Host and bacterial urine proteomics might predict treatment outcomes for immunotherapy in advanced non-small cell lung cancer patients.

INTRODUCTION: Urine samples are non-invasive approaches to study potential circulating biomarkers from the host organism. Specific proteins cross the bloodstream through the intestinal barrier and may also derive from gut microbiota. In this study, we aimed to evaluate the predictive role of the host and bacterial urine extracellular vesicle (EV) proteomes in patients with non-small cell lung cancer (NSCLC) treated with anti-PD1 immunotherapy. METHODS: We analyzed the urine EV proteome of 33 advanced-stage NSCLC patients treated with anti-PD1 immunotherapy with LC-MS/MS, stratifying patients according to long (>6 months) and short (≤6 months) progression-free survival (PFS). Gut microbial communities on a subcohort of 23 patients were also analyzed with shotgun metagenomics. Internal validation was performed using the Random Forest (RF) machine learning (ML) algorithm. RF was validated with a non-linear Bayesian ML model. Gene enrichment, and pathway analysis of host urine proteins were analyzed using the Reactome and Gene Ontology databases. RESULTS: We identified human (n=3513), bacterial (n=2647), fungal (n=19), and viral (n=4) proteins. 186 human proteins showed differential abundance (p<0.05) according to PFS groups, 101 being significantly more abundant in patients with short PFS and n=85 in patients with long PFS. We found several pathways that were significantly enriched in patients with short PFS (vs long PFS). Multivariate Cox regression showed that human urine proteins MPP5, IGKV6-21, NT5E, and KRT27 were strongly associated with long PFS, and LMAN2, NUTF2, NID1, TNC, IGF1, BCR, GPHN, and PPBP showed the strongest association with short PFS. We revealed that an increased bacterial/host protein ratio in the urine is more frequent in patients with long PFS. Increased abundance of E. coli and E. faecalis proteins in the urine positively correlates with their gut metagenomic abundance. RF ML model supported the reliability in predicting PFS for critical human urine proteins (AUC=0.89), accuracy (95%) and Bacterial proteins (AUC=0.74). CONCLUSION: To our knowledge, this is the first study to depict the predictive role of the host and bacterial urine proteome in anti-PD1-treated advanced NSCLC.