Immune checkpoint inhibitors (ICIs) targeting the PD-1/PD-L1 pathway have transformed the treatment of non-small-cell lung cancer (NSCLC), the most prevalent and deadly form of lung cancer worldwide. Since 2020, neoadjuvant ICIs - treatments given before surgery - have become increasingly incorporated into care for patients with resectable early-stage NSCLC.
Response rates remain inconsistent. Only 30%-56% of patients with resectable NSCLC achieve a major pathologic response (MPR) from neoadjuvant immunochemotherapy, while up to 32%-63% experience severe treatment-related adverse events. This wide variability makes predicting who will benefit an urgent clinical need.
Existing biomarkers fall short. PD-L1 expression - the most widely accepted marker - does not correlate with pathologic response in neoadjuvant settings for early-stage NSCLC, even though it predicts response in advanced disease. Other markers like tumor mutation burden and microsatellite instability have also not proven sufficient for early-stage prediction.
The gut microbiome and metabolites have emerged as promising new candidates. Studies in advanced NSCLC show that bacteria like Akkermansia and Bifidobacterium are linked to better immunotherapy outcomes. However, no prior study had examined these associations specifically in early-stage NSCLC patients undergoing neoadjuvant immunochemotherapy.
The LungMark study enrolled 53 NSCLC patients at Sun Yat-sen University Cancer Center between December 2020 and November 2022. All patients had stage IIB-IIIA NSCLC eligible for complete surgical removal (R0 resection). After exclusions for adverse events or ineligibility, 44 patients who completed all planned treatment were analyzed.
Treatment protocol consisted of neoadjuvant tislelizumab (an anti-PD-1 antibody) combined with platinum-based doublet chemotherapy, administered every three weeks for 3-4 cycles before surgical resection. The chemotherapy regimen was tailored to tumor histology: squamous cancers received carboplatin plus paclitaxel, while non-squamous cancers received carboplatin plus pemetrexed.
Samples collected at baseline and before surgery included both fecal and plasma specimens from each patient. The major pathologic response (MPR) was defined as less than 10% viable tumor cells remaining in the surgically removed specimen - a stringent indicator of treatment success.
Multi-omics analyses combined shotgun metagenomic sequencing of fecal DNA to identify gut bacteria and their functional genes, alongside untargeted liquid chromatography-mass spectrometry (LC-MS) metabolomics of both plasma and fecal samples to detect thousands of metabolic compounds.
Metagenomic sequencing used the Illumina NovaSeq 6000 platform to generate paired-end reads from 68 fecal samples. After quality filtering and removal of human DNA contamination, reads were assembled into contigs and annotated against the NCBI database, yielding profiles of 499 bacterial species and 300 metabolic gene functions (KEGG orthologues).
Metabolomic profiling used ultra-high-performance liquid chromatography (UHPLC) coupled with a high-resolution Orbitrap mass spectrometer, detecting 1,498 plasma metabolites and 2,743 fecal metabolites after quality filtering. Quality control samples were inserted throughout each batch to ensure measurement consistency.
Differential feature discovery used three complementary statistical methods - univariate testing, multivariate linear models adjusting for age and tumor histology, and random forest machine learning - to identify bacterial species and metabolites robustly associated with MPR. Only features that passed all three approaches were designated as important differential features.
Predictive modeling used logistic regression to build biomarker panels, evaluated by receiver operating characteristic (ROC) curves, calibration curves, 10-fold cross-validation, and decision curve analysis. Spearman and distance correlation analyses were used to explore microbe-metabolite relationships that might explain the biological connections.
Patients who achieved MPR had greater gut microbial diversity at baseline, measured by both Shannon and Simpson alpha-diversity indices. A small but statistically significant separation in overall microbial community composition (beta-diversity) was also observed between MPR and non-MPR patients before treatment. Notably, these differences disappeared after neoadjuvant therapy.
Specific beneficial bacterial species were enriched in MPR patients at baseline. These included multiple species from the Clostridiales order and Lachnospiraceae and Ruminococcaceae families - particularly Clostridium sp. M62/1, Eisenbergiella tayi, Ruminococcus callidus, Ruminococcus bicirculans, Angelakisella massiliensis, and Ruminococcaceae bacterium D16.
Prevotella species were notably depleted in MPR patients compared to non-responders - a finding that contrasts with results from studies in advanced NSCLC, where Prevotella is associated with better outcomes. The authors propose this discrepancy reflects stage-specific biology, as pro-inflammatory bacteria may promote immunosuppression in early-stage cancer while potentially benefiting advanced disease.
Microbial functional analysis revealed that non-MPR patients had enrichment of genes related to carbohydrate, amino acid, and cofactor metabolism, contributed primarily by Proteus mirabilis and Prevotella species. These metabolic gene functions were also correlated with virulence factors including the Mu toxin and metalloprotease, suggesting pathogenic pathways may impair treatment response.
The plasma metabolomic profile showed modest but statistically significant differences between MPR and non-MPR patients at baseline, while fecal metabolomes showed no significant differences. Overall, 74 plasma metabolites were associated with MPR, with the most affected metabolic pathways being linoleic acid metabolism and aromatic amino acid pathways including tryptophan, tyrosine, and phenylalanine.
