Multi-Center Assessment of Reproducibility of Mass Spectrometry Imaging

Background Classification of tissues based on label-free mass spectrometric phenotypes measured directly from sections is a promising tool for clinical research. However, reproducibly measuring mass spectra can be challenging and comprehensive studies assessing the variation across different sites are largely lacking. In this work we have compared the reproducibility of Matrix-Assisted-Laser-Desorption/Ionization MALDI mass spectrometric imaging (MALDI-MSI) based tissue classifications measured at three different sites. Design A tissue microarray (TMA) was constructed to contain human FFPE samples representing different tumors. The tumors were leiomyoma, seminoma, lymphoma, melanoma, breast cancer and non-small -cell lung cancer. Each tumor type was sampled from 5 different subjects at each three different sites. These samples were prepared for MALDI-MSI and measured at three different sites using a standard protocol. A linear discriminant analysis (LDA) was used to generate a classifier for these 6 tumor types and all pairwise classifications. The accuracy of the LDA models were evaluated by a two-step "leave-one-core-out-leave-one-measurement-out" cross-validation so that the classifier has never seen the individual subject nor the measurement. Unsupervised clustering and principal component analysis (PCA) were used to asses overall spectral similarities. Results The accuracy of the classification on the individual pixel level was 76% for the 6-class classifier and up to 93% for the pairwise classifications. We did not see a systematic influence of the sampling site or the measurement site on the classification accuracy. Unsupervised clustering and PCA analysis of mass spectral similarities also indicated that the biological difference between tissue states showed a larger influence on the spectral phenotypes than sampling site or measurement site. Conclusion To our knowledge this is the first study that has evaluated the site-to-site reproducibility of MALDI-MSI classification from FFPE-tissue using clinical samples. Our initial results indicate that MALDI-MSI can be performed at a level that allows relevant multi-center research studies.