Specific vaginal and gut microbiome and the anti-tumor effect of butyrate in cervical cancer women.

Mengzhen Han: Data curation, Writing – original draft. Na Wang: Data curation. Wenjie Han: Project administration. Xiaolin Liu: Visualization. Tao Sun: Writing – review & editing. Junnan Xu: Validation.

We collected data from studies containing 16S rRNA sequencing data from cervical cancer and associated populations (cervical intraepithelial neoplasias (CIN), HPV + and healthy) published on PubMed.gov. This work includes eight studies with accessible sample metadata and high-throughput sequencing. Raw sequencing data were downloaded from the National Center of Biotechnology Information (NCBI) via the Sequence Read Archive (SRA) tool using the identifiers: PRJNA518153 for Zehra et al. [19] , PRJNA687644 for Li et al. [20] , PRJNA524816 for Wang et al. [21] , PRJNA415526 for Chen et al. [22] , PRJEB7756 for Mitra et al. [23] , PRJNA685389 for Travis et al. [24] , PRJNA636012 for Li et al. [25] and PRJNA637228 for Zhou et al. [26] . In addition, this study also contains three external validation data the identifiers used are PRJNA725946 for Fan et al., PRJNA745060 for Sukyung et al. [27] and PRJNA753920 for Huang et al. [28] . The 16S rRNA data of 11 projects were filtered by data merging, removal of barcodes and primers, and quality control using VSERACH (v2.18.0), and the clean data were merged again, and then the feature table and representative sequences were obtained by redundancy and noise reduction. After duplicate data removal the sequences were amplified subsequence variants (ASV) with nearly 100 % homology, and the upse-out algorithm was used to select paired sequences with 97 % homology to operational taxonomic units (OTUs), selecting OTUs with a mean relative abundance greater than 1/10,000. ASVs were assigned taxonomy using the Ribosomal Database Project (RDP) classififier against the GreenGenes database (v13.8). Alpha diversity indices were calculated using QIIME1 based on feature tables. Chao1, Shannon, and Simpson index were performed between groups. R (v3.6.1) package phyloseq was applied to perform principal coordinate analysis (PCoA) based on Bray-Curtis distance, and groups differences were analyzed by permutation multivariate analysis of variance (PERMANOVA). Statistical significance was assessed by analysis of similarity (ANOSIM). Species difference analysis was performed using linear discriminant analysis (LDA) effect size (LEfSe) at the phylum and genus level to find biomarkers (LDA > 2). Venn diagrams were plotted for marker comparisons between groups. Correlation relationships between core microbes associated with cervical cancer were determined by co-abundance network analysis. Taxa are represented by different node colors, node degrees are represented by node sizes, and correlations are represented by the width of the connecting lines. Networks were generated by calculating associations between taxa through Spearman correlations. Only connections that were significantly correlated ( P < 0.05) and strongly correlated (≥ 0.7) were shown. The network was visualized using Gephi (v0.9). We constructed random forest (RF) models based on genus. Randomly selected 70 % of the data were used as training set to train and construct the model, the rest was used as validation set to verify the accuracy of the model, and the fitted model was validated by area under the Receiver Operating Characteristic (ROC) curve (AUC) analysis. To test the generality and robustness of the model, we performed study-to-study transfer validation and leave-one-dataset-out (LODO) validation on the entire sample, based on the methods of previous researchers [29] . In study-to-study transfer validation, we train the classifier in a single study and evaluate its performance. Meanwhile, we apply a nested cross-validation procedure to the training studies to calculate the accuracy. In LODO validation, the data from one study is set as the test set, while the data from all the remaining studies are combined as the training set. To assess the specificity of the important features for cervical cancer, we examined the performances in two non-cervical cancer diseases: pelvic inflammatory disease (PID) (vaginal data), and endometriosis (EM) (gut data). For each disease, RF models were constructed to discriminate the non-cervical cancer disease from controls. PICRUSt (v2.4.2) was used to predict the functional gene profile based on the full-length 16S rRNA sequence, and to predict the functional gene composition based on OTU. KEGG were used to detect intergroup enrichment pathways. Human cervical cancer cells Hela, Caski were purchased from ATCC. We confirmed that the cell lines used had been tested for mycoplasma. Hela cells were cultured in DMEM (Hyclone, USA) supplemented with 10 % fetal bovine serum (FBS, Hyclone, USA) and 1 % penicillin/streptomycin. Caski cells