Biphasic pattern of microbial dysbiosis

A consistent microbial signature of treatment resistance has emerged from multi-cohort analyses encompassing over 1,200 cervical cancer patients undergoing definitive CRT. Meta-analytic synthesis reveals a biphasic pattern: depletion of protective commensals and pathobiont enrichment. Five independent studies demonstrate that L. crispatus dominance (relative abundance >50%) confers a 68% reduction in recurrence risk (Refs Reference Liao, Chen, Ruan, Wang, Hu, Long, Li, Zhang, Yu, Ming zhang, Zhang and Liao9, Reference Mitra, MacIntyre, Paraskevaidi, Moscicki, Mahajan, Smith, Lee, Lyons, Paraskevaidis, Marchesi, Bennett and Kyrgiou19, Reference Wang, Wang, Yan, Zhao, Wang, Liu, Fan and Xu36, Reference Armstrong and Kaul37, Reference Zheng, Hu, Liu, Zhao, Li, Wang, Zhang, Zhang, Song, Lyu, Cui, Ding and Wang38). A prospective longitudinal study tracking 234 patients throughout CRT demonstrated that individuals maintaining L. crispatus populations achieved a 3-year progression-free survival of 71%, compared to 39% in those experiencing early loss of this species (hazard ratio of 0.43, p = 0.003) (Ref. Reference Dou, Ai, Zhang, Li, Jiang, Wu, Zhao, Li and Zhang34). This protective effect appears mediated by L. crispatus’s ability to generate a low pH environment (<4.5) through d-lactate production and to preserve local IgA-mediated immune surveillance (Refs Reference Zhu, Frank, Radka, Jeanfavre, Xu, Tse, Pacheco, Kim, Pierce, Deik, Hussain, Elsherbini, Hussain, Xulu, Khan, Pillay, Mitchell, Dong, Ndung’u, Clish, Rock, Blainey, Bloom and Kwon26, Reference Breedveld, Schuster, Van Houdt, Painter, Mebius and Van Der Veer39).

Pathobiont enrichment and lactate production

In stark contrast, L. iners abundance exceeding 30% at baseline independently predicts treatment failure, with a hazard ratio of 2.14 (95% CI 1.45–3.16) for recurrence (Refs Reference Dou, Ai, Zhang, Li, Jiang, Wu, Zhao, Li and Zhang34, Reference Wei, Chen, Lu, Cao, Tang and Yang40). The mechanism linking L. iners to resistance involves its prodigious lactate production; this species converts glucose to l-lactate at rates of 15.3 ± 2.7 nmol/10⁶ cells/hour under hypoxic conditions typical of the tumour microenvironment, leading to intratumoural lactate accumulation from baseline 8.4 to 18.7 mM within 10 days (p < 0.001) (Refs Reference Colbert, El Alam, Wang, Karpinets, Lo and Lynn11, Reference Johnston and Bullman18). This concentration saturates monocarboxylate transporters and activates the HCA1 receptor, triggering downstream resistance pathways that reduce radiation-induced DNA damage (Refs Reference Wu, Wang, Ying, Jin, Li and Hu41, Reference Garcia-Flores, Sollome, Thavathiru, Bower and Vaillancourt42). Genomic analyses reveal that L. iners strains isolated from cancer patients uniquely encode a cytolysin operon that disrupts epithelial barrier integrity at bacterial loads exceeding 10⁷ CFU/ml, while downregulating hydrogen peroxide production in HIV-positive individuals, thereby eliminating its protective capacity (Ref. Reference Klein, Kahesa, Mwaiselage, West, Wood and Angeletti27).

Complex pathobiont signatures beyond Lactobacilli

Beyond Lactobacilli, a pathobiont index combining Fusobacterium nucleatum, Sneathia sanguinegens and Atopobium vaginae demonstrates superior predictive accuracy (area under curve of 0.81) compared to single-taxon markers (Refs Reference Sims, El Alam, Karpinets, Dorta-Estremera, Hegde and Nookala43, Reference Tortelli, Contreras, Markovina, Ding, Wylie and Schwarz44). Fusobacterium enrichment exceeding 1% correlates with elevated interleukin-6 and interleukin-8 levels, creating a pro-inflammatory niche that enhances tumour cell survival through NF-κB activation (Refs Reference Yu, Ni, Xu, Liu, Chen, Li, Xia, Diao, Chen, Zhu, Wu, Tang, Li and Ke5, Reference Lin, Zhou, Liu, Nie, Cao, Li, Li, Zhu, Lin, Ding, Jiang, Gu, Xu, Zhao and Cai6).

Temporal dynamics during therapy

Temporal dynamics during therapy reveal distinct trajectories in responders versus non-responders. The acute phase of CRT triggers immediate L. crispatus loss (log2-fold change −3.2 ± 0.8), while responders show parallel expansion of radiation-tolerant Lactobacillus gasseri that preserves lactic acid production and low pH (Ref. Reference Dou, Ai, Zhang, Li, Jiang, Wu, Zhao, Li and Zhang34). Non-responders, however, experience rapid L. iners proliferation (log2-fold change +4.1), correlating with lactate accumulation from 8 to 18 mM (Ref. Reference Colbert, El Alam, Wang, Karpinets, Lo and Lynn11). The subacute phase witnesses radiation-tolerant Enterobacteriaceae proliferation in non-responders, facilitated by lactate-induced epithelial barrier disruption, while microbial diversity declines by 45% in non-responders but remains stable in responders (p < 0.001) (Refs Reference Garcia-Flores, Sollome, Thavathiru, Bower and Vaillancourt42, Reference Jiang, Li, Zhang, Ma, Liu, Liang, Wei, Huang and Wang45). Six months post-treatment, microbial diversity recovery correlates with reduced recurrence risk (r = 0.61, p = 0.03), and patients achieving L. crispatus reconstitution exhibit 5-year overall survival of 78% versus 51% in persistently dysbiotic individuals (p = 0.01) (Refs Reference Wu, Wang, Ying, Jin, Li and Hu41, Reference Garcia-Flores, Sollome, Thavathiru, Bower and Vaillancourt42).

