Introduction

Cervical cancer (CC) stands as one of the most frequently diagnosed malignancies in women.According to recent data from GLOBOCAN in 2020, cervical cancer affected around 604,127 women worldwide and was responsible for over half of the total deaths, resulting in 341,831 fatalities1. Human papillomavirus (HPV) infection is a significant factor in the development of cervical cancer lesions. According to its pathogenicity, HPV is categorized into high-risk and low-risk types.High-risk human papillomavirus (hrHPV) infects metaplastic cells within the cervical transformation zone, resulting in genomic integration into the host cell. This integration leads to the inactivation of tumor suppressor genes, including p53 and Rb, consequently promoting cell proliferation and facilitating the accumulation of genetic mutations2. Finally, a minority of infections persist or progress to precancerous lesions, ultimately culminating in malignancy3. Among the carcinogenic HPV types, HPV16 and 18 are the most prevalent and are linked to roughly 70% of HPV-related cervical cancers4,5. Without intervention, cytological abnormalities resulting from HPV infections span from low-grade CIN1 to high-grade CIN3, with the latter being primarily attributed to high-risk HPVs6. E6 and E7 proteins are the primary regulators contributing to the pathogenicity of the virus, and they are essential during the early stages of infection7.

The gut microbiota is recognized as a pivotal factor in regulating host health and is associated with the development of internal tumors8. Presently, there exists an extensive body of research that has illuminated the connection between vaginal microbiota and the advancement of cervical cancer lesions9,10. However, it is noteworthy that the exploration of the relationship between cervical cancer and intestinal microbiota is conspicuously limited in the existing scientific literature11. Wang et al.12 observed marked differences in gut microbiome composition between cervical cancer patients (N = 8) and controls (N = 5). Notably, cervical cancer patients showed a significantly higher presence of Bacteroidetes, contrasted by a relatively lower abundance of Firmicutes. Kang et al.13. conducted comparative fecal specimen analysis between healthy individuals and early-stage cervical cancer patients, resulting in the identification of seven specific biomarkers and the development of a machine learning-based cervical cancer prediction model. Randomized controlled trials have the potential to establish causal relationships. However, in the context of gut microbiota research, their implementation is challenging due to technological limitations and other constraints. Consequently, the majority of current investigations derive their conclusions from analyzing the gut microbiota composition in patients’ fecal samples through observational studies. Unfortunately, these studies are frequently susceptible to interference from a range of external factors, such as antibiotic usage, chemotherapy, and surgery14,15,16. In conclusion, the causality of the links between gut microbiota and the lesion progression of cervical cancer, as well as the direction of these causal relationships, remains unclear. Therefore, it is imperative to investigate the causal connection between gut microbiota and the lesion progression of cervical cancer.

Mendelian randomization (MR) utilizes genetic variants in non-experimental data to determine the causative influence of an exposure on an outcome. This approach efficiently explores the causal relationship from exposure to outcome in cross-sectional studies while mitigating the impact of potential confounding factors17,18. In principle, MR exhibits similarities with a RCT. Within MR, genetic variables serve as instrumental variables (IVs) and are randomly assigned to “case” or “control” groups at birth, in accordance with Mendel’s second law, maintaining constancy throughout an individual’s lifetime. In this study, we utilized a two-sample Mendelian randomization approach to explore the causal link between the progression of cervical cancer and the composition of the gut microbiota.

Method

Research design

The study was strictly designed according to the STROBE-MR checklist. A diagram depicting the gut-cervix axis is presented in Fig. 1, and Fig. 2 illustrates a flowchart of the two-sample MR analysis employed in this study. The causal relationship between gut microbiota and the progression of cervical cancer lesions was investigated using the two-sample MR method. We separately investigated the associations between HPV16 and HPV18 viral infections, cervical intraepithelial neoplasia, and cervical cancer with gut microbiota to comprehensively explore the role of gut microbiota in the progression of cervical cancer. To ensure the reliability of MR results, three fundamental assumptions need to be met: (i) IVs exhibit a robust correlation with the exposure variable. (ii) There should be no correlation between IVs and any observed or unobserved confounding factors in the exposure-outcome relationship. (iii) IVs should exclusively influence the outcome through the exposure and not through any other pathways.

