Human Microbiome as Clinical Biomarkers for Respiratory Tract Cancers

1. Introduction

Respiratory tract cancers (RTCs), the prevalent types of cancers,[1] include lung cancer (LC), laryngeal cancer (LAC), nasopharyngeal cancer (NPC), etc. Besides these major types, LCs can be classified as small-cell lung cancer (SCLC) and non-small-cell lung cancer (NSCLC). NSCLC can also be subdivided into squamous cell carcinoma (SCC), adenocarcinoma (ADC), large cell carcinoma (LCC), etc.[2] While these cancers show difficulties in early diagnosis and treatment monitoring,[3] with the help of omics, finding associations of cancers with microbiomes is now more feasible than ever. Thus, we can find non-invasive diagnostic/prognostic biomarkers that might have therapeutic application as well.[4] This mini review aims to explore research on human microbiomes as potential diagnostic or treatment biomarkers and their associations with RTCs.

2. Microbiomes and Their Relevance as Biomarkers

The term microbiome/microbiota was coined by Joshua Lederberg in 2001.[5] However, the related research accelerated after the initiation of the Human Microbiome Project in 2008. The contemporary bioinformatics advancements played a crucial role, too. Several microbiomes (gut, tissue, oral, lung, etc.) related to different body parts have been sequenced so far, and the association of gut and lung microbiomes with RTCs is now well known.[6,7]

Existing biomarkers for RTCs include circulating tumor DNA, oncogenic mutations, protein markers, and imaging-based markers, etc.[8,9] While they improved diagnosis and treatment surveillance, but had limited sensitivity in early-stage illness, intrusive sampling procedures, or a lack of specificity, etc. In contrast, the microbiome has emerged as a non-invasive novel biomarker with proven associations to cancer progression, metabolic pathways, immune responses, and treatment monitoring.

2.1. Association of Cancer Pathways and Microbiomes in RTCs

The major pathways include EGFR, ALK, KRAS, STAT3, MAPK, WNT, etc.[10] Most of the cancer diagnoses and treatments are based on these pathways. And they are significantly associated with and affected by the microbial dysbiosis. (Figure 1)

2.1.1. The EGFR (Epidermal Growth Factor Receptor) Pathway

The EGFR gene codes for a tyrosine kinase receptor activated by EGF binding. The NSCLC cases (40-60%) have EGFR mutations, primarily exon-19 deletions (Ex19del) or an exon-21-point mutation (L858R). Exons 18 (e.g., G719X) and 20 (e.g., T790M) are affected by less frequent mutations, resulting in elevated kinase activity and pathway hyperactivation.[11]

The papillomavirus, Bacteroidetes, along with genera like Parvimonas, Prevotella, Mycoplasma, etc., affects the EGFR pathway. The EGFR-TKI treatment can induce skin toxicity, showing increased abundances of species like E. coli, S. haemolyticus, S. aureus, S. enterica, etc., and lower abundances of Corynebacterium sp., Prevotella copri, and S. epidermidis.[12] Other abundant microbes in NSCLC are Rhizopus oryzae, Natronolimnobius innermongolicus, Staphylococcus sciuri, etc.[13] (Table 1)

2.1.2. The KRAS (Kirsten Rat Sarcoma Viral Oncogene Homolog) Pathway

A GTPase protein encoded by the KRAS oncogene controls proliferation, survival, and differentiation. KRAS mutations inhibit GTP hydrolysis and over-activate pathways frequently in lung adenocarcinoma (LUAD) (20–40%). G12D mutation is linked with PI3K-AKT activation and co-mutations in STK11, KEAP1, TP53, and CDKN2A/B.[14]

The association of Streptococcus, Prevotella, and Veillonella with upregulation of ERK and PI3K pathways in NSCLC has been reported.[13,15] (Table 1)

