Cytokit: a single-cell analysis toolkit for high dimensional fluorescent microscopy imaging

Abstract

Background: Multiplexed in-situ fluorescent imaging offers several advantages over single-cell assays that do not preserve the spatial characteristics of biological samples. This spatial information, in addition to morphological properties and extensive intracellular or surface marker profiling, comprise promising avenues for rapid advancements in the understanding of disease progression and diagnosis. As protocols for conducting such imaging experiments continue to improve, it is the intent of this study to provide and validate software for processing the large quantity of associated data in kind.

Results: Cytokit offers (i) an end-to-end, GPU-accelerated image processing pipeline; (ii) efficient input/output (I/O) strategies for operations specific to high dimensional microscopy; and (iii) an interactive user interface for cross filtering of spatial, graphical, expression, and morphological cell properties within the 100+ GB image datasets common to multiplexed immunofluorescence. Image processing operations supported in Cytokit are generally sourced from existing deep learning models or are at least in part adapted from open source packages to run in a single or multi-GPU environment. The efficacy of these operations is demonstrated through several imaging experiments that pair Cytokit results with those from an independent but comparable assay. A further validation also demonstrates that previously published results can be reproduced from a publicly available multiplexed image dataset.

Conclusion: Cytokit is a collection of open source tools for quantifying and analyzing properties of individual cells in large fluorescent microscopy datasets that are often, but not necessarily, generated from multiplexed antibody labeling protocols over many fields of view or time periods. This project is best suited to bioinformaticians or other technical users that wish to analyze such data in a batch-oriented, high-throughput setting. All source code, documentation, and data generated for this article are available under the Apache License 2.0 at https://github.com/hammerlab/cytokit .

Keywords: Automatic image processing; CellProfiler; Data exploration; Data visualization; GPU; Multiplexed fluorescence imaging.

Fig. 1

Seven dimensional multiplexed experiment structure illustrating how 3D single channel images are grouped as cycles, captured as tiles on a grid, and then potentially repeated over multiple fields of view (as “regions”) before being quantified as two dimensional single-cell information

Fig. 2

Processing pipeline overview with operations in original CODEX project replaced by GPU-accelerated equivalents, decoupled from tile assembly, and modified to support labeled object extraction as well as ad-hoc image stacks/montages

Fig. 3

Example iterative pipeline optimization process with CLI commands used to continually refine and expand the scope of experiment processing for large raw image datasets

Fig. 4

Example extraction and montage CLI commands. a CLI commands define slices as names (where relevant) or as lists and ranges of indexes to extract raw, processed, or object image data. b Resulting TIFF files are ImageJ compatible for blending and visualization, like the example shown with human T cells labeled as CD3 (green), DAPI (blue), nuclei boundaries (cyan), and cell boundaries (red)

Fig. 5

Cytokit Explorer screenshot (see screencast for animated version) showing a CODEX sample imaged at 20x. a 1D or 2D plots of expression, morphological, or graphical cell features support box and free-hand gating. b Custom filtering, to ignore a central photobleached region of cells in this case, or summarizations are applied immediately. c Single cell images match current gate and selected channel display settings and can also be buffered onto the page as tiles are selected, or across the entire image grid (not shown). d Gated cell population projected onto selected tile image with current display settings

Fig. 6

T cell CD3 (blue), CD4 (red), and CD8 (green) intensity. a First row of 5 images in 5 × 5 experiment grid. b Single tile image with corresponding cell and nucleus segmentation, where cells are defined as a fixed radius away from the nucleus in the absence of a plasma membrane stain. c Center zoom on (b) showing co-expression of CD3 and CD4 (magenta) and CD3 and CD8 (cyan) as well as debris in DAPI channel

Fig. 7

T cell gating workflow as annotated Explorer screenshots. a Morphological and intensity gates applied to isolate CD3+ cells. b Cytotoxic and helper cell subpopulations. c Individual cell images matched to subpopulations in (b)

Fig. 8

T cell population recovery comparison (notebook). a CD4/CD8 gating results, as determined by automatic gating functions in OpenCyto [25], over two imaging replicates for each of 4 donors. While all images were collected over a field of view of the same size, samples for donors 40 and 41 were prepared at 3x higher cell concentrations to demonstrate that segmentation and intensity measurements are robust to greater image object densities. b Cell population size for both replicates compared to a single flow cytometry measurement for each donor as well as Pearson correlation demonstrating strong agreement between the two (r > 0.99, P < 0.0001, two-tailed t-test). c Cell images from donor 41 showing (from left to right): DAPI (blue) and PHA (red) stain, DAPI and PHA with cell and nuclei segmentations, and DAPI with CD4 (red) as well as CD8 (green). See supplementary file Additional file 1: Figure S1 for a comparison of these results to those from the same workflow without much of the gating used here to remove invalid cells

Fig. 9

Cell diameter recovery workflow as annotated Explorer screenshots. a Unstimulated/naive cells with DAPI stain (gray), binary PHA image (red), and resulting cell segmentation (green). b Projection of modes in phalloidin/DAPI distribution to segmented cells in original image. c Diameter distribution comparison for unstimulated and activated samples along with corresponding single cell images

Fig. 10

T cell size recovery comparison (notebook) showing inferred cell diameter distributions (violin with bars indicating +/− 1 s.d.) vs point estimates of mean diameter from Thermo Fisher cell counter (dot) as well as filtered single cell population sizes (counts); All images taken at 20x magnification except where indicated otherwise. See supplementary file Additional file 2: Figure S2 for a comparison of these results to those from the same workflow without much of the gating used here to remove invalid cells

Fig. 11

BALB/c spleen (slide 1) CODEX image segmentation/quantification results. a ~ 200 cells with DRAQ5 nuclear stain (above) and corresponding cell/nucleus segmentation (below). b ~ 71 k cells in stitched, downscaled 9072 × 9408 pixel image showing IgD (green) and CD90 (red) expression as well as location of CD169+ marginal zone macrophages as white dots. c Double positive cell population rates post cleanup-gating with expected percentages in green