Geographic heterogeneity and host factors

In sub-Saharan Africa, where cervical cancer burden is highest, HIV co-infection drives a 3-fold enrichment of inflammatory pathobionts (P. bivia, G. vaginalis) and depletes protective L. crispatus. This dysbiosis is mechanistically linked to HIV-induced CD4⁺ T-cell depletion (reducing counts by 50–70%), which impairs mucosal IgA secretion – a critical factor for maintaining L. crispatus adhesion to vaginal epithelium. Concurrently, diminished CD4⁺ T-cell surveillance permits expansion of pro-inflammatory anaerobes that elevate local IL-1β and IL-6, further disrupting the mucosal barrier and activating HIF-1α/NF-κB pathways to promote therapy resistance. This dysbiotic signature, present in 68% of HIV+ patients, independently predicts CRT failure (adjusted HR = 2.34) (Refs Reference Klein, Kahesa, Mwaiselage, West, Wood and Angeletti27, Reference Sims, Yoshida-Court, El Alam, Ramatlho, Ketlametswe and Ning28, Reference Gosmann, Anahtar, Handley, Farcasanu, Abu-Ali, Bowman, Padavattan, Desai, Droit, Moodley, Dong, Chen, Ismail, Ndung’u, Ghebremichael, Wesemann, Mitchell, Dong, Huttenhower, Walker, Virgin and Kwon46, Reference Anahtar, Byrne, Doherty, Bowman, Yamamoto, Soumillon, Padavattan, Ismail, Moodley, Sabatini, Ghebremichael, Nusbaum, Huttenhower, Virgin, Ndung’u, Dong, Walker, Fichorova and Kwon47). Recent clinical trials demonstrate that restoring L. crispatus colonization via live biotherapeutics (e.g., LACTIN-V) reduces endocervical CD4⁺ T-cell activation-a key HIV target cell population-and may mitigate dysbiosis-driven treatment resistance (Ref. Reference Hemmerling, Mitchell, Demby, Ghebremichael, Elsherbini, Xu, Xulu, Shih, Dong, Govender, Pillay, Ismail, Casillas, Moodley, Bergerat, Brunner, Liebenberg, Ngcapu, Mbano, Lagenaur, Parks, Ndung’u, Kwon and Cohen35). Latin American cohorts show CST-IV dominance in 68% of high-grade lesions versus 23% of controls, with concurrent elevation of sialidase-producing G. vaginalis (Refs Reference Tortelli, Contreras, Markovina, Ding, Wylie and Schwarz44, Reference Tosado-Rodríguez, Alvarado-Vélez, Romaguera and Godoy-Vitorino48, Reference Nieves-Ramírez, Partida-Rodríguez, Moran, Serrano-Vázquez, Pérez-Juárez, Pérez-Rodríguez, Arrieta, Ximénez-García and Finlay49). These regional variations necessitate population-specific biomarker validation before global implementation.

Establishing causality beyond association

Establishing causality requires experimental validation beyond observational associations. Antibiotic-mediated microbiome depletion followed by L. iners reconstitution in patient-derived xenograft models demonstrates that L. iners colonization alone reduces CRT efficacy (tumour growth inhibition: 42% versus 68% in controls, p = 0.009), with lactate inhibition rescuing this effect, thereby fulfilling modified Koch’s postulates for microbiome-mediated resistance (Ref. Reference Colbert, El Alam, Wang, Karpinets, Lo and Lynn11). Such functional studies bridge the gap between correlation and causation, strengthening the rationale for microbiome-targeted interventions.

Lactate: the master metabolite driving resistance

The molecular machinery underlying microbiome-mediated resistance centres on lactate as a master metabolite that reprograms tumour cells, remodels the stromal microenvironment, and suppresses anti-tumour immunity (Figure 2). Studies have shown that Lactobacillus colonized in tumours can secrete a large amount of l-lactic acid in the tumour microenvironment, and induce metabolic reprogramming of tumour cells through mechanisms such as enhancing glycolysis, tricarboxylic acid cycle activity and REDOX balance. And it activates the HIF-1α signalling pathway, p53/ P73-dependent apoptotic pathway, FGFR signal transduction, etc., accelerating the repair of DNA double-strand breaks caused by radiotherapy, ultimately leading to CRT resistance. It is worth noting that lactic acid is not only a metabolic product but also participates in the metabolic coupling between tumour cells as a ‘metabolic fuel’, further enhancing treatment resistance (Refs Reference Colbert, El Alam, Wang, Karpinets, Lo and Lynn11, Reference Wu, Wang, Ying, Jin, Li and Hu41, Reference Faubert, Li, Cai, Hensley, Kim, Zacharias, Yang, do, Doucette, Burguete, Li, Huet, Yuan, Wigal, Butt, Ni, Torrealba, Oliver, Lenkinski, Malloy, Wachsmann, Young, Kernstine and DeBerardinis50, Reference Liu, Chen, Liu, Liu, Cai, You, Ke, Lai, Huang, Gao, Zhao, Pelicano, Huang, McKeehan, Wu, Wang, Zhong and Wang51, Reference Kantarci, Gou and Riley52, Reference Hsu, Yeh, Tsai, Chen and Tsou53, Reference Rozenberg, Zvereva, Dalina, Blatov, Zubarev, Luppov, Bessmertnyi, Romanishin, Alsoulaiman, Kumeiko, Kagansky, Melino, Ganini and Barlev54, Reference Sedelnikova, Redon, Dickey, Nakamura, Georgakilas and Bonner55, Reference Ziech, Franco, Pappa and Panayiotidis56, Reference Bartesaghi, Graziano, Galavotti, Henriquez, Betts, Saxena, Minieri, A, Karlsson, Martins, Capasso, Nicotera, Brandner, de Laurenzi and Salomoni57). These findings collectively indicate that the metabolites of the microbiome influence the treatment response to cervical cancer through multi-level metabolic reprogramming. It provides a new perspective for enhancing therapeutic effects by targeting the microbiota-metabolic axis.

Figure 2.

The core mechanisms of microbiome-driven resistance in cervical cancer. The tumour microenvironment (TME) exhibits dysbiosis with L. iners enrichment (red) and L. crispatus depletion (fading green). L. iners-derived lactate drives resistance via: (1) Metabolic-epigenetic axis: Lactate influx activates HCA1-PI3K/AKT signalling, enhancing DNA repair and promoting histone hyperacetylation/lactylation. (2) Immune suppression: Lactate inhibits CD8⁺ T cells and expands Tregs; L. iners lipoteichoic acid activates TLR2/NF-κB to upregulate PD-L1. (3) Direct drug metabolism: Tumour-resident Gammaproteobacteria express β-glucuronidase (GUS), inactivating chemotherapy via enzymatic modification.

