A multi-modal image fusion workflow incorporating MALDI imaging mass spectrometry and microscopy for the study of small pharmaceutical compounds

Abstract

Multi-modal imaging analyses of dosed tissue samples can provide more comprehensive insight into the effects of a therapeutically active compound on a target tissue compared to single-modal imaging. For example, simultaneous spatial mapping of pharmaceutical compounds and endogenous macromolecule receptors is difficult to achieve in a single imaging experiment. Herein, we present a multi-modal workflow combining imaging mass spectrometry with immunohistochemistry (IHC) fluorescence imaging and brightfield microscopy imaging. Imaging mass spectrometry enables direct mapping of pharmaceutical compounds and metabolites, IHC fluorescence imaging can visualize large proteins, and brightfield microscopy imaging provides tissue morphology information. Single-cell resolution images are generally difficult to acquire using imaging mass spectrometry, but are readily acquired with IHC fluorescence and brightfield microscopy imaging. Spatial sharpening of mass spectrometry images would thus allow for higher fidelity co-registration with higher resolution microscopy images. Imaging mass spectrometry spatial resolution can be predicted to a finer value via a computational image fusion workflow, which models the relationship between the intensity values in the mass spectrometry image and the features of a high spatial resolution microscopy image. As a proof of concept, our multi-modal workflow was applied to brain tissue extracted from a Sprague Dawley rat dosed with a kratom alkaloid, corynantheidine. Four candidate mathematical models including linear regression, partial least squares regression (PLS), random forest regression, and two-dimensional convolutional neural network (2-D CNN), were tested. The random forest and 2-D CNN models most accurately predicted the intensity values at each pixel as well as the overall patterns of the mass spectrometry images, while also providing the best spatial resolution enhancements. Herein, image fusion enabled predicted mass spectrometry images of corynantheidine, GABA, and glutamine to approximately 2.5 μm spatial resolutions, a significant improvement compared to the original images acquired at 25 μm spatial resolution. The predicted mass spectrometry images were then co-registered with an H&E image and IHC fluorescence image of the μ-opioid receptor to assess co-localization of corynantheidine with brain cells. Our study also provides insight into the different evaluation parameters to consider when utilizing image fusion for biological applications.

Enhancing Spatial Resolution in Tandem Mass Spectrometry Ion/Ion Reaction Imaging Experiments through Image Fusion

“…For example, pan sharpening applications of image fusion fuse high spatial resolution panchromatic images with low spatial resolution multispectral images to generate a high spatial resolution multispectral image. ,− Van de Plas et al demonstrated image fusion using mass spectrometry images by combining molecular information from the mass spectrometer with fine-grained texture information from optical microscopy images to generate a mass spectrometry image predicted to higher spatial resolution (i.e., compared to the microscopy image) . Our group has utilized this image fusion workflow to enhance imaging mass spectrometry spatial resolution for better integration with high spatial resolution imaging modalities (e.g., fluorescence and brightfield microscopy) in neuropharmacology studies . Image fusion has also been used to combine multiple mass spectrometry-based images.…”

“…The accuracy of each model was evaluated based on previously described parameters. , This includes visual inspection of the spatial sharpening of the predicted ion images, calculation of the correlation scores (i.e., the R 2 value obtained by plotting the intensity values from the ion image before image fusion against those of the predicted image using the plotregression function in MATLAB), and calculation of the average absolute residual (eq ): ∑ n = 1 n = N abs false( I ( original ) n − I ( predicted ) n false) N where N is the number of total pixels, I (original) n is the intensity value of the n th pixel for the ion image before image fusion, and I (predicted) n is the intensity value of the n th pixel for the predicted ion image. A smaller average absolute residual indicates the model more accurately predicts the intensity values of the ion image.…”

“…The 2D CNN model is the same model used by our group previously and is optimized in terms of epoch (Supplemental Table 1). The reported spatial resolutions of the predicted ion/ion reaction product ion images are equal to the spatial resolution of the full scan imaging data set. However, our previous work utilizing image fusion demonstrates that different levels of spatial sharpening are achieved with different models; thus, the level of spatial sharpening of the predicted image must be visually inspected.…”