Predicting amyotrophic lateral sclerosis in the pre-symptomatic phase: Insights from SOD1G93A mouse gene expression profiles

Abstract

Amyotrophic lateral sclerosis (ALS) is a fast-paced fatal disease that requires immediate intervention to slow down the course of pathology and improve patients’ quality of life. However, in most cases, ALS is diagnosed too late. For this reason, an accurate diagnostic test is urgently needed to identify ALS patients early, enabling a timely introduction of novel therapeutics and effective monitoring of disease progression. To address this significant unmet medical need, we explored a transcriptome-based signature to predict ALS during the preclinical phase. Using publicly available gene expression profiles from central nervous system (lumbar isolated motor neurons and spinal cord homogenates) of transgenic SOD1G93A mice with different genetic background and their respective control littermates, covering pre-symptomatic to late stages of the disease, we identified 463 differentially expressed genes (DEGs), primarily involved in immune response and metabolic processes. Based on this ALS gene-associated signature, we tested three machine learning binary classifiers (Support Vector Machine, Neural Network and Linear Discriminant Analysis), which demonstrated highly significant predictive power in discriminating mutant SOD1G93A from controls mice, even at pre-symptomatic stages. This was evident in both the discovery cohort and in two additional peripheral cross-tissue validation datasets from preclinical SOD1G93A sciatic nerve and muscles. Our study provides the first proof of concept for early ALS detection using a machine learning-based transcriptomic classifier. This could lead to earlier diagnosis, potentially enabling effective monitoring of disease progression and earlier interventions.