were cultured in RPMI-1640 supplemented with 10 % FBS and 1 % penicillin/streptomycin. Cells were cultured at 37 °C in an incubator containing 5 % CO 2 . Sodium butyrate (B5887) was purchased from Sigma-Aldrich. We treated Hela and Caski cells with a concentration gradient of 0.003 mM, 0.01 mM, 0.03 mM, 0.1 mM, 0.3 mM, 1 mM, 3 mM, and 10 mM butyric acid for 24, 48, and 72 h. Treated cells were incubated with 10 μL CCK-8 (Dojindo Laboratories, China) at 37 °C for 3–4 h. Absorbance was measured at 450 nm. Apoptosis was assessed with FITC Annexin V and PI Apoptosis Kit (GeneCodex, China). Cells were collected after different treatments and washed twice with PBS. Then, cells were resuspended in 100 μL of Binding Buffer and stained with 5 μL of Annexin FITC and 2 μL of PI for 15 min in the dark. The percentage of apoptotic cells was immediately analyzed using a NovoCyte ACEA flow cytometer (Agilent, China). We cultured the cells in 6-well plates (3.0 × 10 5 cells/well) with medium for 24 h. After treatment, cells were washed with 70 % ethanol, collected and fixed, and stored overnight at 4 °C. Subsequently, Tris–HCl buffer (pH 7.4) containing 1 % RNase A solution (Solarbio, China) was added to the cells and they were stained in each well using PI (5 mg/ml). Analysis was performed using NovoCyte ACEA flow cytometry and FlowJo (v10.8.1). Results are shown as means ± standard deviations (SDs). Statistical analyses were carried out by the Student's test using a statistical software package (SPSS, USA). Statistical significance was also taken as * P < 0.05, ** P < 0.01, and *** P < 0.001.

416 samples from the vagina were collected, and we divided the vaginal data into 3 groups, including the Normal group ( n = 180), the CINs group ( n = 152), and the Cervical cancer (Cervical_cancer) group ( n = 84). 116 samples from the gut, which included the Normal group ( n = 51), Cervical cancer (Cervical_cancer) group ( n = 65). Compared to healthy women, HPV infection, CINs and cervical carcinogenesis can indeed cause VM changes (Figure. S1). Although α diversity index showed an increase in diversity with increasing disease severity, the differences among Normal, HPV, and CINs groups were non-significant (Figure. S2), and the effect of HPV infection on VM composition was also not significant (Figure.S1; Figure. S2; Figure. S3). Data from vaginal samples at the phylum level were dominated by Firmicutes, Actinobacteria, Bacteroidetes, Proteobacteria , and Fusobacteria ( Fig. 1 A), and were similar to the results of previous studies. At the genus level, Lactobacillus was significantly lower in the Cervical_cancer group, while the proportion of other genera was generally higher. The difference between the CINs group and the Normal group was similarly insignificant ( Fig. 1 B). A general trend of increasing α diversity was observed with increasing disease severity ( Fig. 1 C). PCoA results shows the VM composition was different among studise ( R = 0.109; P = 0.001) and groups ( P = 0.001) ( Fig. 1 D). Fig. 1 Microbial composition and difference analysis of vaginal samples in Normal group, CINs group and Cervical_cancer group. (A) Stacked and circled graphs of species composition at the phylum level for the three groups of samples. (B) Stacked and circled graphs of species composition at the genus level for the three groups of samples. (C) Boxplots showing Alpha diversity of three groups of vaginal samples using different metrics (Chao1, Shannon and Simpson indices). (D) Principal coordinate analysis (PCoA) based on Bray-curtis distance for the three groups of samples, Anosim results showed different microbial composition between groups ( P = 0.001). (E) Histogram of the difference enriched genera between the three groups of vaginal samples analyzed by linear discriminant analysis (LDA) effect size (LEfSe). (F) Cladogram of differentially enriched taxa among the three groups of vaginal samples analyzed by linear discriminant analysis (LDA) effect size (LEfSe). Fig. 1 Microbial composition and difference analysis of vaginal samples in Normal group, CINs group and Cervical_cancer group. (A) Stacked and circled graphs of species composition at the phylum level for the three groups of samples. (B) Stacked and circled graphs of species composition at the genus level for the three groups of samples. (C) Boxplots showing Alpha diversity of three groups of vaginal samples using different metrics (Chao1, Shannon and Simpson indices). (D) Principal coordinate analysis (PCoA) based on Bray-curtis distance for the three groups of samples, Anosim results showed different microbial composition between groups ( P = 0.001). (E) Histogram of the difference enriched genera between the three groups of vaginal samples analyzed by linear discriminant analysis (LDA) effect size (LEfSe). (F) Cladogram of differentially enriched taxa among the three groups of vaginal samples analyzed by