Seventy (70) fecal and serum samples were prospectively collected from the General Hospital of the Chinese People’s Liberation Army. Participants were excluded if they: 1) had received antibiotic or probiotic treatment within 1 month before the study; 2) had experienced a lung infection within 1 month; 3) had other types of lung cancer or other cancers; or 4) had received long-term immunosuppressive drug treatment. The study population was divided into two groups: BPN (n = 20) and LUAD (n = 50). Detailed information on the participants, including age, body mass index, and smoking history, was recorded. The clinical characteristics are shown in Supplementary Table S1. All procedures involving human participants were performed in accordance with the ethical standards of the institutional and/or national research committee and the Declaration of Helsinki. The study protocol was approved by the Ethics Committee of the General Hospital of Chinese People’s Liberation Army (S2022-407–01). All participants provided informed consent.

DNA was extracted using the HiPure Fecal DNA Kit (Magen, Guangzhou, China) according to the manufacturer’s instructions. DNA concentration and purity were determined using a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA). Fungal internal transcribed spacer (ITS) 2 was amplified using the sequence-specific primers ITS3_KYO2 (5’-GATGAAGAACGYAGYRAA-3’) and ITS4 (5’-TCCTCCGCTTATTGATATGC-3’). PCR amplification of genomic DNA was conducted in a 20 μL reaction volume with the following cycling parameters: initial denaturation at 95°C for 5 min; 33 cycles of denaturation at 95°C for 1 min, annealing at 60°C for 1 min, and extension at 56°C for 1 min; and a final extension at 72°C for 7 min. Amplicons were separated on 2% agarose gels and purified using an AxyPrep DNA Gel Extraction Kit (Axygen Biosciences, Tewksbury, MA, USA). After purification, DNA concentration was measured using a StepOnePlus Real-Time PCR System (Thermo Fisher Scientific) before pooling into libraries. A second round of PCR was then performed in a 50 μL reaction mixture under the following conditions: 95°C for 5 min; 12 cycles of 95°C for 1 min, 60°C for 1 min, and 72°C for 1 min; followed by a final extension at 72°C for 7 min. Subsequently, the polymerase chain reaction products were used for library construction on an Illumina NovaSeq 6000 platform (Illumina, San Diego, CA, USA) and sequenced using a MiSeq PE250 sequencer.

The ACE, Shannon, and Simpson indices were calculated using QIIME v.1.9.1. Wilcoxon rank-sum tests were used to compare alpha diversity indices between groups using the R vegan package v.2.5.3. Principal coordinates analysis (PCoA) based on the Bray–Curtis distance and Adonis tests were performed using the R vegan package v.2.5.3. Venn diagrams were generated to identify unique and shared families, genera, and species between groups. Linear discriminant analysis effect size (LEfSe) v.1.0 was used to identify biomarkers for each group. Circos plots were generated using the OmicStudio online tool (https://www.omicstudio.cn/tool).

Metabolites were quantified using ultra-performance liquid chromatography–tandem mass spectrometry (ACQUITY UPLC-Xevo TQ-S; Waters Corp., Milford, MA, USA) according to the manufacturer’s instructions. A rigorous quality control system was used to ensure the reliability of the results. Raw data were processed using TMBQ v.1.0 (Metabo-Profile, Shanghai, China), including peak integration and quantification. Statistical analyses, including principal component analysis, orthogonal partial least squares discriminant analysis (OPLS-DA), and univariate and pathway analyses, were performed using a self-developed iMAP platform v.1.0 (Metabo-Profile).

The levels of 30 cytokines were assessed by multiplex analysis using a Luminex 200 analyzer with a Procarta Plex Immunoassay kit (Thermo Fisher Scientific), according to the manufacturer’s instructions.

In the demographic analysis conducted using the R software (version 4.3.1), normally distributed continuous variables were expressed as the mean ± standard deviation (SD), and categorical variables were expressed as numbers (percentages). Intergroup comparisons of cytokines were also performed using R. Normally distributed continuous variables were compared using t-tests, whereas non-normally distributed variables were analyzed using the Wilcoxon test. A two-sided chi-square test or Fisher’s exact test was used to compare categorical variables between the two groups. In this study, 11 fungal species and 10 differential metabolites were selected as feature variables, and group status was used as the dependent variable. Using SPSS v.26.0, a diagnostic model was established to differentiate LUAD from BPN. The discriminatory power of fungi and metabolites in diagnosing BPN and LUAD was evaluated using receiver operating characteristic (ROC) curves and the area under the curve (AUC).

The Youden index was calculated as follows:

Fu et al., 2021.

Spearman correlations among key gut fungi, differential metabolites, and cytokines were calculated, and networks were visualized using Cytoscape v.3.10.1. Statistical significance was defined as a two-tailed P-value< 0.05.

The gut fungal data were generated by ITS2 sequencing. Alpha diversity indices, including the ACE, Shannon, and Simpson indices, were used to assess the sample diversity and richness. No significant differences were observed between the LUAD and BPN groups (Supplementary Figure S1). However, significant differences were observed in the gut fungal communities between the two groups based on PCoA at the family, genus, and species levels (P = 0.021, 0.005, and< 0.001, respectively; Figures 1A–C). Venn diagrams showed that the LUAD group included more fungi at the family, genus, and species levels than the BPN group (Figures 1D–F); the two groups shared 74 families, 92 genera, and 90 species.