Fig. 1
figure 1

Diagram illustrating the gut-cervix axis. The left side highlights a potential connection between gut microbiota and the cervix, while the right side outlines the three-stage progression of cervical cancer.

Full size image
Fig. 2
figure 2

A workflow illustrates the process of our two-sample Mendelian Randomization (MR) analysis. MR: Mendelian Randomization; IVs: instrumental variables;.

Full size image

Data sources

Single-nucleotide polymorphisms (SNPs) associated with the human gut microbiome composition were chosen as instrumental variables (IVs) from the MiBioGen consortium (https://mibiogen.gcc.rug.nl/, accessed 12 July 2023). The MiBioGen consortium integrates 16 S RNA gene sequencing profiles and genotyping data from 18,340 samples to explore the relationship between a patient’s gut microbiota and genetic variations19. The MiBioGen group, gathering data from 25 cohorts representing 11 European ancestral nations, involved an exposure dataset featuring 122,110 SNPs related to 211 taxa (131 genera, 35 families, 20 orders, 16 classes, and 9 phyla). In this study, an extensive meta-analysis was conducted, encompassing a wide array of ethnicities, to probe the associations between autosomal genetic variants in humans and the gut microbiome. The investigation delved into the varied composition of the microbiome across different populations, while also scrutinizing the influence of methodological discrepancies on the derived microbiome data. During the analysis, 15 bacterial traits lacking specific species names were omitted, and 196 bacterial traits were selected for further examination.

Table 1 Comprehensive details of studies and datasets employed in MR analyses.
Full size table

We selected serum HPV16 and HPV18 E7 proteins, CIN, and CC GWAS data to illustrate the cervical cancer lesion process. The serum levels of HPV16 and 18 E7 proteins were derived from a GWAS of the plasma proteome20. Genetic variations at the individual level can influence protein levels in the blood. In this study, a highly multiplexed, aptamer-based affinity proteomics platform (SOMAscan) was utilized for GWAS to quantify 1,124 protein levels in plasma samples. Samples from 1,000 individuals in the population-based KORA (Cooperative Health Research in the Region of Augsburg) study were examined, and the results were replicated in 338 Arab and Asian participants of the Qatar Diabetes Metabolomics Study (QMDiab). The data related to CC and CIN were acquired from the fastGWA laboratory led by Dr. Yang.Generalized linear mixed model (GLMM)-based methods are more suitable for genome-wide association (GWA) analysis of binary traits compared to approaches based on linear mixed models (LMMs)21. Jian Yang developed the genome-wide association (GWA) tool called fastGWA-GLMM. This tool was applied to the UK Biobank dataset, which included 456,348 individuals, 11,842,647 variants, and 2,989 binary traits22. The GWAS summary data noted above were accessed from the IEU open GWAS project (https://gwas.mrcieu.ac.uk/, accessed 14 July 2023) and the GWAS Catalog (https://www.ebi.ac.uk/gwas/, accessed 16 July 2023). The specifics of the GWAS datasets we chose are presented in Table 1.

Instrumental variable selection

As shown in Fig. 2, the selection of IVs follows the following quality control steps to ensure the accuracy and authenticity of the causal relationship in the study. The entire gut microbial classification comprised 211 taxa (131 genera, 35 families, 20 orders, 16 classes, and 9 phyla), each of which was treated as an independent exposure in the MR analysis. First, to ensure an adequate number of IVs for our MR analysis, we adjusted the SNP selection threshold to 5 × 10−5, as opposed to the initial threshold of 5 × 10−8 19. The IVs chosen for this study were the SNPs that exhibited significant associations with the gut microbiome. Next, we excluded SNPs with a minor allele frequency (MAF) lower than 0.01 as the second step. In the third step, we employed samples from the 1000 Genomes European Project as a reference to assess the linkage disequilibrium (LD) among IVs. Our criteria involved an r-squared value less than 0.01 and a window size exceeding 10,000 kb, thereby mitigating the influence of LD between IVs23,24.