linear discriminant analysis (LDA) effect size (LEfSe). Comparing the enriched ASV corresponding to the genera in the Cervical_cancer and CINs groups (Figure. S4), we found Anaerococcus, Prevotella, Gardnerella etc. eight genera were enriched in both groups. For in-depth study of the alterations, we used LEfSe to screen out the differential taxa. At phylum level, Bacteroidetes, Cyanobacteria, Actinobacteria were enriched in cervical cancer women (Figure. S5A). At genus, Rhodococcus, Wolbachia, Peptostreptococcus, Gardnerella and 25 other genera were significantly enriched in Cervical_cancer group. While the CINs group was enriched in Atopobium, Sneathia, Shuttleworthia, Lactococcus , and Dialister . The healthy population was characterized by Lactobacillus, Bacillus, Megasphaera etc. ( Fig. 1 E; Fig. 1 F). The gut was also dominated by Firmicutes, Bacteroidetes, Proteobacteria, Actinobacteria etc. ( Fig. 2 A), but the percentages were different. Cervical cancer women Bacteroides, Prevotella etc. increased while Faecalibacterium, Roseburia, Blautia etc. decreased ( Fig. 2 B). The GM abundance and diversity showed an opposite trend to the vagina, and the GM α diversity was significantly lower in the Cervical_cancer group ( Fig. 2 C). In addition, PCoA results shows the gut microbiota composition was different among studise and groups ( R = 0.625; P = 0.001) ( Fig. 2 D). Fig. 2 Analysis of microbial composition and differences between gut samples of Normal group and Cervical_cancer group. (A) Stacked and circled graphs of species composition at the phylum level for the two groups of samples. (B) Stacked and circled graphs of species composition at the genus level for the two groups of samples. (C) Boxplots showing Alpha diversity of two groups of gut samples using different metrics (Chao1, Shannon and Simpson indices). (D) Principal coordinate analysis (PCoA) based on Bray-curtis distance for the 2 groups of samples, Anosim results showed different microbial composition between groups ( P = 0.001). (E) Histogram of the genus of the difference between the two groups of gut samples analyzed by linear discriminant analysis (LDA) effect size (LEfSe). (F) Cladogram of the differentially enriched taxa between the two groups of gut samples analyzed by linear discriminant analysis (LDA) effect size (LEfSe). Fig. 2 Analysis of microbial composition and differences between gut samples of Normal group and Cervical_cancer group. (A) Stacked and circled graphs of species composition at the phylum level for the two groups of samples. (B) Stacked and circled graphs of species composition at the genus level for the two groups of samples. (C) Boxplots showing Alpha diversity of two groups of gut samples using different metrics (Chao1, Shannon and Simpson indices). (D) Principal coordinate analysis (PCoA) based on Bray-curtis distance for the 2 groups of samples, Anosim results showed different microbial composition between groups ( P = 0.001). (E) Histogram of the genus of the difference between the two groups of gut samples analyzed by linear discriminant analysis (LDA) effect size (LEfSe). (F) Cladogram of the differentially enriched taxa between the two groups of gut samples analyzed by linear discriminant analysis (LDA) effect size (LEfSe). We used LEfSe to compare gut samples at the phylum (Figure. S5B) and genus level. 28 differentiated genera (LDA > 2.0) were distributed, of which 16 genera such as Bacteroides, Prevotella, WAL_1855D, Porphyromonas etc. were specific for Cervical_cancer, while 12 genera such as Faecalibacterium, Roseburia, Blautia etc. were gut microbial markers for healthy women ( Fig. 2 E; Fig. 2 F). To construct a noninvasive model for early screening and diagnosis of cervical cancer, we constructed RF models based on genus. The AUC of vaginal samples was 91.58 % in Cervical_cancer group versus Normal group ( Fig. 3 A); and 70.03 % in CINs versus Normal group (Figure. S6). The AUC was 99.95 % with the gut samples ( Fig. 3 B). Fig. 3 Performance of genera in early screening and diagnosis of cervical cancer using Recipient Operating Characteristic (ROC) curve analysis. (A) Cervical cancer subjects and Normal group vaginal genera were tested by ROC analysis. (B) Cervical cancer subjects and Normal group gut genus were tested by ROC analysis. (C) Heat map showing AUROC values for models constructed using genus characteristics in each cohort of the vaginal cervical cancer prediction model. (D) Heat map showing AUROC values for models constructed using genus features in each cohort of the vaginal CIN prediction model. (E) ROC analysis of the vaginal microbiome Cervical_cancer-Normal model in an independent 16S rRNA-seq validation cohort. (F) The comparison of the performances of genera among different microbiome-linked disease models: PID ( n = 41) vs. control ( n = 32) model, endometriosis ( n = 21) vs. control ( n = 20) model. Fig. 3 Performance