Causal relationship between oral/gut microbiota and lung cancer: a two-sample Mendelian randomization study.

Introduction
Globally, lung cancer constitutes a significant public health concern due to its high incidence and death rate, ranking among the primary causes of cancer-associated deaths [1]. Despite continuous improvement of therapy strategy such as radiotherapy, operation chemotherapy, targeted therapy and immune checkpoint inhibitors (ICIs), patients still face challenges of drug resistance and recurrence with metastasis [2].
The advent of the Human Microbiome Project has facilitated a more comprehensive understanding of the human microbiome. The historical perception that the lungs are a sterile environment has been refuted by contemporary research. Utilizing molecular techniques such as Polymerase Chain Reaction (PCR) and Next-Generation Sequencing (NGS), numerous research have now confirmed a close relationship to the oral microbiomes, gut microbiomes and the pulmonary microbiome [3, 4]. The oral microbiota, second to the gut microbiota in diversity, encompasses a broad range of bacterial genera and families [5]. Anatomically, the oral cavity is contiguous with the respiratory tract through the oropharynx, which serves as a passageway to both the upper and lower airways, ultimately reaching the lungs. Growing evidence suggests that oral microbiota can translocate to the lungs, where they may establish colonization and influence the composition and dynamics of the pulmonary microbiome [6, 7]. Moreover, recent research has increasingly emphasized the importance of the pulmonary microbiota within the tumor microenvironment of lung cancer, highlighting its involvement in oncogenesis, tumor progression, and the development of drug resistance [8–10].
Mendelian Randomization (MR) is an epidemiological methodology that leverages naturally occurring and randomly assigned genetic variants as instrumental variables. This approach effectively mitigates confounding variables and reverse causality [11]. MR analyses have been widely utilized to investigate potential causal correlations between gut microbiota and the development of various types of cancer [12, 13]. The two-sample MR analyses increase statistical power and improves the ability to detect causal relationships by combining data from independent studies [14]. In this study by Feng et al., Aggregatibacter and Gemella were positively correlated with a higher risk of lung cancer, whereas Fusobacterium, Streptococcus, Campylobacter A, and members of the Saccharimonadaceae TM7x family were negatively associated with lung cancer risk [15]. While numerous emerging studies suggest the potential role of oral microbiota in cancer [16], there remains a paucity of study elucidating the specific causal relationships between the more comprehensive oral microbiota and lung cancer.
This research endeavored to explore the potential causal relationships between lung cancer and both oral and gut microbiota, with a particular focus on identifying pathogenic microbial clusters. To achieve this, we performed a MR analysis based on the data derived from Genome-Wide Association Studies (GWAS). By treating lung cancer as the outcome and oral and gut microbiota as exposures, we sought to determine whether specific microbial taxa play contributory or protective roles in lung cancer development.

Methods and materials

Study designs
In this work, we conducted a two-sample MR analysis to research the potential causal connection between oral and gut microbiota which derived from the Shenzhen cohort and lung cancer risk in the Japanese cohort. Data were extracted from multiple publicly available repositories. To guarantee the stability and credibility of the outcomes, an array of sensitivity analyses were performed. An overview of the research design is illustrated in Fig. 1.

Source of oral/ gut microbiome data
In our investigation of the oral microbiota, we drew upon data from an extensive GWAS, covering 2017 samples from the dorsal tongue and 1915 saliva samples obtained from 2984 Chinese healthy subjects [17]. We had access to extensive whole-genome sequencing data, which revealed 455 independent associations. These associations involved 340 distinct genetic loci and 385 microbial taxa, all reaching genome-wide significance [17]. The microbial community exhibited high coverage, with 99.7% in tongue dorsum samples and 98.7% in saliva samples, indicating comprehensive representation of the oral microbiota in the dataset [17].
For the gut microbiome component, all Chinese adult participants included in this study were enrolled as part of a multi-omics investigation. The discovery cohort comprised 2002 individuals recruited during routine physical examinations between March and May 2017 in Shenzhen. Among them, blood samples were captured from all volunteers, and fecal samples were available for 1539 individuals. All participants underwent high-depth whole-genome and whole-metagenomic sequencing. For replication, 1430 individuals were recruited across multiple cities in China (e.g., Wuhan, Qingdao), following the same study design but on a smaller scale. Of these, 1006 provided both blood and fecal samples. Sample collection protocols for both blood and stool, as well as sequencing procedures, were consistent with those established in our previous studies [17–22].

