Amlexanox (AA-673)

Identity, and what was actually approved

Amlexanox carries two ATC codes in unrelated therapeutic areas: A01AD07 among stomatological preparations, and R03DX01 among systemic drugs for obstructive airway disease. That is the unusual feature of the record. The identifiers themselves are clean. PubChem returns one entry: CID 2161, CAS 68302-57-8, formula C16H14N2O4, molecular weight 298.29, InChIKey SGRYPYWGNKJSDL-UHFFFAOYSA-N, systematic name 2-amino-5-oxo-7-propan-2-ylchromeno[2,3-b]pyridine-3-carboxylic acid. The FDA Global Substance Registration System holds the same substance under UNII BRL1C2459K, with the same registry number and formula, a calculated weight of 298.294, and the ChEMBL identifier CHEMBL1096. Development codes AA-673 and CHX-3673, and the older name amoxanox, sit on that same record. The molecule is a small heterocycle. It is not a peptide, despite the company it now keeps on research-compound listings.

Two United States approvals exist and neither product is marketed. Aphthasol, a 5 percent dental paste, was approved under NDA 020511 on 17 December 1996 as a Type 1 new molecular entity, with a labelling supplement in June 2002. A 2 mg mucoadhesive patch followed under NDA 021727 on 29 September 2004, classified as a Type 3 new dosage form. Drugs@FDA gives the marketing status of both as Discontinued. A DailyMed query for amlexanox returns no current prescription labelling at all, and the only entries under the ingredient name in the National Drug Code directory are bulk-substance listings for compounding; no finished medicine appears there.

Japanese marketing followed a separate path. Tablets were registered in Japan at 25 mg and 50 mg for bronchial asthma and allergic rhinitis, and a commercial Japanese prescription-drug reference now marks both strengths as discontinued lines. That reference is the only support this record could find for any of those statements, and the failed attempt to corroborate them against PMDA is recorded in the untraced list below. Those tablets supplied the drug for the University of Michigan metabolic trials, where the registry names the product and gives the tablet strength as 25 mg. The compound is often introduced as a long-established approved medicine. Its two approved United States products are listed as discontinued, and the systemic tablet form was never approved in the United States at all.

What the aphthous-ulcer trials measured

Khandwala and colleagues reported four vehicle-controlled, randomised, double-blind, parallel-group, multicentre studies in 1997, involving 1,335 subjects who had one to three aphthous ulcers less than 48 hours old at enrolment. Pastes were applied four times daily until healing or the end of the study. The primary determinant was the percentage of subjects with complete healing and complete resolution of ulcer pain. The authors recorded that the vehicle itself had marginal beneficial effects, as a covering material would, and that its statistical separation from no treatment was inconsistent. An earlier double-blind study by Greer in 1993 applied paste to 32 patients over three days and found group differences significant on every criterion except pain reduction.

The approved label states the pooled figures. Against no treatment, across the two studies with such an arm, the median time to complete healing fell by 1.6 days and the median time to complete pain relief by 1.3 days; after three days of treatment 21 percent versus 8 percent had completely healed ulcers and 44 percent versus 20 percent had complete pain resolution. Against vehicle, across three studies, both medians fell by 0.7 days, and after four days 37 percent versus 27 percent had healed and 60 percent versus 49 percent had pain resolution. Those studies evaluated 464 patients on amlexanox paste, 465 on placebo paste and 195 on no treatment. The same label records that the paste was not shown to be an analgesic in its own right, and that effectiveness in immunocompromised people was not assessed.

Pharmacokinetics were characterised for the topical product only. After a single 100 mg application of the paste, which contains 5 mg of amlexanox, maximum serum concentrations of roughly 120 ng/ml were observed at 2.4 hours, the elimination half-life was 3.5 plus or minus 1.1 hours in healthy individuals, and approximately 17 percent of the dose was recovered in urine as unchanged drug, a hydroxylated metabolite and their conjugates. Most systemic absorption was attributed to the gastrointestinal tract rather than to the ulcer bed. Steady state was reached within a week of four-times-daily application, with no accumulation over four weeks. Carcinogenicity studies were negative over two years in rats and 18 months in mice, and Ames and mouse micronucleus tests were negative.

Comparative work since has been small. Jijin and colleagues compared 50 participants with minor aphthous ulcers allocated to 5 percent amlexanox paste or low-level laser therapy in 2016 and found no difference in pain between groups on the first and seventh days, with both groups improving; the report does not describe a randomisation method, and the PubMed record carries the publication types Clinical Trial and Comparative Study without a randomised-controlled-trial tag. A 2025 systematic review and meta-analysis of 25 randomised trials and 1,474 participants placed amlexanox among the agents with efficacy comparable to 0.1 percent triamcinolone acetonide. The same analysis reported that triamcinolone showed no significant advantage over placebo for pain reduction or healing time, so the comparator anchoring that comparison is itself weakly separated from control.

