Fecal microbiome patterns linked to immune response and antibiotic effects in lung cancer

Over the past decade, immune checkpoint inhibitors (ICIs) targeting PD-1/PD-L1 and CTLA-4 have revolutionized the therapeutic landscape of lung cancer (LC). However, only approximately 20% of patients with locally advanced or metastatic disease derive sustained clinical benefit from immunotherapy (Cascone et al., 2022; Diao et al., 2025; Huang et al., 2022). Therefore, identifying reliable predictive biomarkers and elucidating the mechanisms underlying responses to ICIs remain key scientific challenges in the field of tumor immunotherapy.

Although PD-L1 expression is commonly employed as a predictive biomarker in clinical practice, it has limited prognostic value in squamous cell carcinoma and exhibits significant spatial and temporal heterogeneity between primary and metastatic lesions (Doroshow et al., 2021; Hirsch et al., 2022). Similarly, tumor mutational burden (TMB) has been proposed as an alternative marker; however, the lack of consensus on threshold definitions and technical inconsistencies across sequencing platforms hinder its routine clinical application (Sholl et al., ). Emerging strategies, such as profiling the tumor immune microenvironment, analyzing circulating tumor DNA (ctDNA), and applying high-dimensional single-cell omics, hold promise, but their clinical utility is constrained by high cost, insufficient standardization, and technical complexity (Datar et al., 2019; Leng et al., 2021; Zhang et al., 2020). Therefore, there remains a pressing need to develop low-cost, non-invasive, and biologically stable biomarkers with a robust predictive value for immunotherapy response.

Our previous work demonstrated that LC progression is associated with significant alterations in the composition of the host-associated microbiota, suggesting a potential role of the gut microbiome in LC development and immune regulation (Han et al., 2023). The influence of the GM on cancer immunotherapy has garnered increasing attention in recent years. As early as 2007, murine models first demonstrated that gut microbes can activate antitumor immune responses, marking the inception of microbiota–immunity research (Paulos et al., 2007). In 2015, prospective studies further confirmed a direct association between gut microbial composition and ICI efficacy (Vétizou et al., 2015). In 2018, fecal microbiota transplantation (FMT) from responders was shown to enhance ICI sensitivity in recipient mice (Routy et al., 2018). By 2021, human-to-human FMT was successfully translated into clinical settings, offering a novel approach to overcome resistance to PD-1 blockade (Davar et al., 2021).

Several single-center studies conducted in South Korea, China, the United States, and France have investigated gut microbiome alterations associated with ICI response (Fang et al., 2022; Routy et al., 2018; Lee et al., 2021; Liu et al., 2022). However, the heterogeneity of findings across cohorts raises concerns regarding the reproducibility and translational applicability of candidate microbial biomarkers. Among them, Akkermansia muciniphila (AKK) has been repeatedly reported as a favorable taxon in responders and was highlighted as a top-ranked biomarker in a landmark study published in Science (Routy et al., 2018). Using spatial metabolomics, Zhu et al. demonstrated that Akk can translocate from the gut to the bloodstream, colonize lung tumor tissue, and exert antitumor effects by modulating tumor-associated microbiota and reprogramming tumor metabolism (Zhu et al., 2023). However, contrasting findings by Lee et al. reported the enrichment of Akk in non-responders, further underscoring inconsistencies in the current literature (Lee et al., 2021).

To date, no large-scale, multi-cohort meta-analysis has systematically evaluated GM signatures associated with ICI response in LC. To address this gap, we conducted an integrative analysis of publicly available metagenomic datasets from multiple countries and regions. Using a standardized metagenomic pipeline and machine learning-based modeling framework, we aimed to identify reproducible microbial features strongly associated with immunotherapy response and to discover stable and clinically translatable microbial biomarkers. Our findings provide critical insights into the microbiota–immune axis and establish a foundation for microbiome-informed personalized immunotherapy strategies in LC.

In this meta-analysis, we conducted the largest metagenomic study to date focusing on the gut microbiome in the context of immunotherapy for LC. Participants were recruited from five independent cohorts across the globe, aiming to establish a broader and more reproducible microbiome signature associated with response to ICIs.

Overall, the responders exhibited higher species richness. Although this trend did not reach statistical significance—likely due to inter-cohort heterogeneity—it aligns with previously reported findings from both Chinese and US cohorts (Heshiki et al., 2020; Lee et al., 2021). The observed decrease in α-diversity among non-responders suggests a more profound dysbiosis in this group. Similar patterns of reduced α-diversity have been reported in various metabolic disorders and are closely associated with disease states (Dang and Marsland, 2019; Hakozaki et al., 2020). However, findings across studies remain inconsistent—for instance, in the study by Lee et al., non-responders demonstrated greater microbial diversity, whereas David et al. found no significant differences in α-diversity between responders and non-responders (Lee et al., 2021; Dora et al., 2023). Although reduced α-diversity has been proposed as a characteristic feature in melanoma immunotherapy, its relevance as a predictive biomarker for ICI efficacy in lung cancer remains uncertain and warrants further validation using real-world datasets.

