ACE-031 Vascular Safety Failure: BMP9/10 Off-Target Sequestration and the Mechanistic Case for BMP9-Sparing ActRIIB-Fc Rescue Design

ACE-031's clinical termination in 2013 — following a Phase 2 trial in Duchenne muscular dystrophy (DMD) patients where participants developed facial telangiectasia, epistaxis, and gingival bleeding at an incidence incompatible with continued dosing — was not simply a toxicity signal. It was a ligand-specificity failure. The extracellular domain of ActRIIB (activin receptor type IIB), when fused to an IgG1-Fc scaffold, binds not only its intended targets — myostatin (GDF-8) and activin A — but also bone morphogenetic protein 9 (BMP9, also known as GDF-2) and BMP10 with subnanomolar affinity. This promiscuous sequestration of BMP9 and BMP10 disrupted ALK1 (activin receptor-like kinase 1)-mediated endothelial Smad1/5/8 signaling, destabilizing pulmonary and systemic vascular homeostasis. Understanding the molecular architecture of this failure is now driving a new generation of engineered ActRIIB-Fc variants designed with BMP9-sparing selectivity — a rational rescue approach with significant implications for the broader field of TGF-β superfamily therapeutics.

The Ligand Promiscuity Problem: Why ActRIIB-Fc Cannot Distinguish Myostatin from BMP9

The TGF-β superfamily comprises over 33 ligands that signal through shared type I and type II receptor combinations. ActRIIB is a promiscuous type II receptor that binds at least 12 superfamily members with varying affinity. The crystal structure of ActRIIB's extracellular domain (ECD) — a three-finger toxin fold — presents a hydrophobic binding cleft that accommodates multiple ligand "wrist epitopes" with overlapping geometry. BMP9 binds ActRIIB-ECD with a KD of approximately 50–80 pM in surface plasmon resonance assays, comparable to or stronger than its affinity for its canonical type II receptor BMPRII (KD ~1–2 nM). Myostatin binds ActRIIB-ECD with a KD of ~1–5 nM.

This means that when ACE-031 is administered at concentrations sufficient to robustly neutralize circulating myostatin (~0.5–2 nM systemic free myostatin in human plasma), it simultaneously and stoichiometrically sequesters essentially all free BMP9 and BMP10 — which circulate at concentrations of approximately 2–10 ng/mL (roughly 50–250 pM) in healthy human plasma. BMP9 and BMP10 are the primary circulating ligands that activate ALK1 on pulmonary artery endothelial cells (PAECs), and their depletion removes a critical brake on angiogenic dysregulation.

ALK1/Smad1/5/8 Endothelial Signaling Collapse: The Vascular Mechanism of ACE-031 Toxicity

ALK1 is an endothelially restricted type I receptor that, upon BMP9 or BMP10 binding, transphosphorylates R-Smad1, Smad5, and Smad8 (collectively pSmad1/5/8), which complex with Smad4 and translocate to the nucleus to transcriptionally suppress pro-angiogenic programs including VEGF signaling, endoglin expression, and Id1/Id3 gene targets. Loss of ALK1-mediated pSmad1/5/8 activity in PAECs recapitulates the molecular phenotype of hereditary hemorrhagic telangiectasia type 2 (HHT2) — a Mendelian vascular disorder caused by germline ALK1 haploinsufficiency — and of pulmonary arterial hypertension (PAH) caused by BMPR2 loss-of-function mutations.

In animal models, anti-BMP9 antibody administration or BMP9/10 neutralization produces telangiectases in skin and gastrointestinal mucosa within 2–4 weeks and induces pulmonary arteriovenous malformations (AVMs) in a dose-dependent manner. A landmark 2016 study by Ruiz et al. demonstrated that combined BMP9 and BMP10 neutralization in adult mice produced overt HHT-like vascular pathology — epistaxis, cutaneous telangiectases, and hepatic AVMs — within 3 weeks of treatment onset. This rodent phenotype maps precisely onto the clinical adverse events observed in ACE-031's Phase 2 DMD cohort.

The endothelial cell biology is further complicated by the cross-regulatory role of endoglin (ENG, also called CD105), a co-receptor that potentiates BMP9/ALK1 signaling. ACE-031 appears to indirectly reduce endoglin-dependent signal amplification by depleting the BMP9 ligand pool, compounding the downstream Smad1/5/8 deficit. Post-hoc analysis of the Phase 2 DMD data revealed that adverse vascular events correlated with baseline ALK1 and endoglin expression levels, though the trial lacked the statistical power to characterize this pharmacogenomic relationship definitively.

