Cell line GL261 was bought from Hunan Fenghui Biotechnology Co. Ltd., China. and cultured in DMEM media (Gibco, MA, United States) containing 10% fetal bovine serum (Gibco, MA, United States) and 1% penicillin/streptomycin (Invitrogen, MA, United States) at 37°C, 5%CO₂.

B. breve (BBR-15), B. longum (JBLC-141), B. lactis (JYBR-190), and B. bifidum (JYBB-163) freeze-dried powders were procured from Shandong Zhongke Jiayi Biological Engineering Co. Ltd., China. A concentration of 4 × 10⁹ CFU/0.4 mL (equivalent to 1 × 10⁹ CFU for each Bifidobacterium) was attained by suspending These four types of Bifidobacterium in sterile saline (Wang et al., 2022). The bacterial identification results are presented in Supplementary Tables S1–S5 (Supplementary sequences S1–S4).

Female C57BL/6 mice (4 weeks old) were procured from Chengdu Dashuo Laboratory Animal Company (Chengdu, China) and were maintained at a temperature of 22°C ± 2°C under a 12 h:12 h light–dark cycle within specific pathogen-free facilities. All animal experiments adhered to the ethical regulations stipulated by the Experimental Animal Administration of Sichuan University. Three sets of C57BL/6 mice were employed in this study: The first set was allocated for the construction of a survival curve (n = 6), the second set for indicator detection (n = 12), and the third set for the investigation of liver and kidney function (n = 8).

For survival curve analysis, the first set of mice were distributed into groups at random: the Model (Mod) group and the Model-Bifidobacterium (Mod-BIF) group. The Mod-BIF group received daily gavages of a Bifidobacterium mixture at a dose of 4 × 10⁹ CFU/400 μL (with each dose containing 1 × 10⁹ CFU) from day −14 to day 30 (Wang et al., 2022). The Mod group received a daily equal-volume saline solution from day −14 to day 30. For tumor induction, anesthetized mice of both groups were stereotactically injected with 1 × 10⁵ GL261 cells (1 μL) into the right striatum at a rate of 0.4 μL/min on day 0. The stereotaxic coordinates with respect to the bregma were as follows: 1.0 mm anterior, +1.5 mm lateral, and 3.5 mm ventral (Lam et al., 2018).

For indicator detection, the second set of mice were distributed into groups at random: the Mod group and the Mod-BIF group. The Mod-BIF group received daily gavages of a Bifidobacterium mixture at a dose of 4 × 10⁹ CFU/400 μL (with each dose containing 1 × 10⁹ CFU) from day −14 to day 30. The Mod group received a daily equal-volume saline solution from day −14 to day 30. For tumor induction, anesthetized mice of both groups were stereotactically injected with 1 × 10⁵ GL261 cells (1 μL) into the right striatum at a rate of 0.4 μL/min on day 0. The stereotaxic coordinates with respect to the bregma were as follows: 1.0 mm anterior, +1.5 mm lateral, and 3.5 mm ventral. The mice were euthanized 30 days after the inoculation.

For the investigation of liver and kidney function, the third set mice were distributed into groups at random: the control (Con) group, the Mod group and the Mod-BIF group. The Mod-BIF group received daily gavages of a Bifidobacterium mixture at a dose of 4 × 10⁹ CFU/400 μL (with each dose containing 1 × 10⁹ CFU) from day −14 to day 30. The Mod group received a daily equal-volume saline solution from day −14 to day 30. For tumor induction, anesthetized mice of the Mod and Mod-BIF groups were stereotactically injected with 1 × 10⁵ GL261 cells (1 μL) into the right striatum at a rate of 0.4 μL/min on day 0. The stereotaxic coordinates with respect to the bregma were as follows: 1.0 mm anterior, +1.5 mm lateral, and 3.5 mm ventral. The mice were euthanized 30 days after the inoculation. The mice of the Con group did not receive any treatment.

Cerebral MRI was conducted in vivo on a 7 T MRI scanner designed for small animals (Bruker BioSpec 70/30, Germany). 2% isoflurane was inhaled to induce anesthesia in all of the mice. Imaging was performed utilizing a 2D T2-weighted sequence in the axial orientation with the following parameters: repetition time = 2,625 ms, echo time = 33 ms, number of averages = 4, matrix = 256 × 256, and slice thickness = 0.6 mm. The ITK-SNAP program was utilized to calculate tumor volumes.

