Zurück
  • Poster presentation
  • P-I-0277

Identifying mass spectrometry-based biomarkers for predicting tumor recurrence in lung cancer

Termin

Datum:
Zeit:
Redezeit:
Diskussionszeit:
Ort / Stream:
Clinical Proteomics

Poster

Identifying mass spectrometry-based biomarkers for predicting tumor recurrence in lung cancer

Thema

  • Clinical Proteomics

Mitwirkende

Karim Aljakouch (Heidelberg / DE), Qian-Wu Liao (Heidelberg / DE), Marc Schneider (Heidelberg / DE), Barbara Helm (Heidelberg / DE), Dario Frey (Heidelberg / DE), Christina Schmidt (Heidelberg / DE), Thomas Naake (Heidelberg / DE), Qiuqin Zhou (Mannheim / DE), Gernot Poschet (Heidelberg / DE), Rüdiger Hell (Heidelberg / DE), Hauke Winter (Heidelberg / DE), Julio Saez-Rodriguez (Heidelberg / DE), Wolfgang Huber (Heidelberg / DE), Carsten Hopf (Mannheim / DE), Ursula Klingmüller (Heidelberg / DE), Jeroen Krijgsveld (Heidelberg / DE), Junyan Lu (Heidelberg / DE)

Abstract

Cancer recurrence remains the main reason for cancer-related death, with nearly 50% of non-small cell lung cancer (NSCLC) patients experiencing recurrence after curative resection. As part of the SMART-CARE consortium, we aim to identify proteomic and metabolomic biomarkers to predict surgical outcomes and the necessity for chemotherapy to reduce recurrence risk.

We collected snap-frozen tumor and adjacent tissue samples from 60 stage II adenocarcinoma patients, with half relapsing within two years post-surgery. Longitudinal plasma samples at baseline, at the point of recurrence, and after two years in patients without recurrence were also gathered. Subsequently, proteomic, targeted/untargeted metabolomic, and lipidomic profiles of those samples were acquired with state-of-the-art mass spectrometry technologies. Extensive bioinformatics quality control was performed to ensure high-resolution data free of environmental or technical variations, which are common in clinical omic data. We then applied various machine learning models to predict recurrence using these data modalities.

Comparative analyses showed that linear XGBoost models with proteomic data from tumor-adjacent tissues most effectively predicted recurrence, with a cross-validated AUROC of 0.85, surpassing current genomic or transcriptomic models. Models based on baseline lipidomics performed slightly worse, with an AUROC of 0.78, but could be more accessible for clinical translation. We further applied multi-omics factor analysis (MOFA) to integrate biomarker candidates from all data modalities to understand their functional roles. This analysis revealed a gradient significantly associated with recurrence and shared between proteomic and lipidomic modalities. Enrichment analysis of the proteins associated with this gradient suggested a downregulation in cell-cell adhesion and upregulation in oxidative phosphorylation and fatty acid metabolism in tissues from patients prone to recurrence, highlighting the role of lipid metabolism in lung cancer prognosis. Our study presents viable protein and metabolite biomarkers for clinical use and offers a case study for future mass spectrometry studies aimed at clinical translation.

    • v1.20.0
    • © Conventus Congressmanagement & Marketing GmbH
    • Impressum
    • Datenschutz