ATX-304

AMPK is the enzyme that senses a cell's energy shortage. Exercise and calorie restriction both switch it on. O304 was developed by the Swedish company Betagenon to activate it pharmacologically and is now called ATX-304. A 2025 paper describes it as "ATX-304 (formerly O304)", with the same CAS number, 1261289-04-6 [1]. In diet-induced obese mice it raised glucose uptake into skeletal muscle, reduced stress on insulin-producing β cells and let them rest [2].

It then went into a proof-of-concept phase IIa trial in people with type 2 diabetes who were already taking metformin. It lowered fasting plasma glucose and HOMA-IR, a measure of insulin resistance. It also improved peripheral microvascular perfusion and lowered blood pressure [2]. In aged mice it prevented and reversed age-related insulin resistance and improved cardiac function and exercise capacity [8].

A second human trial followed. In a phase 1b study, 23 adults with obesity and prediabetes were randomized 2:1 to 400 mg of ATX-304 once a day or placebo for 8 weeks, with an optional open-label extension to 16 weeks [3]. The results were shown as a poster at the American Diabetes Association meeting in June 2026 and summarized in a press release from Cambrian Bio, the company now developing the drug, which describes the trial as double-blind [4]. No full paper has been published. The EU registry record, sponsored by Betagenon, describes a two-part study open to healthy volunteers as well as patients, run in the Netherlands and Germany from September 2023 to May 2025, and holds no results [9]. So what follows is the developer's own account, and it covers only the 23 people with obesity and prediabetes:

The p-values, the 8% figure and the triglyceride and weight results come from the press release; the abstract reports the adiponectin, liver fat and visceral fat changes without figures [3][4]. Three things limit how far the table can be read:

• No effect sizes or time points. Apart from resting metabolic rate, neither source says how large any change was, or at which week it was measured. The abstract says only that the changes grew with time on treatment [3].
• Compared with baseline, not placebo. The abstract describes the changes as differences from baseline [3]. Neither source reports how the placebo group changed or tests the drug against it.
• Two figures for metabolic rate. The abstract says resting metabolic rate rose "up to 33%" from baseline; the press release says 8% [3][4]. Neither explains the gap. "Up to" suggests the larger figure is a maximum, not an average.

The company says the minimal weight loss matched what it predicted for this level of exposure. It plans two phase 2 trials: REWIRE-1, testing higher exposures on muscle function and lipid metabolism, and REWIRE-2, testing weight loss in people with obesity [4].

Later mouse studies widened the picture:

• Fatty liver. In a model of progressive fatty liver disease it reduced body fat and blood cholesterol and eased liver steatosis and fibrosis, though unevenly across the liver [1].
• Acute kidney injury. A week of pre-treatment protected against cisplatin-induced kidney injury. The protection was lost in cells lacking AMPK [5].
• Kidney aging. In naturally aged mice it reduced cellular senescence and fibrosis in the kidney [10].
• Aortic aneurysm. In a mouse model it reduced aneurysm formation and blood pressure [11].

ATX-304 activates AMPK across its isoforms. In the heart it acted as exercise does: it raised glucose uptake, lowered glycogen and improved stroke volume in mice, without enlarging the heart [2].

A 2023 study added an important detail. It also acts as a mitochondrial uncoupler. The metabolic demand that creates is what lets muscle take up and burn glucose independently of insulin, rather than storing it as glycogen [12]. In kidney cells it raised basal oxygen consumption by 38% without changing maximal respiration [5]. In the liver it shifted metabolism toward fatty acid oxidation and away from lipid synthesis [1]. In the pancreas it largely prevented the gene expression and chromatin changes that a high-fat diet causes in islets [13].

A 2026 poster abstract from the developer asked whether the AMPK activation is just a by-product of that uncoupling, since a compound that leaks protons can switch AMPK on by draining the cell's ATP. In mitochondria isolated from mouse muscle, ATX-304 raised oxygen consumption and lowered membrane potential, as a proton leak would. In intact cells it did not lower membrane potential at up to 20 µM, and it did not lower ATP in human liver cells. It also activated a mutant AMPK that cannot sense AMP as strongly as it activated the normal enzyme. The authors conclude that AMPK activation by ATX-304 is not secondary to a fall in ATP [14]. These are cell experiments described in an abstract, without figures.

