Advancing the understanding of Aβ plaques through a systematic comparison of statistical methods applied to hyperspectral Raman data
Annika Haessler, Nathalie Jung, Maike Windbergs- Psychiatry and Mental health
- Cellular and Molecular Neuroscience
- Geriatrics and Gerontology
- Neurology (clinical)
- Developmental Neuroscience
- Health Policy
- Epidemiology
Abstract
Background
While fluorescence microscopy is widely applied in Alzheimer’s disease (AD) research, spectroscopic imaging methods like Raman imaging have only emerged recently. As a label‐free, chemically selective method, Raman microscopy can complement established imaging approaches by providing data as hyperspectral images, containing information on the molecular structure, composition, and distribution within a sample. However, brain tissue is a challenging tissue to analyze, since spectral variations in the sample are extremely low. Thus, to analyze pathologic hallmarks such as beta‐amyloid (Aβ) plaques, including the surrounding tissue, a systematic approach to accessing hyperspectral images is needed and herein presented.
Method
Regions of interest containing Aβ plaques in brain sections of 10 month old APP/PS1 (Amyloid‐Precursor‐Protein / Presenilin‐1) mice were imaged using a confocal Raman microscope (WITec GmbH, Ulm, Germany). The hyperspectral Raman data sets were processed using Project 4 software (WITec GmbH, Ulm, Germany), in‐house written Matlab scripts (The Mathworks, Inc., Natick, USA) and the Matlab‐integrated RamanLight App.
Result
By comparing univariate peak integration analysis and multivariate methods, such as hierarchical cluster analysis, spectral unmixing, and 2D correlation spectroscopy on the same hyperspectral Raman data set of an Aβ plaque and its surrounding tissue, the approaches could be systematically studied. pathological structures in the tissue can only be successfully differentiated, if multivariate methods (cluster analysis, spectral unmixing) are used. Nonetheless, univariate peak integration can provide basic spatial and compositional information on the tissue and the Aβ plaque. By applying 2D correlation spectroscopy, interfaces between the Aβ plaque and the surrounding tissue can be thoroughly analyzed.
Conclusion
In this study, a hyperspectral Raman image of an Aβ plaque and the surrounding tissue has successfully been analyzed with diverse explorative statistical methods to access Therefore, this study paves the way for the application of label‐free techniques such as Raman imaging into AD research by serving as a guide for the systematic analysis of hyperspectral images of pathological structures.