An Activin A/BMP2 Chimera, AB204, Displays Bone-Healing Properties Superior to Those of BMP2


  • Byung-Hak Yoon,

    1. Protein Engineering Laboratory, Joint Center for Biosciences at Songdo Global University Campus, Incheon, Republic of Korea
    2. Department of Pharmacology, Seoul National University College of Medicine, Seoul, Republic of Korea
    Search for more papers by this author
  • Luis Esquivies,

    1. Structural Biology Laboratory, The Salk Institute for Biological Studies, La Jolla, CA, USA
    Search for more papers by this author
  • Chihoon Ahn,

    1. Protein Engineering Laboratory, Joint Center for Biosciences at Songdo Global University Campus, Incheon, Republic of Korea
    Search for more papers by this author
  • Peter C Gray,

    1. Clayton Foundation Laboratories for Peptide Biology, Salk Institute for Biological Studies, La Jolla, CA, USA
    Search for more papers by this author
  • Sang-kyu Ye,

    1. Department of Pharmacology, Seoul National University College of Medicine, Seoul, Republic of Korea
    Search for more papers by this author
  • Witek Kwiatkowski,

    1. Protein Engineering Laboratory, Joint Center for Biosciences at Songdo Global University Campus, Incheon, Republic of Korea
    2. Structural Biology Laboratory, The Salk Institute for Biological Studies, La Jolla, CA, USA
    Search for more papers by this author
  • Senyon Choe

    Corresponding author
    1. Protein Engineering Laboratory, Joint Center for Biosciences at Songdo Global University Campus, Incheon, Republic of Korea
    2. Structural Biology Laboratory, The Salk Institute for Biological Studies, La Jolla, CA, USA
    • Address correspondence to: Senyon Choe, PhD, Structural Biology Laboratory, The Salk Institute for Biological Studies, 10010 N. Torrey Pines Rd., La Jolla, CA 92037, USA, E-mail:

    Search for more papers by this author


Recombinant bone morphogenetic protein 2 (rhBMP2) has been used clinically to treat bone fractures in human patients. However, the high doses of rhBMP2 required for a therapeutic response can cause undesirable side effects. Here, we demonstrate that a novel Activin A/BMP2 (AB2) chimera, AB204, promotes osteogenesis and bone healing much more potently and effectively than rhBMP2. Remarkably, 1 month of AB204 treatment completely heals tibial and calvarial defects of critical size in mice at a concentration 10-fold lower than a dose of rhBMP2 that only partially heals the defect. We determine the structure of AB204 to 2.3 Å that reveals a distinct BMP2-like fold in which the Activin A sequence segments confer insensitivity to the BMP2 antagonist Noggin and an affinity for the Activin/BMP type II receptor ActRII that is 100-fold greater than that of BMP2. The structure also led to our identification of a single Activin A-derived amino acid residue, which, when mutated to the corresponding BMP2 residue, resulted in a significant increase in the affinity of AB204 for its type I receptor BMPRIa and a further enhancement in AB204's osteogenic potency. Together, these findings demonstrate that rationally designed AB2 chimeras can provide BMP2 substitutes with enhanced potency for treating non-union bone fractures. © 2014 American Society for Bone and Mineral Research.


In developed countries, a person will sustain an average of two fractures in his or her lifetime and this number will only increase in the near future as the average age of the population increases. Bone morphogenetic proteins (BMPs) regulate bone growth and remodeling,[1, 2] and BMP2 (recombinant human BMP2, rhBMP2) has been used clinically to heal bone fractures in human patients.[3] However, the effectiveness of BMP2 in the bone-healing process can be limited, leaving an unmet medical need.[4] BMP2 is particularly limited in patients with critical size defects (CSDs) that cannot heal spontaneously. To heal such defects, BMP2 is administered in high quantities,[5] but such doses of BMP2 are at the same time associated with undesirable side effects. Therefore, BMP2 substitutes with higher therapeutic potency are needed.