Sources of lung cancer data
GWAS summary statistics for lung cancer were obtained from a large genome-wide association study conducted in an East Asian population (Japanese). The dataset featured a comprehensive genetic association map highlighting the pleiotropic landscape, including key loci within the major histocompatibility complex (MHC) region and fine-mapped human leukocyte antigen (HLA) targets. This non-European cohort comprised a total of 212,453 individuals (4,050 cases and 208,403 controls) and included 8,885,805 single-nucleotide polymorphisms (SNPs). In this data source, lung cancer subtypes were not specifically distinguished. Quality control procedures were applied to address imbalances, and adjustments were made for potential confounding factors. The annotated version was built with HG19/GRCh37.

Instrument selection
Six hundred and forty-two SNPs and 467 SNPs of oral and gut microbiome respectively achieving genome-wide significance (p < 5 × 10^-8) for traits related to oral and gut microbiome groups. We applied a more stringent significance threshold (p < 5 × 10^-9) to select robust genetic IVs, achieving 81 SNPs and 118 SNPs of oral and gut microbiome respectively. These IVs were identified by grouping them according to the reference panel of the Linkage Disequilibrium (LD) from the 1000 Genomes Project, with a threshold of R^2 < 0.001 within a distance of 1,000 kilobases (kb) [23]. To ensure the robustness and reliability of IVs, we retained only those with F-statistics over than 10 which corresponds to a moderate level of weak instrument bias that is generally acceptable [24, 25], thereby identifying them as strong tools for following MR analysis. These selected IVs were then drawn from the GWAS summary statistics for lung cancer. To minimize potential bias from horizontal pleiotropy, SNPs exhibiting direct associations with lung cancer (p < 10^-5) were excluded. These are in accordance with established protocols in previous studies [26]. To maintain consistency in our analysis, the SNPs between the exposure and outcome datasets were synchronized to confirm uniform effect estimates for the same effect allele [23].

Statistical analysis
In this research, a range of individual genetic variants were utilized as IVs, instead of depending solely on aggregated allele scores. This strategy was adopted to enable a more comprehensive assessment of the core assumptions underlying MR as well as to detect and account for horizontal pleiotropy thus enhancing the robustness of sensitivity analyses [27].
To evaluate the robustness of our findings under varying assumptions regarding heterogeneity and pleiotropy, we applied complementary MR methods: inverse variance weighted (IVW; random-effects model), weighted median, MR-Egger regression, and MR-Pleiotropy RESidual Sum and Outlier (MR-PRESSO) analysis. The IVW method, implemented under a random-effects model, was used as the main analytical approach across all four categories of instrumental variables. Heterogeneity among the IVs was assessed using Cochran’s Q statistic to evaluate variability.
To enhance the robustness of our findings, we also conducted analyses under more stringent conditions. While the IVW method presumes that all genetic variants are sound instruments, its estimates may be biased in the existence of substantial horizontal pleiotropy affecting a considerable number of SNPs [28]. Conversely, the weighted median method presumes that at least 50% of the overall weight comes from IVs, making it robust even when up to half of the variants are affected by horizontal pleiotropy [29]. When more than 50% of the variants were potentially affected by horizontal pleiotropy, we evaluated the robustness of the genetic instruments using F-statistics, with a mean F-statistic below 10 considered suggestive of weak instrument bias [30].
In addition, the MR-Egger method was employed to detect potential directional pleiotropy. A statistically significant intercept from the MR-Egger analysis would reveal a violation of the instrumental variable assumptions, suggesting the existence of unbalanced directional pleiotropy [31]. Besides, the MR-PRESSO method was implemented to reduce heterogeneity in causal effect assessments by identifying and excluding outlier SNPs that exerted disproportionate influence on the results (NbDistribution = 1500) [32]. What’s more, steiger filtering was implemented to recognize and exclude genetic variants that showed stronger correlations with the outcome than with the exposure [33].
All statistical analyses were conducted using R version 4.3.1 (R Foundation) and R packages (“TwoSampleMR” and “MR”) [34, 35]. The TwoSampleMR package provided causal estimates from the four MR models (IVW, weighted median, MR-Egger, and MR-PRESSO).