of genera in early screening and diagnosis of cervical cancer using Recipient Operating Characteristic (ROC) curve analysis. (A) Cervical cancer subjects and Normal group vaginal genera were tested by ROC analysis. (B) Cervical cancer subjects and Normal group gut genus were tested by ROC analysis. (C) Heat map showing AUROC values for models constructed using genus characteristics in each cohort of the vaginal cervical cancer prediction model. (D) Heat map showing AUROC values for models constructed using genus features in each cohort of the vaginal CIN prediction model. (E) ROC analysis of the vaginal microbiome Cervical_cancer-Normal model in an independent 16S rRNA-seq validation cohort. (F) The comparison of the performances of genera among different microbiome-linked disease models: PID ( n = 41) vs. control ( n = 32) model, endometriosis ( n = 21) vs. control ( n = 20) model. To test the universality and robustness of the identified significant features, we performed study-to-study transfer and LODO validation on all vaginal samples. In the Cervical_cancer-vs-Normal classification model, the AUC for the study-to-study transfer validation ranged from 0.59 to 0.83, with a mean of 0.725. The AUC for the LODO validation ranged from 0.66 to 0.87 (mean AUC = 0.75) ( Fig. 3 C). In addition, we validated the CINs-vs-Normal model in the same way ( Fig. 3 D). It is easy to see that classifiers performed better in the comparison of Cervical_cancer-vs-Normal than CINs-vs-Normal. Moreover, an independent vaginal microbiome 16S rRNA-seq cohort favorably validated the validity of the vaginal Cervical_cancer-vs-Normal model ( Fig. 3 E). One gynecological disease, PID, was considered in this analysis. The AUC of the PID model were significantly lower than those of the cervical cancer model ( Fig. 3 F). In addition, we validated the GM model using EM gut 16S rRNA data. The AUC were significantly lower ( Fig. 3 F). To understand the interaction of VM in the Normal group, CINs, and Cervical_cancer group, we constructed 3 co-abundance networks. The network complexity of cervical cancer samples was higher than that of non-cancer samples ( Fig. 4 A-C). In Cervical_cancer group, we found SMB53, Turicibacter, Rhodococcus, Propionibacterium , and Proteus as the main hubs or keystone taxa ( Fig. 4 C). Fig. 4 Analysis of vaginal and gut microbial co-abundance network between cervical cancer patients and non-cancerous population. The color of nodes indicates different phylum, node size represents node degree, connecting line indicates the interaction between genera, and width of connecting line represents correlation. (A) Microbial co-abundance network of the vaginal Normal group. (B) Microbial co-abundance network of the vaginal CINs group. (C) Microbial co-abundance network of vaginal Cervical_cancer group. (D) Microbial co-abundance network of the gut Normal group. (E) Microbial co-abundance network of gut Cervical_cancer group. Fig. 4 Analysis of vaginal and gut microbial co-abundance network between cervical cancer patients and non-cancerous population. The color of nodes indicates different phylum, node size represents node degree, connecting line indicates the interaction between genera, and width of connecting line represents correlation. (A) Microbial co-abundance network of the vaginal Normal group. (B) Microbial co-abundance network of the vaginal CINs group. (C) Microbial co-abundance network of vaginal Cervical_cancer group. (D) Microbial co-abundance network of the gut Normal group. (E) Microbial co-abundance network of gut Cervical_cancer group. Similar trends were observed for the GM co-abundance network. First, complexity of the network was higher in the Cervical_cancer group than in the Normal group ( Fig. 4 D-E). Cervical_cancer group Epulopiscium, Peptoniphilus, Halomonas, Pseudomonas , and Anaerococcus had higher degree centrality ( Fig. 4 E). Epulopiscium was the most dominant genus in both groups, whereas the cancer group was more dominant and more strongly associated with other genera. In our study, Firmicutes were dominant in VM and GM from non-cancerous populations at the phylum level, whereas Firmicutes were significantly decreased in Cervical_cancer group ( Fig. 1 A; Fig. 2 A). At the genus level, a decrease in commensal bacteria and an increase in pathogenic bacteria were observed in both ecotopes in the Cervical_cancer group to some extent ( Fig. 1 B; Fig. 2 B). However, species diversity and richness showed opposite trends ( Fig. 1 C; Fig. 2 C). Comparison with existing studies of related diseases revealed that the vaginal disordered state tended to have elevated microbial diversity and decreased Lactobacillus dominance, whereas the gut health state tended to have diverse abundance, interactions, and stable coexistence of microbial members. Venn diagram was used for comparison and it was found that Anaerococcus, Bacteroides, Campylobacter, Finegoldia, Fusobacterium, Peptoniphilus, Porphyromonas , and WAL_1855D as two ecotopic shared markers were enriched in cervical cancer patients (Figure. S7). A total of 146 unique KEGG Orthology (KO) pathways at level 3 were predicted in Cervical_cancer group and Normal group. 