Mapping human microglial morphological diversity via handcrafted and deep learning-derived image features

Area of Science:

• Neuroscience
• Immunology
• Cell Biology

Background:

• Microglia, the brain's immune cells, exhibit diverse activation states crucial for neural health and disease.
• Characterizing microglial heterogeneity at scale is essential but technically challenging.
• Existing methods struggle to capture the continuous spectrum of microglial phenotypes.

Purpose of the Study:

• To develop and validate a scalable imaging and analysis framework for high-resolution mapping of human induced pluripotent stem cell-derived microglia (iMG) activation states.
• To compare different feature extraction and classification methods for quantifying microglial heterogeneity.
• To identify informative features for predicting biological readouts like inflammasome activation.

Main Methods:

• High-content imaging combining immunofluorescence (IF) targeting NF-κB, ASC, and CD45 with Cell Painting (CP) assay.
• Quantification of cell phenotypes using handcrafted and representation-learning features.
• Classification of microglial states using Gaussian Mixture Models (GMMs) for probabilistic assignments and comparison with graph-based methods (Leiden).

Main Results:

• The developed framework successfully mapped iMG activation landscapes at single-cell resolution.
• Deep-learning features from the IF panel provided the most informative classification, correlating strongly with biological readouts.
• Gaussian Mixture Models (GMMs) offered interpretable descriptions of microglial heterogeneity, comparable to graph-based methods.
• NLRP3 inflammasome activation was among the biological readouts effectively correlated with quantified microglial states.

Conclusions:

• The developed platform offers a scalable approach to quantify complex microglial activation states.
• This framework facilitates the discovery of novel compounds that modulate microglial phenotypes for therapeutic applications.
• The study highlights the utility of deep-learning features and GMMs for characterizing cellular heterogeneity in neuroscience research.