Podium Presentation in Room 2 on Thursday at 15:30 (Chair: Livia Eberlin / Anand Mehta)
Authors: Nina Ogrinc (1), Pierre-Damien Caux (1), Alexandru Lintis (2,3), Florence Renaud (3,4), Khawla Seddiki (5), Maxence Wisztroski (1), Arnaud Droit (5), Guillaume Piessen (2,3), Michel Salzet (1) and Isabelle Fournier(1)
Introduction: Gastric cancer (GC) is the fourth most commonly diagnosed cancer. The extent of it can’t be evaluated from the initial CT-scan or endoscopy exam so the surgeons perform an exploratory surgery followed by the intraoperative pathological exam. The exam is usually between 30-40 min and the sub types of GC, such as Poorly Cohesive Carcinoma (PCC), are challenging to identify via histopathology, leading to inaccurate diagnosis. There is an urgent need for technology allowing for real-time diagnostics using molecular signatures while being of minimal invasiveness to the patient’s tissue. The water-assisted laser desorption/ionization MS (SpiderMass), designed for in vivo real time analysis has already demonstrated its capability for correct classification of tumour types and grading. Herein, we present a pipeline for ex vivo human biopsy analysis using the SpiderMass system combined with Mass Spectrometry Imaging to asses and classify sub-types of GC cancer moving us a step closer towards in vivo analysis.
Objectives: Our aim is to improve intraoperative diagnostics of sub-types of gastric cancer.
Methods 54 fresh-frozen biopsies of GC and normal tissue were supplied by the FREGAT national database. The biopsies were sectioned for histopathological examination (5µm), SpiderMass analysis (20µm) and MALDI (12µm) in 3 rounds. The SpiderMass uses a mini invasive IR-laser microprobe under ambient conditions tuned to excite the most intense vibrational band of water. The microprobe is connected to the Xevo (Waters) instrument via a transfer tube. Lipidomic and metabolic profiles were acquired using 10 s irradiation steps in positive and negative ion mode. The data collected was processed via supervised AMX (Waters Research Centre) and Convolutional Neuronal Network based "home-built" software. For MALDI, the sections were coated with Norharmane (7mg/mL in CHCl3:MeOH, 2:1, HTX Imaging TM-sprayer) and analysed using the Bruker RapifleX MALDI Tissue Typer™ at 50 µm.
Results: The data files were directly imported into the AMX software and processed using PCA-LDA analysis. The PCA generated list of features with the largest variance are subjected to supervised LDA through user assigned classes (adenocarcinoma, PCC and normal tissue). The model very well discriminated between the PCC and the rest of adenocarcinoma. Additional classification models were made using CNN transfer learning on unprocessed data. The cumulative transfer learning yielded 99.3 % accuracy on the developed model. Cross-validation with MSI was performed by dual polarity mass spectrometry imaging. The most discriminative peaks in SpiderMass analysis m/z 796.5, 810.6, 742.5 and 766.5 were also localized with MALDI-MSI to reveal tumour heterogeneity within the PCC biopsy, corresponding to cell type differentiation.
Conclusion: The new developed pipeline allows for rapid biomarker detection, guided surgery and real-time analysis of sub-types of GC.
|Planning to mention or discuss specific products or technology of the company(ies) listed above:||