-dependent analysis, the Chao1 and ACE indices gradually decreased with increasing probiotic dosage, whereas the Shannon and Simpson indices significantly increased (Figure 2C, Supplementary Figure S2C).

Phylum-level analysis identified five predominant bacterial taxa: Firmicutes, Bacteroidetes, Verrucomicrobia, Proteobacteria, and Actinobacteria. Compared with the ND and LPH groups, the relative abundance of Firmicutes was significantly increased in the HFD group, whereas that of Bacteroidetes was significantly decreased. The HFD significantly elevated the Firmicutes-to-Bacteroidetes ratio, which was significantly reduced after HM018 intervention (Figures 2D,E).

At the genus level, the heatmap analysis clustered the microbiota into three groups. Muribaculaceae, Clostridia_UCG-014, and Lachnospiraceae_NK4A136_group were the dominant genera in the ND group. Whereas, HFD significantly suppressed the abundance of Muribaculaceae and significantly increased the abundances of Blautia, Hydrogenophaga, Subdoligranulum, Romboutsia, Akkermansia, and [Eubacterium]_coprostanoligenes_group. In the LPH group, the abundances of Parasutterella, Bacteroides, Parabacteroides, and Prevotellaceae_UCG − 001 significantly increased after HM018 intervention (p < 0.05). In addition, UBA1819, Eisenbergiella, Lachnoclostridium, Roseburia, and Phascolarctobacterium abundances also increased (p > 0.05) (Figure 2F).

To delineate gut microbiota-metabolite interactions in hyperlipidemic models, we implemented LC–MS-based untargeted metabolomics profiling across three experimental cohorts (ND, HFD and LPH) Partial least squares-discriminant analysis (PLS-DA) revealed distinct clustering among the three groups, with significant separation between the LPH and HFD groups after HM018 intervention (Figure 3A). Comparative metabolomic analysis identified 1,134 differentially expressed metabolites (DEMs) in the HFD group relative to ND controls, including 288 downregulated and 847 upregulated metabolites. In contrast, the LPH group showed 166 upregulated and 90 downregulated metabolites compared to the HFD group (Figure 3B).

KEGG pathway enrichment analysis indicated that the differential metabolites were significantly enriched in pathways such as bile secretion, nucleotide metabolism, vitamin digestion and absorption, and thiamine metabolism (Figure 3C). Key metabolites involved in bile secretion included morphine-3-glucuronide, pravastatin, 1-methylnicotinamide, L-carnitine, fexofenadine, deoxycholic acid, quinine, hydrocortisone, and levofloxacin (Figure 3D). Additionally, bile acid homeostasis was significantly altered across the three groups: total bile acid levels were markedly elevated after HFD feeding but significantly reduced following HM018 intervention (Figure 3E). Compared to the HFD group, the LPH group exhibited significant downregulation of free bile acids, such as lithocholic acid (LCA), hyodeoxycholic acid (HDCA), chenodeoxycholate (CDCA), ursodeoxycholic acid (UDCA), and deoxycholic acid (DCA), and upregulation of conjugated bile acids such as taurocholic acid (TCA), glycine-conjugated cholic acid (GCA), taurochenodeoxycholic acid (TCDCA), taurodeoxycholic acid (TDCA), and tauroursodeoxycholic acid (TUDCA) (Figure 3F).

To elucidate host gene regulation by HM018-derived metabolites, we implemented transcriptomic sequencing of ileal specimens from three rat groups (15 samples total). Transcriptomic PCA clustering showed LPH group profiles were more similar to ND than HFD (Figure 4A). The LPH-HFD comparison yielded 213 significant DEGs, including 117 upregulated and 96 downregulated genes (Figure 4B).

KEGG pathway enrichment analysis demonstrated that the upregulated genes were significantly enriched in cholesterol metabolism, linoleic acid metabolism, fat digestion and absorption, vitamin digestion and absorption, arachidonic acid metabolism, steroid hormone biosynthesis, carbohydrate digestion and absorption, retinol metabolism, PPAR signaling pathway, and glutathione metabolism (Figure 4C). Notably, the cholesterol efflux-related genes Abcg5 and Abcg8 were markedly upregulated post-intervention, and this result was consistently verified by qPCR (Figure 4D; Supplementary Figure S3). Additionally, key regulators of lipid metabolism, including HDL-associated Apoa1 and intestinal lipid transporter Apoa4, demonstrated significant transcriptional upregulation (Figure 4E).

