AI microscopy can tell ALS neurons from healthy ones, in the lab

The first surprise came from the endoplasmic reticulum, a network of membranes inside every cell. Craig Blackstone of Harvard Medical School and Mass General Brigham showed that parts of it once drawn as flat sheets are in fact networks of tiny tubes, constantly changing shape. Those are the kinds of details that matter in hereditary spastic paraplegias and other disorders linked to protein mutations. "When we can see how structures inside neurons are organized and how they change dynamically in disease, we can begin to connect a genetic or molecular abnormality to what is happening inside the cell," Blackstone said. Neurons are unusually large, highly specialised cells that must keep their internal architecture working over long distances and for a lifetime, which makes such problems hard to catch early.

Live imaging adds the time dimension. Erika Holzbaur of the University of Pennsylvania described watching mitochondria, the parts of the cell that supply its energy, and autophagy, the process by which a cell removes damaged mitochondria and other waste. Her team's work on fragile X syndrome found that without the FMRP protein, mitochondria malfunctioned in both mouse and human neurons, pointing to possible new directions for treatment. "By the time a neuron dies in a patient with ALS, Parkinson's or Huntington's disease, it is too late to save it," Holzbaur said. The aim is to catch the first signs of stress.

ALS, amyotrophic lateral sclerosis, is notoriously hard to diagnose because no definitive test exists. Patients may spend close to a year seeing doctors and specialists before the disease is confirmed, and that delay costs time when the few available treatments and clinical trials are most useful early on. Robotic microscopy paired with AI could eventually offer a faster and more objective way to support diagnosis. These instruments track thousands of individual motor neurons over time, and the AI analyses the images for patterns too subtle for manual review, then designs and runs experiments to decide what to test next.

For patients, the gain would be time: earlier, more reliable signals could bring people to treatment and trials sooner. For researchers, it means a way to test hypotheses on cells taken from patients themselves. Those benefits remain conditional. The AI result comes from laboratory cells, the fragile X findings from mouse and human neurons, and no timeline for use in clinics has been announced. The next step, which the reporting does not describe, is showing that the same patterns hold in larger groups of patients.