Picture the small ritual that happens thousands of times a night in nursing homes: the rattle of the medication trolley, a familiar face at the door, a hand on a shoulder, a pill or a patch, and then, twenty minutes later, a man who had been calling out all evening settles. Families watch this and reasonably conclude that the drug worked. Pharmacologists have mostly concluded the same thing. A study published in Neuron on April 16, 2026 suggests the accounting is more complicated than that, and it does so not by arguing but by taking the brain apart circuit by circuit.
Giulia Livrizzi, Janie Chang-Weinberg and colleagues reverse-translated a human placebo paradigm into mice. Over four days, male mice were conditioned with repeated pairings of morphine (10 mg/kg, injected into the abdomen) and a particular context. Tested later with no drug at all, the context alone produced what the authors quantified as 52.4 percent (plus or minus 10.0) of the increase in paw withdrawal latency that the morphine dose itself had produced. This was not a trick of training the animals to hold still: the conditioning only worked when morphine had been paired with an actual painful stimulus, which the authors read as evidence that the effect is driven by a pain-predictive process rather than a learned motor habit. Using viral tools and chemogenetics, they then showed that descending neurons in the ventrolateral periaqueductal gray, projecting to the rostroventromedial medulla, were necessary for both morphine analgesia and the placebo effect, that endogenous opioids released in that region gated the placebo relief, and that cortical input from medial prefrontal and anterior cingulate cortex was required for the placebo effect but not for morphine's own. The relief also generalised to forms of pain the animals had never been conditioned against, and pre-conditioning healthy animals left them more resilient to pain after a later injury.
That last set of results is what prompted Tor Wager, of Dartmouth, to write a Dispatch in Current Biology on July 20, 2026 under the title "Placebo effects: Rethinking the mechanisms of morphine analgesia". He calls the work "the most comprehensive, mechanistic account to date of how the brain learns to predict and mimic drug effects", and notes that human imaging has already found placebo-related opioid release in the periaqueductal gray, the anterior midcingulate cortex and the ventromedial prefrontal cortex. The uncomfortable implication is arithmetical. If a conditioned context can drive the brain's own opioids into the very brainstem hub that morphine needs, then every time a person takes morphine in a familiar room, from a familiar nurse, after years of learning that this particular ritual brings relief, the measured effect of the drug and the measured effect of expectation are running through overlapping hardware. In a mouse you can switch the cortex off and see the two come apart. In a clinical trial you cannot.
For readers of this paper, the place where that arithmetic bites hardest is the treatment of pain in advanced dementia, where the person cannot reliably report what hurts and clinicians read agitation as a proxy. The landmark trial here remains Bettina Husebo and colleagues' cluster randomised study in The BMJ, published July 15, 2011: 352 residents with moderate to severe dementia and clinically significant behavioural disturbance, across 60 nursing home units in 18 Norwegian homes, randomised to eight weeks of a stepwise individual pain treatment protocol (paracetamol first, then extended release morphine, buprenorphine patch or pregabalin) or to usual treatment and care. Agitation fell in the treated group, and the trial changed practice for the better by making untreated pain visible. But a design that compares active treatment against usual care, as that trial did, cannot separate the pharmacology of an opioid from the daily attention, the examination, the ritual and the staff expectation that come bundled with it. That is not a criticism of the trial; it is a description of what any unblinded comparison can and cannot resolve, and the mouse circuitry now gives a biological reason to take the distinction seriously rather than treat it as a statistical nuisance.
There is a second, sharper question that the Neuron findings raise rather than answer. If the expectation-driven share of relief depends on prefrontal and cingulate input to the brainstem, while morphine's direct effect does not, then what happens to that share in a brain where exactly those frontal regions are damaged? Alzheimer's disease, frontotemporal dementia and vascular disease all take their toll on those circuits, in different patterns and at different speeds. The mice cannot tell us. What they can tell us is that the question is a mechanistic one with a testable answer, not a philosophical aside, and that it applies to the interpretation of every analgesic and anti-agitation trial run in this population.
None of this makes anyone's relief less real. A context that triggers opioid release in the periaqueductal gray is not a placebo in the dismissive sense; in these animals it was measurable pain modulation with a traceable address in the brainstem. If anything, the finding argues that the room, the routine and the person delivering care are part of the treatment rather than the packaging around it. What it does not do is tell a family or a clinician anything about doses, about whether a given person's morphine is working, or about whether to change a single thing on a medication list.