Prevotella bivia promotes cervical cancer progression and modulates macrophage polarization, while Lactobacillus iners suppresses these processes: evidence from multiomics analysis Authors: Ting Zhang, Qiuhong Qian, Qianwei Zhen, Jiaxin Zhang, Yu Sun, Junhua Zhang, Lingyu Guo, Xiaoli Liu, Deyuan Zou, Baokun Zhou, Chang Liu, Shuqi Chi, Gaishuang Shang, Baoxia Cui, Youzhong Zhang, Yunlong Cui, Youming Zhang, Sai Han Published: 2026 ABSTRACT ABSTRACTThe incidence and mortality rates of cervical cancer are high among women. Growing evidence suggests the key roles of the intratumor microbiome in various solid tumors. However, the intratumoral microbiome in patients with cervical cancer has not been well characterized. In this study, 16S rRNA sequencing was performed on 76 tissues to reveal the features of the intratumoral microbiome, and the highly differentially abundant bacteria Lactobacillus iners/Prevotella bivia were selected for functional verification. Flow cytometry, transwell migration, and invasion experiments, among others, were performed in vitro; subcutaneous tumor formation and lung metastasis experiments were performed in vivo. Additionally, macrophages were cocultured with the L. iners/P. bivia supernatant to evaluate how the intratumoral microbiome affects their polarization, and tumor cell and macrophage transcriptome sequencing was subsequently performed to explore the potential molecular mechanisms involved. Metabolomic analysis of tissues and bacterial supernatants was performed to search for potential carcinogens and cancer suppressors. We determined that the abundance of microbes was greater in cervical cancer tissues than in normal cervical and paracancerous tissues. The relative abundance of Prevotella was correlated with deep stromal invasion, tumor size, clinical stage, and poor survival prognosis in cervical cancer patients. L. iners inhibited the proliferation and promoted the apoptosis of tumor cells, whereas P. bivia significantly promoted cervical cancer cell migration and invasion. Mechanistically, the modulation of phosphorylated PI3K/AKT/mTOR signaling may be involved in the observed effects. P. bivia promoted M2 macrophage polarization by activating phosphorylated STAT3 and NF-κB, but the role of L. iners remains unknown. The amino acid metabolism, carbohydrate metabolism, and lipid metabolism pathways were enriched in differentially abundant metabolites such as glycine, which may be a key molecule. This study provides evidence that the intratumoral microbiome, represented by L. iners/P. bivia, affects tumor biology and macrophage polarization in patients with cervical cancer.IMPORTANCEThe microbiome in cervical cancer tissues significantly differed from that in normal cervical tissues and showed significant correlations with clinicopathological features and survival prognosis. The tumor microbiome affects the biological behavior of cervical tumor cells and the polarization of macrophages through metabolite production, thus playing an important role in the occurrence and development of cervical cancer. INTRODUCTION Cervical cancer remains the most prevalent gynecological malignancy and a leading cause of cancer-associated mortality in women worldwide. According to recent WHO data, approximately 600,000 new cases and 340,000 deaths were reported globally in 2022, and the disease burden was particularly high in China (1, 2). The persistent challenges in eliminating cervical cancer cells highlight the urgent need for new studies to advance the understanding of its pathogenesis and identify new potential avenues for intervention. In recent years, accumulating evidence has established the presence of distinct intratumoral microbiomes across a variety of solid tumors—including those of the colon, breast, lung, and pancreas; studies have also revealed that tumor-resident bacteria play key roles in modulating tumor progression, metastasis, prognosis, and even treatment response (3–6). For example, in breast tumors, unique bacteria such as Staphylococcus and Lactobacillus drive metastatic colonization (4), and the presence of Fusobacterium in the tumor microenvironment of colorectal cancers is correlated with poor survival and altered immune infiltration (6). Mechanistic studies indicate that these bacteria may influence cancer biology through the modulation of tumor metabolism, immune responses, and cell signaling pathways. Despite such progress in other malignancies, the significance and characteristics of the intratumoral microbiome remain largely undefined in the context of cervical cancer. Recent studies have focused on cervical and vaginal microbial communities in the context of HPV infection, a well-established etiological factor in cervical carcinogenesis, and have revealed shifts in the vaginal microbiome that are associated with disease progression (7). However, few investigations have directly addressed the microbiome composition within cervical tumor tissues, its differences from