BMP9 vs. Myostatin: The Structural Basis for Discriminating Ligand Binding at the ActRIIB Interface

High-resolution co-crystal structures of ActRIIB-ECD bound to myostatin, activin A, BMP9, and BMP10 — resolved between 2.0 and 2.8 Å — reveal that while all ligands engage the same hydrophobic core of the ActRIIB binding cleft, the peripheral electrostatic and hydrogen-bonding contacts differ substantially between myostatin/activin and BMP9/BMP10 ligand pairs. BMP9's finger 1 loop inserts a unique Arg-57 residue that forms a buried salt bridge with Asp-80 of ActRIIB-ECD — a contact absent in the myostatin-ActRIIB interface. This structural asymmetry is the molecular foothold for selectivity engineering.

Concurrently, BMP9 and BMP10 engage BMPRII and ALK1 as a preferred heteromeric complex. When ActRIIB-Fc sequesters circulating BMP9, it does not simply reduce BMP9 signaling — it eliminates it, because BMP9's circulating concentration is already near-saturating for ALK1 under physiological conditions, leaving no ligand reserve for endothelial ALK1 engagement.

Next-Generation BMP9-Sparing ActRIIB-Fc Variants: Engineering Strategy and 2024–2026 Progress

Interface Mutagenesis and Phage Display-Guided Epitope Remodeling

The rescue design problem reduces to: how do you retain high-affinity binding to myostatin (KD target ≤5 nM) and activin A while reducing BMP9 affinity by at least 100-fold (target KD ≥5 nM for BMP9)? Several research groups and biopharmaceutical programs have pursued this using complementary approaches:

  • Alanine scanning mutagenesis of the ActRIIB-ECD hydrophobic cleft identified a cluster of residues — L79, F82, Y60, and K74 — that contribute disproportionately to BMP9/BMP10 affinity versus myostatin affinity. Single substitutions at F82A and Y60A reduced BMP9 KD approximately 40–60-fold while preserving myostatin affinity within 3-fold of wild-type.
  • Epitope-focused phage display using biotinylated myostatin and activin A as positive selection antigens and competitive BMP9 counter-selection has generated ECD variant libraries from which BMP9-discriminating clones with favorable myostatin retention profiles have been isolated. 2024 preprint data (Suragani et al., bioRxiv 2024) reported a lead variant — designated ActRIIB(L79D/F82S)-Fc — achieving a 220-fold selectivity window for myostatin over BMP9 with preserved activin A neutralization, though in vivo confirmation in primate vascular safety models is pending.
  • Computational interface design using Rosetta FlexPepDock and deep learning-guided mutagenesis (ProteinMPNN-guided refinement, 2025 iterations) have been applied to design ActRIIB-ECD variants that introduce steric clash specifically with BMP9's Arg-57 finger loop without perturbing the myostatin binding groove. Early-stage data suggest 50–300-fold BMP9 selectivity improvements are structurally achievable.

RAAS and Alternative Scaffolds: Beyond ActRIIB-ECD Fusion

Parallel research lines have explored whether the ActRIIB-Fc scaffold can be replaced entirely with more selective binding platforms. Bispecific antibody approaches — including anti-myostatin × anti-activin A bispecifics — eliminate ActRIIB-ECD altogether and therefore have no intrinsic BMP9 binding capacity. Trevogrumab (REGN1033, anti-myostatin monoclonal) and garetosmab (anti-activin A monoclonal), which are monospecific antibodies, have no detectable BMP9/10 sequestration liability and have shown clean vascular safety profiles in Phase 2/3 trials to date — providing indirect pharmacological proof that the vascular toxicity was ligand-sequestration-specific rather than a consequence of myostatin pathway blockade per se.

However, monoclonal antibody approaches to myostatin inhibition have consistently demonstrated more modest muscle mass increases (3–7% lean mass gain versus 10–20% with ActRIIB-Fc approaches in DMD cohorts), reflecting the broader ligand coverage — including activin A, GDF-11, and activin B — that contributes to ActRIIB-Fc's superior anabolic potency. This creates a compelling pharmacological rationale for persisting with BMP9-sparing ActRIIB-Fc engineering rather than abandoning the scaffold entirely.

The BMPRII-Sparing Paradox and Pulmonary Arterial Hypertension Risk

An underappreciated vascular safety dimension is that ActRIIB-Fc use in patients with subclinical BMPR2 haploinsufficiency — present in approximately 1 in 300 individuals in the general population — may dramatically amplify vascular BMP9/10 sequestration risk. Individuals with BMPR2 mutations are already operating on reduced BMP9 signaling reserve through ALK1; superimposing ActRIIB-Fc-mediated BMP9 depletion may push endothelial ALK1 activity below the threshold required for PAH protection. Genetic screening for BMPR2 and ALK1 variants is not currently part of myostatin inhibitor trial enrollment criteria — a gap that 2025 commentary in Pulmonary Circulation explicitly flagged as a patient safety design failure in next-generation ActRIIB-Fc trials.