Linoleic acid (LA) and its derivatives were elevated in non-MPR patients. Specifically, plasma LA, fecal gamma-linolenic acid (GLA), and both plasma and fecal 9-hydroxyoctadeca-10,12-dienoic acid (9-HODE) were significantly higher in non-responders. These omega-6 fatty acids are known to be pro-inflammatory and tumorigenic, potentially counteracting the immune-activating effects of anti-PD-1 therapy.
Tryptophan metabolism metabolites told a more nuanced story. L-kynurenine was elevated in MPR patients, while quinolinic acid (QA) was paradoxically higher in non-MPR patients. Indole-3-acetic acid (IAA) and its oxidative derivative oxindole-3-acetic acid (OxIAA) were both significantly enriched in MPR patients, in both plasma and fecal samples. Higher baseline IAA was also associated with complete pathologic response and longer disease-free survival.
The biological logic behind these metabolite signals relates to immune regulation. IAA and OxIAA may promote anti-tumor immunity through neutrophil-driven reactive oxygen species production that inhibits cancer cell proliferation. In contrast, QA helps tumor cells resist oxidative stress-induced cell death and promotes macrophage polarization toward immune tolerance, effectively shielding tumors from immune attack.
Significant correlations between specific gut bacteria and metabolites support a mechanistic link. Clostridium sp. M62/1 was positively correlated with plasma IAA and OxIAA - consistent with the known ability of Clostridium species to produce indolic compounds from aromatic amino acids. Conversely, linoleic acid metabolites were negatively correlated with butyrate-producing bacteria including Butyricicoccus pullicaecorum and Eisenbergiella tayi.
Prevotella species showed negative correlations with kynurenine and positive correlations with quinolinic acid, suggesting that Prevotella-dominated microbiomes may promote a metabolic environment favoring tumor immune escape through quinolinic acid-mediated suppression of anti-tumor immunity.
Two integrated prediction models achieved excellent performance. The integrated microbial model - combining just two species, Clostridium sp. M62/1 and Eisenbergiella tayi - reached an AUC of 0.89. The integrated metabolomic model - combining linoleic acid, oxindole-3-acetic acid, and quinolinic acid - reached an AUC of 0.85. Both models passed goodness-of-fit tests and maintained good accuracy in 10-fold cross-validation (82% and 77%, respectively).
Decision curve analysis showed that both models offered meaningful clinical benefit compared to treating all patients or no patients. The metabolomic model using plasma samples from all 44 patients covers a broader range of clinical scenarios, while the microbial model provides particularly high net benefit at intermediate decision thresholds. A Cox regression model using LA, OxIAA, and QA also predicted 1-year and 2-year disease-free survival with AUCs of 0.79 and 0.74, respectively.
A key insight from this study is that the role of specific gut bacteria in immunotherapy response appears to depend on cancer stage. Prevotella bacteria have been linked to better outcomes in advanced NSCLC in prior research, but this study finds them associated with non-response in early-stage disease. The authors suggest this reflects different immune microenvironments: pro-inflammatory signals that help activate immunity in immunosuppressed advanced tumors may worsen outcomes in early-stage tumors where inflammation already promotes tumor growth.
The well-established microbiome biomarkers for advanced NSCLC - Akkermansia muciniphila, Bifidobacterium adolescentis, and Bifidobacterium longum - showed consistent enrichment trends in MPR patients in this study but did not reach statistical significance, and adding them did not improve model performance. This is consistent with the idea that disease-stage context matters when interpreting microbiome-immunotherapy relationships.
The linoleic acid finding connects diet and gut microbiota metabolism to treatment outcomes. LA is an essential dietary fatty acid that can be co-metabolized by host and gut bacteria. Its downstream conversion to arachidonic acid drives pro-inflammatory pathways, and its oxidative derivatives (9-HODE, 13-HODE) can affect tumor proliferation - a mechanism that may explain why high baseline LA levels predict poor immunotherapy response.
Plasma metabolites outperformed fecal metabolites as predictors, likely because plasma reflects the systemic impact of gut microbial metabolism after intestinal absorption. Many microbially derived metabolites - especially lipids, bile acids, and amino acid derivatives - are primarily absorbed in the small intestine and reach systemic circulation, making plasma a more integrative window on microbe-host metabolic interactions than fecal samples alone.
This is the first study to systematically examine gut microbiome and metabolome signatures in early-stage NSCLC patients receiving neoadjuvant anti-PD-1 immunochemotherapy. The baseline microbial and metabolic profiles - collected before any treatment - can reliably predict which patients will achieve major pathologic response, enabling prospective patient stratification.
The simplicity of the predictive panels is clinically attractive. Just two bacterial species or three plasma metabolites can achieve prediction accuracy (AUC 0.85-0.89) that exceeds conventional biomarkers like PD-L1 expression, tumor mutation burden, and neutrophil-to-lymphocyte ratio, all of which showed no significant association with MPR in this cohort.
Limitations require acknowledgment. The cohort was relatively small (44 patients), and the study had not yet reached median disease-free survival as of March 2025, making DFS-based conclusions preliminary. Dietary data - which heavily influence both linoleic acid levels and gut microbiome composition - were not collected, and independent external validation is needed before clinical implementation.
The findings open a path toward microbiome-based interventions. If Clostridiales-family bacteria and tryptophan metabolites like IAA truly enhance immunotherapy response, probiotic supplementation or dietary manipulation targeting these pathways could potentially improve outcomes for patients unlikely to respond. This represents a novel, mechanistically grounded strategy for personalizing neoadjuvant immunotherapy in lung cancer.