2.1.3. The MAPK (Mitogen-Activated Protein Kinase) and ERK Pathway

A serine/threonine kinase related to MAPK, regulating cell survival, differentiation, and proliferation, is encoded by the BRAF gene. Melanoma, colorectal (CRC), thyroid, ovarian, and NSCLC are associated with BRAF mutations, particularly the V600E. Protein interactions, disrupted by the mutation, lead to pathway overactivation and worse outcomes for NSCLC patients, such as lower progression-free survival (PFS) and overall survival (OS) rates. Additional inactivating mutations (D594X, T599I, and G466X) and activating mutations (G469X, K601E, and L597X) have also been found. It is yet unknown how BRAF mutations affect the way targeted treatments respond.[16]

Lower airway microbiota, especially Veillonella parvula, are associated with the upregulation of IL-17, PI3K, MAPK, and ERK pathways and influence chemokine signaling pathway, TNF, JAK-STAT, and PI3K-AKT signaling pathways. Several viruses [human T-cell leukemia virus 1 (HTLV-1), human papillomavirus (HPV), Epstein-Barr virus (EBV), and SARS-CoV-2] trigger the MAPK, non-canonical WNT, and IFN pathways. TNF ⍺ is associated with Bifidobacterium, while Wolbachia, Prevotella, and Streptococcus upregulated ERK and PI3K pathways in LC.[17,18,19] (Table 1)

2.1.4. The VEGF (Vascular Endothelial Growth Factor) Pathway

With the help of VEGFR-1 and VEGFR-2 receptors, VEGF drives endothelial cell proliferation, blood supply, angiogenesis, and vasculogenesis. Although RTCs are minimally dependent on angiogenesis, NSCLC, SCLC, and other adenocarcinomas all show aberrant VEGF expression. With prostaglandin E2 (PGE2), raised cyclooxygenase-2 (COX-2) levels in NSCLC promote EGFR signaling, which stimulates VEGF expression and angiogenic potential. In NSCLC, overexpression of VEGF is a prognostic marker that is linked negatively to a reduced survival rate. Angiogenesis is directed by several microbes including Helicobacter pylori, Akkermansia muciniphila, Enterococcus hirae, and HPV. [20] (Table 1)

2.1.5. The WNT Pathways

Wnt/β-Catenin (Canonical Wnt pathway) is part of the major WNT pathways. WNT, a protein family, is involved in cell survival, formation, and renewal of organs and stem cells. In the canonical pathway, inhibition of β-Catenin degradation by active WNT signaling results in the regulation of several genes. Non-canonical WNT pathways are β-Catenin independent. The Wnt/Calcium pathway specially Wnt5a is involved in LC progression.[21,22]

In CRC, Bacteroides fragilis and Fusobacterium nucleatum can activate Wnt/β-catenin signaling.[23] The WNT pathways also influence Notch and Sonic Hedgehog pathways.[21,22] The Notch pathway is associated with lung cell development, and cell fate determination. The concrete evidence of associations between these pathways and microbes is still missing. (Table 1)

2.2. Association of Transcription Factors (TFs) and Microbiomes in RTC

TFs are pivotal in tumor development and progression, immunological response, immune cell infiltration, and the cell cycle etc. Recent studies showed TFs can predict patient prognosis, regulate drug resistance, and associate with gene expressions.[24,25,26] A study on NSCLC patients identified 725 associated TFs, including 200 upregulated TFs. Among them, five TFs (SETDB2, SNAI3, SCML4, ZNF540, and ETV1) were identified as potential prognostic markers (Survival-related TFs).[24]

Although the association of the microbiome (E. coli, butyrate-producing bacterium SM4/1, and a species of Oscillatoria) with expression of SNAI3 is reported, other TFs’ associations have not been reported yet.[27]

Another study found seventeen overregulated and sixteen downregulated TFs in LC. According to the transcriptional regulatory networks of LC, the main positive regulators are SOX4, FOXM1, ETV4, HOXC6, and E2F3, and the main negative regulators are SOX17, KLF4, and ZBTB16. These TFs (SOX4, TCF3, ETV4, and FOXM1) are found to be associated with HTLV-1, HPV, EBV, and SARS-CoV-2 in different cancers, including LC.[17] Additionally, Fusobacterium nucleatum is found to be associated with MNDA, HOXC6, SOX4, FOXF1, and TAL1. [28]