Fresh fecal pellets were collected and immediately frozen at −80°C on day 30 post-inoculation. Following the experiments, 1% pentobarbital sodium (40 mg/kg) solution was used to anesthetize all mouse. All mice were subsequently subjected to intracardial perfusion with cold, sterilized saline. Tissues from the brain, liver, kidney, ileum, and colon were harvested. Some portions of these tissues were snap-frozen in liquid nitrogen and stored at −80°C, while others were fixed using a 4% paraformaldehyde solution and subsequently embedded in paraffin.

Serum was obtained from mice by centrifugation at 3,500 rpm for 10 min, following a 30-min incubation in a 37°C oven. Serum levels of AST, ALT, BUN, and Cr were assessed using respective kits as per the manufacturer’s instructions (Nanjing Jiancheng Bioengineering Institute, China).

The paraffin-embedded tissues were sectioned into 3-μm-thick slides. According to standard histological protocols, hematoxylin and eosin (HE) staining was conducted. For immunohistochemistry (IHC), the following primary antibodies were used to incubate the slides, respectively, after blocking: bacterial lipopolysaccharide (LPS; HycultBiotech #HM6011, Netherlands), lipoteichoic acid (LTA; Santa Cruz, #sc-57752, United States), occludin (Servicebio #GB111401, China), and zonula occludens 1 (ZO-1; Servicebio #GB111981, China). IHC quantification involved the calculation of positive areas (occludin and ZO-1) using Aipathwell software (Version 2.0, Servicebio, China).

The total RNA was isolated using TRIzol reagent (Thermo Fisher Scientific, MA, United States) and subsequently reverse-transcribed using the RevertAid First Strand cDNA Synthesis Kit (Thermo Fisher Scientific). Real-time polymerase chain reaction was conducted according to established protocols from prior studies, and data were analyzed utilizing the 2^−ΔΔCT method. Primer sequences for the target genes are listed in Table 1.

Total proteins were extracted using RIPA buffer (Beyotime, Shanghai, China) according to the manufacturer’s instructions. The protein lysate was subjected to electrophoresis and transferred onto a PVDF membrane (Bio-Rad, CA, United States). The following primary antibodies sourced from CST (Shanghai, China): p-MEK (Ser217/221; 1:1000; 9154), MEK (1:1000; 8727), p-ERK1/2 (Thr202/Tyr204; 1:2000; 4370), and ERK (1:1000; 4695), PD-1 (1:1000; 84651), PD-L1 (1:1000; 60475), and p-NF-κB (1:1000; bs-0982R) from Bioss (Beijing, China), NF-κB (1:1000; sc-8008) from Santa Cruz Biotechnology (Texas, United States) were used to incubate the membrane, respectively. The proteins were detected using ECL Plus Reagent (Beyotime) and quantified using Image Lab software.

RNA-seq analysis was performed by LC-Bio (Hangzhou, China). Briefly, glioma tissues were collected and processed using TRIzol reagent. Utilizing UMI technology, each sequence fragment was tagged with unique identifiers, thereby mitigating the impact of PCR-induced duplications on transcriptome quantification accuracy. The RNA-seq reads were aligned to the mouse genome (GRCh37/hg19) using the Hisat2 software (version 2.0.4). Transcript abundance was quantified in terms of fragments per kilobase of exon per million fragments. Significantly differential expression was defined with a threshold of p < 0.05 and |log2(fold change)| ≥ 1 The raw sequence data have been submitted to the NCBI Gene Expression Omnibus (GEO) with accession number GSE246028.

The frozen glioma samples were processed using the CTAB method. Subsequently, the 16S rRNA gene was amplified and sequenced, involving the simultaneous amplification of five regions of the 16S rRNA gene. The resulting libraries were sequenced using the Illumina NovaSeq 6000 system. Reads were demultiplexed per sample, filtered, and aligned to the five amplified regions according to the primer sequences. To generate coherent profiling results, the Short MUlitiple Regions Framework method was employed, addressing a maximum likelihood problem by combining read counts from the five regions. Reference data were obtained from the GreenGenes database (May 2013 version, with some improvements). The raw sequence data have been submitted to the NCBI Short Read Archive (SRA) with accession number PRJNA1021034.

DNA from fecal samples was extracted following the manufacturer’s CTAB instructions. The bacterial and archaeal 16S rRNA gene’s V3-V4 region was amplified using the 341F primer (5′-CCTACGGGNGGCWGCAG-3′) and the 805R primer (5′-GACTACHVGGGTATCTAATCC-3′).