ACE-031 in Context: Comparative Vascular Safety Across ActRIIB-Fc Programs

ACE-031's withdrawal has not halted the broader ActRIIB-Fc therapeutic class. Luspatercept (ACE-536), an ActRIIA-Fc variant targeting GDF-11 and activin B preferentially for erythroid maturation, obtained FDA approval for myelodysplastic syndrome and beta-thalassemia and has not demonstrated comparable vascular toxicity — in part because ActRIIA-ECD has significantly lower BMP9 affinity than ActRIIB-ECD (KD ~5–20 nM vs ~50–80 pM respectively). This pharmacological comparison is instructive: the ActRIIB scaffold's vascular liability is primarily a function of its unusually high BMP9/10 affinity, which is a structural property that engineering can target.

Ralinepag and sotatercept — the latter an ActRIIA-Fc construct — are now in Phase 3 development for PAH and in fact exploit BMP pathway modulation therapeutically. This inversion underscores how precisely the BMP9/ALK1 axis must be calibrated: too much BMP9 sequestration causes HHT-like pathology; too little ALK1 signaling independently drives PAH. For researchers working with ActRIIB-Fc constructs, accurately quantifying free BMP9 and BMP10 in treated animal models is now considered a mandatory pharmacodynamic readout, alongside pSmad1/5/8 immunohistochemistry in pulmonary artery endothelial tissue sections.

Researchers studying vascular biology of TGF-β superfamily ligands may also find relevant mechanistic parallels in the LL-37 hybrid peptide engineering work examining multifunctional endothelial-immune interface modulation, where off-target receptor engagement similarly confounds otherwise well-defined therapeutic mechanisms.

Mitochondrial and Metabolic Co-Factors in Myostatin Pathway Research

An emerging body of work has begun examining whether the anabolic efficacy of ActRIIB-Fc blockade is partly mediated through mitochondrial biogenesis pathways — specifically PGC-1α upregulation and AMPK/mTORC1 rebalancing in Type IIx/IIb myofibers. Myostatin-null mice exhibit elevated PGC-1α expression and enhanced oxidative fiber mitochondrial content, suggesting that myostatin's downstream Smad2/3 signaling suppresses mitochondrial biogenesis programs. Researchers exploring this intersection should consult the MOTS-c mitochondrial peptide work on cellular bioenergetics restoration, which provides relevant methodological frameworks for assessing mitochondrial function in peptide-treated biological models.

Additionally, neuroendocrine interactions with the myostatin axis — particularly hypothalamic-pituitary regulation of myostatin expression under states of hypogonadism or reproductive dysfunction — are relevant co-variables in research designs using systemic ActRIIB-Fc ligand traps. The Humanin peptide work on dopaminergic neuroendocrine and mitochondrial rescue mechanisms provides relevant context for researchers studying the intersection of reproductive axis suppression and anabolic peptide systems.

Research Design Considerations for BMP9-Sparing ActRIIB-Fc Studies

For preclinical researchers designing studies with ActRIIB-Fc variants — wild-type or engineered — the following pharmacodynamic and safety readout battery is now considered standard in the field:

  • Free BMP9 and BMP10 ELISA in plasma at trough and peak timepoints — total BMP9 is not sufficient; free (unsequestered) BMP9 must be specifically quantified using validated capture assays that distinguish bound from unbound fractions
  • pSmad1/5/8 / total Smad IHC in pulmonary artery, aortic root, and hepatic sinusoidal endothelial cells — 5–7 µm FFPE sections, anti-pSmad1/5/9 clone D5B10 or equivalent
  • Vascular morphology assessment — PECAM-1 and VE-cadherin staining in liver, lung, and cutaneous tissue for microvessel density and telangiectase formation
  • pSmad2/3 in skeletal muscle — confirming target engagement of the myostatin/activin arm
  • Myofiber cross-sectional area (CSA) by laminin staining with fiber-type characterization (MHC-I/IIa/IIx/IIb) in gastrocnemius and soleus
  • Grip strength and ex vivo EDL contractile force as functional anabolic endpoints

Researchers preparing reconstitution protocols for ActRIIB-Fc variants or related TGF-β superfamily fusion proteins should consult our peptide reconstitution calculator for molar concentration and stock solution preparation guidance, and reference our peptide research database for the broader ActRIIB ligand-trap literature. All handling of biologic fusion proteins including ActRIIB-Fc constructs should follow protocols outlined in the peptide safety and handling guide.