102 pathways were enriched in Cervical_cancer including Biotin metabolism (ko00780), C5-Branched dibasic acid metabolism (ko00660), Fatty acid biosynthes (ko00061), etc. 43 pathways were enriched in Normal group including Phosphotransferase system (ko02060), d-Alanine metabolism (ko00473), Fructose and mannose metabolism (ko00051), etc. ( Fig. 5 A). Fig. 5 The analysis of functional annotation of the KEGG database combined with the relative abundance of vaginal and gut microbes. (A) Welch's test shows that Cervical_cancer group is significantly changed in level 3 KEGG pathway in vaginal samples. (B) Welch's test shows that Cervical_cancer group is significantly changed in level 3 KEGG pathway in gut samples. (C) Overlap of vaginal and gut Cervical_cancer group enriched pathway in Venn diagram. Fig. 5 The analysis of functional annotation of the KEGG database combined with the relative abundance of vaginal and gut microbes. (A) Welch's test shows that Cervical_cancer group is significantly changed in level 3 KEGG pathway in vaginal samples. (B) Welch's test shows that Cervical_cancer group is significantly changed in level 3 KEGG pathway in gut samples. (C) Overlap of vaginal and gut Cervical_cancer group enriched pathway in Venn diagram. A total of 152 unique KO pathways at level 3 were predicted in Cervical_cancer group and Normal group. Among them, 97 pathways were significantly different. 58 pathways were enriched in Cervical_cancer including Lipoic acid metabolism (ko00785), Lipopolysaccharide biosynthesis (ko00540), Folate biosynthesis (ko00790), etc. 39 pathways were enriched in Normal group including Biosynthesis of ansamycins (ko01051), Bacterial chemotaxis (ko02030), etc. ( Fig. 5 B). We compared the vaginal and gut microbial functions using Venn diagrams and found that 25 pathways co-enriched in VM and GM of women with cervical cancer. Such as Lipopolysaccharide biosynthesis (ko00540), Other glycan degradation (ko00511), Fatty acid metabolism (ko00071), Toluene degradation (ko00623), Biotin metabolism (ko00780), etc. ( Fig. 5 C). The effects of butyric on the proliferation of cervical cancer cells, including the Hela and Caski cell, were evaluated. The result showed that butyric were able to significantly inhibit Hela and Caski cell viability in a concentration- and time-dependent manner ( Fig. 6 A). To evaluate apoptosis of cervical cancer cells, the Hela and Caski cell lines were treated with butyrate at 1 mmol/L for 48 h. As shown in Fig. 6 B-C, the results showed that butyrate induced apoptosis in cervival cancer cells. In addition, we observed that butyrate significantly increased the percentage of G1 phase cells and decreased the percentage of S phase cells ( Fig. 6 d-E). These results suggest that butyrate induces cell cycle arrest in the G1 phase of cervical cancer cells. Fig. 6 Effects of butyric on proliferation, apoptosis and cell cycle of cervical cancer cells (Hela and Caski cells). (A) Effects of different butyric concentrations and treatment time on the cell viability of Hela and Caski cells. (B) Gating strategy for cell apoptosis assessment. Representative plots were obtained from 1 mM butyrate-treated Hela cells and 1 mM butyrate-treated Caski cells. (C) Histogram of apoptosis rate of Hela cells and Caski cells treated with 1 mM butyrate for 48 h. (D) Flow cytometry analysis of cell cycle phases in Hela cells and Caski cells following treatment with 1 mM butyrate for 48 h. (E) Percentage of cells in each phase of the cell cycle. Fig. 6 Effects of butyric on proliferation, apoptosis and cell cycle of cervical cancer cells (Hela and Caski cells). (A) Effects of different butyric concentrations and treatment time on the cell viability of Hela and Caski cells. (B) Gating strategy for cell apoptosis assessment. Representative plots were obtained from 1 mM butyrate-treated Hela cells and 1 mM butyrate-treated Caski cells. (C) Histogram of apoptosis rate of Hela cells and Caski cells treated with 1 mM butyrate for 48 h. (D) Flow cytometry analysis of cell cycle phases in Hela cells and Caski cells following treatment with 1 mM butyrate for 48 h. (E) Percentage of cells in each phase of the cell cycle.