Fungal proteins from Hericium erinaceus have antitumor effects through dysbacteriosis, causing inhibition of some bacterial growth, while increasing the growth of probiotic bacteria (Bifidobacterium, Gemellales, Blautia, Sutterella, Anaerostipes, Roseburia, Lachnobacterium, Lactobacillus, and Desulfovibrio). The fungus influences the expression of FOXM1, too.[29]

Other than these TFs, nuclear factor kappa-B (NF-KB) and nuclear factor (erythroid-derived 2)-like 2 (Nrf2) are also associated with LCs, and Faecalibacterium prausnitzii was found to inhibit NF-KB activation and IL-8 secretion.[18] These associations between TFs and microbiomes show promise for combined biomarker detection. (Table 1)

2.3. Association of Metabolites and Microbiomes in RTC

Microbial metabolites, such as Serum Metabolites (glycerophospholipids, glycerolipids, lysophospholipids, etc.), Short Chain Fatty Acids (SCFAs), Polyamines, Reactive Oxygen Metabolites (ROS), Aldehydes, Terpenes, Ketones, Secondary Bile Acids (SBAs), Volatile Organic Compounds (VOCs), Sphingosine, etc., are involved in inflammation, immune responses, cell growth, apoptosis, tumor progression, and thus regulate the RTCs.[30,31]

Zhao et al. (2021), revealed that Acylcarnitines, Lysophospholipids, Beta-Santalyl Acetate, Xanthines, and Theobromine have associations with 21 genera, including Synergistes, Megasphaera, Clostridioides, Prevotellaceae, Halocella, etc. Zhao et al. also revealed an association of glycerophospholipids with microbes such as Erysipelotrichaceae_UCG_003, Clostridium, and Synergistes in LC[32].

Another study on NSCLC patients showed that SCFAs (pentanoic and butyric acids) have a strong association with A. muciniphila, Rikenellaceae, Bacteroides, Peptostreptococcaceae, Mogibacteriaceae, and Clostridiaceae [abundant in healthy controls (HCs)]. On the other hand, patients undergoing anti-PD1 treatment had a different microbial richness, including a higher incidence of Rikenellaceae, Prevotella, Streptococcus, Lactobacillus, Bacteroides plebeius, Oscillospira, and Enterobacteriaceae in stool. The presence of SCFAs (propionic, butyric, acetic, and valeric acids), lysine, and nicotinic acid, and bacteria like Granulicatella are associated with better treatment response. In non-responders, metabolites such as tridecane and 2-pentanone were potential indicators. In addition, Butanal is a common gut aldehyde formed by the bacterial breakdown of leucine and has a strong association with inflammation and cancer.[33,34,35]

Qian at el. used Fecal Microbiota Transplant (FMT) and identified genera like Prevotella, Gemmiger, and Roseburia as significantly upregulated. And nervonic acid/all-trans-retinoic acids are negatively associated with Prevotella. When FMT replenished the Prevotella community, the levels of the nervonic acid/all-trans-retinoic acid were improved. Moreover, eight differentially expressed proteins were associated with IL-8 and NF-KB pathways, and CRP, LBP, and CD14 were reported as NSCLC biomarkers. The analysis also showed 49 significant metabolites changed between NSCLC and HCs. In NSCLC, m-Coumaric acid, 13-l-hydroperoxylinoleic acid, allocholic acid, 13S-hydroxyoctadecadienoic acid, palmitic acid, 5-hydroxyindoleacetic acid, eicosadienoic acid, pyroglutamic acid, 3-(2-hydroxyphenyl)propanoic acid, nervonic acid, arachidic acid, l-glutamic acid, and oxoadipic acid exhibited lower concentrations. In NSCLC tumor samples, hippuric acid, adipic acid, 5-amino pentanoic acid, 16-hydroxy hexadecanoic acid, and citric acid were also reduced significantly. However, the levels of p-aminobenzoic acid, azelaic acid, 4-hydroxycinnamic acid, L-lactic acid, N-acetyl-α-neuraminic acid, gluconic acid, hydroxyphenyl acetic acid, gamma-aminobutyric acid, adenyl succinic acid, citramalic acid, glutaric acid, and propionic acid were increased.[36]