BMPs and Activins are dimeric TGF-β superfamily ligands that signal by binding and assembling type I and type II transmembrane serine/threonine receptor kinases.[6] After ligand-induced assembly of two type I and two type II receptors, the constitutively active type II receptor kinases phosphorylate and activate the type I receptors, or Activin-like kinases (Alks), which in turn phosphorylate and activate cytoplasmic Smad proteins that enter the nucleus to regulate the transcription of target genes.[7-12] BMPs selectively bind the type I receptors Alk1, Alk2, Alk3, and Alk6 with high affinity and the type II receptors ActRII, ActRIIb, and BMPRII with low affinity, leading to the assembly of receptor complexes that activate Smads 1, 5, and 8.[13-15] By contrast, Activins bind the type II receptors ActRII and ActRIIb with very high affinity, allowing subsequent recruitment of the type I receptors Alk4 and Alk7 and activation of Smads 2 and 3.[16-18]

In addition to their distinct receptor and Smad specificities, BMPs and Activins also have profound structural differences. BMP2 adopts an extended, rigid butterfly conformation that has also been observed in other BMPs.[19-22] By contrast, Activin possesses a level of flexibility not found in BMPs[23-25] and has the ability to exhibit a more closed conformation than that of BMPs or other TGF-β superfamily members.[25] The ternary complex structure of BMP2 bound to Alk3 (BMPRIa) and ActRII shows that the receptors' extracellular domains do not make physical contact with one another, indicating that BMP2 binding to its receptors is a proximal element mediating interactions between receptors' cytoplasmic domains.[26] It is not yet known how BMPs compare with Activins in this regard because the structure of the ternary complex of an Activin together with its type II and type I receptors has not yet been solved.

Despite their differences, BMPs and Activins both bind the type II receptors ActRII and ActRIIb and do so in almost exactly the same spatial configuration.[23, 26] This led us to hypothesize that chimeric ligands possessing the type I receptor specificity of BMP2 and the high-affinity type II receptor binding properties of Activin A may have enhanced BMP2-like signaling properties. We tested this in a previous study where we replaced the type II receptor epitope of BMP2 with that of Activin A to create a chimeric ligand, which we named AB204.[27] In support of our hypothesis, we found that AB204 utilizes the same signaling receptors and Smads as BMP2 but that its activity is enhanced relative to BMP2, as demonstrated in in vitro signaling assays.[27] This suggested that AB204 could also have biological effects such as osteogenic and bone-healing properties that are superior to those of BMP2.

In the present study, we provide in vitro and in vivo evidence demonstrating that AB204 indeed has osteogenic and bone-healing activity greatly exceeding that of BMP2. We solve the crystal structure of AB204 that provides detailed mechanistic insight into the basis for the increased activity of AB204 relative to BMP2. This structure also reveals an amino acid residue derived from Activin A, which when mutated to the corresponding BMP2 residue results in a version of AB204 with increased affinity for the type I receptor BMPR1a and further enhanced osteogenic potency in vitro. Together, our results indicate that rational, structure-based design of Activin A/BMP2 chimeras can lead to highly effective BMP2 substitutes with superior bone induction properties and therapeutic potential.

Materials and Methods

Protein preparation

AB204 was prepared as previously described.[27] In brief, AB204 was expressed in Escherichia coli and chemically refolded. After the purification steps of heparin affinity and C4 reverse phase chromatography, the refolded protein was lyophilized for storage. rhBMP2 was purchased from National Institute for Biological Standards and Control and joint Protein Central ( Before use, the lyophilized proteins were reconstituted in 1 mM hydrochloric acid (HCl) in small volume before diluting by at least a factor of 100 in a relevant final buffer or media including PBS.

Osteoblast cell culture

MC3T3-E1 cells (pre-osteoblast cell line, originated from a newborn mouse calvaria, purchased from ATCC, Manassas, VA, USA) were maintained in basal media (BM): α-MEM supplemented with 10% fetal bovine serum, 100 units/mL of penicillin and 100 µg/mL of streptomycin. Osteogenic medium (OM) is BM with 10 mM β-glycerol phosphate and 50 µg/mL of ascorbic acid. Cells were grown in a humidified atmosphere of 5% CO2 at 37°C. Cells were tested for mycoplasma contamination by Mycoplasma Detection kit (Invitrogen, Carlsbad, CA, USA). The culture medium was refreshed every 3 days.

Cell viability assay

Cell viability was determined by the MTT colorimetric assay (ATCC). For the assay, 1 × 104 MC3T3-E1 cells per well were plated into a 96-well plate with OM (5% CO2, 37°C) supplemented with 3 ng/mL to 300 ng/mL of BMP2 or AB204 and cultured for 0 to 7 days.