Result
Following a series of quality control procedures, 85,043 SNPs were retained for analysis including 84,578 associated with the oral microbiome and 465 with the gut microbiome. All selected IVs demonstrated F-statistics exceeding 10 thus indicating no evidence of weak instrument bias. Eventually, we selected 42,353 SNPs from the oral microbiome and 121 SNPs from the gut microbiome for further analysis. In the oral microbiome, we chose 968 SNPs representing genera such as Aggregatibacter, Pseudopropionibacterium, and Capnocytophaga. In the gut microbiome, we selected 6 SNPs representing species Enterococcus faecalis and three metabolic pathways.

Causal effect of oral microbiota on lung cancer
Utilizing the IVW method in MR analysis, we identified that increased abundance of several bacterial taxa, as genetically predicted, was correlated with a reduced risk of lung cancer. Specifically, these included Pauljensenia (OR: 0.777, 95% CI 0.651–0.927, p = 0.0052), Capnocytophaga (OR: 0.748, 95% CI: 0.580–0.966, p = 0.0260), Centipeda (represented by specie unclassified mgs 2230: OR: 0.809, 95% CI 0.669–0.977, p = 0.0279), and Aggregatibacter which includes A. segnis 'mgs_2462' (OR: 0.818, 95% CI 0.673–0.994, p = 0.0432), 'sp000466335_mgs_1474' (OR: 0.770, 95% CI 0.651–0.910, p = 0.0021) and 'sp000466335_mgs_2558' (OR: 0.746, 95% CI 0.602–0.926, p = 0.0078) and so on. These results are shown in Fig. 2.
In contrast, increased abundance of certain bacterial taxa was connected to an elevated risk of lung cancer. Specifically, these included Prevotella which includes P. baroniae 'mgs_143' (OR: 1.294, 95% CI 1.048–1.598, p = 0.0012) and P. buccae 'mgs_3394' (OR: 1.479, 95% CI: 1.071–2.044, p = 0.0037), Granulicatella which includes G. elegans 'mgs_1090' (OR: 1.210, 95% CI: 1.019–1.436, p = 0.0064) and the unclassified species 'mgs_2338' (OR: 1.192, 95% CI: 1.010–1.406, p = 0.0160), Streptococcus which includes S. oralis C 'mgs_62' (OR: 1.298, 95% CI: 1.004–1.679, p = 0.0292), the unclassified species 'mgs_1416' (OR: 1.279, 95% CI: 1.072–1.526, p = 0.0446), and the unclassified species 'mgs_2628' (OR: 1.179, 95% CI: 1.031–1.348, p = 0.0370), Veillonella (OR: 1.304, 95% CI: 1.016–1.673, p = 0.0140), and Fusobacterium (OR: 1.296, 95% CI: 1.023–1.641, p = 0.0043) and so on. These findings are shown in Fig. 3.
Notably, bacteria within the genera Pauljensenia, Streptococcus, Centipeda, Saccharimonadaceae TM7x, Haemophilus D and other four genera exhibited species-level heterogeneity on lung cancer risk. Specifically, different species within the same genus were associated with increased or decreased risks of lung cancer. To clearly illustrate this phenomenon, we summarized these species-level heterogeneity bacteria in Fig. 4.

Causal effects of gut microbiota on lung cancer
Compared with the oral microbiota, certain features of the gut microbiota were correlated with lung cancer risk. Specifically, the metabolic pathways xylose degradation (OR: 0.490, 95% CI 0.268–0.893, p = 0.0199) and threonine degradation II (OR: 0.419, 95% CI 0.233–0.755, p = 0.0038) were associated with a reduced risk of lung cancer. Conversely, Enterococcus faecium (OR: 1.180, 95% CI 1.030–1.353, p = 0.0173) and the acetyl-CoA synthetase pathway (OR: 1.884, 95% CI 1.121–3.168, p = 0.0169) were positively associated with a higher risk of lung cancer. These findings are shown in Fig. 5.