that in adjacent or normal cervical tissues, or its impact on tumor biology, immune cell function, and clinical outcomes. Furthermore, the potential interplay between the tumor microbiome and key factors unique to cervical cancer, such as persistent HPV infection, remains unexplored. Given the diversity and tumor-type specificity of intratumoral bacteria revealed in other cancers, determining whether cervical cancers harbor a unique microbiome that may shape their progression through distinct mechanisms is critical. This is a particularly important task in cervical cancer research, given the unique dependence of this cancer on HPV-driven oncogenesis and the increasing recognition of the tumor immune microenvironment—including the role of tumor-associated macrophages in immune escape (8). The possibility that tumor-resident bacteria interact with these cervical cancer-specific factors to modulate disease progression, immune evasion, and prognosis represents an important but unresolved theory in cervical cancer research. In this study, we performed 16S rRNA sequencing on tumor and adjacent tissues from cervical cancer patients and normal cervical tissues and used computational biology methods to quantify the bacteria in the tissues and analyze their relationships with the clinicopathological features of cervical cancer patients. We also confirmed the presence of intracellular bacteria in cervical cancer tissues and adjacent tissues using tissue microarrays and an EdU probe. Furthermore, we demonstrated the biological functions of intratumoral bacteria in cervical cancer in vitro and in vivo and analyzed the impact of intratumoral bacteria on macrophages in a coculture model. Finally, we studied the potential mechanisms involved via transcriptomic and metabolomic approaches. In summary, our research provides a comprehensive overview of the intratumoral microbiome in patients with cervical cancer and investigates its role in the occurrence and development of this disease. MATERIALS AND METHODS Patient selection and sample collection Tissue samples were collected from 23 normal controls and 40 cervical cancer patients (13 of whom had paracancerous tissues) admitted to the Department of Obstetrics and Gynecology, Qilu Hospital of Shandong University, from 2018 to 2021. All clinical data, including age and pathological diagnosis, were collected, and the patients were staged according to the 2018 FIGO staging guidelines. The present study was approved by the Ethics Committee of Qilu Hospital. In this study, all human tumor and normal cervical tissue samples were freshly collected during surgery. Immediately after excision, the samples were placed into sterile tubes and rapidly snap-frozen in liquid nitrogen to preserve microbial DNA integrity and minimize potential contamination. Formalin-fixed paraffin-embedded (FFPE) samples were not used. 16S rRNA sequencing and bioinformatics analysis During sample processing and DNA extraction, strict aseptic techniques were employed using sterilized instruments and consumables under sterile conditions to prevent contamination. Negative controls, such as negative control samples, were incorporated during the PCR amplification step, and nuclease-free water provided in commercial nucleic acid extraction kits was used as a template to monitor potential contamination during PCR. Genomic DNA was extracted from the tumor samples using hexadecyl trimethyl ammonium bromide (CTAB). The rDNA of the bacterial ribosome subunit 16S (V4 region) was amplified using the universal primers 515 F (5′-GTGCCAGCMGCCGCGGTAA-3′) and 806 R (5′-GGACTACHVGGGTWTCTAAT-3′). After the samples were compiled, inspected for quality, quantified, and mixed in equal proportions, they were sequenced using the NovaSeq 6000 platform (Illumina, San Diego, CA, USA). The paired-end reads were subsequently merged using FLASH (V1.2.11, http://ccb.jhu.edu/software/FLASH/), and quality filtering was performed using fastp software (version 0.23.1) to obtain high-quality clean reads (9, 10). For the effective reads obtained previously, denoising was performed with DADA2 or the deblur module in QIIME2 software (version QIIME2-202202) to obtain the initial amplicon sequence variants (ASVs) (DADA2) (11). The top axis of each sample at each taxonomic rank (phylum, class, order, family, genus, and species) was used to plot a distribution histogram of relative abundance using the SVG function in Perl. The abundances of the top taxa of each sample at each taxonomic rank were used to construct a heatmap. Venn diagrams were produced in R with the VennDiagram() function. The 100 most abundant genera were subjected to sequence alignment, and a phylogenetic tree was constructed using the SVG function in Perl. The alpha diversity was calculated from seven indices, observed_otus, Chao1, Shannon, Simpson, dominance, Good’s coverage, and Pielou_e, using QIIME2. To evaluate the complexity of the community composition and evaluate the differences