The 16S rRNA amplicons were sequenced on an Illumina NovaSeq platform following LC-Bio’s instructions. Paired-end reads were merged with FLASH. Raw reads underwent quality filtering, applying specific filtering conditions to obtain high-quality clean tags using fqtrim (v0.94). Chimeric sequences were removed using Vsearch software (v2.3.4). After dereplication using DADA2, feature tables and feature sequences were generated. α diversity and β diversity were calculated by randomly normalizing the sequences. Feature abundance was normalized using SILVA (release 138) classifier. QIIME2 was employed for α diversity analysis, assessing species diversity complexity. β diversity was calculated using QIIME2, and R packages were utilized to generate corresponding graphs. Sequence alignment was performed via BLAST, and feature sequences were annotated using the SILVA database for each representative sequence. Additional diagrams were generated using the R package (v3.5.2). The raw sequence data have been submitted to the NCBI Short Read Archive (SRA) with accession number PRJNA1021780.

The metabolic extracts from serum were analyzed using LC-Bio. Serum samples were collected, and metabolites were extracted with a 50% buffer solution. Pooled quality control samples were prepared by combining 10 μL from each extraction mixture. Chromatographic separations were performed using an ultra-performance liquid chromatography (UPLC) system (SCIEX, UK) with an ACQUITY UPLC T3 column (100 mm*2.1 mm, 1.8 μm, Waters, UK) for reversed-phase separation. The detection of eluted metabolites from the column was performed using a high-resolution tandem mass spectrometer, the Triple TOF5600 Plus (SCIEX). The acquired MS secondary data were subsequently matched with an in-house metabolite standard secondary profile library to identify differential metabolites.

Statistical analyses were carried out with the SPSS 26.0 software. Log-rank test was used for Kaplan–Meier survival curves. Except as otherwise indicated, statistical difference between 2 groups was identified Statistical analyses were conducted using SPSS 26.0 software. The log-rank test was applied to Kaplan–Meier survival curves. In cases where not otherwise specified, statistical differences between two groups were determined using an independent sample t-test. For multiple comparisons, a one-way analysis of variance (ANOVA) was employed, followed by a Tukey post-hoc test. Significance was attributed to differences at p < 0.05. In figures, “*” indicated p < 0.05, “**” indicated p < 0.01.

To ascertain the effectiveness of Bifidobacterium in inhibiting glioma growth in mice, we administered the Mod-BIF group with 400 μL of Bifidobacterium solution (4 × 10⁹ CFU) daily through intragastric gavage from day −14 to day 30. On day 0, mice of the Mod and Mod-BIF groups were implanted with 1 × 10⁵ GL261 cells into the right striatum (Figure 1A). Subsequently, we employed cerebral T2-weighted MRI sequences to evaluate tumor volume on day 30 (Figure 1B). The results indicated a significant reduction in tumor volume in the Mod-BIF group compared with the Mod group (46.35 ± 16.34 mm³ vs. 21.71 ± 10.75 mm³, p < 0.01; Figure 1C). This observation was further supported by HE staining of glioma tissues (Figure 1D). As depicted in Figure 1E, the administration of Bifidobacterium resulted in a slight extension of median survival time in mice with glioma (42 days vs. 52 days, p < 0.05; Figure 1E). To delve deeper into the effect of Bifidobacterium on liver and kidney function, we assessed the organ indices of the liver and kidney, as well as the serum concentrations of ALT, AST, BUN, and Cr, alongside HE staining of liver and kidney of mice. The results showed no significant differences in the organ indices of the liver and kidney, nor in the levels of serum ALT, AST, BUN, and Cr among the three groups of mice (Figures 1F,G). Moreover, HE staining indicated the absence of noticeable lesions in the liver and kidneys of the mice (Figure 1H). Therefore, it can be concluded that Bifidobacterium effectively inhibits glioma growth without causing hepatic or renal toxicity.

To ascertain the presence of microbiota within tumor tissues, we conducted IHC, utilizing antibodies specific to bacterial LPS and LTA while employing PBS to mitigate non-specific staining. As shown in Figure 2A, bacterial LPS was detected in both the Mod and Mod-BIF groups, whereas LTA showed negative results, consistent with findings from Nejman et al. (2020). These results collectively corroborated the existence of microbiota in gliomas. Subsequently, we conducted 16S rDNA 5R sequencing to investigate alterations in the tumor microbiota in glioma mice. Three metrics (Chao1, Shannon, and Simpson indices) were used to analyze the α-diversity of tumor microbiota. The results showed that Bifidobacterium administration led to an elevated Chao1 index trend and a significant increase in both the Shannon and Simpson indices (p < 0.01; Figure 2B). Furthermore, nonmetric multidimensional scaling (NMDS) based on Bray–Curtis distance illustrated distinct differences in the composition of tumor microbiota between the Mod and Mod-BIF groups (Figure 2C).