At the cellular level, lactate uptake via monocarboxylate transporter 1 induces a pseudohypoxic state through stabilization of HIF-1α, even under normoxic conditions. In cervical cancer cell lines, 15 mM lactate increases HIF-1α protein half-life from 8 to 35 min (p = 0.002), leading to transcriptional upregulation of glycolytic enzymes and downregulation of oxidative phosphorylation genes (Ref. Reference Lu, Jin, Tang, Zhou, Xu, Su, Wang, Xu, Zhao, Yin, Zhang, Jia, Peng, Zhou, Wang, Chen, Wang, Yang, Chen and Chen24). Metabolic flux analysis using [¹³C₆]-glucose reveals that lactate-exposed cells redirect 62% of glucose carbon towards ribose synthesis, facilitating DNA repair and cell cycle progression post-irradiation (Ref. Reference Bartesaghi, Graziano, Galavotti, Henriquez, Betts, Saxena, Minieri, A, Karlsson, Martins, Capasso, Nicotera, Brandner, de Laurenzi and Salomoni57). This metabolic shift establishes a bioenergetic state that confers survival advantage during CRT. Additionally, Lactate-mediated cancer-associated fibroblast (CAF) transformation represents a critical resistance mechanism. Cervical CAFs exposed to 15 mM lactate upregulate α-SMA expression by 3.2-fold and secrete elevated levels of HGF, FGF2 and TGF-β (Ref. Reference Lin, Zhou, Liu, Nie, Cao, Li, Li, Zhu, Lin, Ding, Jiang, Gu, Xu, Zhao and Cai6). This creates a fibrotic, immunosuppressive extracellular matrix that physically impedes drug penetration and reduces oxygen diffusion, increasing hypoxia-induced radioresistance (Ref. Reference Gopalakrishnan, Spencer, Nezi, Reuben, Andrews, Karpinets, Prieto, Vicente, Hoffman, Wei, Cogdill, Zhao, Hudgens, Hutchinson, Manzo, Petaccia de Macedo, Cotechini, Kumar, Chen, Reddy, Szczepaniak Sloane, Galloway-Pena, Jiang, Chen, Shpall, Rezvani, Alousi, Chemaly, Shelburne, Vence, Okhuysen, Jensen, Swennes, McAllister, Marcelo Riquelme Sanchez, Zhang, le Chatelier, Zitvogel, Pons, Austin-Breneman, Haydu, Burton, Gardner, Sirmans, Hu, Lazar, Tsujikawa, Diab, Tawbi, Glitza, Hwu, Patel, Woodman, Amaria, Davies, Gershenwald, Hwu, Lee, Zhang, Coussens, Cooper, Futreal, Daniel, Ajami, Petrosino, Tetzlaff, Sharma, Allison, Jenq and Wargo58).

Epigenetic modulation represents another layer of lactate-mediated resistance. Lactate functions as an endogenous histone deacetylase inhibitor (Ki = 4.2 mM), competing with acetyl-lysine substrates and promoting global histone H4 hyperacetylation that correlates with intratumoural lactate levels (r = 0.68, p < 0.001) (Refs Reference Yu, Ni, Xu, Liu, Chen, Li, Xia, Diao, Chen, Zhu, Wu, Tang, Li and Ke5, Reference Ciszewski, Sobierajska, Stasiak and Wagner21). This hyperacetylation accelerates recruitment of DNA repair proteins including 53BP1 and RAD51, reducing radiation-induced γH2AX foci persistence by 45% at 24 h (p = 0.003) (Ref. Reference Sedelnikova, Redon, Dickey, Nakamura, Georgakilas and Bonner55). Lactate-derived histone lactylation, a novel post-translational modification, directly activates NDRG1 transcription, a known radioresistance driver (Ref. Reference Zhang, Fu, Leiliang, Qu, Wu, Wen, Huang, He, Cheng, Liu and Cheng59). These epigenetic changes create a durable resistance phenotype that persists beyond the initial microbial stimulus.

Microbial drug metabolism

Microbial expression of drug-metabolizing enzymes directly alters chemotherapeutic pharmacokinetics within the tumour microenvironment, which affects the bioavailability and toxicity of chemotherapy drugs. In patients with cervical cancer, the correlation between tumour and gut microbiome characteristics and the response to radiotherapy and chemotherapy has been observed. The gut microbiota can produce GUS, thereby affecting drug metabolism. Taking irinotegan as an example, its active metabolite SN-38 is detoxified by hepatic dinucleotide phosphoglucoside transferase into SN-38G and excreted into the intestine. However, GUS can re-dissociate SN-38G into the toxic SN-38, causing severe diarrhoea, which is dose-limiting toxicity (Refs Reference Nagar and Blanchard22, Reference Cheng, Tseng, Tzeng, Leu, Cheng, Wang, Chang, Lu, Cheng, Chen, Cheng, Chen and Cheng60, Reference Chen, Xu, Piao, Chang, Liu and Kong61, Reference Lin, Chen, Lin, Yeh, Hsieh, Gao, Burnouf, Chen, Hsieh, Dashnyam, Kuo, Tu, Roffler and Lin62, Reference Ting, Lau and Yu63). While GUS activity primarily impacts irinotecan toxicity, cervical cancer biopsies with high enzyme activity show reduced cisplatin-DNA adduct formation (r = −0.52, p = 0.008) and decreased apoptotic index (caspase-3+ cells: 12% versus 28% in low-activity tumours, p = 0.004) (Ref. Reference Nagar and Blanchard22). Metagenomic analysis identifies E. coli strains harbouring the gus operon in 34% of treatment-resistant tumours versus 9% of sensitive tumours (p < 0.001) (Ref. Reference Cheng, Tseng, Tzeng, Leu, Cheng, Wang, Chang, Lu, Cheng, Chen, Cheng, Chen and Cheng60). Co-administration of the specific GUS inhibitor TCH-3562 (10 mg/kg) restores cisplatin sensitivity in patient-derived xenografts, reducing tumour volume by 58% versus 23% with cisplatin alone (p = 0.009) (Ref. Reference Chen, Xu, Piao, Chang, Liu and Kong61). Tumour-enriched Gammaproteobacteria similarly express cytidine deaminase that metabolizes topotecan, reducing its cytotoxicity in vitro (IC₅₀ increase from 2.1 to 8.7 μM, p < 0.001), an effect reversible by tetrahydro-uridine (Ref. Reference Geller, Barzily-Rokni, Danino, Jonas, Shental, Nejman, Gavert, Zwang, Cooper, Shee, Thaiss, Reuben, Livny, Avraham, Frederick, Ligorio, Chatman, Johnston, Mosher, Brandis, Fuks, Gurbatri, Gopalakrishnan, Kim, Hurd, Katz, Fleming, Maitra, Smith, Skalak, Bu, Michaud, Trauger, Barshack, Golan, Sandbank, Flaherty, Mandinova, Garrett, Thayer, Ferrone, Huttenhower, Bhatia, Gevers, Wargo, Golub and Straussman23). Studies have shown that the microbial drug metabolism enzyme system is of universal importance in tumour treatment resistance (Ref. Reference Song, Yang, Ma, Wang and Leung64). These enzyme systems may alter the metabolokinetics of drugs under specific tumour microenvironmental conditions through spatiotemporal specific activation patterns.