The study on Ginseng polysaccharides (GPs) as therapeutic interventions for LC revealed that GPs can increase the SCFA-producing bacteria Muribaculum in mice models and maintain gut homeostasis. The degradation of polysaccharides by microbes produces the SCFAs, and they facilitate communication between the microbiome and immune cells. Besides SCFAs, metabolites such as L-tryptophan and L-kynurenine are also influenced by GPs. Increased and decreased production of L-tryptophan and L-kynurenine, respectively, and kynurenine/tryptophan ratio were observed, indicating the influence of GPs in tryptophan metabolism with the help of gut microbes. In anti-PD-1 treatment responses, responders showed an abundance of bacteria such as Bacteroides vulgatus, Parabacteroides distasonis, bacterium LF-3, and Sutterella wadsworthensis, and had better survival rates.[37]

Akkermansia muciniphila-influenced symbiotic bacteria (Bacteroides, Dubosiella, Methylovirgula, and Acidobacteriaceae) exhibited significant correlations with discriminative metabolites in the respective pathways; the pathway of Gln (glutamic acid, succinic acid, and malic acid) and adenosine (AMP, ADP, UMP, GMP, and uric acid) metabolism of the metabolic network in Lewis LC tissue.[38]

Another Review mentioned the effect of bacterial metabolites and toxins on inducing genomic instability in the host. Butyrate, Folate, Propionate, Acetaldehyde, Deoxycholic Acid, and Biotin can potentially modify epigenetics. Lactobacillus, Bifidobacteria, Streptococcus pneumoniae, Klebsiella pneumoniae, Prevotella, and Fusobacterium are required for the formation and activities of the metabolites mentioned earlier. In NSCLC patients receiving anti-PD-1 immunotherapy, the responders exhibited higher serum acetic acid, propionic acid, and butyric acid.[39]

Lipid metabolites (Leukotriene B4 and prostaglandin E2) are upregulated in mice with LC. In addition, high levels of adrenic, palmitic, stearic, and oleic acids were also seen. Moreover, isohumulone (peroxisome proliferator-activated receptor gamma activator) and resolvin (an ω-3 polyunsaturated fatty acid) both have anti-cancer effects. The abundance of Lachnospiraceae_UCG-006 and the decrease of L-valine showed a strong association with LC patients, indicating a potential biomarker.[40,41]

Fecal metabolites (quinic acid, 3-hydroxybenzoic acid,1-methyl hydantoin,3,4-dihydroxydrocinnamic acid, and 3,4-dihydroxy benzene acetic acid) showed reduced levels in LC. The study also found correlations with most metabolites, where Ruminococcus gnavus showed a negative correlation, and Lachnospira, Firmicutes, and Fusicatenibacter showed positive correlation[42].

Another interesting study on Time-restricted feeding (TRF) and LC provided valuable insights. TRF increased Lactobacillus and Bacillus, which were correlated with most differential metabolites and six genera (Lactobacillus, unclassified_c__Bacilli, Bifidobacterium, Bacillus, Ileibacterium, and Candidatus_Saccharimonas).[43]

A separate study reported 40 metabolites showing different abundances between HCs and LC patients. Out of them, 14 were upregulated and 26 were downregulated in LC. Finally, they suggested three metabolites (Cysteinyl-Valine, 3-Chlorobenzoic acid, and 3,4-Dihydroxyphenyl ethanol) and nine species (s_Lactobacillus murinus, s__uncultured bacterium g__norank f__Muribaculaceae, s__Pelomonas saccharophila, s__uncultured Olsenella_sp g__Atopobium, s__Corynebacterium glutamicum, s__uncultured rumen_bacterium g__Prevotella, s__Massilia timonae, s__Lactobacillus delbrueckii subsp__bulgaricus, and s__Lactobacillus reuteri) as potential biomarkers. Additionally, Sphingosine inhibited the growth of S. aureus, A. baumannii, H. influenzae, E. coli, F. nucleatum, S. sanguinis, etc.[44].

