Chemical tissue heterogeneity of young Arabidopsis stems revealed by Raman imaging combined with multivariate data analysis

[Display omitted] •Raman imaging revealed chemical heterogeneity in context with plant microstructure.•Spectra analysis included 3 different algorithm and a plant spectral database.•Primary and secondary plant cell walls as well as the cell contents were investigated.•Fitting the spectrum of every pixel by reference spectra discriminated hemicelluloses. Structural and chemical tissue heterogeneity in plant stems is essential to fulfil the many different functions for growth and survival. Cutting microsection of young developing stems of the model plant Arabidopsis opened the view on vascular bundles (transport of water, nutrients, food, and other chemicals), interfascicular fibers (mechanical support), parenchyma (production, storage) and the epidermis with cuticle (protection, barrier, exchange,..). Mapping such a cross-section with a Confocal Raman microscope resulted in hyperspectral datasets, which are the basis to image chemical heterogeneity with a spatial resolution of 300 nm in context with the microstructure. We generated the images based on three different multivariate approaches: unmixing to find the most pure components, cluster analysis so segment the dataset into clusters with spectral similarity, and fitting the original spectra at every pixel by a linear combination of plant component reference spectra. All three visualized chemical heterogeneity and confirmed in a complementary way the distribution of carbohydrates and aromatic components. The true component analysis was superior to cluster analysis in specificity and added information on lipids and starch distribution. Due to the multicomponent nature of plant tissues no “pure” components were retrieved, wherefore a subsequent mixture analysis (orthogonal matching pursuit) of extracted component spectra with a reference database followed. This led to details on the molecular composition of the spectra and tissues and was essential input for a final reference component fit at every pixel. By the last analysis, different aromatic components and hemicelluloses were discriminated and a similarity of their distribution patterns elucidated. Insights into starch and lipid distribution as well as the aromatic and protein co-location (mixtures) of cell contents were gained. The fit of the latter lumen content was inferior to the other fits due to noisier spectra (higher fluorescence background) and relevant protein and enzyme spectra were missing in our database. The gained comprehensive view on p