A computer model of damaged brain cells offers an explanation for why the sleeping pill zolpidem can briefly wake a small number of patients with brain injuries, researchers reported in the Proceedings of the National Academy of Sciences.

A 2017 review estimated that only 5 to 7 percent of patients with disorders of consciousness respond to the drug, which is designed to bring on sleep. The puzzle dates to a 2000 case report in the South African Medical Journal describing a young man in a semi-comatose state who woke for three to four hours after being given zolpidem.

Francesco Achilli, Antoine Cautru, Jean-Pierre Changeux and Guillaume Dumas, working across institutions in Milan, Montreal and Paris, built a model of 149 simulated neurons and damaged it deliberately. Their results suggest that in some injured brains the cells that act as brakes fail, and zolpidem helps by strengthening the brakes that remain.

Most neurons are excitatory, prompting other cells to fire, while the rest are inhibitory, restraining activity. According to the study, “a reduction in excitatory tone was traditionally considered the sole mechanism” behind disorders of consciousness. Because zolpidem strengthens inhibition, it should by that logic deepen unresponsiveness, which is why its effect was called a paradox.

One earlier answer, Nicholas Schiff’s mesocircuit hypothesis, traces some cases to a failed brake that lets a second brake suppress a deep brain loop involved in wakefulness. Another holds that some injured brains tip into activity that spreads too easily. Both involve too little braking, and the authors read them as two descriptions of one failure.

The team expanded a model Changeux and Dumas had helped build in 2022, drawing on the global neuronal workspace theory, which holds that information becomes conscious when it is broadcast and sustained across a wide network. The network performs a simple memory task, carrying a number across a brief silence, as a stand-in for conscious access. A trained network answers correctly about 94 percent of the time; a badly damaged one drifts toward 50 percent, the level of guessing.

Tests simulated clustered damage, like a stroke, and scattered damage, like traumatic brain injury. The engine cells proved the more fragile: losing 7 percent of them from one spot cut the score to about 83 percent, while losing 43 percent of the brake cells from one spot had the same effect. Because brake cells reach across the whole network, “a few evenly distributed inhibitory neurons can preserve this regulatory function,” according to the study.

To mimic zolpidem, the researchers strengthened every connection leaving a surviving brake cell, by 5 to 100 percent. With scattered damage that removed 86 percent of brake cells, a 70 percent boost lifted the score from near guessing to about 73. With 71 percent of the brakes lost in one spot, a 10 percent boost brought the score to about 68. Neither network returned to 94.

With 96 percent of the brakes gone, only one brake cell remained and no dose helped; after damage to the engine cells or to both cell types, the boost produced no recovery. Scores peaked when the spread of activity settled slightly above its pre-injury level. The result suggests “successful network recovery relies on tuning the network into a precise dynamical window,” the authors wrote.

If the model is right, zolpidem helps so few patients because it works only in a brain short on braking with enough braking left to strengthen, depending on “the pattern and severity of inhibitory network damage.” Several patients who responded in case reports had damage in the deep brain loop pointed to by the mesocircuit hypothesis, and other brake-strengthening drugs such as the anesthetic propofol are being studied for similar effects.

The authors suggest brain-wave recordings could one day reveal runaway activity and scans could map where the brakes were hit, helping doctors identify likely responders before giving the drug.

They caution that the model is small, lacks the deep brain loop, and scores a network holding a number rather than a patient waking. They offer the simulations as “tools to test mechanistic hypotheses” that remain unchecked in patients.