among samples (groups), beta diversity was calculated on the basis of weighted and unweighted UniFrac distances using QIIME2. A cluster tree was constructed on an unweighted pair group arithmetic average (UPGMA) diagram using the upgma.tre function within QIIME2. Nonmetric multidimensional scaling (NMDS) was implemented with the ade4 and ggplot2 R packages. Differences in community structure were revealed via ANOSIM, Adonis, the multiresponse permutation procedure (MRPP), SIMPER, t tests, metagenomeSeq, and linear discriminant analysis (LDA) effect size (LEfSe) analyses. Anosim, Adonis, MRPP, and SIMPER analyses were performed with the vegan and ggplot2 packages within R; metagenomeSeq was performed with the metagenomeSeq package in R; and the exclusive package lefse was utilized for LEfSe. For the LDA and LEfSe analyses, the power comparison control strategy was set as one against one, the LDA threshold was set to 4, and the classification level threshold was set to 0.05 (12). Transcriptome sequencing and bioinformatics analysis RNA integrity was assessed using an RNA Nano 6000 Assay Kit and a Bioanalyzer 2100 system (Agilent Technologies, CA, USA). After library preparation, clustering and transcriptome sequencing were performed. Clustering of the index-coded samples was performed on a cBot Cluster Generation system using a TruSeq PE Cluster Kit v3-cBot-HS (Illumina) according to the manufacturer’s instructions. The generated clusters and library preparations were subsequently sequenced on an Illumina NovaSeq platform, and 150 bp paired-end reads were generated. Analysis of the differentially expressed factors between the two groups was performed using the DESeq2 package in R (1.20.0). The obtained P values were adjusted using Benjamini and Hochberg’s approach to control the false discovery rate. Genes with an adjusted P value ≤0.05 as determined by DESeq2 were considered differentially expressed. We then used the clusterProfiler package in R to evaluate the pathways with significant enrichment of differentially expressed genes using the Kyoto Encyclopedia of Genes and Genomes (KEGG) database (http://www.genome.jp/kegg/). Metabolomic sequencing and bioinformatics analysis A nontargeted metabolomics study was conducted using a liquid chromatography-mass spectrometer (LC-MS/MS) (13, 14). For these experiments, the metabolites were extracted from the samples and subjected to LC-MS/MS and data analysis. Three main parameters were considered when screening differentially abundant metabolites: the variable importance in projection (VIP), the fold change (FC), and the P value. The VIP refers to the first principal component from the partial least squares discriminant analysis (PLS-DA) model and represents the contribution of the metabolites to the clustering; the FC was calculated as the ratio of the mean value of all quantified metabolites from each replicate in the comparison group, and the P value, which indicates the significance of the differences among the data, was calculated by a t test. The thresholds for the differentially abundant metabolites were a VIP >1.0, FC >1.2 or an FC <0.833 and a P value <0.05. The identified metabolites were subjected to functional annotation using mainly the KEGG database, as mentioned above. Cell lines and culture The human cervical cell lines HeLa and SiHa, and the monocytic THP-1 cell line, which were purchased from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China), were used in this study. The cells were maintained in RPMI-1640 medium supplemented with 10% fetal bovine serum (FBS; Biological Industries, USA) and incubated under standard culture conditions (5% CO2, 37°C). Extraction of bacterial supernatant In this study, bacterial culture supernatants from both in vitro and in vivo experiments were sterile-filtered through 0.22 µm membranes to remove all viable bacteria. The supernatants were subsequently adjusted with sterile phosphate-buffered saline (PBS) to achieve physiological osmolarity (~290 mOsm/kg) and a neutral pH (~7.4) to minimize nonspecific effects. Given that Prevotella bivia is a gram-negative bacterium capable of releasing lipopolysaccharides (LPS), we acknowledged that residual endotoxins might induce cellular stress responses and potentially confound data interpretation. Therefore, endotoxin levels in the collected supernatants were quantified using the limulus amebocyte lysate (LAL) assay, which revealed that endotoxin concentrations were below the thresholds commonly recognized to significantly affect mammalian cells. Flow cytometry To detect apoptosis and the cell cycle, we established three groups of cell samples: control cells, SiHa cells treated with Lactobacillus and Prevotella supernatants, and HeLa cells treated with the same supernatants. Apoptosis: The control group and the SiHa and HeLa cells treated with Lactobacillus and Prevotella supernatants were digested using trypsin without EDTA. Next, the cells were stained with an Annexin V-FITC/PI