The kinase result that reopened the file

Reilly and colleagues reported in Nature Medicine in 2013 that amlexanox inhibits TBK1 and IKK-epsilon with half-maximal inhibitory concentrations of approximately 1 to 2 micromolar, competitively with respect to ATP, without effect on IKK-alpha or IKK-beta and without hits across a broad kinase panel at those concentrations. Micromolar is the operative word. Beyett and colleagues later solved crystal structures of TBK1 with amlexanox and a series of analogues, and reported that installing a tetrazole bioisostere of the carboxylic acid improved potency to 200 nanomolar toward IKK-epsilon and 400 nanomolar toward TBK1, while no analogue produced a greater response than amlexanox in adipocytes.

The same 2013 paper reported rodent results across three arms, all by daily oral gavage. In a prevention arm run alongside 12 weeks of high-fat feeding, male C57BL/6 mice given 25 or 100 mg per kg held weights equivalent to control-diet mice, with five animals per group; that arm is also where the authors placed the glucose finding, an approximate 30 to 40 percent reduction in the area under the curve on glucose tolerance testing. In a separate treatment arm begun after 12 weeks of high-fat diet, the authors described roughly 10 grams of weight loss after four weeks at 25 mg per kg, with seven animals per group. A third arm gavaged ob/ob mice on normal chow at 100 mg per kg and reported 7 to 8 grams over the same period, with eight per group.

Mechanism was refined afterwards by the same laboratory. A 2015 Nature Communications paper attributed the acute glucose effect to raised cAMP in subcutaneous fat, adipocyte secretion of IL-6, and hepatic STAT3 phosphorylation suppressing gluconeogenic gene expression. A 2021 Journal of Clinical Investigation paper reported that the long-term weight effect in mice required FGF21 and proceeded through beiging of white adipose tissue, with adipocyte-derived FGF21 acting locally. Notably, that paper's abstract describes the inhibitor as producing weight loss in obese animals and patients, which is not what the human trial recorded; the next section sets out the trial as posted.

The 42-patient trial, and what it did not show

ClinicalTrials.gov holds three amlexanox studies. NCT01842282 was an open-label study at the University of Michigan, terminated when funding ran out. Seven participants were enrolled, and all three posted outcome measures carry a denominator of six: change in glycated haemoglobin at 12 weeks, posted as 0; change in hepatic steatosis by magnetic resonance imaging, minus 2.2 percentage points; and change in weight, minus 2.2 kg. There was no control group. NCT01975935 was the randomised, double-blind, placebo-controlled study reported by Oral and colleagues in Cell Metabolism in 2017: 42 adults with type 2 diabetes, body mass index between 27 and 45, and non-alcoholic fatty liver disease. Twenty of 21 on drug and 18 of 21 on placebo completed. Registry-posted mean baseline age was 55.81 years on amlexanox and 59.24 on placebo.

Registry-posted results give the arm means. Glycated haemoglobin moved from 7.68 to 7.44 percent on amlexanox and from 7.77 to 7.76 percent on placebo, with a two-sided t-test p value of 0.05. Body weight moved from 100.95 to 100.81 kg on amlexanox and from 100.24 to 97.94 kg on placebo, p equals 0.51. Hepatic fat by magnetic resonance imaging, measured in the 20 participants who were scanned, moved from 18.0 to 16.22 percent on amlexanox and from 12.82 to 12.88 percent on placebo, p equals 0.38. On weight, the placebo arm fell further than the drug arm.

Publication and registry do not state the same primary figure. The paper describes glycated haemoglobin as reduced close to 0.5 percent more on average in the drug-treated group than placebo; the registry-posted arm means imply a difference in change of about 0.23 percentage points. Both records describe one trial of 42 people, and the analysis behind the published figure is not posted. The paper also reports a responder analysis: seven participants in the treatment arm reached a reduction of at least 0.5 percent against one on placebo, p equals 0.045, and reductions in hepatic fat were confined to that subgroup, with a difference in percent total and truncal body fat between responders and non-responders at p equals 0.06. Those are post hoc comparisons inside a trial of 42.

Publication and registry also differ on the regimen and on who was enrolled. The paper states 50 mg of amlexanox three times daily across twelve weeks. The registry's posted arm description instead records a lower dose for the first two weeks before the higher one, which the publication does not mention. The paper's Methods give an age criterion of 18 to 60 years; the registry posts eligibility as 18 or over with no upper limit, and its own posted baseline table places seven of the 42 participants at 65 or over. Both records are set down here, and no attempt is made to reconcile either disagreement into a single figure.

Rash accounted for most of the adverse-event discussion, and the paper draws on two studies for it. In the earlier open-label study, two of six patients developed a pruritic, erythematous generalised rash; both were biopsied and showed perivascular inflammation. In the 42-patient trial the paper reports seven cases of rash, one of which reached the severity seen in those two earlier patients. The registry posts skin rash as 4 of 21 on amlexanox and 3 of 21 on placebo, one serious adverse event in each arm and no deaths; the one non-ST-elevation myocardial infarction occurred in the placebo arm at week 11. Energy expenditure was not measured by calorimetry in these participants. The third registered study, a 58-participant randomised comparison of a 0.5 percent oral rinse against vehicle for radiation-associated oral mucositis, completed in June 2001 with no posted results and no publication retrievable in PubMed.