The main fungal compositions of the BPN and LUAD groups are shown in Circos diagrams (Figures 1G–I). At the family level, the BPN and LUAD samples mainly comprised Saccharomycetales fam. Incertae sedis, Saccharomycetaceae, and Trichosporonaceae. At the genus level, the two groups mainly comprised Candida, Saccharomyces, Apiotrichum, Vanrija, and Cutaneotrichosporon. At the species level, the two groups mainly comprised Candida albicans, Apiotrichum montevideense, Vanrija humicola, Candida glabrata, Brassica carinata, and Cutaneotrichosporon guehoae. LEfSe revealed 11 fungal genera and 11 species that differed between the BPN and LUAD groups (Figure 1J). Four species, Hanseniaspora uvarum, Cucumis sativus, Mucor circinelloides, and Talaromyces stollii, were enriched in the LUAD group compared with the BPN group.

Targeted metabolomic analysis of 201 microbial-related and host-specific metabolites detected 127 metabolites (Supplementary Table S2). The metabolites were classified into 13 categories: short-chain fatty acids, phenylpropanoids, phenols, organic acids, indoles, imidazoles, fatty acids, carnitines, carbohydrates, bile acids, benzoic acids, benzenoids, and amino acids (Figure 2A). In the LUAD group, amino acids, benzoic acids, and imidazoles accounted for a relatively large proportion (P = 0.046, 0.043, and 0.007, respectively), whereas carbohydrates accounted for a relatively small proportion (P = 0.002).

Univariate analysis identified 10 differential metabolites between the two groups (P < 0.05; Figure 2B; Supplementary Table S3). OPLS-DA, a supervised multivariate method, revealed significant differences in metabolite profiles between the paired groups (Figures 2C, D). Forty-two metabolites were selected based on the criteria of variable importance in projection (VIP) > 1 and |P(correlation)|< 0.5 (Supplementary Table S4). Furthermore, 10 shared metabolites were identified from the results of univariate and multivariate analyses, which were mainly involved in valine, leucine, and isoleucine biosynthesis (Figure 2E; Supplementary Figure S2B). The VIP scores of the 10 metabolites are shown in Supplementary Figure S2A; the top three were ornithine, indole-3-propionic acid (IPA), and valine. In the LUAD group, the levels of IPA and docosapentaenoic acid n-6 (DPAn-6) were relatively high, whereas the levels of eight metabolites, such as ornithine, isoleucine, and valine, were relatively low (Figure 2F).

We selected 11 fungal species from the LEfSe analysis, including H. uvarum, C. sativus, M. circinelloides, T. stollii, Hypoxylon begae, Auxenochlorella protothecoides, Setophoma terrestris, Amaurodon mustialaensis, Mrakia frigida, Malassezia restricta, and Vanrija humicola (Figure 1J), and 10 differential metabolites (Figure 2F) as feature variables, and established an ROC model through binary logistic regression analysis to distinguish LUAD from BPN. Figure 3 presents the ROC curve results. The AUC of the fungus model was 0.867 [95% confidence interval (CI): 0.771–0.963; P < 0.001]. The Youden index (Equation 1) was used to determine the highest sensitivity and specificity. The corresponding cutoff value for the diagnosis of the fungal model was 0.67, and the sensitivity and specificity were 92% and 75%, respectively. The AUC of the metabolite model was 0.894 [95% CI: 0.811–0.977; P < 0.001]. The corresponding cutoff value for the diagnosis of differential metabolites was 0.566, and the sensitivity and specificity were 90% and 75%, respectively. The AUC of the integrated model with fungal species and metabolites was 0.944 [95% CI: 0.872–1; P < 0.001]. The corresponding cutoff value for the integrated diagnosis was 0.67, and the sensitivity and specificity were 92% and 75%, respectively.

Abnormal expression of cytokines occurs in LUAD (Xu et al., 2022). In the present study, the levels of interferon-γ (IFN-γ)-induced protein 10 (IP-10) and interleukin (IL)-31 were higher (P = 0.0041 and 0.0488, respectively), whereas the levels of IL-15, IL-27, and IL-17 were lower (P = 0.0248, 0.0462, and 0.0226, respectively) in the LUAD group than in the BPN group (Figure 4A, Supplementary Table S5). The interactions among microbiota, metabolites, and cytokines play significant roles in cancer. Therefore, a correlation network comprising 11 fungal species, 10 metabolites, and 5 cytokines was established (Figure 4B). The abundance of H. begae was positively correlated with lactic acid, leucine, and ornithine levels (P = 0.015, 0.021, and 0.018, respectively). The abundance of T. stollii was positively correlated with IPA levels (P = 0.011) and negatively correlated with alpha-linolenic acid levels (P = 0.01). Gluconolactone levels were positively correlated with the abundance of V. humicola (P = 0.034) and negatively correlated with the abundance of H. uvarum (P = 0.015). High IP-10 levels were associated with low abundances of H. begae and A. mustialaensis (P = 0.015 and 0.037, respectively).