Classification of small blue round cell tumors by integrating peptide and N-glycan mass imaging spectrometric...

Small blue round cell tumor (SBRCT) is an umbrella term for several tumor types of completely different cellular origin and differentiation. Members of this tumor group share a common morphological phenotype, which is characterized by small round cells with high nuclear-cytoplasmic ratio. The diagnosis cannot be made based on histomorphology alone, instead requiring a range of immunohistochemical and molecular analyses. Accurate diagnosis is essential for the correct choice of treatment and proper prognostic assessment, but can be challenging on small pre-therapeutic biopsies.

In this proof-of-concept study matrix-assisted laser desorption/ionization mass spectrometry imaging (MALDI-MSI) of peptides and N-glycans from a TMA (tissue microarray) of 26 SBRCT samples comprising Ewing sarcoma (n = 5), rhabdomyosarcoma (n = 5), neuroendocrine carcinoma (n = 5), acute lymphoblastic leukemia (n = 5), nephroblastoma (n = 3), and neuroblastoma (n = 3) was used to create classification algorithms to distinguish between the tumor types. We trained and tested four different classification algorithms (gradient boosting (GB), support vector machine (SVM), k-nearest neighbor (KNN), linear discriminant analysis (LDA)) on pixel-level achieving accuracy values between 76 % and 100 %. The integration of peptide and N-glycan mass spectrometry in one dataset led to a 4-5 % increase in accuracy compared to the results on the individual datasets. GB was the most robust classifier. Tentative identification of discriminatory tryptic peptide and N-glycan signatures revealed proteins involved in epigenetic regulation and organization of the cytoskeleton as well as fucosylated N-glycans.

Peptide and N-glycan MSI has huge potential and is a promising tool to discriminate different SBRCT entities. Combining mass spectrometry imaging data from peptides and N-glycans even enhanced classification performance. At the same time, the results underline the importance of an accurate spatial mapping to create such integrated data sets.