Advanced Registration and Analysis of MALDI Imaging Mass Spectrometry Measurements through Autofluorescence Microscopy

Abstract

The correlation of imaging mass spectrometry (IMS) with histopathology can help relate novel molecular findings obtained through IMS to the well-characterized and validated histopathology knowledge base. The quality of correlation between these two modalities is limited by the quality of the spatial mapping that is obtained by registration of the two image types. In this work, we develop novel workflows for MALDI IMS-tomicroscopy data registration and analysis using nondestructive IMScompatible wide field autofluorescence (AF) microscopy combined with computational image registration. First, a substantially automated procedure for high-accuracy registration between IMS and microscopy data of the same section is described that explicitly links the MALDI laser ablation pattern imaged by microscopy to its corresponding IMS pixel. Subsequent examination of the registered data allows for high-confidence colocalization of image features between the two modalities, down to single-cell scales within tissue. Building on this IMS-microscopy spatial mapping, we furthermore demonstrate the automated spatial correlation between IMS measurements from serial sections. This AF-registration-driven inter-section analysis, using a combination of nonlinear AF-to-AF and IMS-to-AF image registrations, can be applied to tissue sections that are prepared and imaged with different sample preparations (e.g., lipids vs proteins) and/or that are measured using different spatial resolutions. Importantly, all registrations, whether within a single section or across serial sections, are entirely independent of the IMS intensity signal content and thus unbiased by it.

Discovering new lipidomic features using cell type specific fluorophore expression to provide spatial and biological specificity in a multimodal workflow with MALDI IMS

“…Sections were analyzed for GCs in spleens of mice, immunized or not, and bearing or lacking an AID-GFP transgenic fluorophore. The same tissue sections were then used for IMS, while serial sections were used for IF and H&E. This method was applied to an investigation of the differences between lipids associated with GCs and other regions in spleens using non-destructive Fem as a mono-modal registration medium 19 (Fig. 1e).…”

“…Image registration techniques were performed according to previously published methods 19 , however, rather than using AF images for registration we used Fem / AF images.…”

“…To develop this technology, we analyzed the spleens of unimmunized and immunized mice using a well-characterized tracking allele that encodes green fluorescent protein (GFP) to enable high accuracy image registration and provide biological context 19 . Data mining strategies such as manual interpretation 9,11 , standard segmentation 20 , and data-driven image fusion 21 were subsequently applied to determine whether lipids could be mapped to a feature of normal microanatomy in immune responses.…”

Lipid Landscape of the Human Retina and Supporting Tissues Revealed by High Resolution Imaging Mass Spectrometry

“…Registration of the IMS data to microscopy was performed using an explicit IMS pixelto-laser ablation mark registration method described previously 33 Targeted: Mass spectra were collected over a scan range of 500 -1600 m/z, at a resolution of 30,000. MS/MS data were collected using parallel reaction monitoring at a resolution of 15,000 to isolate and fragment several ions of interest (see Supplemental Table 1).…”

Unsupervised machine learning for exploratory data analysis in imaging mass spectrometry

“…More precisely, this review attempts to collect representative examples of unsupervised machine learning algorithms and their applications in an IMS context. This means that work related to preprocessing (e.g., normalization (Deininger et al, 2011;Fonville et al, 2012;Källback et al, 2012), baseline correction (Coombes et al, 2005;Källback et al, 2012), peak picking and feature detection (McDonnell et al, 2010;Alexandrov et al, 2010;Bedia, Tauler, & Jaumot, 2016;Du, Kibbe, & Lin, 2006), data formats (Schramm et al, 2012;Rübel et al, 2013;Verbeeck et al, 2014a;Verbeeck, 2014b;Verbeeck et al, 2017), spatial registration (Schaaff, McMahon, & Todd, 2002;Abdelmoula et al, 2014a;Anderson et al, 2016;Patterson et al, 2018aPatterson et al, , 2018b, and supervised methods such as classification (Luts et al, 2010) and regression (Van de Plas et al, 2015) do not fall within the scope of this review, unless there is a substantial contribution to their analysis pipeline by an unsupervised machine learning algorithm. Our focus will lie on three particular subbranches within unsupervised methods, namely (i) factorization methods, (ii) clustering methods, (iii) manifold learning methods, and any hybrid methods that feature a strong relationship to these approaches.…”