Finally, a crucial aspect of MR is ensuring that the alleles of SNPs affecting the exposure match those influencing the outcome.To prevent potential issues related to strand orientation or allele coding, we excluded palindromic SNPs, such as those with A/T or G/C alleles25.

MR analysis

We conducted MR analysis to investigate potential causal relationships between microbiome features and the lesion progression of cervical cancer. When assessing features with a single instrumental variable (IV), we utilized the Wald ratio test to estimate the association between the identified IV and each specific cancer26. This method derives causal estimates by dividing the genetic effect on the outcome by the genetic effect on the exposure, under the key assumption that IVs influence outcomes exclusively through the target exposure. Its simplicity minimizes overfitting risks in single-IV scenarios while requiring strong instrument validity (F-statistic > 10) to ensure reliability. For features with multiple IVs, we employed five well-established MR methodologies: the IVW test27, simple mode28MR‒Egger regression29, the weighted median estimator30, and the weighted mode31. The IVW method was prioritized as it aggregates causal estimates from multiple IVs using precision-weighted averaging, thereby maximizing statistical efficiency when majority IVs satisfy validity assumptions. It is important to note that, under certain conditions, the IVW method has been reported to exhibit slightly superior statistical power in comparison to the other methods30. Therefore, when addressing features with multiple IVs, we primarily relied on the IVW method, complemented by the use of the other four methods for sensitivity analyses.

For the primary MR results, we evaluated the significance of multiple tests at various feature levels using Bonferroni correction (p < 0.05/n), where ‘n’ denoted the number of bacterial taxa within each feature level32. Consequently, the significance thresholds for multiple testing were determined as follows: 5.56 × 10−3 for phylum, 3.13 × 10−3 for class, 2.5 × 10−3 for order, 1.56 × 10−3 for family, and 4.20 × 10−4 for genus. Furthermore, we also considered a nominal significance level for the MR estimates, requiring p-values to be less than 0.05.

We carried out various sensitivity analyses to assess the robustness of the results. One of these involved performing a leave-one-out analysis, which aimed to determine if a single SNP was the primary driver of the causal signal. The Cochran Q test was employed to assess heterogeneity, and significant Q statistics at a p-value less than 0.05 suggest the existence of heterogeneity33. The MR-Egger intercept was employed to investigate potential pleiotropy within the IVW model34. Furthermore, MR-PRESSO was used to detect horizontal pleiotropy by identifying and removing instrumental variables that displayed outlier behavior35. In cases where the MR-PRESSO global test yielded a p-value below 0.05, a systematic procedure was enacted. Initially, SNPs were sorted in ascending order according to their MR-PRESSO outlier test p-values. Subsequently, SNPs were methodically removed one at a time, and following each elimination, the MR-PRESSO global test was performed on the remaining SNPs. This iterative process continued until the p-value for the global test became statistically insignificant (p > 0.05).

Statistical analyses were conducted using R packages, including two-sample MR(version 0.5.7)36and MR-PRESSO(version 1.0)28.

Results

Instrumental variables

A comprehensive set of 196 bacterial taxa, spanning 9 phylum, 16 classes, 20 orders, 32 families, and 119 genera, was taken into account for MR analysis. As shown in the flowchart of MR analysis in Fig. 2, we conducted a series of quality control procedures for all SNPs. We removed all SNPs with an F-statistic less than 10 to ensure the absence of weak instrument bias (Table S1). Following the removal of pleiotropic SNPs as detected by the MR-PRESSO outlier test and MR‒Egger regression, there was no evidence of horizontal pleiotropy in the instrumental variables (IVs). This absence of horizontal pleiotropy was substantiated by both the MR-PRESSO global test (p > 0.05) and MR Egger regression (p > 0.05) (Table S2-S5).