In a Mendelian Randomization Study, fifteen plasma metabolites were found to be associated with LC, including eight and seven metabolites with a positive and protective effect, respectively.[45] In addition, Docosapentaenoic acid (DPA) was found to be associated with major LC subtypes. Finally, the study found associations of 13 taxa and 15 metabolites in LC risk, where eight taxa and fourteen plasma metabolites were associated with LUAD, four taxa and ten metabolites with SCC, and seven taxa and sixteen metabolites with SCLC. The metabolites include DPA, Serine, Pyridoxate, etc., and taxa include Bacteriodales, Ruminococcaceae, Collinsella, Lentisphaerae, etc. [45]

Another study on a postbiotic (MS-20) (comprised of microbial metabolites) showed inhibitory actions against CRC and LC growth in combination with PD-1 antibody in mice. After the treatment with MS-20, the abundances of Ruminococcaceae and Bacteroidaceae increased, and Clostridium colinum, Tyzzerella spp, and Clostridium leptum were decreased[46].

Early-stage LC patients showed a high abundance of Bacillus and a lower abundance of Bifidobacterium and Faecalibacterium. [18].

Finally, VOCs are also found to be associated with microbiome dysbiosis and cancer detection. VOCs such as methyl butane and Acetone were significantly decreased, while Cyclohexane and Xylene were increased in patients. Another study showed that Cyclohexane levels and infections by E. coli, P. aeruginosa, and S. aureus are associated.[47] Microbial-related VOCs like Isopropanol, Butane, Butanol, Methanol, Methyl acetate, Butandione, Cyclohexanone, Toluene, Benzaldehyde, Propanal, and O-Xylene are also associated with LC.[35] (Table 1)

3. Microbiomes as Diagnostic and Treatment Biomarkers: Recent Empirical Data

As of February 2025, we found only 174 papers in PubMed on the “microbiome” as a “diagnostic biomarker” for LC. Out of them, only 63 original articles were finally included for this study. The rest of them are either reviews, only abstracts, not covering LC or microbiome, only protocol papers, or inaccessible to us.

For LAC and NPC, we included only 1 and 3 papers from PubMed after initial exclusion. And no original studies were found related to Bronchial, Sinus, Pleural, and Tracheal cancers. The summary of these papers is shown in Table 2.

The majority of studies used the gut microbiome. Several studies combined the data from microbiome, metabolites, tumor markers, etc., to obtain better performance. Some even tried multiple microbiomes together, too.[48,49]

For presenting the prognostic/diagnostic efficacy, AUC values were used predominantly along with “r”, “HR”, “LDA”, “R²”, “C-index”, or simple “p values”. An AUC value greater than 0.7 indicates an acceptable value, and 0.9 indicates a highly reliable value for any diagnostic tool. Only two studies had AUC values less than 0.7 [50,51], and several studies achieved AUC values equal to or greater than 0.9 as well. [18,19,44,52,53,54,55] Some studies found higher AUC (~0.9) in the discovery data set, but lower AUC (~0.7) in the independent or validation dataset.[18,56]

Several machine learning models (MLMs) were also used in these studies, and the random forest (RF) model was the most frequent. The other MLMs, such as SVM, Lasso, XGBoost, etc., were also used individually or together.