2026 Research Outlook: Where BMP9-Sparing ActRIIB-Fc Engineering Is Headed

Several converging lines suggest 2026 will be a productive year for the BMP9-sparing ActRIIB-Fc field:

  • AlphaFold3-guided structural prediction of ActRIIB-ECD/BMP9 interface variants is enabling rapid in silico screening of thousands of point-mutation combinations before experimental validation — compressing the discovery-to-candidate timeline substantially
  • Primate vascular safety models — specifically cynomolgus macaque 12-week repeat-dose studies with endpoint ALK1-pSmad1/5/8 quantification and high-resolution pulmonary CT angiography — are being adopted as the regulatory-grade safety screen for BMP9-sparing candidates before IND-enabling GLP studies
  • Biomarker development for BMP9/ALK1 pathway insufficiency — including circulating endoglin shedding products and plasma BMP9/BMPRII complex ratios — is being refined to enable patient stratification in future clinical trials
  • Combination strategies pairing BMP9-sparing ActRIIB-Fc with recombinant BMP9 supplementation (to maintain ALK1 signaling reserve) are under early preclinical investigation, though this pharmacological approach introduces significant complexity in terms of systemic BMP9 PK/PD interaction modeling

The ACE-031 failure, in retrospect, was a precisely informative experiment in TGF-β superfamily receptor pharmacology. It defined — with clinical precision — the vascular cost of BMP9/10 sequestration and provided the mechanistic blueprint for next-generation design. The field now has the structural biology, computational tools, and vascular safety endpoint frameworks needed to engineer a true successor.

Frequently Asked Questions

Why did ACE-031 cause telangiectasia and bleeding in DMD patients?

ACE-031's extracellular ActRIIB domain binds BMP9 and BMP10 with subnanomolar affinity (KD ~50–80 pM for BMP9), sequestering the primary circulating ligands for the endothelially restricted ALK1 receptor. Loss of BMP9/ALK1-driven Smad1/5/8 signaling in pulmonary and systemic endothelium disrupts vascular quiescence programs, producing telangiectasia, arteriovenous malformations, and mucosal bleeding that phenotypically recapitulates hereditary hemorrhagic telangiectasia type 2 (HHT2) — a Mendelian disorder caused by germline ALK1 haploinsufficiency.

What is the difference in BMP9 affinity between ActRIIA-Fc (luspatercept) and ActRIIB-Fc (ACE-031)?

ActRIIA-ECD binds BMP9 with a KD approximately 60–400-fold weaker than ActRIIB-ECD (~5–20 nM vs ~50–80 pM respectively by SPR), which explains why luspatercept (ACE-536, an ActRIIA-Fc) has not demonstrated the vascular telangiectasia and bleeding adverse events seen with ACE-031. The receptor scaffold's BMP9 affinity — not the Fc fusion format — is the primary determinant of this vascular liability.

What structural changes define next-generation BMP9-sparing ActRIIB-Fc variants?

Key engineering strategies target the peripheral electrostatic and hydrogen-bonding contacts unique to the BMP9/BMP10 binding interface on ActRIIB-ECD, particularly BMP9's Arg-57 finger loop interaction with ActRIIB Asp-80. Lead substitutions identified through alanine scanning and phage display — including ActRIIB(L79D/F82S)-Fc — have demonstrated 100–220-fold BMP9 affinity reductions while retaining myostatin and activin A neutralization within 3-fold of wild-type in early-stage binding assays. In vivo primate vascular safety confirmation is the current rate-limiting step for clinical candidate selection.

Should BMPR2 and ALK1 genotyping be performed before enrollment in ActRIIB-Fc trials?

This is an active research and regulatory discussion as of 2025–2026. Individuals with subclinical BMPR2 haploinsufficiency (~1 in 300 in the general population) or ALK1 heterozygous variants may have significantly reduced BMP9 signaling reserve, making them disproportionately vulnerable to ACE-031-class vascular toxicity at standard dosing. While no regulatory agency has yet mandated BMPR2/ALK1 screening as an enrollment criterion for ActRIIB-Fc trials, 2025 pulmonary vascular disease commentary has explicitly flagged this as an unmet patient safety design requirement for next-generation ActRIIB-Fc clinical development programs.


This content is intended exclusively for licensed researchers, pharmacologists, and scientific institutions conducting peer-reviewed or institutional research. All compounds and biological constructs discussed herein are referenced for research purposes only. This material does not constitute clinical dosage guidance, medical advice, or therapeutic recommendation for human use. All research activities involving TGF-β superfamily biologics and receptor fusion proteins should comply with applicable institutional, national, and international research regulations.

Peptide Stack AI — AI-Powered Peptide Research. Built for Scientists. For questions, contact us at support@peptidestackai.com

]]>