Surgical outcomes and the microbiome

Studies have found that microbiome dysbiosis can affect postoperative recovery and treatment outcomes through multiple mechanisms. Emerging data from colorectal surgery suggests microbiome composition critically influences postoperative outcomes, with direct implications for cervical cancer surgery. In colorectal cancer, faecal microbiota transplantation (FMT) from anastomotic leak patients impairs wound healing in germ-free mice, associated with increased matrix metalloproteinase-9 activity and reduced collagen deposition (Ref. Reference Hajjar, Gonzalez, Fragoso, Oliero, Alaoui, Calvé, Vennin Rendos, Djediai, Cuisiniere, Laplante, Gerkins, Ajayi, Diop, Taleb, Thérien, Schampaert, Alratrout, Dagbert, Loungnarath, Sebajang, Schwenter, Wassef, Ratelle, Debroux, Cailhier, Routy, Annabi, Brereton, Richard and Santos65). Alistipes onderdonkii directly degrades extracellular matrix components, increasing leak risk (RR = 3.1, p = 0.01) (Refs Reference Hajjar, Gonzalez, Fragoso, Oliero, Alaoui, Calvé, Vennin Rendos, Djediai, Cuisiniere, Laplante, Gerkins, Ajayi, Diop, Taleb, Thérien, Schampaert, Alratrout, Dagbert, Loungnarath, Sebajang, Schwenter, Wassef, Ratelle, Debroux, Cailhier, Routy, Annabi, Brereton, Richard and Santos65, Reference Li, Zhang, Fu, Jiang, Zhang, Zhang, Chen, Tao, Chen and Zeng66). Translating to cervical cancer, radical hysterectomy patients with preoperative vaginal dysbiosis (Nugent score > 7) demonstrate increased postoperative cuff cellulitis (18% versus 6%, p = 0.02) and delayed wound healing (median 21 versus 14 days, p = 0.004) (Ref. Reference Ohigashi, Sudo, Kobayashi, Takahashi, Nomoto and Onodera67). These findings are not only applicable to colorectal cancer surgery but also provide important implications for the surgical treatment of cervical cancer. Microbial environment imbalance may also lead to local immune environment imbalance by disrupting mucosal barrier function, promoting chronic inflammatory responses and regulating immune cell function, increasing the risk of postoperative infection and delaying wound healing (Refs Reference Hajjar, Gonzalez, Fragoso, Oliero, Alaoui, Calvé, Vennin Rendos, Djediai, Cuisiniere, Laplante, Gerkins, Ajayi, Diop, Taleb, Thérien, Schampaert, Alratrout, Dagbert, Loungnarath, Sebajang, Schwenter, Wassef, Ratelle, Debroux, Cailhier, Routy, Annabi, Brereton, Richard and Santos65, Reference Li, Zhang, Fu, Jiang, Zhang, Zhang, Chen, Tao, Chen and Zeng66, Reference Ohigashi, Sudo, Kobayashi, Takahashi, Nomoto and Onodera67). Perioperative microbiome regulation for high-risk patients may become a new strategy to improve surgical outcomes.

Surgical site infections are similarly microbiome-dependent. Preoperative vaginal Staphylococcus aureus colonization increases post-hysterectomy infection risk 4.7-fold, with antibiotic prophylaxis efficacy reduced in the presence of biofilm-forming Enterococcus faecalis (Ref. Reference Di Modica, Gargari, Regondi, Bonizzi, Arioli and Belmonte68). Metagenomic analysis of infected surgical sites reveals enrichment of bacterial genes encoding antibiotic resistance and virulence factors, likely transferred from the vaginal reservoir (Ref. Reference Chen, Wei, Zhao, Zhou, Wang, Zhang, Zuo, Dong, Zhao, Hao, He and Bian69). Broad-spectrum prophylaxis reduces surgical site infections but causes long-term vaginal dysbiosis, with L. crispatus recovery taking >6 months in 68% of patients (Ref. Reference Zhang, Fu, Leiliang, Qu, Wu, Wen, Huang, He, Cheng, Liu and Cheng59). A pilot study implementing preoperative vaginal probiotic supplementation (L. crispatus suppositories, 10⁹ CFU twice daily for 7 days) reduced postoperative infections from 14 to 4% (p = 0.03) and accelerated microbiome normalization, suggesting perioperative microbiome optimization could improve both infectious and oncologic outcomes (Ref. Reference Mitra, MacIntyre, Paraskevaidi, Moscicki, Mahajan, Smith, Lee, Lyons, Paraskevaidis, Marchesi, Bennett and Kyrgiou19).