Downregulated genes were primarily enriched in pathways such as systemic lupus erythematosus, transcriptional dysregulation in cancer, alcoholism, and neutrophil extracellular trap formation (Supplementary Figure S3). To evaluate intestinal barrier integrity, we analyzed the expression of tight junction components (ZO-1, Claudin-4, Occludin) and the goblet cell marker Muc2. Both Occludin and Muc2 were significantly upregulated after HM018 intervention (Figure 4F).

To explore the effect of HM018 on rat livers, we performed RNA-seq analysis on liver samples from three groups: ND, LPH, and HFD. PCA revealed that the LPH and HFD groups exhibited more dispersed distributions than the ND group, indicating that both HFD and HM018 intervention significantly altered the hepatic transcriptomic profiles (Figure 5A). Further analysis identified 24 upregulated and 35 downregulated genes in the LPH group compared to those in the HFD group (Figure 5B). Additionally, Pathway enrichment profiling identified key lipid metabolism-related signaling cascades: JAK–STAT, TGF-β, FoxO transcriptional regulation, insulin signaling, adipocytokine communication, and diabetes pathways (Figure 5C; Supplementary Figure S4).

Among the genes related to lipogenesis, although not statistically significant, their transcriptional levels generally showed a downward trend after HM018 intervention (Supplementary Figure S4). Meanwhile, analysis of glucose metabolism-related genes revealed that HM018 intervention significantly upregulated glucose-6-phosphatase (G6pc) expression, while downregulating suppressor of cytokine signaling 2/3 (Socs2 and Socs3) and cytokine-inducible SH2-containing protein (Cish). Quantitative PCR (QPCR) results further confirmed that the expression of Socs2 and Cish in the LPH group was significantly reduced, validating this regulatory trend (Figure 5D; Supplementary Figure S4). Notably, HM018 intervention restored the expression of BMP and activin membrane-bound inhibitor (Bambi), growth arrest and DNA damage-inducible alpha (Gadd45a), and Apoa1 to levels comparable to those in the ND group. In contrast, Apob (low-density lipoprotein metabolism regulator) exhibited no significant expression differences across experimental groups (Figure 5E). These results suggest that HM018 may ameliorate hepatic metabolic disorders by multi-target regulation of key genes involved in glucose and lipid metabolism.

We first conducted a joint analysis of the four significantly altered bacterial phyla (Firmicutes, Bacteroidetes, Verrucomicrobia, and Proteobacteria) post-HM018 intervention using lipid profiles across the three experimental groups. The results showed that, except for Bacteroidetes, the relative abundances of Firmicutes, Verrucomicrobia, and Proteobacteria were inversely related to HDL levels, while demonstrating positive correlations with triglyceride, total cholesterol, and LDL levels (Supplementary Figure S5A). Genus-level analysis identified positive correlations between Blautia, Romboutsia, Subdoligranulum abundances and dyslipidemia severity scores, whereas Muribaculaceae, Incertae-Sedis, Lachnospiraceae_NK4A136_group, Ruminococcus, and Prevotellaceae_UCG-001 were significantly correlated with improved hyperlipidemia (Figure 6A).

In the joint analysis of differential metabolites and lipid parameters, β-sitosterol, eicosenoic acid, and N-palmitoyl-D-sphingosine were strongly negatively correlated with TG, TC, and LDL (r < −0.7; p < 0.05), but strongly positively correlated with HDL (r > 0.7; p < 0.05). Increased concentrations of these metabolites were associated with better prognostic outcomes, whereas other metabolites were significantly enriched in hypercholesterolemic rats (|r| > 0.7) (Supplementary Figure S5B).

A combined analysis of differential metabolites and bacterial genera demonstrated that Muribaculaceae, a dominant genus in the ND group, showed strong positive correlations with eicosenoic acid, β-sitosterol, and N-palmitoyl-D-sphingosine, but negative correlations with other compounds (|r| > 0.7). In the intervention group, the increased abundance of Prevotellaceae_UCG-001 weakly correlated with eicosenoic acid and ω-3 arachidonic acid methyl ester (0.5 > |r| > 0.3). Notably, the dominant genera in the model group, Blautia and Subdoligranulum, displayed moderate negative correlations with β-sitosterol (−0.6 > r > −0.7) and positive correlations with compounds such as empenthrin (r > 0.3) (Figure 6B).

Furthermore, HM018 intervention significantly altered the composition of the bile acid pool in rats. Joint analysis of differential bile acids and their derivatives with bacterial genera revealed that Muribaculaceae was negatively correlated with most bile acids, whereas Blautia, Subdoligranulum, and Romboutsia showed significant positive correlations with bile acids (Figure 6C). These findings suggest that HM018 may ameliorate dyslipidemia by modulating the interaction between the gut microbiota and bile acid metabolism.