A second literature: nonsense-mutation readthrough

Gonzalez-Hilarion and colleagues reported in 2012 that amlexanox, recovered from a screen for inhibitors of nonsense-mediated mRNA decay, both raised the amount of nonsense-containing mRNA in treated cells and produced full-length protein from three patient-derived cell lines carrying nonsense mutations, and that the proteins were functional. The dual activity, decay inhibition plus readthrough, is what distinguishes this strand from aminoglycoside readthrough work.

Replication has been in cells rather than in people. Banning and colleagues reported in 2018 that amlexanox produced full-length aspartylglucosaminidase polypeptide and increased enzyme activity from a Trp168X allele in aspartylglucosaminuria, with a synergistic effect in compound heterozygous cells. Similar cell-level reports followed for COL7A1 in recessive dystrophic epidermolysis bullosa (Atanasova 2017, PMID 28549954), for BBS2 and ALMS1 in ciliopathies (Eintracht 2021, PMID 34365092), for PAX6 in aniridia models (Lima Cunha 2023, PMID 37483273), for GDAP1 in Charcot-Marie-Tooth neuronal cultures (Benslimane 2023, PMID 37513945), and in a murine cystic fibrosis model where readthrough and decay inhibition were combined (McHugh 2020, PMID 33396210).

Negative results are part of the same record. Tarraso and colleagues tested amlexanox, ataluren, RTC13 and G418 across three cellular models of the Pygm p.R50X mutation in McArdle disease in 2020 and found no evidence of readthrough at detectable levels in any model. Dabrowski and colleagues tested non-aminoglycosides including amlexanox against six premature termination codons found in primary ciliary dyskinesia patients in 2021, reported readthrough at lower efficiency than aminoglycosides, and located the advantage in tolerability. No human trial of amlexanox for any nonsense-mutation disorder is registered.

Older mechanisms, and a recent complication

Before the kinase work, the mechanism on file was mast cell mediator release. Makino and colleagues reported in 1987 that amlexanox at 10 to the minus 8 through 10 to the minus 6 molar roughly doubled cAMP content in rat peritoneal mast cells, that combination with isoproterenol raised it 15-fold to 88-fold depending on concentration, and that cAMP phosphodiesterase was inhibited with a half-maximal concentration of 1.4 times 10 to the minus 5 molar, about half the activity of IBMX. Pretreatment produced tachyphylaxis. The approved label states plainly that the mechanism by which the paste accelerates ulcer healing is unknown.

Separate binding work identified S100 proteins as targets. Shishibori and colleagues used three structurally unrelated anti-allergic drugs as affinity-chromatography probes in 1999 and found that native and recombinant S100A12 and recombinant S100A13 bound immobilised amlexanox. Rani and colleagues published the solution structure of the S100A13-amlexanox complex in 2010, reporting that the drug binds at the S100A13-FGF1 interface and prevents assembly of the complex through which acidic fibroblast growth factor leaves the cell by the non-classical route.

Human data outside dentistry are thin and old. Imokawa and colleagues randomised 15 asthmatic patients, eight with aspirin-induced asthma and seven without, to oral amlexanox or lactose placebo in 1993 and reported that FEV1 improved significantly after amlexanox in the aspirin-induced group only. A 2026 mouse study complicates the picture. Mungo and colleagues treated animals with established plaques and hepatic steatosis; plaque progression was not halted, and hepatic steatosis improved substantially in male mice while female mice showed worsened lipid accumulation. A compound with reported activity at kinases, phosphodiesterase and S100 proteins, all at micromolar concentrations, leaves open which target any given whole-animal result belongs to.

What is not known

The systemic form has never been characterised in people to modern standards. Published human pharmacokinetics cover the topical paste only, at a 5 mg delivered dose; no dose-ranging, absorption, distribution or steady-state study of the oral tablet is retrievable in English, and the only recent pharmacokinetic work is a 2022 rat assay method. The metabolic evidence in people rests on a single 12-week trial of 42 participants at one centre, never replicated, with its responder analysis performed after the fact in a group of seven, and with publication and registry disagreeing on the primary figure, the regimen and the ages of those enrolled. Energy expenditure, the mechanism the rodent literature points to, was never measured directly in a person. Nothing is established about administration beyond 12 weeks, or about effects in people without diabetes or fatty liver. Rash occurred in both arms of that trial, posted to the registry as 4 of 21 on amlexanox and 3 of 21 on placebo, a split too small and too close to separate drug from background. The readthrough literature is entirely cellular, includes one clear null in three McArdle models, and has produced no registered trial. The 2026 report that hepatic steatosis improved in male mice while worsening in female mice has no human counterpart, since the 2017 trial did not report outcomes by sex. Finally, a 58-participant oral-mucositis trial completed in 2001 has neither posted results nor a publication.

Questions