Radiotherapy outcomes and microbiome

Pelvic intensity-modulated radiation therapy combined with cisplatin-based chemotherapy represents the standard of care for locally advanced cervical cancer; however, radioresistance remains a major constraint on therapeutic efficacy. Longitudinal 16S rRNA sequencing analyses reveal significant dynamic shifts in the microbial community structure of both the vagina and local tumour sites during radiotherapy. Notably, the relative abundance of taxa such as Gammaproteobacteria and Acinetobacter increases progressively over the course of treatment (Ref. Reference Jiang, Li, Zhang, Ma, Liu, Liang, Wei, Huang and Wang45). Concurrently, reduced microbial diversity within cervical tumours is observed, with specific taxa like Bifidobacteriaceae showing a positive correlation with radiosensitivity. A higher abundance of these beneficial bacteria is found in patients exhibiting favourable responses to radiotherapy, suggesting their potential utility as predictive biomarkers for treatment outcome (Ref. Reference Dou, Ai, Zhang, Li, Jiang, Wu, Zhao, Li and Zhang34).

These microbial dynamics are closely linked to the modulation of the local immune microenvironment. A proposed mechanistic hypothesis suggests that intratumoural microbiota influence radiosensitivity by regulating immune cell infiltration, inflammatory cytokine production and the DNA damage repair capacity of tumour cells. For instance, certain bacterial species may enhance radiation-induced apoptosis by activating TLR/NF-κB signalling pathways, thereby augmenting tumour response to radiotherapy (Ref. Reference Zhou, Li, Ren, Wang and Wang70). Furthermore, microbial metabolites, including SCFAs, are implicated in modulating tumour cell responses to radiation by influencing metabolic reprogramming and immune regulation (Ref. Reference Miya, Marima, Damane, Ledet and Dlamini71).

Critically, the composition of the tumour-associated microbiome not only affects immediate radiotherapy response but is also associated with patient prognosis and recurrence risk. Overall, high diversity in both the gut and tumour microbiome is generally correlated with better treatment response and improved survival rates. Conversely, an increased abundance of specific pathobionts, such as Fusobacterium, may promote therapeutic resistance and tumour recurrence (Refs Reference Sims, El Alam, Karpinets, Dorta-Estremera, Hegde and Nookala43, Reference Tortelli, Contreras, Markovina, Ding, Wylie and Schwarz44).

In summary, microbial shifts are not merely epiphenomena of radiotherapy but likely function as active modulators of treatment efficacy. Future research should prioritize elucidating the precise molecular mechanisms orchestrated by key microbial species and their metabolites. This knowledge is essential for developing microbiome-based predictive models and targeted interventions aimed at enhancing the efficacy of CRT and improving patient outcomes in cervical cancer.

The L. iners paradox: a Janus-faced commensal

L. iners, as an important strain in the vaginal microecology, shows a complex duality in its role in cervical cancer and related diseases. Studies have shown that it has a certain protective effect in maintaining the balance of vaginal microecology. Under vaginal health conditions, L. iners are common and abundant. They can regulate the pH of the local environment by producing metabolic products such as lactic acid, and inhibit the colonization and growth of pathogenic bacteria. L. iners exemplifies the challenge of binary classification, as this species contributes to ecological stability through hydrogen peroxide production and NOD2-dependent antimicrobial peptide secretion in homeostatic environments (Refs Reference Zhu, Frank, Radka, Jeanfavre, Xu, Tse, Pacheco, Kim, Pierce, Deik, Hussain, Elsherbini, Hussain, Xulu, Khan, Pillay, Mitchell, Dong, Ndung’u, Clish, Rock, Blainey, Bloom and Kwon26, Reference Armstrong and Kaul37). In HPV-infected women, L. iners co-dominance correlates with increased viral clearance (OR = 1.73, 95% CI 1.12–2.68) (Ref. Reference Fan, Wu, Li, An, Yao, Wang, Wang, Yuan, Jiang, Li and Li72). However, in oncogenic contexts, the species transforms into a pathobiont, with cancer-derived strains showing 3.5-fold higher lactate production and upregulated cytolysin expression that disrupts epithelial barriers (Ref. Reference Colbert, El Alam, Wang, Karpinets, Lo and Lynn11). Studies have found through 16S sequencing analysis that the abundance of L. iners in the vaginal microbiota of patients with recurrent cervical cancer is significantly increased. As the most important microbial marker, it can be used to predict the risk of cancer recurrence before chemotherapy and radiotherapy (Ref. Reference Wang, Wang, Yan, Zhao, Wang, Liu, Fan and Xu36). This context-dependent pathogenicity precludes universal recommendations and demands personalized risk stratification integrating immune status and microbial load.

L. crispatus probiotics: bridging the efficacy gap

The microbiota structure mainly composed of L. crispatus in the vaginal microbiota is closely related to the cervical health status. Compared with the state of vaginal microbiota imbalance, the vaginal environment dominated by L. crispatus exhibits stronger immunomodulatory characteristics and can maintain a more favorable reproductive tract health state (Ref. Reference Armstrong and Kaul37). L. crispatus probiotics represent the most intuitive intervention, yet clinical validation remains incomplete. Preclinical data show that L. crispatus supernatant reduces L. iners-mediated lactate production by 45% through competitive glucose consumption and enhances epithelial barrier integrity via Wnt signalling activation (Refs Reference Dou, Ai, Zhang, Li, Jiang, Wu, Zhao, Li and Zhang34, Reference Fan, Wu, Li, An, Yao, Wang, Wang, Yuan, Jiang, Li and Li72). A phase I trial in 45 patients with CIN3 demonstrated achievable colonization in 73% of participants receiving vaginal suppositories (10⁹ CFU twice daily), with a non-significant trend towards reduced treatment-related adverse events (Grade 3+ toxicity: 31% versus 44% in historical controls, p = 0.19) (Ref. Reference Zeng, An, Li and Zhang73). The 2-year recurrence rate of 18% versus 29% in matched controls failed to reach statistical significance (p = 0.09), reflecting inadequate statistical power (Ref. Reference Zeng, An, Li and Zhang73). Dose-escalation studies reveal a narrow therapeutic window, with 10⁹ CFU superior to 10⁷ CFU for colonization (68% versus 28%), but 10¹⁰ CFU provoking inflammatory cytokines without added benefit (Refs Reference Zeng, An, Li and Zhang73, Reference Wang, Wang, Yu, Zhang, Hu, Xu, Zhang, Tian, Zheng, Lu, Hu, Guo, Cai, Geng, Zhang, Xia, Zhang, Li, Liu and Zhang74). Probiotics can enhance anti-cancer immune function (Refs Reference Gopalakrishnan, Spencer, Nezi, Reuben, Andrews, Karpinets, Prieto, Vicente, Hoffman, Wei, Cogdill, Zhao, Hudgens, Hutchinson, Manzo, Petaccia de Macedo, Cotechini, Kumar, Chen, Reddy, Szczepaniak Sloane, Galloway-Pena, Jiang, Chen, Shpall, Rezvani, Alousi, Chemaly, Shelburne, Vence, Okhuysen, Jensen, Swennes, McAllister, Marcelo Riquelme Sanchez, Zhang, le Chatelier, Zitvogel, Pons, Austin-Breneman, Haydu, Burton, Gardner, Sirmans, Hu, Lazar, Tsujikawa, Diab, Tawbi, Glitza, Hwu, Patel, Woodman, Amaria, Davies, Gershenwald, Hwu, Lee, Zhang, Coussens, Cooper, Futreal, Daniel, Ajami, Petrosino, Tetzlaff, Sharma, Allison, Jenq and Wargo58, Reference Ting, Lau and Yu63, Reference Matson, Fessler, Bao, Chongsuwat, Zha, Alegre, Luke and Gajewski75) by reducing Treg levels, promoting the activation of CD8 + T cells, the differentiation of CD4 + T cells and the infiltration of NK cells within tumours. L. crispatus supplementation can restore vaginal microecology, enhance local immunity, and improve the response to radiotherapy and immunotherapy. Applying Grading of Recommendations, Assessment, Development and Evaluation (GRADE) criteria, the certainty of evidence for L. crispatus probiotic efficacy is low, derived from small pilot studies with indirect endpoints, warranting only conditional recommendations within clinical trial settings.

