Predictive Cellular Signatures from Live Human Motor Neurons Distinguish TDP-43 ALS and Enable ALS Subtype...

Area of Science:

• Neuroscience
• Computational Biology
• Genetics

Background:

• Amyotrophic lateral sclerosis (ALS) is a fatal neurodegenerative disease with unknown causes in 90% of cases.
• TDP-43 protein pathology is a common hallmark, suggesting shared mechanisms in genetic and sporadic ALS (sALS).
• ALS heterogeneity necessitates understanding individual mechanisms for effective subgroup-specific therapies.

Purpose of the Study:

• To systematically decode cellular signatures of ALS using machine learning (ML).
• To identify discriminating cellular signals and validate their association with ALS pathology.
• To explore early pathogenic events and neurodevelopmental changes preceding motor neuron degeneration.

Main Methods:

• High-content imaging of human induced pluripotent stem cell-derived motor neurons (iMNs) from ALS patients and gene-edited lines.
• Training shallow connected ML algorithms (SMLs) and deep convolutional neural networks (DNNs) to distinguish mutant and control iMNs.
• Employing explainability methods to identify key cellular features and a time-interaction ML model for dynamic analysis.

Main Results:

• ML models accurately distinguished mutant and control iMNs, with nuclear area signals being most discriminative.
• TDP-43 mutant iMNs showed altered nucleocytoplasmic shuttling and cellular integrity.
• ML uncovered dynamic morphological transitions preceding degeneration and identified overlapping/distinct signatures for C9orf72 and sALS iMNs.

Conclusions:

• ML-driven phenotypic profiling is a powerful tool for stratifying ALS patients and disentangling disease heterogeneity.
• The study highlights nuclear alterations and early dynamic changes as key features in ALS pathogenesis.
• This scalable ML approach offers a paradigm for uncovering early disease mechanisms in ALS and other neurodegenerative disorders.