In people, the rise in resting metabolic rate reported from the phase 1b trial is in line with a drug that raises energy expenditure [3][4].

How much of the drug reaches the blood depends heavily on its form. That is covered in its own section below.

ATX-304 has been given to people in two forms: the original molecule (the "free" or non-salt form) as a liquid suspension, and a finely milled sodium salt in capsules or tablets. The developer describes the sodium salt in two patent filings. Both are company data and not peer reviewed, but they contain the only published comparisons of the two forms [6][7].

Why a new form was needed. In the earlier phase IIa trial (TELLUS), 65 people with type 2 diabetes took 1000 mg a day of the non-salt drug as a suspension for 28 days. Their average peak blood level at day 28 was 55 µg/mL [6]. The later patent's stated aim is a "dose-efficient" form that reaches the same blood levels from a smaller dose, which it argues should also mean fewer side effects [6].

The rabbit studies (2019 patent). The first patent compared salt form, milling and capsule coating in single-dose rabbit studies [7]:

• Materials. The non-salt drug, its sodium salt and its potassium salt. Each was air-jet milled twice, bringing 90% of particles under about 8 to 12 µm (from 55, 32 and 205 µm before milling).
• Capsules. Ready-made gelatin capsules, either plain or enteric-coated, each holding the milled material with small amounts of mannitol, a binder (PVP K30) and a surfactant (docusate, 1% of the drug's weight).
• Reference. The non-salt drug as a suspension in methylcellulose, given by stomach tube. The patent does not describe this material as milled, and the suspension had no surfactant.
• Design. Male New Zealand White rabbits, 3 per group, each given one oral dose of 90 mg/kg. The salts were dosed by the weight of salt, so they delivered slightly less parent drug (about 85 mg/kg for the sodium salt). Blood was sampled for 72 hours.
• Measure. Total exposure, as the area under the blood-level curve up to the last sample (AUC).

AUC is in h·µg/mL. Read down the table and three separate effects appear [7]:

• Milled drug in capsules raised exposure about 3 to 3.6 times over the suspension, before any salt was involved. That step changed particle size, dosage form and excipients (including the surfactant) at once, so it cannot be put down to milling alone.
• The sodium salt raised it a further 1.7 to 1.9 times over the equally milled non-salt drug in the same kind of capsule; the patent describes this as a 50 to 100% increase. The potassium salt added only about 1.3 times. Because the salts were dosed by salt weight, these figures slightly understate the gain per milligram of parent drug.
• Enteric coating made little difference: with each material, exposure from enteric capsules was 5 to 13% lower than from plain capsules, well within the spread between rabbits, and peak levels were almost the same.

The headline "up to six-fold" gain compares the milled sodium salt in capsules with the suspension, so it bundles the salt, milling and capsule effects together. The capsules peaked late (32 to 48 hours, against 19 for the suspension) and blood levels were still high at 72 hours, while the suspension had largely cleared. So exposure up to the last sample, used here, if anything understates the capsules' advantage. The patent's figures for exposure extrapolated to infinity rest mostly on estimates beyond the sampling window (up to 89% of the total) and give implausible half-lives of up to 15 days, so this entry does not use them [7].

Magnesium oxide. Two further rabbit studies used enteric capsules of the sodium salt with magnesium oxide, an alkaline excipient [7]:

• Controlled comparison, 25 mg/kg. With magnesium oxide, exposure was about 16% higher than the same capsule made with mannitol (AUC 1,704 versus 1,473), with almost the same peak level (50 versus 47 µg/mL).
• Single group, 53 mg/kg (equal to 50 mg/kg of parent drug). With magnesium oxide, AUC was 8,926 and the peak level 153 µg/mL. That is higher than the 90 mg/kg enteric capsule without it (5,708), but the two were separate experiments, so it is not a controlled comparison.

