Code and data for validation of CyCIF multiplex imaging platform
Multiplex imaging technologies are increasingly used for single-cell phenotyping and spatial characterization of tissues; however, quantitative, reproducible analysis is a technical and computational challenge. We developed an open-source python-based image analysis tool, mplex-image, to achieve fully-reproducible multiplex image visualization and analysis. We deploy this tool in the accompanying Jupyter notebooks to validate specificity, sensitivity, reproducibility and normalization of the multiplex imaging platform cyclic immunofluorescence (CyCIF).
Through our work, we learned general principles of antibody staining performance, signal removal and background removal, and developed new methods, summarized below:
1. Signal removal using hydrogen peroxide
- Inceased concentrations above 3% hydrogen peroxide do not improve speed of signal removal.
- Increased incubation times of improve signal removal somewhat, but do not result in complete signal removal.
- Increased heat of quenching solution results in complete signal removal but must be balanced with increased tissue loss.
2. Background autofluorescence removal
- A single quenching of 3% H2O2 applied for 15 - 30 minutes dramatically reduced tissue autofluorescence and this "pre-quenching" step should be performed before staining.
- Additional rounds of quenching decrease the bright autofluorescent cells linearly, while most cells show little additional decrease.
- Therefore, we recommend taking a blank image after pre-quenching and and blank image after all staining has completed, and combining these two images (weighted by round) for autofluorescence subtraction.
3. Antibody staining optimization and reproducibility
- CyCIF method shows differences in dynamic range, but similar signal-to-background as standard IF.
- Antibodies applied ealier in the panel are brighter, some later antibodies can show decreased signal, while others show no change.
- Antibodies applied earlier may show non-specific staining; these non-specific staining artifacts are abrogated by applying antibodies later.
- Antibodies in adjacent channels can show channel bleed-through, so adjacent channel antibodies must be matched in signal intensity
- Technical replicates show differences in dynamic range, but replicates show similar signal-to-background ratios.
- Given a stable signal-to-background ratio, batch normalization can be performed by dividing each marker's signal by its background. We show this improves unsupervised cluster analysis
4. Methods
- Images published in the literature are not reproducible without documentation of ROI selection, display range, and scale; we have developed code for reproducible image visualization.
- Similarly, we have code for reproducible multi-color overlays
- Finally, we share our image processing pipeline for QC, registration, segmentation and cell-type calling.
