A simple blood test may soon help doctors diagnose amyotrophic lateral sclerosis (ALS) more quickly, potentially transforming care for patients facing one of medicine’s most devastating neurodegenerative diseases. Researchers at the University of Michigan have developed a gene-based test that identified ALS patients with nearly 90% accuracy, even when applied to completely independent patient groups.
ALS is notoriously difficult to diagnose, often taking 5 to 15 months after symptoms appear, with some patients waiting up to 19 months for a definitive answer. During this period, the disease progresses, patients see multiple specialists, and misdiagnoses are common. Early diagnosis is critical because existing treatments are more effective when started sooner, and many clinical trials exclude patients with advanced disease.
Gene Activity Tested in Blood Samples
The study, published in Nature Communications, analyzed blood samples from 422 ALS patients and 272 healthy controls collected between 2011 and 2021 at Michigan’s Pranger ALS Clinic. Using advanced RNA sequencing, researchers measured the activity of more than 22,000 genes simultaneously, creating a detailed molecular profile of each participant’s blood cells.
Patients represented typical ALS demographics, with a median age of 65, 58% male, and 87% reporting no family history. All major ALS subtypes were included: 26% with bulbar onset, 32% with cervical onset, and 39% with lumbar onset. Functional impairment varied, with a median ALS rating scale score of 37 out of 48.
Analysis revealed 3,640 genes with significant expression differences between ALS patients and controls—1,999 more active and 1,641 less active in ALS patients. Many were related to immune function, supporting evidence that immune dysregulation contributes to disease progression. Others affected cellular waste recycling, muscle maintenance, intracellular transport, and programmed cell death.
Machine Learning Improves Accuracy
Seven machine learning models were trained to recognize ALS patterns in gene expression data. The best performer, XGBoost, achieved an area under the curve of 0.91, indicating high diagnostic accuracy. To make the test clinically feasible, researchers distilled thousands of genes into smaller panels of 27, 29, 30, and a combined 46 genes. These streamlined panels performed consistently well, with 91% accuracy, sensitivities up to 94%, and specificities around 87%.
Validation on an independent dataset of 86 ALS patients and 46 controls confirmed the results. The 46-gene panel achieved an area under the curve of 0.894, demonstrating robustness across patient populations. Earlier blood-based ALS tests had much lower accuracy in independent trials, around 63%, far below clinical standards.
Predicting Survival and Understanding Disease Biology
Beyond diagnosis, combining gene expression with clinical variables—age at onset, sex, and symptom location—helped predict patient survival, separating individuals into short, intermediate, and long survival groups. This method outperformed models using only clinical data, potentially offering patients clearer expectations about disease progression in the future.
The team also identified biological pathways most disrupted in ALS, including oxidative phosphorylation, protein processing, and pathways shared with Parkinson’s and Huntington’s diseases. Core genes altered in both blood and affected nerve tissue were highlighted, providing targets for computational drug screening. This analysis suggested eight compounds, including FDA-approved drugs for other conditions, that may reverse ALS-related gene expression changes—though laboratory testing is still required.
Advantages Over Single-Biomarker Approaches
ALS patients typically survive only 2 to 4 years after diagnosis. Delays in diagnosis reduce opportunities for treatment and trial participation. Unlike single biomarkers such as neurofilament light chain, which can rise in multiple neurological conditions, gene panels may offer greater specificity for ALS.
The technology to bring these findings into clinical practice already exists. Similar gene expression panels are FDA-approved for cancer subtype classification, demonstrating regulatory and commercial feasibility. The study’s large sample size, advanced RNA sequencing, and rigorous machine learning methods overcome the limitations of previous ALS blood tests, suggesting readiness for the next phase: testing against ALS-mimicking conditions and presymptomatic carriers of ALS-related mutations.
Related topic：
