Spatial and Spectral Correlations in MALDI Mass Spectrometry Images by Clustering and Multivariate Analysis

Abstract

Matrix-assisted laser desorption/ionization mass spectrometry imaging is a technique for direct analysis of tissue sections without the use of molecular tags or contrast agents. The combination of spatial and mass resolution results in large and complex data sets that require powerful and efficient analysis and interpretation tools. Conventional images, derived from a specific analyte mass, do not identify the spatially localized correlations between analytes that are latent in the data. A new approach to find and visualize these correlations is presented. Clustering methods are used to classify pixels by spectral similarity, facilitating definition of distinct spatial regions. Principal component and discriminant analyses are combined to comprehensively identify changes in the mass spectra between regions. Images are generated by projecting the spectra of each pixel on the discriminant spectra; contrast is then a function of multiple correlated peaks.

MALDI Imaging Combined with Hierarchical Clustering as a New Tool for the Interpretation of Complex Human Cancers

“…Such effects have been reported in the literature. 15 McCombie et al 15 discouraged the use of PCA scores for direct generation of images, because of the effect of inhomogeneities of the sample preparation. In recent years, however, the sample preparation in MALDI imaging was significantly improved, and with the imagePrep device used in this study, a homogeneous coating of the sample was regularly achieved.…”

“…[11][12][13] It has also been reported in the context of MALDI imaging as a tool to find characteristic masses for specific tissue regions 14 or as a data pretreatment for multivariate classifications. 15 However, it has not found widespread application in the direct reconstruction of images in MALDI imaging. There are also some limitations to this approach, namely, the scores of spectra can be similar for one principal component, while they can be different for other principal components.…”

“…This can be achieved by hierarchical clustering, 16 which was previously employed in MALDI imaging data sets to generate images based on spectra similarity. 15 However, until now, it has not reached widespread use, because a convenient user interface was unavailable to gain full advantage of the clustering results. Here we present hierarchical clustering as a tool for the easy and convenient interpretation of single imaging data sets.…”

The C-terminal fragment of the immunoproteasome PA28S (Reg alpha) as an early diagnosis and tumor-relapse biomarker: evidence from mass spectrometry profiling

“…A decade of developments in instrumentation and chemistry has been required to achieve optimal extraction, detection, and spatial resolution when mapping compounds at the tissue level (Franck et al 2009a, b;Lemaire et al 2006a, b;2007a;van Remoortere et al 2010). These developments, combined with using the appropriate data-processing tools, gave rise to the discovery of biomarkers for diverse pathologies (Bonnel et al 2011;Djidja et al 2010;McCombie et al 2005;Stauber et al 2008), of which cancer is the most studied. MALDI mass spectrometry imaging (MALDI-MSI) technology makes it possible to study the molecular profiles of the benign and malignant portions of a solid tumor.…”

Imaging mass spectrometry: principle and application

“…In contrast, the concentration of the C-18 species decreased in the molecular layer (ML)/SLM with aging (arrows). Reprinted from Sugiura et al (2008) pons/medullar pons/medullary PCA provides information regarding which molecules are altered in tissue microdomains and is very helpful in analyzing large datasets of imaging MS (Altelaar et al 2007;Denkert et al 2006;Lapolla et al 2007;McCombie et al 2005;Zaima et al 2009b).…”