Faecal microbiota transplantation: compartmental failure and safety risks

FMT has shown potential value in regulating the response to tumour treatment. Animal experiments have confirmed that FMT treated with antibiotics or from donors receiving antibiotic treatment can significantly weaken the anti-tumour activity of trastuzumab, manifested as impaired recruitment of CD4 T cells and granzyme B positive cells within the tumour (Ref. Reference Di Modica, Gargari, Regondi, Bonizzi, Arioli and Belmonte68). In the stress-induced colorectal cancer model, both FMT and antibiotic treatment can eliminate the stimulating effect of chronic stress on tumour progression, and the abundance of Lactobacillus johnsonii in stressed mice is significantly reduced (Ref. Reference Cao, Zhao, Su, Liu, Lin and Da76). Studies on liver injury induced by PFOS have shown that FMT can verify the causal role of the gut microbiota in mediating liver injury. After transplantation, the recipient mice exhibited similar characteristics of microbiota imbalance and amino acid metabolism disorders to the donors (Ref. Reference Song, Yang, Ma, Wang and Leung64). In terms of metabolic improvement, FMT from bariatric surgery mice can significantly improve the metabolic parameters of recipient mice, but this effect is strain-specific (Ref. Reference Liu, Tu, Shi, Fang, Fan, Zhang, Ding, Chen, Wang, Zhang, Xu, Sharma, Gillece, Reining, Jin and Huang77). FMT enhances the response rate of radiotherapy, chemotherapy and immunotherapy for cervical cancer by reshaping the intestinal/tumour microbiota, regulating the function and metabolic pathways of immune cells, and provides new ideas for overcoming drug resistance.

However, FMT faces unique barriers in cervical cancer due to anatomic compartmentalization and safety concerns. A pilot study administering oral FMT capsules to 12 platinum-resistant patients increased gut microbiome diversity but showed only transient vaginal L. crispatus elevation in 3 patients, with no objective tumour responses and one serious adverse event of donor-derived Enterococcus faecium bacteraemia (Ref. Reference Di Modica, Gargari, Regondi, Bonizzi, Arioli and Belmonte68). The mechanistic disconnect between gut and vaginal compartments, with faecal-vaginal microbial transmission efficiency below 0.1%, undermines FMT rationale (Ref. Reference Chen, Wei, Zhao, Zhou, Wang, Zhang, Zuo, Dong, Zhao, Hao, He and Bian69). Direct vaginal FMT carries unacceptable transmission risks for HPV and HIV, with no safety data in immunocompromised patients (Ref. Reference Khoruts and Sadowsky78). Current evidence supports conditional use of FMT only within clinical trials employing rigorous donor screening and direct vaginal application protocols.

Antibiotic stewardship: balancing infection control with microbiome preservation

Antibiotic stewardship presents a double-edged sword, as broad-spectrum antibiotic exposure within 30 days before CRT initiation reduces progression-free survival (pooled HR = 1.45, 95% CI 1.18–1.78) and overall survival (HR = 1.62, 95% CI 1.29–2.03) by depleting butyrate-producing taxa and impairing CD8+ T-cell function (Refs Reference Elkrief, Pidgeon, Maleki Vareki, Messaoudene, Castagner and Routy12, Reference Zhou, Li, Ren, Wang and Wang70). Conversely, targeted pathobiont eradication – such as metronidazole depletion of Fusobacterium – enhances radiation efficacy in preclinical models without compromising systemic immunity (Ref. Reference Gopalakrishnan, Spencer, Nezi, Reuben, Andrews, Karpinets, Prieto, Vicente, Hoffman, Wei, Cogdill, Zhao, Hudgens, Hutchinson, Manzo, Petaccia de Macedo, Cotechini, Kumar, Chen, Reddy, Szczepaniak Sloane, Galloway-Pena, Jiang, Chen, Shpall, Rezvani, Alousi, Chemaly, Shelburne, Vence, Okhuysen, Jensen, Swennes, McAllister, Marcelo Riquelme Sanchez, Zhang, le Chatelier, Zitvogel, Pons, Austin-Breneman, Haydu, Burton, Gardner, Sirmans, Hu, Lazar, Tsujikawa, Diab, Tawbi, Glitza, Hwu, Patel, Woodman, Amaria, Davies, Gershenwald, Hwu, Lee, Zhang, Coussens, Cooper, Futreal, Daniel, Ajami, Petrosino, Tetzlaff, Sharma, Allison, Jenq and Wargo58). Similarly, bacteriophage therapy targeting L. iners reduces tumour lactate by 55% and restores cisplatin sensitivity (Ref. Reference Zheng, Hu, Liu, Zhao, Li, Wang, Zhang, Zhang, Song, Lyu, Cui, Ding and Wang38). These findings support a nuanced approach avoiding empiric broad-spectrum antibiotics while reserving targeted eradication for pathobiont enrichment documented by microbiome sequencing.

