This post is part of a two-part series on Akoya’s software tools for spatial analysis. Part 2 covers the inForm Tissue Analysis software.
With the right analysis tools, researchers can move from multiplexed imaging data to visualizing the spatially resolved cellular structure and cellular phenotypes of the tissue microenvironment. As part of the CODEX® solution for spatially resolved, multiplexed immunofluorescence, the CODEX software suite includes powerful, easy-to-use tools for spatial biomarker analysis.
The CODEX Software Analysis Suite is comprised of three core components:
- Data Collection — The CODEX Instrument Manager (CIM) is used to control the fluidics instrument and integrates with the microscope control software for image acquisition
- Data Processing — The CODEX Processor formats images for downstream analysis and is used to perform drift compensation, background subtraction, deconvolution, cell segmentation and clustering
- Data Interpretation — The Multiplex Analysis Viewer (MAV) is an ImageJ plugin used to visualize, annotate, and analyze cell populations
Since we’re focusing on the analysis capabilities of the CODEX software suite in this post, we’ll cover some of the main features of the Multiplex Analysis Viewer (MAV).
Gating cell populations
CODEX can image over 40 biomarkers in a single tissue section, so it’s understandable if you’re overwhelmed with data upon completing your experiment. This is where gating comes in. Gating is the process of setting parameters to identify groups of cells for further analysis. For example, we can gate cells based on their expression patterns for a single marker. Users can then create and save that group as a cell population within MAV.
Once we’ve used gating to identify a specific cell population – say, one that’s DAPI-positive – it’s ready for further analysis. Through clustering, we can identify sub-populations expressing different biomarkers within the DAPI-positive population.
Unsupervised clustering to identify unique cell populations
To characterize the tissue microenvironment with multiplex imaging data, it’s necessary to identify unique cell subpopulations. This is achieved via cluster analysis. MAV uses X-shift clustering, an unbiased, k-means, density-based clustering algorithm for multidimensional single-cell data. X-shift clustering has been shown to reliably identify known cell populations as well as previously undiscovered populations.
Running the X-shift clustering tool in MAV produces a clustering panel, which shows the different biomarker densities present in each cluster.
Once clusters are generated, they can be visualized as cell populations directly on the tissue image in MAV. Colored dots in the tissue image represent different clusters. This enables users to easily identify regions with similar expression patterns, an important first step in determining the phenotype for each cell population.
Determining phenotype with T-SNE plots
T-SNE plots allow us to see biomarker expression in a graphical format, where cells are organized in two-dimensional space according to their similarity. In MAV, users can toggle back and forth between T-SNE plots and the original morphological data, to simultaneously examine spatial context and associations between cells.
The T-SNE plot combines two forms of high-dimensional analysis in one graph. Each two-dimensional region on the plot represents a distinct population of cells. When we color points in the graph by population, we can see the effects of cluster analysis, with each color representing a cluster as it does in the tissue image. The range of biomarker expression present in an image can also be reflected in the T-SNE plot by coloring points by density.
The two-dimensional information and biomarker intensities from the T-SNE plot and clustering enable us to identify and annotate cells within the image. Gating is used to separate regions in the T-SNE plot based on their biomarker expression profiles.
By visualizing specific biomarkers on the tissue image (with spatial context), we can draw conclusions about cell types and states. Box plots can be used to confirm these conclusions by comparing biomarker expression across different clusters.
Visualize spatial interactions with CIRCOS Plots
In MAV 1.4, we introduced CIRCOS plots to replace the spatial network graph for improved graphical representation of spatial interactions between populations. CIRCOS plots convey these interactions clearly, in a publication-friendly format.
MAV defines interactions by spatial proximity — users define the distance used to calculate the number of interacting cells. An interacting cell will be any cell that falls within a predefined spatial distance range (e.g., 5-10 μm).
In the example below, populations are shown along the outside of the plot, and are connected by colored ribbons. The graph shows the number of interactions both numerically, listed next to each ribbon, and visually, based on the color and size of the ribbon. Blue indicates low population interaction and yellow indicates high interaction. The CIRCOS plot also presents the biomarker characteristics for each population, to provide a sense of the different populations that are interacting with each other, and what markers they express.
The MAV software is a robust visualization solution for sophisticated algorithmic clustering of cellular phenotypes. For more information about the features we described, watch our on-demand webinar, “A Deep Dive on the Multiplex Analysis Viewer (MAV) to Analyze CODEX Images” to see Bioinformatics Field Application Scientist, Grady Carlson, PhD, demonstrate how MAV’s various features can be used in combination to identify and annotate cell populations.