For CC, we detected 102, 180, 217, 357, and 1240 SNPs at the phylum, class, order, family, and genus levels, respectively. In the case of cervical intraepithelial neoplasia (CIN), we found 101, 180, 217, 355, and 1236 SNPs at the phylum, class, order, family, and genus levels, respectively. In the context of HPV E7 Type 16, we identified 12, 22, 27, 47, and 177 SNPs at the phylum, class, order, family, and genus levels, respectively. Similarly, for HPV E7 Type 18, we discovered 12, 22, 27, 47, and 177 SNPs at the phylum, class, order, family, and genus levels, mirroring the findings for Type 16. All detailed information of the SNPs retained after filtering for MR analysis can be found in Additional file 1: Table S6-S9.

HPV infection

We used the positivity of the E7 protein of HPV16 and HPV18 in serum to refer to the exposure factor of HPV infection. The association between genus Adlercreutzia, genus Dorea, genus Lactobacillus and HPV E7 Type 18, as well as family Porphyromonadaceae, genus Adlercreutzia, genus Marvinbryantia and HPV E7 Type 16, is being examined. Due to the limit in the number of instrumental variables, only one SNP is used for the MR analysis, therefore the Wald ratio test will be used to examine the association. As shown in Table 2, we found that class.Coriobacteriia (IVW, OR = 0.30, 95% CI 0.137–0.658, P = 2.65 × 10−3), family.Coriobacteriaceae (IVW, OR = 0.30, 95% CI 0.137–0.658, P = 2.65 × 10−3), order.Coriobacteriales (IVW, OR = 0.30, 95% CI 0.137–0.658, P = 2.65 × 10−3), genus.Ruminococcusgnavusgroup (IVW, OR = 0.48, 95% CI 0.233–0.987, P = 4.59 × 10−2), family.Porphyromonadaceae (Wald ratio, OR = 0.12, 95% CI 0.019–0.780, P = 2.61 × 10−2) and genus.Marvinbryantia (Wald ratio, OR = 0.20, 95% CI 0.051–0.799, P = 2.27 × 10−2) have protective effect against high-risk HPV infection, while class.Bacteroidia (IVW, OR = 4.37, 95% CI 1.209–15.783, P = 2.44 × 10−2),genus.Eubacteriumfissicatenagroup (IVW, OR = 1.91, 95% CI 1.049–3.486, P = 3.43 × 10−2), genus.Bilophila (IVW, OR = 3.12, 95% CI 1.053–9.252, P = 4.00 × 10−2), genus.Faecalibacterium (IVW, OR = 3.56, 95% CI 1.414–8.969, P = 7.04 × 10−3), order.Bacteroidales (IVW, OR = 4.37, 95% CI 1.209–15.783, P = 2.44 × 10−2), phylum.Bacteroidetes (IVW, OR = 4.41, 95% CI 1.215–15.997, P = 2.40 × 10−2), genus.Adlercreutzia (Wald ratio, OR = 4.40, 95% CI 1.398–13.864, P = 1.13 × 10−2), genus.Dorea (Wald ratio, OR = 10.91, 95% CI 2.066–57.589, P = 4.88 × 10−3), genus.Lactobacillus (Wald ratio, OR = 4.09, 95% CI 1.34–12.464, P = 1.33 × 10−2) and genus.Adlercreutzia (Wald ratio, OR = 4.12, 95% CI 1.306–13.019, P = 1.57 × 10−2) promote a higher likelihood of high-risk HPV infection.

Table 2 MR analysis results for the association between gut microbiota and cervical cancer lesion progression.
Full size table