Engineered bacteria: translational promise meets clinical reality

With the deepening understanding of the tumour microbiome (TM), the utilization of microorganisms themselves as a strategy for tumour treatment has shown great potential, among which engineered bacteria therapy is a cutting-edge direction in tumour microbiological therapy. Engineered bacteria refer to non-pathogenic bacteria that have been genetically modified and possess specific anti-tumour functions. These non-toxic and highly selective engineered microorganisms can enhance the efficacy of radiotherapy and chemotherapy. Such bacteria as Bifidobacterium, Salmonella and Clostridium can be modified to target the hypoxic and necrotic regions specific to solid tumours (Refs Reference Sims, Yoshida-Court, El Alam, Ramatlho, Ketlametswe and Ning28, Reference Tosado-Rodríguez, Mendez, Espino, Dorta-Estremera, Aquino, Romaguera and Godoy-Vitorino30, Reference Tosado-Rodríguez, Alvarado-Vélez, Romaguera and Godoy-Vitorino48). Colonization of engineered bacteria can increase the number of CD4+ and CD8+ T cells in tumours, reduce the proportion of FOXP3+ regulatory T cells, enhance the effector function of T cells (such as TNF production) and reduce the co-expression of PD-1/LAG-3. Meanwhile, the combination of engineered bacteria with PD-1/PD-L1 monoclonal antibodies and others can improve the penetration of drugs in deep tumours and overcome drug resistance, demonstrating a synergistic effect (Refs Reference Sims, Biegert, Ramogola-Masire, Ngoni, Solley, Ning, el Alam, Mezzari, Petrosino, Zetola, Schmeler, Colbert, Klopp and Grover29, Reference Nieves-Ramírez, Partida-Rodríguez, Moran, Serrano-Vázquez, Pérez-Juárez, Pérez-Rodríguez, Arrieta, Ximénez-García and Finlay49). The above findings indicate that engineered bacteria therapy can regulate the metabolism of the tumour microenvironment, thereby enhancing the efficacy of immunotherapy. E. coli Nissle 1917 engineered to express lactate oxidase reduces intratumoural lactate by 70% and enhances radiation sensitivity, while L. crispatus engineered with cytosine deaminase achieves 20-fold higher local 5-FU concentration (Refs Reference Lim, Yin, Ye, Cui, Papachristodoulou and Huang79, Reference Brevi and Zarrinpar80).

Even engineered bacteria offer theoretically precise interventions but remain confined to preclinical development. Firstly, ensuring the genetic stability and controllable reproduction of engineered bacteria in vivo is the key to preventing them from developing drug resistance or causing infections (Refs Reference Lim, Yin, Ye, Cui, Papachristodoulou and Huang79, Reference Xie, Fan, Cheng, Yin, Li, Wegner, Chen and Zeng81). Secondly, clinical transformation is restricted by multiple aspects such as safety, immune rejection reactions, and regulatory supervision. A large number of basic and clinical studies are still needed to verify its efficacy and safety (Refs Reference Brevi and Zarrinpar80, Reference Sharma, Sharma and Gogoi82). In addition, different tumour types and individual differences among patients also have a significant impact on the therapeutic effect of engineered bacteria. It is necessary to precisely screen suitable strains and design individualized treatment plans (Refs Reference Deb, Wu, Coker, Harimoto, Huang and Danino83, Reference Radford, Vrbanac, De Nys, Worthley, Wright and Hasty84). So, no engineered bacteria have entered cervical cancer trials, with genetic instability, immune clearance and regulatory hurdles posing formidable barriers (Refs Reference Xue, Wang and Liu85, Reference Chang, Liu, Wang, Ma, Liang and Li86). Technology readiness levels remain at 3–4, restricting recommendation to research settings only.

Multi-omics integration and predictive model development

Current research has confirmed that the tumour microbiome can influence the signalling, metabolism and proliferation processes of cancer cells by generating active metabolites (Refs Reference Song, Yang, Ma, Wang and Leung64, Reference Fernandez, Wargo and Helmink87). The immediate research priority involves constructing integrated multi-omics frameworks that transcend single-layer microbiome characterization. Current studies demonstrate that tumour microbiome features alone explain approximately 30–40% of variance in CRT response, indicating that microbial signatures must be contextualized within host genetics, immune status and metabolic networks to achieve clinically actionable predictive power (Refs Reference Dai, Tan, Qiao and Liu7, Reference Wahid, Dar, Jawed, Mandal, Akhter, Khan, Khan, Jogaiah, Rai and Rattan10). Machine learning algorithms capable of integrating vaginal microbiome sequencing data, metabolomic profiling of lactate and short-chain fatty acids and immune checkpoint expression (PD-L1, MAdCAM-1) represent a critical near-term objective (Refs Reference Liu, Chen, Zhang and Dong13, Reference Boesch, Horvath, Baty, Pircher, Wolf, Spahn, Straussman, Tilg and Brutsche88).

Prospective validation of such models requires standardized data acquisition protocols across geographically diverse cohorts. The cervical cancer microbiome consortium, comprising institutions from sub-Saharan Africa, Latin America and Asia-Pacific, has initiated harmonized sampling procedures using unified collection devices, DNA extraction kits, and bioinformatic pipelines to reduce technical heterogeneity that currently confounds meta-analyses (Ref. Reference Mitra, MacIntyre, Paraskevaidi, Moscicki, Mahajan, Smith, Lee, Lyons, Paraskevaidis, Marchesi, Bennett and Kyrgiou19). This effort must incorporate host genotyping for HLA-DRB1 and other alleles that modulate microbiome resilience, as emerging data indicate host genetic factors contribute significantly to microbial community stability during therapy. Importantly, these predictive models must be trained on hard clinical endpoints – 3-year progression-free survival rather than surrogate markers like microbial diversity indices – to ensure clinical relevance.