Among microbial populations, Akkermansia muciniphila was associated with HCs or better treatment outcomes, even restoring antibiotic-induced dysbiosis along with E. hirae.[57] One study[33] even suggested the bacterium as a probiotic, too. Besides, Prevotella, Bifidobacterium, Faecalibacterium, etc., are also associated with positive outcomes. Based on available omics data, some studies used certain probiotics and analyzed their effects on treatment or survival. For example, Takada et al. (2022)[58] used Clostridium butyricum, Bifidobacterium sp., and antibiotic-resistant lactic acid bacteria as probiotics, finding moderate C-index values (0.57–0.62). Similarly, Qian et al. (2022) used Prevotella copri as an intervention.[36]

In treatment outcome, Alistipes indistinctus, Actinomycetota, Euryarchaeota, Bacillota, Bifidobacterium, and Collinsella showed effects on ICI treatment.[57] Other microbes (Alistipes shahii, Alistipes finegoldii, Barnesiella viscericola, Streptococcus salivarius, Streptococcus vestibularis, Parabacteroides, Methanobrevibacter, Veillonella, Selenomonadales, Negativicutes, Desulfovibrio, Actinomycetales, Bifidobacterium, Odoribacteraceae, Anaerostipes, Rikenellaceae, Faecalibacterium, etc.) showed association with better survivals (OS or PFS), or better response to chemotherapy, or association with anti-PD1 immunotherapy, etc.[59,60]

Overall, microbes increased and associated to LC include different species (Bacteroides thetaiotaomicron, Eubacterium spp.), genera (Acidovorax, Agathobacter, Bacteroides, Bifidobacterium, Blautia, Capnocytophaga, Cetobacterium, Clostridioides, Clostridium, Enterococcus, Faecalibacterium, Fusicatenibacter, Fusobacterium, Granulicatella, Haemophilus, Klebsiella, Lactobacillus, Lactococcus, Megasphaera, Mycobacterium, Pedobacter, Prevotella, Ruminococcus, Streptococcus, Veillonella, Xanthomonas), family (Christensenellaceae), order (Oscillospirales), and phylum (Proteobacteria, Verrucomicrobia, Firmicutes) etc.[19,32,44,48,61,62,63,64,65,66]

Moreover, NPC is associated with Corynebacterium, Staphylococcus, Prevotella, Porphyromonas, Epulopiscium, Terrisporobacter, Turicibacter, Pseudomonas, Cutibacterium, Finegoldia, and Paracoccus, and LAC with Fusobacterium, Acinetobacter, Pseudomonas, Klebsiella, Flavobacterium, Mycoplasma, Ralstonia, Streptococcus, Rothia, and Lactobacillus in tumor tissues and Saccharopolyspora, and Actinobacillus in oral rinse.[67,68,69]

4. Emerging Next-Generation Approaches

Besides omics approaches, some other emerging approaches are coming forward. Two such approaches are Liquid Biopsy, or using circulating microbial DNA (c-mDNA) or cell-free RNA (cfRNA) from blood samples [20,31,34,55], and analyzing the mycobiome. [54,70]

Nowadays, imaging approaches (tomography of gut microbiota) are getting polished for diagnostic use. These include positron emission tomography (PET), single-photon emission computed tomography (SPECT), and computed tomography (CT). While CT alone cannot detect microbiota, but with PET and SPECT imaging can establish a non-invasive biomarker technique both for diagnosis and treatment monitoring. These assays use tracers [¹⁸F-FDS (targets sugar metabolism of Enterobacteriaceae), ¹⁸F-FDG, D-[¹⁸F]-CF₃-ala (targets cell wall peptidoglycan), ²⁻¹⁸F-PABA (targets folate biosynthesis), etc.] for detection. [71,72,73,74,75]

5. Conclusion

Human microbiome as biomarkers shows high diagnostic potential for differentiating HCs and RTCs, cancer types, and monitoring treatment. Standardized protocols for sample collection, extraction, sequencing, access to easy-to-use analytic pipelines/tools, open-access databases, etc., are essential for clinical adaptation and reproducibility. However, the lack of reference ranges and the high population variations call for future cohort studies to determine whether microbes contribute to or originate from tumor development.

In conclusion, integrating microbiome data into cancer diagnostics and treatment holds great promise. Although we couldn’t access some original studies, the overall evidence suggests they would have shown a similar picture.