Apoptosis Kit (Elabscience, Wuhan, China) for 30 minutes at room temperature in the dark. Cell cycle: The cells were digested with trypsin containing EDTA and then fixed in 75% ethanol. After overnight fixation, PI/RNase Staining Buffer (BD Biosciences, Franklin Lakes, NJ, USA) was added to the respective fixed samples. The samples were incubated for 15 minutes at room temperature in the dark. Apoptosis and the cell cycle were detected by flow cytometry (Cytoflex, USA), and quantitative analysis was performed using FlowJo v10 and CytExpert. Transwell migration and invasion assays After coculture, 5 × 104 SiHa or HeLa cells were digested, suspended in 100 µL of serum-free 1640 medium, and seeded into the upper chamber of a Transwell system (8.0-µm pore size; Costar, Cambridge, MA, USA) in the absence or presence of 100 µL of Matrigel (diluted 1:8 in serum-free medium; Corning, NY, USA). The experiments were performed as previously described (15). Reverse transcription real-time PCR Cellular RNA was extracted using a SPAReasy cell RNA rapid extraction kit (SparkJade, Jinan, Shandong, China), and cDNA was synthesized from 5 µg of RNA using M-MLV reverse transcriptase (Invitrogen, Shanghai, China) according to the manufacturer’s instructions. In addition, primer–probe sets were designed for each gene for qPCR using PrimerBank and are listed in Table S1. The data were collected using the StepOnePlus software (Applied Biosystems, Shanghai, China) and quantified using the 2−ΔΔCt method. Western blotting Total protein was extracted from the cells as previously described (15). The primary antibodies used were anti-p-PI3K (1:1,000; Cell Signaling Technology, #4228T), anti-p-AKT (1:1,000; Cell Signaling Technology, #4060S), anti-p-mTOR (1:1,000; Cell Signaling Technology, #2971S), anti-p-4E-BP1 (1:1,000; Cell Signaling Technology, #2855T), anti-p70 S6 kinase (49D7) (1:1,000; Cell Signaling Technology, #2708T), anti-p-STAT3 (1:1,000; Abcam, ab46154), anti-p-NF-κB (1:1,000; Cell Signaling Technology, #3033T), anti-IL-6 (1:1,000; Immunoway, #YT5348), and anti-β-actin (1:80,000; ABclonal, #AC026); notably, β-actin is considered a very stable internal control. The secondary antibodies used were anti-rabbit (1:5,000; Immunoway, #RS0002) and anti-mouse (1:2,000; Cell Signaling Technology, 7076S) IgG peroxidase conjugates. A Tanon 4800 multifunctional imaging system (Tanon Life Science Co., Ltd., Shanghai, China) was used for protein detection, and the results were analyzed with ImageJ software (NIH, Bethesda, MD, USA). Tumor cell growth and metastasis in athymic nude mice Twelve 4-week-old female athymic nude mice purchased from Weitong Lihua Biotechnology Co., Ltd. (Beijing, China) were used in this study and randomly divided into three groups (4 mice/group). Upon reaching 80% confluence, the HeLa cells were digested with 0.25% trypsin and collected by centrifugation at 1,000 rpm for 5 min. A cell suspension with a concentration of 5 × 106 cells/mL was then subcutaneously injected into the mice. The tumor volume and size were measured as previously described (16, 17). Additionally, the supernatants of the two test species and phosphate-buffered saline (PBS) were injected into the abdominal cavity every 2 days. After 18 days, the xenografts were removed from the mice and photographed. The lung metastasis model also included 12 female athymic nude mice randomly divided into three groups (4 mice/group). For these experiments, HeLa cells (1 × 106 cells/mL) were injected into the mice through the tail vein. The mice received injections of the supernatant of the two test bacterial species and PBS for two months, after which the lungs were removed for hematoxylin and eosin (HE) staining to identify tumor nodules. Statistical analysis Statistical analyses were conducted with GraphPad Prism 8.0 software (GraphPad Software Inc., San Diego, CA, USA). All the experiments were repeated at least three times, and the results are presented as the means ± standard deviations (SDs). Student’s t test, the Pearson r test, the chi-square test, and Yate’s continuity-corrected chi-square test were performed to analyze the data from the different groups. Finally, the Kaplan–Meier method and the log-rank test were used to analyze the survival data. P < 0.05 was considered to indicate statistical significance. RESULTS Patients We collected 76 clinical samples from patients (40 cervical cancer tissues [C] plus 13 paracancerous tissues [Np] and 23 normal cervical tissues [Nc]). The clinicopathological features of the cervical cancer patients are presented in Table 1. TABLE 1Association between Lactobacillus/Prevotella relative abundance and the clinicopathological factorsaVariableNo.Expression of LactobacillusExpression of PrevotellaHighLowP-valueHighLowP-valuedAge0.73570.7357 ≤45137676 >452713141314Histology0.71501.000 Adenocarcinoma106455 SCC3014161515Differentiation1.0000.5186 Low168879 Moderate/high2412121311Clinical stage0.1769 0.0428b ≥IB32711161710