At the phylum level, as shown in Figure 2D, the tumor microbiota in Mod and Mod-BIF groups was primarily dominated by Proteobacteria. Compared with the Mod group, the Mod-BIF group exhibited a higher abundance of Firmicutes, Bacteroidetes, Cyanobacteria, and Actinobacteria, and a lower abundance of Proteobacteria. At the genus level, the Mod group displayed a relatively high abundance of Raoultella and Hydrogenophilus, while Corynebacterium, Acinetobacter, and others were relatively abundant in the Mod-BIF group (p < 0.05, Figure 2E). Significance analysis revealed that Enterobacter, Citrobacter, Acinetobacter, Wautersiella, Enhydrobacter, and Escherichia/Shigella were significantly more abundant in the Mod-BIF group (p < 0.05, Figure 2E). Linear discriminant analysis (LDA) effect size (LEfSe) was employed to further identify specific microbiota, and the results indicated a notable difference in tumor microbiota between the two groups (Figure 2F). Most of the differential bacteria were enriched in the Mod-BIF group, suggesting their potential significance in this group. Furthermore, we utilized Phylogenetic Investigation of Communities by Reconstruction of Unobserved States (PICRUSt) to predict the functional potential of bacterial genes. PICRUSt2, which boasts an expanded reference genome database compared with the PICRUSt1(Langille et al., 2013; Douglas et al., 2019) was employed. As Figure 2G illustrates, the Mod-BIF group exhibited upregulation in several pathways compared with the Mod group, including aerobic respiration I (cytochrome c) and fatty acid salvage (mean proportion > 0.6%).

The intestinal barrier protects the organism against pathogenic invasions. Its primary physical barrier is upheld by intercellular tight junctions, with occludin, claudin, ZO, and junctional adhesion proteins constituting these junctions (Suzuki, 2020). Prior laboratory investigations have shown that gliomas can induce modest impairment of the intestinal barrier in mice (Wang et al., 2022). Consequently, we aimed to ascertain whether Bifidobacterium can ameliorate such damage through HE and IHC staining. As shown in Figure 3A, there were no discernible anomalies observed in the ileum and colon tissues of mice in both experimental groups. Subsequent IHC staining revealed no significant alterations in the protein expression levels of occludin and ZO-1 in the ileum and colon tissues (Figures 3B,C). Thus, to some extent, Bifidobacterium may not alleviate the glioma-induced slight damage to the intestinal barrier.

To compare the differences in gut microbiota composition between the Mod and Mod-BIF groups, we performed 16S rDNA sequencing using fecal samples from mice. The α-diversity of the gut microbiota remained unaffected by Bifidobacterium in glioma mice (Supplementary Figure S1A). Furthermore, β-diversity analysis revealed distinctions in the gut microbiota composition between the two groups (Supplementary Figure S1B). At the phylum level in both groups, Bacteroidetes and Firmicutes were the predominant bacterial taxa (Figure 4A). The Mod-BIF group exhibited higher levels of Actinobacteriota and lower levels of Myxococcota compared with the Mod group (p < 0.05, Figure 4C). Interestingly, upon closer examination at the genus level, administration of Bifidobacterium significantly augmented the abundance of Bifidobacterium (p < 0.05, Figures 4B,D). Similarly, the results from LEfSe indicated that Bifidobacterium dominated the microbiota in the Mod-BIF group (Figure 4E). A Manhattan plot further revealed that the administration of Bifidobacterium may primarily influence the levels of Bacteroidetes and Firmicutes in the gut microbiota of glioma mice (Figure 4F).