Alizarin Red S staining

Confluent cultures of MC3TC-E1 cells (>5 × 104 cells/well of a 24-well plate) in OM were exposed to 30 to 90 ng/mL of BMP2 and AB204 with and without Noggin for 3 days. The cells were subsequently washed with PBS, fixed with ice-cold 70% EtOH for 10 minutes, and incubated with 2% Alizarin Red S (Sigma, St. Louis, MO, USA) pH 4.1 to 4.3 for 10 minutes, followed by several PBS washes. Observations were carried out at 40× and 100× magnifications with an Olympus (Tokyo, Japan) CKX41 microscope. For quantification, the cells stained with Alizarin Red S were destained with ethylpyridium chloride, and the 550-nm absorbance of the extracted stain solution was measured using a Du-730 spectrophotometer (DU-730, Beckman Coulter, Brea, CA, USA). Experiments were performed in triplicate wells and were repeated three times.

Von Kossa staining

Confluent cultures of MC3T3-E1 cells (2.5 × 104 cells/well in a 48-well plate) in OM were exposed to 30 ng/mL of BMP2, AB204, or AB204 (I103Y) for 1, 3, 5, or 7 days. To detect mineralized nodules in vitro, cultures were fixed in 4% paraformaldehyde (cat # P2031, Biosesang Inc., Seongnam, Korea) for 10 minutes, stained with silver nitrate (Sigma) for 60 minutes under the UV light, washed 3 times with distilled water, then the reaction was stopped with 5% sodium thiosulfate for 5 minutes and washed again with running tap water. For staining the nucleus, 0.1% nuclear fast red solution was applied to the wells for 5 minutes on the bench and then the plates were dehydrated quickly through 3 changes of fresh absolute alcohol. Each well was photographed using an Olympus CKX41 microscope under 40× magnification. Triplicate wells of calcium nodules were quantified by Motic Image 2.0 and NIH image software. Statistical analyses of the data were performed using analysis of variance (ANOVA).

Alkaline phosphatase (ALP) staining

Confluent cultures of MC3TC-E1 cells (>5 × 104 cells/well of a 24-well plate) were exposed to 30 ng/mL of BMP2 and AB204 with or without Noggin for 3 days, and the cells were washed with PBS and fixed with 4% paraformaldehyde buffer at room temperature for 10 minutes. The ALP staining was done with a freshly prepared 48-mL solution of Fast Blue RR1 (Sigma) with 2 mL of Naphthol AS-MX phosphate (Sigma) and incubated in the dark for 1 hour. ALP activity was quantified using the ALP detection kit (APF, Sigma). Experiments were performed in triplicate wells and were repeated three times.

RNA extraction and reverse transcription-PCR

A total of 1 × 106 of MC3T3-E1 cells were plated at 60 × 15 mm plates (SPL Co., Pocheon, Korea) with BM. After 3 days, the medium was exchanged to OM or OM supplemented with 30 ng/mL of BMP2 or AB204. The culture medium was refreshed every 3 days. At day 7, the cells were lysed, and total RNA was harvested using an RNeasy Mini Kit (Qiagen, Inc., Seoul, Korea). Total RNA yield was verified by measuring optical density with a spectrophotometer (DU-730, Beckman Coulter). A complementary DNA library was then constructed with an Accupower reverse transcription kit (Bioneer Co., Daejeon, Korea). Oligo-primers for alkaline phosphatase (ALP, F: GTTGCCAAGCTGGGAAGAACAC, R: CCCACCCCGCTATTCCAAAC), osteocalcin (OCN, F: AGGGAGGATCAAGTCCCG, R: GAACAGACTCCGGCGCTA), osteopontin (OPN, F: GACCACATGGACGACGATG, R: TGGAACTTGCTTGACTATCGA), collagenIα (ColIα, F: TCTCCACTCTTCTAGTTCCT, R: TTGGGTCATTTCCACATG), and GAPDH (F: CCCTGTTGCTGTAGCCGTA, R: CCGGTGCTGAGTATGTCG), as an internal standard, was obtained from Bioneer Synthesis Service. The primers were then used to amplify their corresponding cDNA targets by polymerization in a C1000 thermal cycler (Bio-Rad, Hercules, CA, USA).