The patent also gives two example recipes for a 350 mg tablet holding 106 mg of the sodium salt (equal to 100 mg of parent drug), optionally enteric-coated. No data are reported for them [7].

The human studies (2021 patent). The second patent reports two studies of the sodium salt in healthy volunteers [6]:

• Capsule study (enteric capsules). An open study: 24 volunteers were randomized to 200, 400 or 800 mg of the sodium salt (one, two or four capsules) once a day for 18 days. Each capsule was an enteric (acid-resistant) Vcaps shell holding 200 mg of milled salt (90% of particles under 8 µm) with mannitol and two surfactants, sodium lauryl sulfate and docusate. From day 19 to day 25, dapagliflozin 10 mg a day was added.
• Tablet study (uncoated tablets). Also open: 20 volunteers were randomized to 212 or 424 mg of the sodium salt (equal to 200 and 400 mg of the parent drug) once a day for 21 days; 9 per group have pharmacokinetic results. The tablet was a plain compressed tablet holding 424 mg of salt with lactose, microcrystalline cellulose, a disintegrant (sodium starch glycolate), silica and magnesium stearate, and the patent lists no enteric or other coating. It was scored so it could be split for the lower dose. The patent does not give the salt's particle size for the tablets.

The first thing both studies show is slow build-up. A first capsule took about 15 hours to reach its peak, and levels kept climbing with daily dosing until close to steady state before day 18 [6]:

With tablets, blood was not sampled at the day-1 peak, so the patent compares the level 24 hours after a dose. It rose from 11.4 to 80.4 µg/mL at 200 mg (about 7-fold) and from 24.9 to 129 µg/mL at 400 mg (about 5-fold) between day 1 and day 21 [6].

What the build-up says about half-life. Neither the patent nor the trial paper reports a half-life, and no study sampled blood after dosing stopped. The build-up gives an estimate: with one dose a day, levels rise by a ratio R that depends only on the half-life, so half-life = 24 hours × ln 2 ÷ −ln(1 − 1/R). Applied to the published figures [2][6]:

These are this entry's own calculations, not figures from either source. How to read them:

• They are effective half-lives: the figure that sets how long build-up takes, not a measured elimination half-life.
• The AUC-based capsule figures are probably too long. Levels were still rising 24 hours after the first capsule, so the day-1 AUC misses part of that dose and inflates the ratio. The peak and 24-hour-level ratios are less affected, which puts the salt at about 3 to 4.5 days.
• The two suspension estimates come from different studies and methods but agree (3.7 and 4.0 days). The trial paper says only that the long half-life delays steady state to about day 14 [2]. The patent says steady state takes 4 to 5 half-lives and typically about two weeks, which implies about 3 to 3.5 days [6].
• The salt does not appear to change the half-life. Its advantage is in how much drug is absorbed.
• At steady state the daily curve is nearly flat, with the peak only 7 to 13% above the 24-hour level, so the steady-state time to peak (means of 8 to 11.5 hours with capsules and about 11 hours with tablets, individual values 0 to 24 hours) means little [6].
• The ratio falls as the dose rises, in line with exposure rising less than in step with dose (below), so the formula is only approximate at higher doses. Each figure rests on the group means of 5 to 9 volunteers with pharmacokinetic results, or on 65 patients split between drug and placebo for TELLUS.

The second finding is how the salt compares with the old suspension. The patent lines up these studies with five earlier multiple-dose studies of the non-salt suspension. The fairest comparison is with 800 mg of suspension taken once a day in healthy volunteers, measured at steady state [6]:

AUC is in h·µg/mL over one 24-hour dosing interval. "As taken" divides by the whole weight of compound swallowed, which for the salt includes the sodium; "parent drug" divides by the ATX-304 it contains. The patent's own table mixes the two, dividing the capsule AUCs by the weight of salt (5.08, 3.82 and 3.21) but the tablet AUCs by the weight of parent drug (9.71 and 7.74), which slightly overstates the tablets' lead over the capsules. Either column gives the same ranking. Read plainly:

• A quarter of the dose, the same exposure. One 200 mg salt capsule (189 mg of parent drug) matched 800 mg of suspension on both peak and total exposure. That is the source of the patent's "up to five-fold" claim.
• Uncoated tablets beat enteric capsules. At about the same dose of parent drug (200 versus 189 mg), the tablet gave 1.9 times the daily AUC of the capsule, or 1.8 times per milligram on either basis; the patent calls it "up to two-fold". Per milligram the tablets were about 6 to 7 times the suspension as taken, or 6 to 8 times per milligram of parent drug.
• Why the tablets did better is not known. The two products differed in more than the coating: the capsule was an enteric shell with surfactants, the tablet an uncoated tablet with a disintegrant, made by a different manufacturer and tested in a separate study. The patent does not attribute the difference to any one of these. In rabbits, putting the same material in enteric rather than plain capsules made little difference (see above), which argues against the coating being the main cause, but that was never tested in people.
• Exposure does not rise in step with dose. Per milligram it fell from 5.1 to 3.2 as taken (5.4 to 3.4 per milligram of parent drug) as the capsule dose went from 200 to 800 mg, so doubling the dose less than doubles the blood level.
• Compared with TELLUS. The lowest capsule dose came close to the peak level that 1000 mg of suspension produced in TELLUS (49.9 versus 55 µg/mL); the higher capsule doses and both tablet doses exceeded it. TELLUS measured people with type 2 diabetes at day 28, so this is a rough comparison.

The phase 1b trial. The 2026 trial in obesity and prediabetes gave 400 mg once a day [3], and its registry record names sodium salt tablets [9]. That is the same nominal dose as the higher arm of the patent's tablet study (424 mg of salt, 400 mg as parent drug), but no source says whether the trial's 400 mg is the weight of the salt or of the parent drug, or whether the tablets were the same product. No blood levels from the trial have been published. The company says only that exposure reached the level its animal work predicted was needed for the metabolic effects [4].

What this does and does not show.

• The patent studies are exposure studies. None measured blood sugar, heart function or any other clinical outcome, and the patent describes safety assessments without reporting their results.
• The human comparison with the suspension is across separate studies run at different times, not a head-to-head trial, and so is the tablet-versus-capsule comparison. In people, the salt was never compared with equally milled non-salt drug. That like-for-like comparison exists only in rabbits, where the salt's own contribution was about 1.7 to 1.9-fold.
• The studies are small (3 rabbits per group; 20 to 24 volunteers split across dose groups) and run by the developer.
• The practical upshot is that milligrams are not interchangeable across forms. Salt versus non-salt, milled versus not, and tablet versus capsule each shifted exposure, and research-chemical products rarely state any of the three.

Later work from the developer's group describes it as well tolerated in humans [8], but the published human data is brief and small. In mice and rats it did not increase heart weight, a concern with long-term AMPK activation in the heart [2]. Because it partly uncouples mitochondria [12], the heat and energy-wasting risks that uncouplers carry are worth keeping in mind, though none have been reported.

The phase 1b trial looked for them directly. Core body temperature was monitored continuously and heart rate over 24 hours, and the company reports no increase in either and no adverse events involving body temperature, flushing or heart rate. Adverse events on ATX-304 were mostly mild and about as frequent as on placebo, with no new safety signals in the open-label extension to 16 weeks [4]. The abstract says only that the drug was safe and well tolerated [3]. No adverse-event counts have been published, and the trial covered 23 people at one dose.

The patent's volunteer studies included safety assessments, but it does not report their results in a form that can be cited as a safety finding [6].

The formulation findings matter for safety in their own right (see Formulation). The same number of milligrams can give very different blood levels depending on the form, and levels keep rising for two to three weeks of daily dosing [6], so effects and side effects may not show up in the first few days.

The phase IIa trial was run on top of metformin, so that combination has been studied [2]. In the developer's capsule study, volunteers added the SGLT2 inhibitor dapagliflozin (10 mg a day) for a week once ATX-304 levels were steady, and the patent claims the combination [6]. The patent does not report how dapagliflozin changed ATX-304 levels. No peer-reviewed interaction data exists beyond that.