Metabolites originating from the gut microbiota can enter the bloodstream and exert regulatory functions in distal organs (Chen et al., 2022). Therefore, we performed an untargeted metabolomic analysis by LC-MS on serum samples collected from mice in both the Mod and Mod-BIF groups. A total of 138 differential metabolic ions were identified, with 56 upregulated ions and 82 downregulated ions meeting the criteria of having a variable importance in projection score of ≥1, a p value of <0.05, and a fold change of ≥1.5 or ≤1/1.5 (Figure 4G). Subsequently, we correlated these differential ions to secondary metabolites, resulting in the identification of five differential metabolites (Figure 4H). Among these metabolites, compared with the Mod group, the Mod-BIF group exhibited upregulated 1-amino-propan-2-ol, tyrosine, and 4-methylene-2-pyrrolidinecarboxylic acid levels and downregulated phosphocholine and 1,2,3,4-tetrahydroxybutane levels in the serum. Analysis of these metabolites using the Human Metabolome Database revealed their primary enrichment in organonitrogen compounds (Supplementary Figure S2). Furthermore, results from Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment demonstrated that these differential metabolites were mainly enriched in metabolic pathways, melanogenesis, thiamine metabolism, and phenylalanine, tyrosine, and tryptophan biosynthesis (Figure 4I).

By jointly analyzing differential microbiota and metabolites with a rho >0.5 and p < 0.05, as the threshold values indicated in Figure 5A and Supplementary Table S6, it is evident that tyrosine emerges as a crucial metabolite exhibiting strong correlations with most differential microbiota at the genus level. Furthermore, Bifidobacterium, which exhibited a significant increase following treatment, demonstrated a positive association with tyrosine.

To investigate the relationship between the abundance of gut microbiota and tumor microbiota, we conducted a genus-level analysis of all bacteria present in both groups. In both gut and tumor tissues, we identified 57 shared bacterial species (Figure 5B). Among these 57 species of bacteria, we observed that the abundance of 20 bacterial species was upregulated, the abundance of 10 bacterial species was downregulated, and the abundance of 27 species exhibited opposite changes (Figure 5C). Surprisingly, we found that Bifidobacterium exhibited increased abundance in both gut and tumor tissues (Figure 5D), suggesting a potential influence of alterations in gut microbiota on the composition of tumor microbiota.

After initially establishing that Bifidobacterium could inhibit glioma, we proceeded to investigate changes in the glioma genes through transcriptome sequencing. Differentially expressed genes (DEGs) between the Mod and Mod-BIF groups were identified based on a significance level of p < 0.05 and |log2 fold change| ≥ 1. We found 441 upregulated and 520 downregulated DEGs (Figure 6A). Subsequently, we conducted a Gene Ontology (GO) enrichment analysis on the downregulated DEGs. This analysis revealed an enrichment of downregulated DEGs in the GO term 0070734, specifically related to the positive regulation of the ERK1 and ERK2 cascades. To visually represent the differences between these genes in the two groups, we conducted a heatmap analysis of the downregulated DEGs associated with this GO term (Figure 6B).

The MEK/ERK cascade plays a pivotal role in the mitogen-activated pathway, facilitating the transmission of growth signals from the cell surface to the nucleus. Furthermore, the MEK/ERK pathway modulates the transcription of numerous genes to bolster cell proliferation (Choudhury et al., 2014). Moreover, the activation of the MEK/ERK cascade is associated with the invasive behavior of glioma cells (Huang et al., 2023). Western blot analysis was employed to assess the protein levels of MEK and ERK. The results revealed a significant reduction in the phosphorylation levels of MEK and ERK1/2 proteins in glioma tissues upon treatment with Bifidobacterium (p < 0.05; Figures 6C–E). This suggests that Bifidobacterium may exert an inhibitory effect on MEK and ERK1/2 protein phosphorylation in glioma tissues, thereby suppressing the MEK/ERK cascade. Original western blot images are presented in Supplementary Figure S3.

Additionally, through a protein–protein interaction (PPI) network established via the STRING database for the analysis of interactions among the downregulated DEGs, we discerned that Wnt5a served as the central hub gene (Figure 6F). Wnt5a, a member of the WNT family, has been previously demonstrated to induce rapid glioma growth and migration while also being associated with the survival of tumor-associated microglia (Xu et al., 2020). Subsequently, we assessed the differential expression and prognostic significance of Wnt5a by utilizing the Cancer Genome Atlas (TCGA) and Genotype-Tissue Expression (GTEx) datasets through the Gene Expression Profiling Interactive Analysis platform. In both GBM and low-grade glioma (LGG), the expression of Wnt5a was significantly increased in glioma tissues compared with normal tissues (p < 0.05; Figure 6G). However, the expression level of Wnt5a was associated with overall survival only in LGG (Figure 6H). In the context of our study, qRT-PCR results demonstrated a reduction in the mRNA expression of Wnt5a as a consequence of Bifidobacterium treatment (Figure 6I).