At the taxonomic level, intergroup differences were primarily observed at the species level, and the use of ATBs appeared to influence the abundance of specific key taxa. Regarding the debated enrichment status of the “signature” bacterium AKK, our study demonstrated that AKK was relatively enriched in responders across both genus and species levels in the overall population and the ATB-exposed subgroup. In contrast, in participants who had not received ATBs, AKK was enriched in non-responders. These findings suggest that, in responders, ATBs may exert a stronger suppressive effect on non-AKK taxa—particularly those associated with non-response—thereby indirectly enhancing the relative abundance of AKK. Consequently, a key question for future clinical research is whether targeted antibiotic interventions can modulate the abundance balance between responder- and non-responder-associated microbiota to improve the efficacy of immunotherapy. This represents an important direction for mechanistic and translational investigation.

The R-specific strain B. caccae and the NR-specific strain P. copri identified in our study have been previously reported in the context of cancer immunotherapy. Both B. caccae and AKK exhibit ecological variability and have been associated with favorable outcomes in response to ICIs (Liu et al., 2023). Enrichment of B. caccae has also been observed in responders with melanoma (Frankel et al., 2017). A systematic review by Mariam et al., encompassing 18 studies across five types of solid tumors, reached a similar conclusion—B. caccae was enriched in patients with better prognoses. Whether B. caccae produces known immunomodulatory metabolites remains unknown. Future research should employ metabolomics, along with in vitro co-culture systems or organoid models, to uncover potential immune-regulatory mechanisms associated with this strain. In contrast, the pathogenic potential of P. copri has been more extensively studied. Xiang et al. demonstrated that GM transplantation from patients harboring P. copri induced severe inflammation and immune dysregulation in murine models (Qian et al., 2022). Furthermore, P. copri has been implicated in the pathogenesis of carboplatin-induced intestinal mucositis, and the abundance of the Prevotella genus has been linked to tumor invasiveness in melanoma (Yu et al., 2019; Vitali et al., 2022). Despite these pro-inflammatory and pro-tumorigenic associations, some studies have reported a higher abundance of P. copri in responders (Jin et al., 2019; Robinson et al., 2023). Whether this strain contributes to limiting the long-term benefits of ICI therapy remains an open question, requiring further investigation through mechanistic and in vivo studies.

In recent years, numerous high-impact studies have demonstrated the feasibility and clinical potential of microbiome-based predictive models for therapeutic response—for instance, Huang et al. developed a cross-cohort, pan-cancer immune response prediction model achieving an AUC of up to 0.89 (Huang et al., 2023). A more recent study published in Nature Medicine further confirmed that strain-level abundance differences derived from metagenomic approaches can be utilized to predict immunotherapy response across multiple cancer types (Gunjur et al., 2024). In the context of LC, existing models primarily focus on distinguishing patients with long versus short PFS. Notably, Liu et al. reported that models based on differential microbial taxa yielded suboptimal performance, whereas those leveraging microbial functional protein profiles achieved superior predictive accuracy, with an AUC of 0.95 (Liu et al., 2022). In addition, Lisa et al. introduced a threshold-based stratification using the abundance of Akk, defining patients as positive or negative subgroups, further highlighting the potential of this species in predicting immunotherapy response (Derosa et al., 2022).

Although models constructed using WGS sequencing data exhibit robust classification performance, our approach faced limitations during external validation. To maximize data utilization, all available WGS samples were included in the training set. Consequently, an independent validation cohort at the genus level was retrieved through systematic data mining, relying on 16S rRNA sequencing data. Due to inherent differences in taxonomic resolution and annotation depth between WGS and 16S platforms, the model’s AUC dropped to 0.61 in the 16S cohort, suggesting limited cross-platform generalizability. Nonetheless, this result remains informative, especially in the absence of large-scale independent WGS cohorts. Moving forward, we will continue monitoring the release of new WGS datasets and explore the integration of multi-omics data, including metagenomics, metabolomics, and transcriptomics, to enhance model robustness and cross-platform adaptability. As global, multi-center metagenomic datasets continue to grow, we anticipate further validation and refinement of our findings and models in broader population contexts.