Western blot analysis

Lysates were extracted from MC3T3-E1 preosteoblast cells after they were stimulated with 30 ng/mL of BMP2 or AB204 for 0, 15, 30, 60, 180, 360, and 720 minutes. And the lysates was mixed with 1× SDS-PAGE loading buffer (2% SDS, 2 M urea, 10 mM DTT, 10% Glycerol, 10 mM Tris HCl, 0.002% bromophenol blue) in order to separate on 10% SDS-PAGE gel followed by electrotransferred to nitrocellulose membrane (Bio-Rad). After blocking with 5% nonfat dry milk/TBS-T, the membranes were incubated with antibody to detect phospho Smad1/5/8 (cat. #9511S, Cell Signaling, Danvers, MA, USA), Smad 2/3 (cat. #5678S, Cell Signaling), p-Smad 2/3 (cat. #SC-11769, Santa Cruz Biotechnology, Dallas, TX, USA), or actin (clone AC-74, cat. #A2228, Sigma-Aldrich) at the 1:200 dilution for 1 hour. The bound antibodies were detected with secondary anti-rabbit IgG-HRP-conjugated antibody (1:1500, cat. #7074S, Cell Signaling). The blots were then washed, and the signal was visualized by enhanced chemiluminescent assay (Biosesang Co.) according to the manufacturer's protocol.

Operative procedures

All the procedures were performed in Lee Gil Ya Cancer and Diabetes Institute, Gachon University in South Korea and approved by the committee on Center of Animal Care. Six mice per group were used for the tibial and three mice per group for the calvarial defect recovery study. No additional mice were included and no mice were excluded during the experiment. Six-month-old C3H/HeN male mice were segregated into groups based on weight, with the average weight within the group being approximately the same to minimize group variation. No blinding protocol was used for this comparative study. Mice were sedated with the intraperitoneal administration of 20 mg/kg of Zoletil 50 (Virbac, S.A, Carros, France) and 10 mg/kg of rompun 2% solution (Bayer, Leverkusen, Germany) before surgery. In tibial defect models, mice were divided into four groups of 6 mice each: naive, collagen sponge (CS) only, CS + BMP2, and CS + AB204. In calvarial defect models, mice were divided into six groups of 3 mice each: naive, defect only, CS only, CS + 1 µg BMP2, CS + 0.1 µg AB204, and CS + 1 µg AB204.

To create a model of a partial tibia defect, the limb to be operated was shaved and a lateral approach to the tibia was performed without damaging the tibialis anterior muscle. The periosteum was lifted and the partial defect was made in a proximal region of the mouse tibia using a 1mm bone drill (Saeshin Co., Daegu, Korea).

To create an initial model of non-union CSD tibial defect, we followed a published procedure.[28] In short, the limb to be operated on was shaved and a lateral approach to the tibia was performed without damage to the tibialis anterior muscle. The periosteum was lifted only enough to allow the pin (d = 0.2 mm and length 5 to 10 mm) to lie directly on one side of the bone. After fixing the pin with surgical sutures and tape, we made a 2-mm osteoperiosteal segmental defect using an oscillating bone saw on the midshaft of diaphysis with continuous saline irrigation. To stabilize the fracture, the second pin was then applied and fixed on the opposite side of the prefixed pin.

To create a second non-union CSD tibial defect model, we followed another published procedure.[29] In this case, a 0.5-mm hole was made by bone drill above the tibial tuberosity, and a stainless pin was introduced into the intramedullary canal of the tibia. The 2-mm segmental defect was made as described above.

To create a model of CSD in the calvaria, a skin incision was made aseptically in a 6-month-old C3H mouse along the bilateral line and the middle of the forehead, and the dissection was continued to the calvaria. The periosteum of the calvaria was ablated and a full-thickness standardized trephine defect, 4 mm in diameter, was made in the calvaria using a Trephin Burr (Dentech Co., Tokyo, Japan) under continuous PBS irrigation. In all models, a porcine collagen sponge (CS) containing 0.1 to 1.0 µg of BMP2 or AB204 was implanted into the defect sites.


