MIBI EXPRIMENTAL AND ANALYTICAL WORKFLOW
(A) Experimental workflow for multiplexed imaging. Clinical FFPE tissue specimens are placed on a slide and stained using a mixture of commercially available antibodies against desired antigens, labeled with elemental mass tags. The slide is placed in the MIBI machine, and subjected to a rasterizing oxygen duoplasmatron primary ion beam. As the ion beam strikes the sample, lanthanide adducts of the bound antibodies are liberated as secondary ions, which are recorded by a time-of-flight (TOF) mass spectrometer. For each physical pixel in the tissue, a mass spectrum is recorded, representing the abundance of the antigens in that location. Spectral data is converted to n-dimensional tif files, with each image depicting the expression of one protein in the FOV. (B-G) Analytical pipeline for multiplexed imaging. (B) Cell segmentation is performed using a convolutional neural net. Training data was generated by manual annotation, and a network was trained to classify each pixel in the image as either ‘Nuclei interior’,’Nuclear border’ or ‘background’. The network was applied to segment new images, not used for training, and was shown to achieve high accuracy across multiple tissue specimens. (C) Following segmentation, cellular features are extracted. These include morphological features (e.g. size), and expression values for each one of the proteins. Cells are clustered and assigned to distinct cell populations (e.g. tumor cells, endothelial cells, B cells). (D) Cell populations are annotated within the context of the tissue using pseudo-coloring. (E) For each patient a spatial enrichment analysis is performed to identify spatial organization patters within the tissue and cell types that are either enriched or depleted for residing in close proximity. Using these distance matrices, locations of spatial features such as necrotic regions and the tumor-immune border are identified. (F) For each patient, quantitative information regarding cell types and their locations is extracted. (G) Patient are clustered using the extracted features to identify similar subtypes. Clinical parameters, such as progression free survival, are evaluated for the different groups.