Chemo-informatic strategy for imaging mass spectrometry-based hyperspectral profiling of lipid signatures in colorectal cancer

Abstract

Mass spectrometry imaging (MSI) provides the opportunity to investigate tumor biology from an entirely novel biochemical perspective and could lead to the identification of a new pool of cancer biomarkers. Effective clinical translation of histology-driven MSI in systems oncology requires precise colocalization of morphological and biochemical features as well as advanced methods for data treatment and interrogation. Currently proposed MSI workflows are subject to several limitations, including nonoptimized raw data preprocessing, imprecise image coregistration, and limited pattern recognition capabilities. Here we outline a comprehensive strategy for histology-driven MSI, using desorption electrospray ionization that covers (i) optimized data preprocessing for improved information recovery; (ii) precise image coregistration; and (iii) efficient extraction of tissue-specific molecular ion signatures for enhanced biochemical distinction of different tissue types. The proposed workflow has been used to investigate region-specific lipid signatures in colorectal cancer tissue. Unique lipid patterns were observed using this approach according to tissue type, and a tissue recognition system using multivariate molecular ion patterns allowed highly accurate (>98%) identification of pixels according to morphology (cancer, healthy mucosa, smooth muscle, and microvasculature). This strategy offers unique insights into tumor microenvironmental biochemistry and should facilitate compilation of a large-scale tissue morphology-specific MSI spectral database with which to pursue next-generation, fully automated histological approaches.

Fig. 1.

MS-based imaging technology in clinical settings.

Fig. 2.

Impact of variance-stabilizing normalization on information recovery via PCA. (A) Coselection of morphologically homogeneous tissue regions. (B) SD vs. rank of the mean diagnostic plots for heteroscedastic noise structure. (C) PCA before and after variance-stabilizing normalization. Bicross validation has been performed to confirm that PCs capture biologically relevant information not attributable to noise (40).

Fig. 3.

Image coregistration, feature coselection, and multivariate analysis. (A) Automatic image transformation for accurate coregistration of biochemical and histological features. (B) High-resolution optical image of H&E tissue section with regions of tumor (red boxes), muscle (green boxes), and healthy mucosa (blue boxes) selected. Shown is aligned DESI-MSI image with automated coselection of pixels corresponding to defined regions of interest. (C) Discriminatory analysis using the RMMC method with leave-region-out cross-validation for enhanced separation of tissue classes based on biochemistry.

Fig. 4.

Chemical reconstruction of tissue regions of interest using multivariate molecular ion patterns. (A and B) Optical H&E-stained image (A) with aligned DESI-MSI RGB image (B). (C) Reconstruction of three distinct histological regions (smooth muscle, blood vessels, and colorectal adenocarcinoma) based on molecular ion patterns extracted by “one against all” RMMC methodology.

Fig. 5.

Overall computational workflow for exploration of region-specific lipid biochemistry using MSI platforms.