Characterizing artifacts in multiplex images of tissue.
Pathologists have historically relied on qualitative descriptions of multicellular structures and individual cell states in healthy and diseased human tissue. Now prodigious technologies for multiplex, antibody-based tissue imaging like CODEX1, CyCIF2,3, and mIHC4 enable unprecedented quantitative spatial analysis of expressed protein in two- and three-dimensional tissue specimens. Such analysis requires computational algorithms capable of learning complex features among millions of cells labeled with dozens to hundreds of antibodies, far surpassing what can be discerned by eye.
While a skilled pathologist can effectively identify and disregard artifacts in histology images, computer algorithms make no such distinction. Thus, while computational analyses are crucial for uncovering complex histological patterns, they are susceptible to visual and image processing artifacts.
In our Nature Methods Article, we provide a detailed account of recurrent artifacts in multiplex immunofluorescence (IF) images of tissue and demonstrate their detrimental impact on derived single-cell data (Fig. 1). Our findings highlight the pervasive nature of artifacts in multiplex tissue imaging and underscore the need to account for them to ensure accurate biological interpretation.
Fig. 1 | Microscopy artifacts and their impact on tissue-derived, single-cell data. (a) Examples of recurrent visual and image processing artifacts in multiplex IF images of tissue. (b) Illumination aberration in the pCREB channel of an IF image of liposarcoma demonstrating that artifactual per cell signals can reach an order of magnitude above background. (c) Artifacts in the CD163 channel of an IF image of colorectal tissue. Shown at right are UMAP embeddings of 21-channel, single-cell data derived from the image at left colored by CD163 (left UMAP) and inclusion in one of the ROIs (right UMAP) demonstrating that cells affected by bright artifacts form discrete clusters in feature space.
An interactive quality control tool for multiplex microscopy.
While quality control (QC) is well established in other single-cell workflows such as scRNA-seq, QC tools for quantitative multiplex tissue imaging are lacking. Accurate study of tissue-derived, single-cell data requires QC methods capable of addressing microscopy artifacts arising as early as tissue slide mounting and as late as cell segmentation (i.e., the process of identifying individual cells in a histology image). To tackle QC challenges in the area of multiplex bioimaging, we have developed CyLinter, an interactive software program designed to identify and eliminate cells affected by a wide range of visual and image processing artifacts through computer-assisted human review. CyLinter represents one of the first QC tools to integrate visual review of pixel-level images with derived single-cell data.
As a plugin for the popular Napari multi-channel image viewer5, CyLinter comprises multiple software modules that facilitate QC analysis of raw single-cell data (Fig. 2a). The pipeline is optimized for batch image processing and supports progress bookmarking within and across modules. CyLinter takes four input files for each tissue specimen in an analysis: (1) a stitched and registered multiplex image (TIFF/OME-TIF), (2) a cell identification mask generated by a segmentation algorithm, (3) a binary image showing the boundaries between segmented cells and (4) a spatial feature table6 in comma-separated values (CSV) format comprising the location and computed signal intensities for each cell (Fig. 2b-e).
Fig. 2 | CyLinter workflow and input data files. (a) Schematic representation of the CyLinter workflow with QC modules colored by type; the program returns as output a multi-specimen data table free of noisy cells. (b-e) Required CyLinter input files (per specimen).
To enhance processing efficiency, CyLinter provides automated region of interest (ROI) curation and intelligent data thresholding using Gaussian mixture modeling. We also introduce a proof-of-concept deep learning model for automated artifact detection for integration into future versions of the program as part of our effort to advance QC scalability as we move towards cloud-based infrastructures.
CyLinter recovers biologically relevant cell types from noisy microscopy data.
In our study, we applied CyLinter to seven different multiplex imaging datasets including a cohort of 25 triple-negative breast cancer (TNBC) tissue specimens from patients enrolled in the TOPACIO clinical trial of “Niraparib in Combination with Pembrolizumab in Patients with Triple-negative Breast Cancer or Ovarian Cancer” (NCT02657889)7. Our results demonstrate that CyLinter effectively salvages otherwise uninterpretable clinical imaging data severely affected by microscopy artifacts (Fig. 3). These findings underscore the importance of integrating artifact removal as a standard practice in multiplex bioimage analysis and advocate for the inclusion of QC reports (such as those generated by CyLinter) with all publicly deposited datasets to promote scientific transparency and reproducibility.
Fig. 3 | Unsupervised clustering analysis of cells in 25 TNBC tissue specimens from patients enrolled in the TOPACIO clinical trial before and after CyLinter QC. After artifact removal, CyLinter reveals 43 bona fide cell states (left) among 492 noisy pre-QC clusters (right). Shown for each clustering are UMAP embeddings of 3.0x106 cells colored by HDBSCAN cluster (a, d), clustered heatmaps of mean marker intensity per cluster (b, e), and image galleries showing example cells for three clusters highlighted in their respective UMAP embeddings (c, f).
Open-source software for the spatial biology community.
Open-access science is critical to advancing experimental and translational research. In support of this principle, we provide CyLinter as a freely accessible tool. The program is designed to handle various types of multiplex imaging data and is compatible with Mac, PC, and Linux operating systems. CyLinter is version controlled with Git/GitHub (https://github.com/labsyspharm/cylinter)8 and can be installed via the Anaconda package manager using instructions provided at the tool's dedicated project website (https://labsyspharm.github.io/cylinter).
CyLinter has significantly enhanced our ability to gain new insights into spatial biology; we hope that it will provide similar benefits for your research!
References
- Goltsev, Y. et al. Deep profiling of mouse splenic architecture with CODEX multiplexed imaging. Cell 174, 968–981.e15 (2018).
- Lin, J.-R. et al. Highly multiplexed immunofluorescence imaging of human tissues and tumors using t-CyCIF and conventional optical microscopes. eLife 7, e31657 (2018).
- Lin, J.-R., Fallahi-Sichani, M. & Sorger, P. K. Highly multiplexed imaging of single cells using a high-throughput cyclic immunofluorescence method. Commun. 6, 8390 (2015).
- Tsujikawa, T. et al. Quantitative multiplex immunohistochemistry reveals myeloid-inflamed tumor-immune complexity associated with poor prognosis. Cell Rep. 19, 203–217 (2017).
- Chiu, C.-L., Clack, N. & Community, T. N. napari: a Python multi-dimensional image viewer platform for the research community. Microanal. 28, 1576–1577 (2022).
- Schapiro, D. et al. MITI minimum information guidelines for highly multiplexed tissue images. Methods 19, 262–267 (2022).
- Vinayak, S. et al. Open-label clinical trial of niraparib combined with pembrolizumab for treatment of advanced or metastatic triple-negative breast cancer. JAMA Oncol. 5, 1132–1140 (2019).
- Baker, G. CyLinter. Zenodo https://doi.org/10.5281/ZENODO.7186909 (2021).
Please sign in or register for FREE
If you are a registered user on Research Communities by Springer Nature, please sign in