Microbes are inextricably linked to cervical carcinogenesis. Microbiome is thought to alter host immunity and influence cancer risk and outcome, and we speculate that microbial factors influence cervical carcinogenesis. In addition, several studies have shown that dysbiosis affects HPV infection, CIN production and regression, and even cervical cancer, and that cervical cancer also disrupts the ratio between commensal and pathogenic microbes, leading microenvironment changes. However, there is no consensus on the role played by the microbiome in cervical cancer. Not only that, the low prevalence of cervical cancer screening is an important factor affecting the global cervical cancer burden, and there is an urgent need for a convenient noninvasive screening diagnostic method that can be accepted by women in all regions. Our study revealed variations in the species composition of women with cervical cancer in vaginal samples. However, a cervical cancer metagenomic sequencing in 2019 showed a significant decrease in Proteobacteria in cervical cancer patients [30] , possibly due to differences in clinical indicators and different analytical tools. α diversity index results were in good agreement with previous studies [31] , suggesting that changes in community diversity in VM may be related to pathological conditions, with greater community diversity likely to represent a higher degree of dysbiosis. β diversity analysis also showed significant differences among the three groups. In addition, key taxa were identified using LEfSe. Cervical_cancer group enriched 25 genera, suggesting that they may be associated with cervical carcinogenesis. In terms of abundance, GM most predominant phylum in healthy women were Firmicutes and cervical cancer women were Bacteroidetes . Some researchers have suggested that elevated Firmicutes / Bacteroidetes ratio is a sign of GM dysbiosis [32] , which is contrary to our findings. The variations in phylum are in good accordance with related studies [33] . Moreover, it has been shown that more Bacteroides may contribute to carcinogenic effects [34] , as shown by our results. Not only that, most of the butyrate-producing microbes belong to Firmicutes [35] . α diversity showed an opposite trend to the vagina, with significantly lower gut microbial diversity in Cervical_cancer group. It indicates that in the GM environment, a more complex flora is associated with a healthier gut state. Such an opposite trend could also be found in an experiment. Using antibiotics to treat mice, nearly half of the taxa enriched in vagina were significantly reduced in gut, while taxa depleted in the vagina were significantly enriched in the gut [36] . Antibiotics exerted negative changes on GM accompanied by positive changes in VM, including enrichment of beneficial bacteria such as Bacteroides, Ruminococcaceae , and Lachnospiraceae , and mice with high diversity of intravaginal flora were able to suppress tumorigenesis. That is not consistent with our results and may be due to the fact that both harmful and commensal bacteria were reduced in antibiotic-affected mice, with harmful bacteria being more affected. However the main reason for the increase in VM diversity in women with cervical cancer is the increase in pathogenic bacteria. This likewise points to the exchange of in two ecological niches flora. LEfSe analysis showed that 16 genera including Bacteroides, Prevotella, Porphyromonas and Fusobacterium were enriched in cervical cancer patients. These markers of microbial origin may be useful for the noninvasive diagnosis and may be a target of cervical carcer inhibition. The AUC of the models based on vaginal and gut genus were 0.92 and 0.99, respectively, indicating that VM and GM have good ability to predict cervical cancer. Besides, we know that SCFA such as butyric acid of microbial origin are involved in cancer inhibition to some extent [37] . We synthesized the results of 16S rRNA analysis of gut and found an interesting phenomenon. In our results, the abundance of SCFA producing bacteria Firmicutes, Roseburia, Ruminococcus, Blautia , and Faecalibacterium decreased in Cervical_cancer group. Not only that, in the results of LEfSe, in addition to the above bacteria, SCFA producers such as Bifidobacterium, Coprococcus and Clostridium were significantly enriched in Normal group. This suggests that the appropriate level of SCFA is important for maintaining homeostasis, while the decrease of SCFA may have some pro-cancer effect and reduce the protective effect. In addition, these results further support the possibility that bacteria and metabolites may migrate from the gut to vagina, with beneficial or harmful effects occurring at the cervix. Functional prediction pathways results suggested the potential role of VM and GM in cervical cancer development. The gut microbiome Lipopolysaccharide (LPS) biosynthesis was significantly enriched in the Cervical_cancer group. It is a consensus among scholars that LPS is closely related to the cancer process [38] . Fatty acid biosynthes and metabolism pathways were enriched in the Cervical_cancer group, suggesting that the concentration of fatty acids in the environment affects the role they play. In addition to the above-mentioned inhibitory effects of SCFA such as butyric acid on cancer, fatty acids are also closely associated with