KEGG-based functional profiling revealed a significant enrichment of metabolism-related pathways across all three groups, suggesting a common microbial contribution to metabolic reprogramming associated with tumor progression and survival (Li and Zhang, 2016). Notably, in immunotherapy responders, pathways involved in amino acid biosynthesis (e.g., tryptophan and branched-chain amino acids) and carbohydrate metabolism (e.g., glycolysis and the pentose phosphate pathway) were markedly upregulated (highlighted in orange bars), indicating a more active microbial role in metabolite production and energy provision (Tufail et al., 2024). Importantly, several microbial-derived metabolites—such as short-chain fatty acids and tryptophan derivatives—are known immunomodulatory agents that can enhance T cell activation or suppress pro-inflammatory signaling, potentially promoting improved therapeutic outcomes. In contrast, non-responders exhibited a significant upregulation of cancer-associated pathways, particularly those involved in PD-L1 expression and PD-1 checkpoint signaling, suggesting that certain microbial taxa may facilitate immune evasion mechanisms. These findings underscore the potential of leveraging microbial functional signatures to identify key taxa whose targeted modulation may reduce tumor immune escape and enhance immunotherapy responsiveness.

Our findings are highly consistent with a recent metatranscriptomic study conducted in a LC immunotherapy cohort, collectively highlighting the pivotal role of GM in modulating responses to immune checkpoint blockade (Dora et al., 2024). Both studies reported significantly higher gut microbial diversity in responders compared to non-responders, suggesting a potential link between microbial ecological complexity and enhanced therapeutic efficacy. Moreover, the consistent enrichment of B. caccae and Akk in responders across studies reinforces the stability and generalizability of B. caccae as a potential predictive biomarker. Future efforts should prioritize integrative multi-omics approaches—including metagenomics, metatranscriptomics, and metabolomics—to elucidate the mechanistic underpinnings of microbe-mediated immune modulation, ultimately advancing personalized prediction and clinical application of immunotherapy.

In addition, we constructed microbial co-occurrence networks to preliminarily investigate potential microbe–microbe interactions associated with immune response, aiming to identify key taxa that may exert antagonistic or synergistic effects in responders. Given the large number of annotated species in the dataset, our analysis was restricted to the top 50 most abundant taxa to enhance robustness. However, it should be noted that correlation-based co-occurrence network methods are inherently limited by the compositionality and sparsity of microbiome data, which may introduce biases in inferring microbial associations. Moreover, since the interactions identified through co-occurrence analysis have not yet been validated by independent analytical approaches, we did not incorporate these results into the final biomarker prediction model to maintain scientific rigor. Future studies may benefit from applying more advanced microbial network inference tools, such as SPIEC-EASI, CoNet, and FlashWeave, to improve the accuracy of interaction inference and enhance biological interpretability.

With the growing body of evidence linking the gut microbiome to cancer immunotherapy outcomes, microbiome profiling is emerging as a promising avenue to advance personalized immunotherapy in lung cancer, with considerable translational potential. First, specific microbial biomarkers—such as B. caccae and Akk—may be assessed through non-invasive fecal testing prior to ICI administration, enabling more accurate patient stratification, enhancing treatment response rates, and reducing unnecessary exposure for likely non-responders (Chen et al., 2019; Han et al., 2023). Second, GM data can inform more rational antibiotic stewardship, including optimization of timing, drug selection, and withdrawal windows, in order to preserve immunologically relevant commensal taxa. Finally, microbiome-informed personalized interventions—such as targeted probiotic supplementation, FMT, or microbial metabolite administration—offer a novel adjunctive strategy to improve ICI efficacy (Lu et al., 2022; Yadegar et al., 2024). However, these approaches must be pursued with rigorous safety evaluation and mechanistic validation. Collectively, these considerations establish a feasible framework for integrating microbiome analysis into the clinical management of lung cancer and advancing precision immuno-oncology.

In future studies, we will focus on the key gut microbial taxa identified in this work to investigate their roles in modulating antitumor immunity. Using in vitro co-culture systems, tumor organoid models, and humanized mouse models, we will systematically explore three mechanistic dimensions: microbial metabolite production, immunomodulatory capacity, and mucosal adhesion or translocation properties. To elucidate functional links between the microbiota and host immunity, we will integrate fecal metabolomic and serum immune transcriptomic data to reconstruct microbe–metabolite–immune signaling networks. This approach will help identify key metabolic mediators driving systemic immune responses and reveal potential causal axes underlying microbiota–host interactions. Additionally, we will expand real-world LC cohorts, particularly those with WGS data, to enable strain-level microbiome profiling and robust external validation across diverse populations stratified by ethnicity, age, and treatment background. Finally, we will integrate microbiome features with clinical parameters, immune phenotypes, and genomic alterations to build high-resolution, multi-dimensional models for predicting immunotherapy response.

We established the first global cross-cohort gut microbiota landscape associated with immunotherapy response and developed a machine learning-based predictive model with potential translational relevance. These findings offer new insights into the design of microbiome-informed immunotherapeutic strategies and the identification of novel adjuvants or therapeutic targets.