Cervical intraepithelial neoplasia

Table 1; Fig. 3 provide convincing evidence suggesting that the genus.Escherichia.Shigella (IVW, OR = 0.68, 95% CI 0.466–0.998, P = 4.87 × 10−2) and genus.Lachnospira (IVW, OR = 0.52, 95% CI 0.313–0.868, P = 1.22 × 10−2) reduce the risk of CIN. However, genus.Eubacteriumventriosumgroup (IVW, OR = 1.55, 95% CI 1.112–2.158, P = 9.65 × 10−3), genus.Ruminococcustorquesgroup (IVW, OR = 1.50, 95% CI 1.012–2.221, P = 4.35 × 10−2), genus.Anaerostipes (IVW, OR = 1.56, 95% CI 1.068–2.292, P = 2.17 × 10 − 2), genus.Faecalibacterium (IVW, OR = 1.56, 95% CI 1.119–2.171, P = 8.70 × 10−3), genus.Paraprevotella (IVW, OR = 1.34, 95% CI 1.045–1.73, P = 2.11 × 10 − 2) and genus.Ruminiclostridium5 (IVW, OR = 1.79, 95% CI 1.178–2.708, P = 6.35 × 10−3) are risk factors contributing to CIN.

Fig. 3
figure 3

A forest plot presenting the causal association between the 9 genetically identified microbial taxa and CIN.

Full size image

Cervical cancer

Utilizing the IVW method, we identified an association suggesting that family.Defluviitaleaceae (IVW, OR = 0.46, 95% CI 0.231–0.937, P = 3.23 × 10−2), genus.RuminococcaceaeUCG011 (IVW, OR = 0.43, 95% CI 0.238–0.79, P = 6.03 × 10−3) and genus.DefluviitaleaceaeUCG011 (IVW, OR = 0.40, 95% CI 0.178–0.877, P = 2.25 × 10−2) were associated with a reduced occurrence of cervical cancer. Conversely, the genus.Eubacteriumventriosumgroup (IVW, OR = 3.25, 95% CI 1.317–8.001, P = 1.05 × 10−2), genus.Intestinimonas (IVW, OR = 2.34, 95% CI 1.23–4.438, P = 9.57 × 10−3), genus.LachnospiraceaeNC2004 group (IVW, OR = 2.17, 95% CI 1.068–4.428, P = 3.22 × 10−2) and genus.RuminococcaceaeUCG003 (IVW, OR = 4.18, 95% CI 1.738–10.066, P = 1.41 × 10−3) were linked to a high risk of cervical cancer. (Table 2; Fig. 4)

Fig. 4
figure 4

A forest plot presenting the causal association between the 7 genetically identified microbial taxa and CC. IVs: instrumental variables.

Full size image

Sensitivity analysis

No significant heterogeneity was observed among the selected instrumental variables and cancer outcomes, based on the Cochran’s Q test results (p > 0.05). Additionally, the robustness of these findings was validated by conducting a leave-one-out sensitivity test, revealing that the exclusion of any individual SNP had no discernible effect on the overall results (Additional file 2: Supplementary Fig. 1). During horizontal pleiotropy testing, distinct patterns emerged across outcomes. When cervical cancer was analyzed as the outcome, the MR-PRESSO global test identified 4 microbial features with significant pleiotropic bias (P < 0.05). For these features, we iteratively excluded outliers following the SNP screening protocol described in the Methods until non-significant P-values (≥ 0.05) were achieved. However, MR-Egger regression detected invalid instruments (intercept P < 0.05) for 8 additional microbial associations, which were consequently excluded (Table S2). For cervical intraepithelial neoplasia (CIN), the MR-PRESSO test flagged 10 microbial features with pleiotropy (P < 0.05); these underwent identical outlier removal procedures to ensure validity, while MR-Egger intercepts rejected 4 further associations post-correction (Table S3). Notably, all HPV16/18 infection-related analyses passed both pleiotropy tests (MR-PRESSO and MR-Egger P ≥ 0.05), confirming instrument validity for these exposures (Table S4,5).

Discussion

The progression of cervical cancer goes through the stages of HPV infection, intraepithelial neoplasia, and eventually cervical cancer, constituting a dynamic process.The gut microbiota has been confirmed to be closely related to the development of cervical cancer. However, there is still a lack of research confirming the association between the two, and we are the first to explore the relationship between the progression of cervical cancer lesions and the gut microbiota using Mendelian randomization. We identified different microbial communities that may play protective or risk factors in the progression of cervical cancer lesions. Surprisingly, we also found that some microbial communities may have a sustained effect at different stages of cervical cancer lesions.

