MCMICRO: a scalable, modular image-processing pipeline for multiplexed tissue imaging

Abstract

Highly multiplexed tissue imaging makes detailed molecular analysis of single cells possible in a preserved spatial context. However, reproducible analysis of large multichannel images poses a substantial computational challenge. Here, we describe a modular and open-source computational pipeline, MCMICRO, for performing the sequential steps needed to transform whole-slide images into single-cell data. We demonstrate the use of MCMICRO on tissue and tumor images acquired using multiple imaging platforms, thereby providing a solid foundation for the continued development of tissue imaging software.

Fig. 1. Overview of the MCMICRO pipeline and key data types.

Modules highlighted in bold red were developed and/or containerized as part of this study. a, A schematic representation of a canonical workflow for end-to-end image processing of multiplexed whole-slide and TMA data using MCMICRO. Shown is a flow of inputs (pink rectangles) from imaging instruments (yellow rectangles) through image-processing steps (white rectangles) that are implemented in software modules (puzzle pieces) to produce key data types (green rectangles). Data flows associated with a whole-slide image and TMA are represented with black and red arrows, respectively. b–e, Highlights of individual software modules incorporated into MCMICRO. b, ASHLAR is used to stitch and register individual CyCIF image tiles with subcellular accuracy (yellow zoom-in). This panel depicts a whole-slide, 484 tile (22 × 22) mosaic t-CyCIF image of a human colorectal cancer in four channels: Hoechst 33342-stained nuclear DNA (blue) and antibody staining against α-smooth muscle actin (α-SMA; red), the Ki-67 proliferation marker (green) and cytokeratin (white). c, Two different segmentation masks computed by UnMICST (blue) and Ilastik (red) overlaid on an image of nuclei from an EMIT TMA core (single experiment). d, SCIMAP enables single-cell clustering, neighborhood analysis and cell-type assignment on the basis of patterns of marker expression. e, A CyCIF image of an EMIT TMA dearrayed using Coreograph to identify individual cores, which are subsequently extracted and analyzed in high resolution. Below, a five-color image of a single lung adenocarcinoma core is shown for channels corresponding to Hoechst 33342-stained DNA (white), cytokeratin (orange), the immune-cell marker CD45 (green), α-SMA (magenta) and Ki-67 (red).

Fig. 2. Exemplary whole-slide images processed using MCMICRO.

a, Selected channels are shown from three exemplary high-plex images from colorectal cancer using CODEX, MxIF and CyCIF; pan-cytokeratin is depicted in blue; CD4 in yellow; CD8 in green and CD20 in red. b, Selected channels are shown from a high-plex image of human tonsil using mIHC with Hoechst 33342 stain in blue (representing DNA), CD20 in orange, keratin in green and CD8 in red. c, Upper panels: selected fields of view from mIHC, CODEX and CyCIF images of the tonsil specimen shown in b (the selected field is highlighted by the red rectangle). Lower panels: centroids for the single-cell segmentation mask for the three fields of view shown above and colored by marker expression to identify cell types. Epithelial cells of the tonsil mucosa stain positive for pan-cytokeratin (green), cytotoxic T cells stain positive for CD8 (red) and B cells stain positive for CD20 (blue). d, Schematic of the comparative analysis of a single tonsil specimen and two tSNE plots for CODEX, CyCIF and mIHC data (shown in c) demonstrating clustering by marker expression (left) but not imaging technology (right).