The Opioid Built to Dodge Overdose Looks, in Primates, Like Buprenorphine

SR-17018 arrived in the pharmacology literature in 2017 with a number attached to it, and the number was the story. Its calculated bias factor — the ratio of G protein signaling to beta-arrestin-2 recruitment at the mu opioid receptor — was 102, the highest reported for any ligand of that receptor. In mice it produced morphine-level analgesia, crossed into the brain, and left breathing largely intact. If signaling bias was the lever that separated pain relief from overdose, this was the molecule holding it.

Nine years and one species later, the lever looks shorter than advertised. In a study published November 11 in Frontiers in Drug Discovery, a team from Wake Forest University School of Medicine and the University of Pittsburgh put SR-17018 into rhesus macaques for the first time and compared it against morphine, heroin, buprenorphine and the high-efficacy research agonist DAMGO across four assays at once: pain response, itch, voluntary drug-taking, and continuous respiratory monitoring. The drug behaved, in nearly every measure, like buprenorphine — a partial agonist that has been on pharmacy shelves since 2002.

That is not a failure. It is a different kind of finding than the one the field was promised, and it lands in the middle of the most consequential argument in addiction medicine: whether better molecules will solve an overdose crisis that is currently being addressed with molecules that already exist.

What the monkeys were actually asked to do

Twenty-four adult rhesus macaques, ten to eighteen years old and drawn from the national primate centers, were tested in a within-subject, counterbalanced design, so each animal served as its own comparison across drugs. The team ran four assays. In the warm-water tail-withdrawal test, a monkey pulls its tail from 46°C water, and the delay before it does so measures antinociception. In the capsaicin assay, topical capsaicin drops withdrawal latencies from a twenty-second cutoff to two or three seconds, a chemically induced hypersensitivity that opioids reverse. Scratching was counted from fifteen-minute video sessions by observers blinded to treatment, because spinal opioid-induced itch is a reliable behavioral readout of how hard a drug is pushing the receptor. Voluntary drug-taking was measured under a progressive-ratio schedule, in which each successive infusion costs the animal more lever presses — a standard proxy for how much reinforcement a drug delivers. And implanted telemetry transmitters recorded respiratory rate, minute ventilation, heart rhythm and body temperature continuously for two hours after dosing.

That four-way design is the reason this paper matters more than a single-endpoint rodent study. It measures analgesia, the signature side effect, abuse liability and the specific physiological event that kills people, in the same animals, in the same week.

The potency nobody put in the headline

Given intrathecally, straight to the spinal cord where pharmacokinetics cannot muddy the comparison, SR-17018 did produce dose-dependent reversal of capsaicin allodynia. It simply needed a great deal more drug to do it. The ED50 values tell the story in micrograms:

Drug	Intrathecal ED50 (95% CI)
DAMGO	0.8 µg (0.4–1.4)
Buprenorphine	5.2 µg (4.7–5.9)
Morphine	11.0 µg (6.8–17.9)
SR-17018	83.2 µg (49.5–140.0)

Roughly eight times morphine and sixteen times buprenorphine. Subcutaneously, 1 mg/kg reached near-maximal antinociception within half an hour, held for about three hours and returned to baseline by five or six — an ordinary duration for a short-acting opioid, achieved at a dose that would be unremarkable if the compound were being sold on potency.

Low potency is not disqualifying in itself. It is a warning that the molecule is not doing much at the receptor, which is exactly what the rest of the data confirmed.

Itch was the tell

Spinal opioids make people scratch, and the intensity of that scratching tracks intrinsic efficacy closely. DAMGO, a full agonist, produced robust scratching that persisted through the three-hour session. Morphine produced moderate, statistically significant scratching. SR-17018 at 300 micrograms — a dose that had already produced maximal antiallodynia — produced scratching indistinguishable from the vehicle control, in both male and female animals. Buprenorphine, at its much smaller doses, looked the same.

The paper's authors read that pattern correctly and without spin. A drug that cannot generate a side effect requiring high receptor activation is not a high-efficacy agonist, whatever its in vitro signaling profile says.

The reinforcing effect that survived

Here the compound's advocates lose their cleanest claim. Under the progressive-ratio schedule, heroin generated the strongest responding, with the highest dose approaching the remifentanil positive control. SR-17018 was self-administered across its full tested range, from 3 to 30 micrograms per kilogram per infusion, at rates clearly above saline — and at rates statistically comparable to buprenorphine.