The human gastrointestinal tract harbors approximately 1014microbes, with a genetic content in these microbial communities estimated to be approximately 100 times greater than that observed in the human genome)12.Their primary roles encompass the metabolism of dietary constituents, synthesis of vitamins, defense against pathogens, preservation of the integrity of the intestinal epithelial barrier, elimination of harmful substances, and modulation of inflammatory and immune responses11. Research has indicated that an imbalance in the gut microbiome, known as dysbiosis, can contribute to an increased susceptibility to specific malignancies and can impact the body’s reaction to various cancer treatments, including chemotherapy, radiotherapy, and immunotherapy37,38,39. For example, a study has shown that the gut microbiota can yield a set of biomarkers for the prediction, assessment of disease activity, and the selection of treatments in cases of radiation enteritis40.

Our study has revealed that certain conclusions drawn from our investigation into the relationship between gut microbiota and cervical cancer lesions using Mendelian randomization are consistent with some findings in existing research. The conclusion drawn suggests that genus Bilophila and genus.Faecalibacterium are high-risk factos for HPV infection in patients. Notably, Bilophila is a sulfur-reducing bacterium known to produce hydrogen sulfide (H₂S), a metabolite that can induce intestinal barrier dysfunction and systemic inflammation, potentially promoting immune dysregulation and facilitating viral persistence41. Recent studies highlight that Bilophila wadsworthiaencodes the enzyme isethionate sulfite-lyase (Isla), which produces genotoxic H₂S through sulfonate metabolism, a mechanism linked to colorectal carcinogenesis and potentially relevant to HPV persistence42. In contrast, Faecalibacterium, a prominent butyrate producer in the gut, exhibits anti-inflammatory effects under healthy conditions. However, its enrichment in HPV infection might reflect a compensatory mechanism of weakened mucosal immunity, or contextual shifts in microbial interactions that compromise pathogen clearance43. A study conducted a systematic comparative analysis of the gut microbiota in two groups of mice exhibiting distinct antitumor effects induced by an HPV E7 peptide-based vaccine. It was observed that Faecalibacterium, Lachnospira, Anaerostipes and Bilophila exhibited significantly lower levels in mice with suppressed tumor growth compared to those with persistent tumor growth44. Additionally, this research found that the relationships observed between gut microbiota members and T cells revealed that the proportions of CD4 + T cells and CD8 + T cells were inversely correlated with the abundance of Bilophila. Conversely, the abundance of Bilophila exhibited a positive correlation with the proportions of myeloid-derived suppressor cells (MDSCs), regulatory T cells (Tregs), and type 2-polarized tumor-associated macrophages (M2-TAMs). Furthermore, the ratio of Bilophila is diminished in the gut microbiota of postmenopausal women45. Our research has also revealed that the genera Faecalibacterium and Anaerostipes may act as risk factors for CIN, while the genus Lachnospira seems to provide protection against CIN. This divergence could be attributed to Lachnospira’s ability to generate short-chain fatty acids (SCFAs) like propionate, which enhance intestinal barrier integrity and suppress pro-inflammatory pathways,, as supported by its reduced abundance in IBD patients with colorectal neoplasia46,whereas Anaerostipes, despite producing beneficial SCFAs such as acetate, may contribute to dysbiosis in specific contexts through cross-feeding interactions with pathobionts47. Therefore, we conclude that Faecalibacterium in the gut microbiota continues to be a detrimental factor in the transition from HPV infection to CIN. Routy et al. analyzed the gut microbiome of patients undergoing PD-1 blockade therapy revealed that nonresponders had an abundance of Bacteroidales11. Intriguingly, Bacteroides fragilis (order Bacteroidales) is critical for the efficacy of CTLA-4 blockade immunotherapy, as its polysaccharides or adoptive T cell transfer restored antitumor responses in germ-free mice48, yet in our research, we found that the presence of order Bacteroidales is a risk factor for HPV18 infection49. Members of Bacteroidales are capable of degrading mucus glycoproteins, which may compromise gut barrier function and indirectly impair systemic immune surveillance against HPV infection through a “leaky gut” mechanism. Wang et al.12discovered that at the phylum level, Bacteroidetes constituted the predominant phylum, accounting for 54.19% and 51.96% of the gut microbiota in the CC group and the group of healthy female controls (HCs), respectively. Furthermore, they observed notable distinctions in the relative abundance of genus Dorea and genus Escherichia.Shigella between the CC and control groups (all ρ < 0.05), suggesting their potential utility as biomarkers for CC. Our analysis identified the phylum Bacteroidetes and the genus Dorea as risk factors for HPV18 infection, and the genus Escherichia.Shigella as risk factors for CIN. This aligns with observations in lung cancer, where Escherichia-Shigella is enriched in specific microbiota clusters linked to altered fecal metabolites50.While Bacteroidetes generally contribute to polysaccharide metabolism and immune tolerance, their overrepresentation might disrupt microbiota equilibrium, favoring pro-inflammatory responses51. Similarly, Dorea species are associated with lactate production and metabolic competition, potentially altering niche availability for protective commensals. Escherichia.Shigella, known for lipopolysaccharide (LPS) production, may directly activate Toll-like receptor 4 (TLR4) signaling, exacerbating chronic inflammation linked to CIN progression52.