malignant processes such as cancer cell growth, proliferation, migration and invasion [39] . Scholars generally agree that dysregulation of lipid metabolism is an important phenotype that distinguishes cancer cells from normal cells, so fatty acid synthesis and metabolic pathways can be potential therapeutic targets for cervical cancer, and precise control of the process can help in cancer treatment. In addition, we found a large number of pathways co-enriched in the Cervical_cancer group in VM and GM, suggesting that cervical malignancy interacts with the metabolism of the flora in vivo to a certain extent, and the co-enriched pathways may serve as potential targets for future cervical cancer control. In this study, we also revealed the antitumor effects of butyrate on human cervical cancer cells, including inhibition of proliferation, induction of apoptosis and cell cycle arrest. Our results showed that different cancer cell line types responded differently to butyrate, presumably with different metabolic types of cells dominating the differences, suggesting that butyric acid can potentially interfere with cancer cell metabolism. And the effects on cell proliferation were dose- and time-dependent. It has been shown that the anticancer effect of butyric is attributed to its function as an inhibitor of histone deacetylase (HDAC) [40] . In various cancer cell lines, HDAC inhibitors inhibit cell proliferation through cell cycle arrest, induce differentiation and apoptosis, reduce angiogenesis and modulate the immune response. This suggests that modulation of the GM, especially for targeted metabolites such as butyrate, may be one of the future avenues for cervical cancer treatment. This is the first study to characterize microbial features and diagnostic models related to cervical cancer from both vaginal and gut ecological sites, and to integrate the results of VM and GM changes for comprehensive analysis. Our cervical cancer diagnostic models constructed in both vaginal and gut showed good diagnostic value and were externally validated interproject and specificity in order to demonstrate the good performance. However, there are still many shortcomings. First, the data used were from different regions, and differences in sampling practices, etc. are inevitable. Second, due to the limitations of the public data information on the clinical stage, histological typing, HPV infection type, vaginal community state types (CST) and sexual life were lacking, and the possible effects of these factors could not be accurately assessed. In addition, the vaginal Normal group involved multiple types of HPV infection women, and the small number of Cervical_cancer groups were key factors contributing to the error in the results. Finally, our sample of women for the cervical cancer developmental dynamics study is not from the same individual and there is no way to avoid inter-individual differences. However, it is our inadequate consideration of the confounding factors that allows our model to be applied to a wider range of patients with different clinical indications as a more general approach to adjuvant cervical cancer screening in the future. Since our study was limited to 16S rRNA data, the analysis result were limited to the genus level, while the species may be more representative, which is the result we lack. Future research should make more use of metagenomics, but attention needs to be paid to the cost of predictive models based on metagenomics sequencing and price acceptability. Secondly, microbes are not the same in healthy populations, and the influencing factors range from individual genotype to daily habits and environment. Larger studies involving more multifaceted factors, applicable to a larger population, should be conducted in the future. From the reverse perspective, future early screening and diagnosis should focus on individualized screening and finding the appropriate screening modality for each individual. Finally, the impact of microbes on cervical cancer and the role they play go far beyond cancer prediction screening; there are links to treatment efficacy, drug side effects, and patient prognosis, and the drug effects of microbes and interventions for cancer patients' flora should also be a research priority in the future. In conclusion, our study demonstrated the changes in both vaginal and gut ecological niches during cervical carcinogenesis. The models for adjuvant early screening based on the genus has good diagnostic value. We also confirmed the uniqueness of flora interactions and functional pathways in women with cervical cancer. Cervical cancer microbiome-specific metabolite butyric acid can inhibit cervical cancer cell proliferation, induce apoptosis and trigger cell cycle arrest. However, the relevance of the specific microbes obtained to cervical cancer requires further validation experimental studies in a broader population.