Synthetic biology: engineering next-generation microbiome therapeutics

Synthetic biology offers theoretically precise interventions, but current designs remain confined to preclinical development requiring substantial safety engineering before human application. The most promising near-term application involves developing ‘smart probiotics’ with enhanced colonization capacity and controlled metabolite production. Chromosomal integration of lactate oxidase or butyrate synthesis operons into L. crispatus chassis strains could circumvent plasmid instability issues that cause 15–20% gene loss per generation in current designs (Ref. Reference Lim, Yin, Ye, Cui, Papachristodoulou and Huang79). However, such engineering must be coupled with robust biocontainment strategies – auxotrophic dependencies that restrict growth to the tumour microenvironment (e.g., requirement for exogenous d-alanine) have proven effective in colorectal cancer models but remain untested in vaginal applications (Ref. Reference Brevi and Zarrinpar80).

Engineered bacteriophage therapy represents a parallel strategy with potentially superior safety profiles. Lytic phages specific for L. iners (e.g., φLinq1) can selectively deplete pathogenic populations without disrupting beneficial commensals, reducing tumour lactate by 55% in preclinical xenografts (Ref. Reference Zhang, Fu, Leiliang, Qu, Wu, Wen, Huang, He, Cheng, Liu and Cheng59). Unlike bacterial therapies, phages are self-limiting and do not persist after target elimination, minimizing ecological disruption. However, phage development is hindered by narrow host ranges and rapid bacterial resistance evolution, necessitating cocktail formulations that increase manufacturing complexity. Regulatory pathways for phage products remain undefined, with current FDA guidance requiring demonstration of purity, consistency and absence of endotoxin contamination that many academic laboratories cannot achieve (Ref. Reference Lee, McClure, Weichselbaum and Mimee89).

Clinical trial standardization and adaptive design

The heterogeneity across current microbiome intervention trials – encompassing disparate patient populations, dosing regimens and outcome measures – precludes meaningful synthesis and delays progress. Establishing an international master protocol for adaptive platform trials is imperative. Such trials should stratify patients by baseline microbiome signatures (L. iners-dominant versus L. crispatus-dominant) and randomize to microbiome modulation arms (probiotic, phage, or engineered bacteria) versus standard-of-care (Ref. Reference Fernandez, Wargo and Helmink87). Primary endpoints must include progression-free survival at 2 years, with mandatory serial microbiome and metabolome sampling to capture dynamic changes. Exploratory endpoints should integrate patient-reported quality-of-life metrics, as vaginal dysbiosis correlates with dyspareunia and treatment-related discomfort that traditional oncologic outcomes overlook (Ref. Reference Liao, Chen, Ruan, Wang, Hu, Long, Li, Zhang, Yu, Ming zhang, Zhang and Liao9).

Standardization extends beyond trial design to sample processing and data analysis. The International Cancer Microbiome Consortium has proposed guidelines requiring parallel 16S rRNA and shotgun metagenomic sequencing, with mandatory reporting of absolute bacterial load via spike-in controls rather than relative abundance alone (Ref. Reference Bhattarai, Du, Zeamer, Morzfeld, Kellogg and Firat90). Bioinformatic pipelines must be benchmarked against synthetic mock communities to ensure inter-laboratory comparability. These technical standards should be enforced by regulatory agencies as prerequisites for Investigational New Drug applications for microbiome-based therapies.

Critical controversies and cautionary principles

Several foundational controversies must be addressed to prevent premature translation that could harm patients and discredit the field. First, the causality question remains incompletely resolved. While PDX models and antibiotic depletion studies support causal roles for specific taxa, these experiments cannot fully exclude host factors that may independently drive both dysbiosis and resistance. Mendelian randomization studies using host genetic variants that influence microbiome composition may help disentangle causation from confounding, but such analyses require massive sample sizes currently unavailable (Ref. Reference Javdan, Lopez, Chankhamjon, Lee, Hull, Wu, Wang, Chatterjee and Donia91).

Second, ecological complexity warnings against oversimplified ‘good bacteria/bad bacteria’ frameworks. The L. iners/L. crispatus ratio, while predictive, captures only a fraction of microbial interactions. Keystone species like A. vaginae may exert disproportionate effects through cross-feeding networks, and elimination of one pathobiont may enable replacement by another equally deleterious species (Ref. Reference Cao, Zhao, Su, Liu, Lin and Da76). This ecological unpredictability necessitates whole-community intervention approaches rather than single-species targeting.

Third, cost-effectiveness and equity considerations loom large. Microbiome sequencing and targeted probiotics add approximately $800–1,200 to treatment costs per patient, a significant burden in low-resource settings where cervical cancer mortality is highest (Ref. Reference Deb, Wu, Coker, Harimoto, Huang and Danino83). Without demonstrated cost-effectiveness compared to standard supportive care, such interventions risk exacerbating health disparities. Implementation research must prioritize affordable diagnostics, such as point-of-care lactate assays, and locally produced probiotics to ensure global applicability. To reduce costs, alternative diagnostic methods like 16S rRNA sequencing could be used to identify key microbial markers at a lower cost. Additionally, partnerships with local manufacturers could enable the production of cost-effective probiotics, ensuring broader access in underserved populations.

Finally, regulatory science lags behind basic discovery. The FDA and EMA have yet to establish quality metrics for ‘live biotherapeutic products’ specific to vaginal applications, including acceptable limits on gene transfer, environmental persistence and immunogenicity (Ref. Reference Lee, McClure, Weichselbaum and Mimee89). Without clear regulatory pathways, academic discoveries will stall in translation, and patients may turn to unproven commercial probiotics with unsubstantiated claims. Urgent dialogue between researchers, regulators and patient advocacy groups is needed to create fit-for-purpose approval frameworks.

In immunocompromised patients, such as those with HIV co-infection, microbiome interventions must be approached with caution due to the heightened risk of infection and immune dysregulation. These patients are more susceptible to opportunistic infections, and microbiome modulation may inadvertently introduce harmful pathogens or exacerbate imbalances in the microbiota. Additionally, interventions like FMT or probiotics could trigger unwanted immune responses. Thus, strict safety protocols, including comprehensive screening and monitoring, are essential when applying microbiome therapies to immunocompromised individuals. Preclinical studies and close clinical supervision are critical to ensuring the safety and efficacy of such interventions.