MSACL 2019 EU Abstract

Chemometric Strategies for Sensitive Regions of Interest Annotation in Complex Imaging Mass Spectrometry Data

Patrick M. Wehrli (1), Wojciech Michno (1,) Kaj Blennow (1,2), Henrik Zetterberg (1,2,3,4) and Jörg Hanrieder (1,3)
(1) Department of Psychiatry and Neurochemistry, the Sahlgrenska Academy at the University of Gothenburg, S-431 80 Mölndal, Sweden (2) Clinical Neurochemistry Laboratory, Sahlgrenska University Hospital, S-431 80 Mölndal, Sweden (3) Department of Neurodegenerative Disease, UCL Queen Square Institute of Neurology, University College London, London WC1N 3BG, United Kingdom (4) UK Dementia Research Institute at UCL, London WC1N 3BG, United Kingdom

Warning: Undefined variable $headshot in /var/www/html/view_abstract/view_abstract_in_program.php on line 704

Patrick Wehrli (Presenter) University of Gothenburg

Presenter Bio: Patrick Wehrli is a postdoctoral research fellow in the analytical neurochemistry group at the University of Gothenburg, Sweden headed by Assoc Prof. Dr. Dipl Chem Jörg Hanrieder. He has a background in organic and medicinal chemistry from undergraduate studies at the University of Basel and industry experience at Bachem. He defended his PhD medicinal chemistry in 2016 at the University of Gothenburg, working on delineating bacterial lipid metabolism with ToF-SIMS imaging. He then joined the Department of Psychiatry and Neurochemistry where his current research is focused on biomolecular mechanisms that underlie neurodegenerative disease pathology by means of multimodal imaging mass spectrometry.

Relevant Financial Disclosures (within past 24 months) No relevant financial relationship(s) to disclose. Discussing Products? NO, presenter will NOT present educational content related to products/services of the ineligible company(ies) listed above. Conflict mitigation NOT REQUIRED.

Introduction
Imaging mass spectrometry (IMS) is used to visualize the spatial distribution of molecules on sample surfaces. In particular, matrix-assisted laser desorption/ionization (MALDI) IMS is a popular approach for studying biochemical processes related to disease pathology, pharmacotherapy and drug metabolism. Biochemical distribution maps obtained from the analysis of biological tissues can be matched to histological features, the technique has thus also been referred to as molecular histology.
A number of studies demonstrated the effects of data processing for interpretation and improved visualization of IMS data. Commonly, data processing methods are chosen based on rational motivations rather than comparative metrics and no quantitative measures to asses and compare processing options have been suggested. In here, we propose an approach to supply the demand for unbiased strategies to evaluate IMS data processing and its effects on multivariate image analysis including objective identification of histological ROI and associated chemical co-localizations.

Methods
We applied the MALDI imaging workflow as previously described. Briefly, fresh frozen brain tissue from control mice was cut into 12 μm thick sections in a cryostat microtome and collected onto conductive glass slides. The matrix was applied immediately prior to MALDI MS analysis using a dedicated matrix sprayer. For lipid analysis, 1,5-diaminonaphthalene was used as matrix. IMS data acquisition in negative ion mode was performed in reflector mode. Subsequently, the matrix was removed through ethanol washing for follow up H&E staining.
MALDI IMS data were processed in MATLAB (MathWorks). Evaluated spectral preprocessing steps included log10, ln and square root transformations along with their combination with common normalization methods including total ion current (TIC) normalization, median/mean normalization, root mean square normalization (RMS), and normalization to the maximum peak. Multivariate modelling was performed in SIMCA (Sartorius Stedim Biotech, Sweden).

Results & Discussion
Multivariate image analysis of the MALDI IMS date using principal component analysis (PCA) provided structural features. This in turn allowed precise segmentation of anatomical features on the PCA data scale using a region-based active contour segmentation algorithm. This class information was supplied to OPLS-DA modelling, which provided quality metrics i.e. predictive ability of OPLS-DA models to quantitatively compare the applied processing methods. Logarithmic transformation reduced the heteroscedastic noise structure of the data resulting in improved multivariate models and the visualization of even delicate anatomical features (i.e. Purkinje cell layer in cerebellum).