Cervical cancer is a serious threat to women's health. Its incidence and mortality rate are consistently among the top gynecologic malignancies [1] , but vary widely in developed and developing countries. Microorganisms present in the vagina can influence host-related physiological processes, resident flora can prevent or hinder many urinary tract and sexually transmitted diseases [ 2 , 3 ]. Dysbiosis of vaginal microbes (VM), on the other hand, decreases cervicovaginal barrier function [4] , increases adhesion, invasion and colonization by abnormal flora, alters metabolic profile and is associated with increased risk of inflammation, human papilloma virus (HPV) infection, premature birth, miscarriage and even gynecological cancers [5] . There is also a strong correlation between gut microbes (GM) and gynecologic diseases. GM can regulate the estrogenic enterohepatic cycle [6] , which affects target organs such as the breast and cervix [7] . There is increasing evidence that GM also play an important role in cervical carcinogenesis, participating in tumorigenesis, treatment, and prognosis. GM may also indirectly influence cervical carcinogenesis by affecting VM. First of all, the gut may be the origin of many VM. Both were dominated by Firmicutes, Bacteroidetes and Proteobacteria [8] . The number and species of GM are higher than those of vagina, and because of the anatomical connection, the migration and exchange of bacteria occur during the daily excretion and cleaning process. Secondly, a research team collected vaginal and rectal swabs from pregnant women at 35–37 weeks of gestation and found that 44 % of bacteria were present in both the vagina and the gut, and 68 % of the 50 strains in the subjects' vaginal and gut pairable samples had the same genotype [9] . Moreover, not only do microorganisms co-exist in the vagina and gut, but there is also a strong correspondence between the bacterial density of the two ecological niches [10] . Dysbiosis of GM is also associated with pregnancy outcomes in women [11] and oral probiotics improve VM dysbiosis to help alleviate diseases such as bacterial vaginitis (BV) [12] , suggesting a potential link between GM and VM. Also, the vagina and the oral have similar histological structures and microorganisms such as G.vaginalis have been detected in both environments [13] and HPV has been shown to be associated with oral cancer [14] . Further revealing that microbes may migrate and colonize unidirectionally or bidirectionally through certain pathways. The microbiota may play a role in disease progression through metabolic pathways. Some metabolites are derived solely from the microbiota, not the host, and are dominated by the GM [15] . The presence of GM and its nutritional short-chain fatty acid (SCFA) metabolites promotes the growth of intestinal epithelial cells, modulates immune and inflammatory responses, improves intestinal barrier function, increases vascular flow and motility, and regulates their differentiation and repair [ 16 , 17 ]. The ability of butyric acid to act as an anticancer agent by inhibiting growth and inducing apoptosis has been demonstrated in multiple studies [18] , but has been little studied in cervical cancer. In this study, we collected published data related to cervical cancer, and dug deeper into it. We obtained the microbial community diversity and taxonomic composition related to cervical cancer risk. In addition, we evaluated the anticancer effects of butyrate on human cervical cancer cells, including inhibition of proliferation, promotion of apoptosis, and effects on the cell cycle. In order to provide a general direction for in-depth research and mechanism excavation of microbial factors, to provide a theoretical basis for future flora intervention therapy and the development of anticancer agents, and to provide a potential adjunctive diagnostic modality.

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.