Sensitivity analysis
A set of sensitivity analyses were performed to evaluate the presence of heterogeneity and horizontal pleiotropy among the selected IVs. Horizontal pleiotropy was specifically analyzed using the MR-Egger intercept test. All p values derived from the MR-Egger intercepts were greater than 0.05, suggesting no significant proof of directional horizontal pleiotropy.
Cochran’s Q test was applied to evaluate heterogeneity among the selected SNPs. As presented in Table 1, the Q_pval values for both the IVW and MR-Egger methods exceeded 0.05, indicating no significant heterogeneity and suggesting that the results were unlikely to be affected. Owing to the relatively limited number of SNPs available for the gut microbiome analysis, sensitivity analyses could not be reliably performed for that dataset. Detailed scatter plots for each MR method are presented in Figure S1. These results demonstrated consistent directions and reinforced the reliability of our findings.
Leave-one-out analysis revealed that sequential removal of individual SNPs did not result in substantial changes to the overall causal estimates. This suggests that no single SNP exerted a disproportionate influence and no influential outliers were identified. These results are illustrated in Figure S2.

Discussion
To our understanding, this is the first study which utilized the two-sample MR approach to systematically evaluate the potential causal relationships between oral/gut microbiota and lung cancer. Our findings identified several novel microbial taxa potentially increased or decreased risk of lung cancer. What’s more, we summarized several specific bacterial genera in which different species have different species-level effects on lung cancer development. Collectively, these findings deepen our understanding of the lung cancer microbiota axis as well as providing potential evidence for future clinical translation and microbiome-targeted research.
MR analysis reduces confounding factors such as smoking status, living habit and other social behaviors by utilizing IVs, which are common limitations in observational research. Additional sensitivity further validated the consistency of our results to strengthen the credibility of our conclusions. This integrative MR framework allowed us to gain clearer insights into the potential roles of microbiota from distinct anatomical sites in the etiology of lung cancer.
Our research identified that certain specific bacteria within genera Pauljensenia, Centipeda and Aggregatibacter are significantly associated with reduced lung cancer risk. In contrast, higher abundances of bacteria within the genera Prevotella, Granulicatella, Streptococcus, Veillonella, Fusobacterium, Saccharimonadaceae TM7x, Neisseria, and Haemophilus D were associated with increased lung cancer risk. Those discoveries have important implications for diagnostic biomarkers, as it suggests that these bacteria could potentially serve as indicators for lung cancer risk assessment.
Our study findings are in accordance with some of the previous research. Former studies suggested that approximately 25% of all cancers are etiologically associated with chronic inflammation and infection [36]. The risk for cancer of the respiratory system is positively associated with inflammatory diseases [36]. Among the bacteria significantly linked to reduced lung cancer risk, the study by He et al. found that Pauljensenia and Capnocytophaga are associated with decreased incidence of bronchitis and tonsillitis, as well as the inhibition of pneumonia and bronchitis [37]. The abundance of Aggregatibacter has been shown to be negatively associated with inflammatory markers in sputum like interleukin-8 (IL-8) and interleukin-1β (IL-1β), and demonstrates anti-inflammatory properties in lower respiratory tract samples from individuals with chronic airway disease (CAD) [38].
Conversely, among the bacteria of increased lung cancer risk, Huang et al. found that bacteria of the genus Granulicatella are significantly enriched in the lung microbiota of patients with lung cancer [39]. Wang et al. discovered that increased abundance of Granulicatella is linked to the transition from neutrophilic to eosinophilic chronic obstructive pulmonary disease (COPD) [40], which is often considered as a common risk factor for lung cancer [41]. The genus Streptococcus has been shown to be enriched in NSCLC patients [42]. Mao et al. reported a significant rise in the abundance of Saccharimonadaceae TM7x phylum bacteria in patients with lung cancer particularly in those who were diagnosed with lung squamous cell carcinoma and lung adenocarcinoma [43]. The Saccharibacteria bacteria and its subgroup c_TM7-3 were significantly increased in bronchoalveolar lavage fluid (BALF) samples from persons with lung cancer and potentially serving as reliable biomarkers [44]. Hosts with higher abundance of Neisseria are more susceptible to environmental damage, resulting in an increased risk of respiratory cancer [45]. In the study by Feng et al. [15], Fusobacterium showed a protective effect against lung cancer, conflicting with our results. We think this stems from differing bacterial strains: the strain Fusobacterium animalis 'mgs_312' in their study versus the unclassified Fusobacterium sp. 'mgs_2195' in ours. This strain-specific effect is consistent with our findings.
