Epetraborole in treatment-refractory MAC: a Phase 2 PRO signal that Phase 3 erased

What was tried

AN2 Therapeutics ran EBO-301 (NCT05327803), a Phase 2/3 randomized, double-blind, placebo-controlled study of oral epetraborole 500 mg once daily added to an optimized background regimen in adults with treatment-refractory Mycobacterium avium complex (MAC) lung disease. The trial enrolled 177 patients, started in May 2022, and reached its actual primary completion in December 2024. Epetraborole is a benzoxaborole that inhibits bacterial leucyl-tRNA synthetase. The program went after an indication with almost no oral options and a hard ceiling on cure rates from the standard macrolide-based regimen. AN2 terminated the study and reported that the truncated Phase 3 portion (n=97) missed its primary endpoint and could not confirm the efficacy seen in Phase 2 (AN2 Therapeutics, May 2025).

The biological hypothesis

Leucyl-tRNA synthetase charges transfer RNA with leucine and proofreads mischarged products at a separate editing site. Epetraborole forms a boron adduct in that editing pocket, traps the tRNA, and stalls protein synthesis. The enzyme is essential and bacterial, so the mechanism is orthogonal to macrolides and aminoglycosides. In vitro the drug is active against MAC, with a reported MIC50 of 2 mg/L and MIC90 of 16 mg/L and the strongest kill among the mycobacteria tested (Singh et al., 2025). The clinical bet was that a mechanistically novel oral agent, added to a failing background regimen, would convert sputum cultures and ease respiratory symptoms in patients who had exhausted standard therapy.

What actually happened

The Phase 2 portion never produced a clean efficacy signal on its own clinical-response measure. Clinical response favored epetraborole by 13.9 percentage points, but the 95% confidence interval ran from -6.8 to 33.8 and the result was not significant (p=0.19). The signal that justified Phase 3 came from a patient-reported outcome. The QOL-B respiratory domain improved by 6.90 points over placebo in Phase 2 (95% CI 0.45 to 13.36, p=0.037). In the truncated Phase 3 portion that same outcome, now the primary endpoint, fell to a 0.98-point difference (95% CI -5.28 to 7.25, p=0.76). Clinical response went to a risk difference of -0.2 points (95% CI -20.8 to 18.6, p=0.98). Phase 3 culture conversion favored placebo by 2.0 points (p=0.74). Across every endpoint the estimates regressed to zero (all figures from the ClinicalTrials.gov results record).

Failure mechanism, best guess

This was an efficacy failure built on a fragile Phase 2 inference, not a mechanism that stopped working. Three factors compounded. The Phase 2 go decision rested on a single nominally significant respiratory PRO in a small sample, with the harder microbiological and clinical-response endpoints already null. The Phase 3 population was sicker, with advanced cavitary or fibrocavitary disease and multidrug resistance at baseline, a group where any add-on faces a low ceiling. And the mechanism carries a known liability. Epetraborole's 2017 Phase 2 trial in complicated urinary tract and intra-abdominal infection was stopped for rapid emergence of resistance during treatment, driven by mutations in the editing domain (Nguyen et al., 2023). On a partially active background regimen in a chronic infection treated for six months, editing-site resistance is the failure mode most likely to erode a real signal.

How to prevent this next time

The decision error was advancing on a secondary PRO while the endpoints that map to bacterial killing stayed null. Two quantitative checks would have reframed the go decision.

First, a power audit. The truncated Phase 3 randomized roughly 59 patients to drug and 23 to placebo for clinical response. Detecting a 15-percentage-point improvement (50% versus 35%) at two-sided alpha 0.05 with 80% power requires about 170 patients per arm. The study was underpowered for its own clinical endpoint by a factor of three before the first patient enrolled, so a null result carried little information.

Second, a Bayesian predictive probability of success computed from the Phase 2 posterior rather than the point estimate.

PP(success)=∫θ​P(final success∣θ,interim data)⋅π(θ∣interim data)dθ

Feeding the wide Phase 2 intervals into that integral, with a prior anchored to the historical base rate for infectious-disease programs (roughly 25% likelihood of approval from Phase 1, against 13.8% overall, per Wong et al., 2019), would have returned a low predictive probability and marked the program as not yet de-risked. A biomarker-enrichment strategy that screened out cavitary, resistant disease at entry would also have raised the achievable effect ceiling.

The single highest leverage change would have been to gate the Phase 3 go decision on a prespecified microbiological endpoint, powered to detect culture conversion, rather than on a nominally significant respiratory PRO.

What this means for similar programs

Leucyl-tRNA synthetase remains a valid antibacterial target, and epetraborole still has a rationale in Mycobacterium abscessus, where AN2 has signaled it will continue. The lesson is narrower. For chronic mycobacterial infection, PRO endpoints are noisy and populations are heterogeneous in ways that swamp add-on effects. Any program against this target should treat editing-domain resistance as a primary design constraint, monitor for it across the full treatment window, and power the pivotal trial on a microbiological endpoint. The Claidex Mechanism Risk Score for this target now sits at 37 of 100 (yellow band), reflecting one documented late-phase efficacy failure against a target with no human genetic anchor and limited programmatic depth.

Open questions

Did resistance emerge during the six-month Phase 3 window, and if so at what frequency? Would a less advanced population, screened to exclude cavitary disease, have separated on culture conversion? And does the Mycobacterium abscessus indication, where in vitro potency is higher at a MIC50 near 0.25 mg/L (Singh et al., 2025), change the achievable effect enough to justify a fresh pivotal design?

Sources

1. AN2 Therapeutics. Reports Data from the Phase 3 Portion of Previously Terminated EBO-301 Study. May 2025. https://investor.an2therapeutics.com/news-releases/news-release-details/an2-therapeutics-reports-data-phase-3-portion-previously.

2. ClinicalTrials.gov. NCT05327803, EBO-301, results record. National Library of Medicine. https://clinicaltrials.gov/study/NCT05327803?tab=results.

3. Singh S, Boorgula GD, Galvez RA, Heysell SK, McShane PJ, Devine M, Gumbo T, Srivastava S. Mycobacteria susceptibility to epetraborole, how far can it go. Int J Tuberc Lung Dis. 2025, 29(10), 447-454.

4. Nguyen TQ, Heo BE, Hanh BTB, Jeon S, Park Y, Choudhary A, et al. DS86760016, a Leucyl-tRNA Synthetase Inhibitor, Is Active against Mycobacterium abscessus. Antimicrob Agents Chemother. 2023, 67(6), e0156722.

5. Shafiee A, Chanda S. In Vitro Evaluation of Drug-Drug Interaction Potential of Epetraborole. Pharmaceuticals (Basel). 2024, 17(1), 120.

6. Wong CH, Siah KW, Lo AW. Estimation of clinical trial success rates and related parameters. Biostatistics. 2019, 20(2), 273-286.