The vaginal microbiota of women in good health demonstrates restricted microbial diversity and is predominantly inhabited by Lactobacillus species, which sustain a mutually advantageous association with the female host53,54. In the presence of estrogen, vaginal epithelial cells accumulate glycogen. Lactobacillus can metabolize this glycogen into lactic acid, which subsequently leads to a decrease in vaginal pH and a reduction in the colonization of pathogenic bacteria55. However, we observed that the genus Lactobacillus in the gut microbiota promotes the infection of high-risk HPV. This paradoxical role may stem from metabolic differences between vaginal and gut Lactobacillus. While vaginal strains protect against pathogens via acidification, gut-resident Lactobacillus may indirectly modulate systemic immunity through γ-aminobutyric acid (GABA) or other immunoregulatory metabolites, inadvertently promoting immune tolerance to HPV56. In a study investigating the impact of gut microbiota on patients undergoing radiotherapy and chemotherapy for CC, the compositional variations among patients were found to be associated with both short-term and long-term survival outcomes. Particularly, the family Porphyromonadaceae was significantly enriched in fecal samples from short-term survivors57. Through MR analysis, we discovered that the family Porphyromonadaceae serves as a high-risk factor for HPV16 infection. This aligns with Porphyromonadaceae’s propensity to degrade host glycans and generate secondary bile acids, which are implicated in chronic inflammation and immunosuppressive microenvironments58.

In addition to the documented roles of gut microbiota in the cervical cancer lesion process, we have identified new microorganisms that act as either protective or detrimental factors during cervical cancer lesion development. For instance, genus Adlercreutzia is associated with increased risk for both HPV16 and HPV18 infections. Meanwhile, the genus Eubacterium ventriosum group was found to elevate the likelihood of occurrence in both CIN and CC based on MR analysis. This group’s ability to produce acetate and formate may fuel oncogenic pathways or alter redox states in precancerous lesions. Consequently, the genus Eubacterium ventriosum group consistently exerts a detrimental influence during the transition from CIN to CC.

