MSACL 2023 Abstract

Aim:
With about 951,000 cases each year, esogastric cancer (EC) is the fifth most often-diagnosed cancer worldwide. EC encompasses a variety of different carcinoma subtypes among which the Poorly Cohesive Carcinoma (PCC) is a very aggressive type occurring in young patients. PCC represents a diagnostic challenge because of its diffuse character and therefore, is difficult to distinguish intraoperatively leading to a relapse of a half of the patients. Our objective is to setup an accurate intraoperative diagnostic to assist surgeons in the uptake of PCC using the SpiderMass technology. To this end MALDI-MSI was used to depict the tissue heterogeneity and reveal specific lipid profiles of PCCs by comparison to other ECs using SpiderMass.

Methods:
A cohort of EC excised tissues were sectioned on a cryostat (Leica CM 1510S). Three sections were collected, one (5 µm) for the H&E staining, the second (20 µm) for SpiderMass analysis and the last one (12 µm) for the MALDI-MSI. The MALDI-MSI analysis was performed in both polarities on a MALDI-TOF (Rapiflex, Bruker) at 50 µm spatial resolution using norharmane as matrix (7 mg/mL). The SpiderMass analysis was performed on a Q-TOF instrument (Xevo, Waters) in both polarities. The data were processed using SCILS (MALDI-MSI and SpiderMass) and AMX (SpiderMass). LDA was used to build classification models and the model was assessed by interrogation in blind on a validation cohort of 50 tissues.

Results:
The 136 (68 healthy and 68 cancers) EC sections that were analyzed using MALDI-MSI showed a different and specific molecular profile for each type of tissue, enabling a clear discrimination between them. Because of the strong similarity between the MALDI-MSI and the SpiderMass-MSI data (93% in negative ion mode), the same cohort was analyzed with the SpiderMass. SpiderMass images were very like the MALDI ones leading to the identification of the same discriminative markers. The markers discriminating the different types and subtypes were identified by MS2 and manually annotated. The SpiderMass data were then used to build classification models for typing and subtyping the EC using machine-learning and deep-learning. The models were further validated in blind on an extra cohort of 50 tissues.

Discussion:
Thanks to the training cohort (136 tissues) and the validation cohort (50 tissues), we were able to build a very trustful classification model for discriminating cancer from healthy tissues first, but also to subtype the esogastric cancer. Thanks to this proof-of-concept we will be able to further evaluate the SpiderMass as a fast intraoperative diagnostic and prognostic tool in the pathology lab for direct comparison to the gold standard histology.