More importantly, we identified certain genera, such as Streptococcus, Saccharimonadaceae TM7x, and Haemophilus, where different species within the same genus have varying effects on lung cancer. Within the genus Streptococcus, some species of Streptococcus were found to mediated the anticancer effects through β-galactosidase which activates oxidative phosphorylation and downregulate the Hippo pathway kinases [46]. However, no prior studies have confirmed its carcinogenic effects on lung cancer, which warrants further investigation in future research. Certain Streptococcus species have been shown to accelerate the cell cycle while preventing apoptosis in lung cancer cells, thereby facilitating the occurrence of lung cancer. Studies have shown that increased abundance of Haemophilus is associated with exacerbated inflammatory responses [47] R. However, the presence of Haemophilus is also correlated with increased abundance of the antimicrobial peptide secretory leukocyte protease inhibitor (SLPI), which is one of the key factors in maintaining respiratory homeostasis [48]. This varying regulatory capacity suggests that different bacterial strains within the same genus may have diverse biological functions and collectively influence disease development. Our study results advocate microbiota research should be detailed to the species level and highlight the importance of microbial community diversity in maintaining normal oral and respiratory functions, as well as the necessity of dynamically monitoring the composition of the upper respiratory tract microbiota for its significance in the development of lung cancer.
Compared with the oral microbiota, we identified fewer gut microbial taxa associated with altered lung cancer risk. From a physiological standpoint, the oral and respiratory microbiota directly constitute the pulmonary microbiome, thereby more directly influencing the physiological and immune functions of the lungs. Nilsson et al. demonstrated that carbohydrates conjugated with xylose-naphthalene selectively suppress tumor cell growth [49]. Various studies have shown that xylitol exhibits selective anticancer activity, potentially through mechanisms involving the inhibition of glucose utilization [50] and dose-dependently inhibits cell proliferation in A549 cells [51]. The activity of ATP-citrate lyase (ACLY), the key enzyme catalyzing citrate conversion to acetyl-CoA, correlates with cancer cell proliferation, survival, and poor prognosis [52]. ACLY inhibition significantly suppresses lung cancer cell growth and proliferation [52]. However, the gut microbiota analysis in our study is limited by a relatively small number of SNPs (121) and we were unable to perform advanced sensitivity analyses such as MR-Egger and MR-PRESSO, which may reduce statistical power, acknowledging the potential impact on result reliability.
However, several limitations should be acknowledged. Firstly, the study population was predominantly composed of Asian descent, which may restrict the generalizability of the results to other ethnic groups. Although the study population is from East Asia, there may still be some heterogeneity across regional subgroups, which could potentially introduce confounding effects in our research. Secondly, the analysis did not distinguish between specific subtypes of lung cancer. Given the heterogeneous nature of lung cancer, future studies are warranted to explore subtype-specific associations. Additionally, personal behavior such as smoking status is not documented in GWAS dataset so that we cannot completely exclude the potential impact of smoking status on our findings. Lastly, this study did not include experimental validation to confirm the biological mechanisms underlying the observed associations. Therefore, further research including studies involving diverse populations, subtypes, and experimental models, exploring the potential microbiome biomarker and accessible therapeutic target, is essential to elucidate the effect of microbiota in lung cancer pathogenesis.

Conclusion
In summary, this study is the first to systematically investigate the potential causal relationships between microbiota from distinct anatomical sites and lung cancer using a two-sample MR analysis. The findings suggested that specific oral and gut microbial taxa may contribute to lung cancer development, offering valuable insights for future clinical diagnostics and basic research. Further studies are advocated to validate these results in more diverse populations and to elucidate the underlying biological mechanisms by which microbiota from different body sites may influence lung carcinogenesis and strategies for prevention and early intervention.

Supplementary Information