Earlier observational studies have revealed a connection between the gut microbiota and cervical cancer. Nevertheless, these results are insufficient to establish a direct causal relationship due to the potential impact of confounding variables, such as environmental and dietary factors. In cases where it is not feasible to explore the relationship between gut microbiota and the progression of cervical cancer lesions using RCTs, we employ Mendelian randomization as a viable approach to investigate the association between the two, which can provide robust evidence. These findings may translate to clinical applications in two key aspects: First, the identified microbiota signatures could complement existing screening strategies to stratify high-risk populations by assessing gut microbiome profiles from non-invasive fecal samples. Second, targeting detrimental taxa through dietary interventions or probiotics may serve as adjuvant strategies to conventional therapies. Notably, microbes like Bilophila demonstrating persistent risk effects across HPV infection, CIN, and cancer progression stages demand priority investigation into their sustained pathogenic mechanisms. Moreover, optimizing combined interventions that target stage-specific microbiota dynamics, such as suppressing high-risk taxa during precancerous phases while simultaneously enhancing protective commensals, could significantly amplify therapeutic synergy. Conversely, protective taxa might be harnessed to enhance mucosal immunity against HPV. Beyond the gut, emerging evidence suggests that intratumoral microbiota may play a role within cervical cancer lesions. Recent studies utilizing spatial transcriptomics and single-cell sequencing revealed that tumor-associated bacteria exhibit nonrandom organization in hypoxic, immune-suppressive microniches characterized by reduced vascularization and increased recruitment of myeloid cells59. Similar bacterial microniches in cervical tumors may promote cancer progression by inducing DNA damage, impairing antitumor immunity through pathways like STING suppression or PD-L1 upregulation, and metabolizing chemotherapeutic agents60,61. Furthermore, longitudinal analyses in metastatic tumors suggest that intratumoral microbiota could drive resistance to immune checkpoint inhibitors by fostering an anti-inflammatory microenvironment enriched in immunosuppressive neutrophils62. These findings align with our observations of gut microbiota-host interactions and underscore the need to explore whether cervical intratumoral microbiota synergistically interacts with systemic gut microbial signatures to shape disease outcomes.

Future studies should prioritize validating these microbial markers in diverse cohorts and exploring mechanistic links, such as how gut-derived metabolites systemically modulate cervical immune microenvironments. Additionally, spatial mapping of intratumoral microbes in cervical lesions could reveal whether bacterial “hotspots” correlate with regions of immune evasion or treatment resistance. Targeting such niches via localized antimicrobial agents or engineered probiotics might disrupt pro-tumorigenic bacteria while preserving protective commensals. It is also critical to dissect how continuously harmful microbes interact with host factors during multistep carcinogenesis, which may open avenues for precision microbiota modulation. While experimental validation of specific strains is warranted, our MR framework provides a prioritization roadmap for translating microbiota-host interactions into clinical management of cervical cancer.

We must acknowledge that our study has several limitations. Firstly, due to the lack of GWAS data from other regions, our analysis could only be based on GWAS summary data from European lineages, which limits the generalizability of this study. Second, while examining our results, we observed that the majority of p-values did not meet the Bonferroni correction threshold. However, the potential influence of nominal significance should not be underestimated. Third, we relaxed the p-value threshold when selecting the genetic instruments to include a larger number of SNPs. While this approach might potentially increase the risk of violating the first assumption in MR design, it’s important to note that we ensured the robustness of our instrument selection. The F-statistic for each SNP was consistently above 10, indicating the absence of weak instrumental variables in our MR estimation. Finally, while Mendelian randomization reduces confounding through genetic instrumental variables, residual bias arising from unmeasured environmental or genetic factors, such as diet, antibiotic exposure, and host epigenetics, may still persist. This could reduce the variability explained by genetic instruments and potentially influence the accuracy of causal inference. Furthermore, our findings are primarily hypothesis-generating and lack experimental validation; mechanistic studies such as microbial culture-based assays or animal models are needed to confirm the causal roles of specific taxa and clarify their biological pathways.

In conclusion, we utilized Mendelian randomization analysis to explore the inherent connections within the gut-cervix axis.(Fig. 1)We identified microorganisms that play distinct roles at different stages of cervical cancer development. Some of these have been previously reported but exhibit differences in causal associations compared to our findings, while others are newly discovered as protective or risk factors during the cervical cancer lesion process. Importantly, we observed that different microorganisms are involved at various stages of cervical cancer evolution, suggesting a stage-specific microbiota in cervical cancer lesions. Furthermore, we found overlapping microorganisms that exert consistent effects between different stages, from HPV infection to CIN and further progression to CC. Therefore, the results of our study are of great significance for future research in the screening, prevention, and treatment of cervical cancer.