So SR-17018 is not non-addictive. It is less reinforcing than heroin and about as reinforcing as the medication used to treat heroin addiction. For a drug that was discussed in the preclinical literature as a way to keep analgesia while discarding abuse liability, "as abusable as buprenorphine" is a real and measurable attenuation, and also a ceiling on the claim.

Breathing held, for a reason that undercuts the marketing

Morphine at 1.8 mg/kg suppressed respiratory rate and minute ventilation across the full 120-minute observation window. Buprenorphine at 0.1 to 0.3 mg/kg did not significantly change either measure. SR-17018 did not either — at its antinociceptive dose of 1 mg/kg, or at three times that dose.

This is the finding that will get quoted, and it deserves the quote. A mu opioid agonist that produces full antinociception in a primate without depressing ventilation is genuinely notable. What the paper argues, though, is that the mechanism is not the one the molecule was designed around. The authors conclude that SR-17018 behaves in vivo as a low-efficacy agonist, and that reduced intrinsic efficacy — not the 102-fold bias factor — explains both the preserved breathing and the modest reinforcement. Bias and efficacy are different properties. A molecule can be exquisitely biased in a cell assay and still be a weak agonist at the receptor, and it is the weakness that the animal notices.

A graveyard the field keeps revisiting

SR-17018 is the third high-profile entrant in this lineage and the third to shrink on contact with a primate.

Oliceridine, formerly TRV130, looked in rodents like morphine's equal without the respiratory penalty. In humans it was non-inferior for analgesia with similar subjective central nervous system effects and no significant difference in respiratory depression at analgesic doses — and it retained abuse liability. PZM21, designed computationally, showed an improved side-effect profile in mice, then significant respiratory depression in follow-up mouse work, and in primates abuse liability and itch indistinguishable from conventional opioids. Neither compound had a quantitative bias factor calculated at the time it was developed, which is how a field ends up optimizing a ratio it never measured.

What bias factor measured, and what it missed

The beta-arrestin hypothesis came from knockout mice that kept morphine analgesia while losing tolerance and several side effects. That was a real signal. The trouble is that beta-arrestin-2 deletion removes a protein entirely, while a biased ligand only leans the receptor's signaling one way — and the downstream arithmetic is not the same. Nine years of ligands later, the cleanest summary of the primate data is that intrinsic efficacy has done more explanatory work than bias, and that rodent promise has repeatedly failed to survive the translation.

The authors are explicit that their result does not close the program. Mixed mu/nociceptin receptor agonists such as cebranopadol are in late-stage development on a different logic, and work continues on downstream targets including GIRK channel signaling. Their own sentence on the field's status is the one to keep: a compound that fully dissociates potent analgesia from abuse liability and respiratory depression, in primates or in humans, remains elusive.

What this means for treatment

There is an uncomfortable practical implication in a paper like this, and it belongs in the article rather than in a footnote. The drug that SR-17018 most closely resembles is already approved, already generic, already on the World Health Organization's list of essential medicines, and still drastically under-prescribed relative to the need it meets. For someone with opioid use disorder, buprenorphine's ceiling on respiratory depression is not a research finding waiting on a better molecule — it is the clinical fact that makes take-home dosing safe. The barrier has never been pharmacology. It has been prescriber limits, prior authorization, pharmacy stocking, and the fact that a partial agonist requires a patient to be in mild withdrawal before the first dose.

A molecule that delivered the same safety profile with full-agonist analgesia would be a genuine advance for surgical and cancer pain, where partial agonists have real ceilings on effect. That is what this line of research is for, and it is worth pursuing. What the primate data argues against is the framing that has attached itself to every biased agonist since 2016 — that the overdose crisis is a drug-design problem that a cleverer ligand will solve. Roughly 120 million opioid prescriptions were dispensed through U.S. pharmacies in 2024, when about 4.8 million Americans met criteria for opioid use disorder and overdose killed around 75,000 people. Those numbers are not waiting on SR-17018.

The study's limits are the ordinary ones for primate pharmacology and should be stated plainly: twenty-four animals, four or more per assay, a single species, acute dosing without a tolerance or withdrawal phase, and doses that do not translate into human milligrams. It cannot tell anyone what SR-17018 would do in a patient over six months. It can tell them what nine years of rodent data could not — that the highest bias factor ever reported produced, in the closest available model to a human, the efficacy profile of medication-assisted treatment rather than the profile of a drug that ends the overdose era.

For the field, that is a useful correction. For anyone waiting on a non-addictive opioid, it is a reminder that the ceiling on this generation of molecules is a property of the receptor, and the receptor has been studied for fifty years.