The mice with CSD tibial and calvarial defects were euthanized after 1 month. The harvested samples were first examined grossly, photographed, radiographed by Fx-pro (Carestream, New Haven, CT, USA) and then sawed apart through the center of the defect and subjected to µCT scanning and histochemical analysis. A µCT scan (system operates at 65 kVp, 60 µA, and pixel size 18 µm) was performed on mice with a calvarial CSD and partial tibial defect, and new bone volume to total tibia volume (NBV/TTV) was calculated for each partial tibial defect. The occupancy, defined as the ratio of the new bone volume to the volume of the segmental defect (NBV/VSD), and bone mineral density (BMD) of the new bone in the defect areas were measured using the Carestream imaging program and the ImageJ software provided by NIH for calvarial and tibial CSDs. The degree of bone formation was examined using µCT (NFR Polaris-G90, NanofocusRay Co. Ltd., Jeonju, Korea).

To obtain histochemical data, isolated tibial and calvarial bones were fixed in 10% formaldehyde for 24 hours, then demineralized with 10% EDTA solution for 3 days at 4°C. After paraffin embedding, tissues were sectioned at 3 µm, and the sections were stained with hematoxylin and eosin (H&E) and Villanueva bone stain solution.

Statistical analysis

Quantitative data were obtained in triplicate and were repeated three times, giving the total of 9 data points, and reported as means ± standard deviations. Comparisons to a control were made using one-way ANOVA and Dunnett's multiple comparison test in the Prism software.

Crystal structure determination

AB204 was resuspended in 10 mM Na Acetate pH. 4 at 10 mg/mL and crystallization trials were conducted with several commercial screening kits, Crystal Screen and Crystal Screen 2 (Hampton Research, Aliso Viejo, CA, USA), Wizard Screens I and II (Emerald BioStructures, Bainbridge Island, WA, USA), and PEG/ION and Nextal Classics suite (Qiagen, Inc., Valencia, CA, USA) using the Mosquito crystallization robot (TTP Labtech, Cambridge, MA, USA). The trials yielded several crystallization hits, which were subsequently optimized. The final crystal was obtained using the hanging-drop vapor diffusion method, in which 1 µL of 10 mg/mL protein was mixed with 2 µL of a reservoir solution of 0.2 M LiSO4, 0.1 M Tris 8.5, and 30% PEG 4000. Crystals appeared after 3 weeks at 15°C and were frozen in liquid N2. Diffraction data were collected in house using a Micromax-007HF X-ray generator (Rigaku, The Woodlands, TX, USA) to a resolution of 2.3 Å. The space group is P3121 with the following cell dimensions: a,b = 32.615, c = 147.358, and α,β = 90, γ = 120. Data were processed using XDS software,[30] and the structure was solved using molecular replacement with BMP2 as a model. Model building was done in COOT version 0.6.2.[31] Refinement of the structure was completed using REFMAC, version 5.5.0109[32] to R and Rfree of 0.25 and 0.28, respectively. The root mean squared deviations in bond lengths and bond angles are 0.006 Å and 1.13°, respectively. All structure figures were generated using PyMOL[33] and Molscript version 2.1.2.[34] Coordinates and structure factors have been deposited in the PDB with accession code 4MID.

Mutant generation

An AB204 mutant was made by PCR-based site-directed mutagenesis (Clontech, Mountain View, CA, USA).

Luciferase reporter assay

The C2C12 cells were cultured in DMEM supplemented with 10% FBS and antibiotics. For luciferase reporter assays, cells were trypsinized, washed with DPBS, and plated into 96-well plates with OptiMEM/1% FBS medium. Cells were transfected with -1147Id1-luciferase construct containing the Smad binding sites (Id1-Luc), Smad1 expression constructs, and a CAGGS-LacZ plasmid using Fugene6 (Promega, San Luis Obispo, CA, USA) according to the manufacturer's protocol. Luciferase activity was measured 16 hours after stimulation with ligands, and the values were normalized for transfection efficiency using beta-galactosidase activity. The activity of the luciferase reporter is expressed in relative units after beta-galactosidase normalization.

Surface plasmon resonance (BIAcore) affinity studies

The affinity of NB250 to ActRII-ECD was measured using a Biacore 3000 (GE Healthcare, Piscataway, NJ, USA). Using primary amine coupling, the receptor ECD was immobilized on a CM5 chip independently using flow cell 2. No protein was immobilized on flow cell 1 as a negative control. Experiments were performed with running buffer of 0.36% CHAPS, 0.005% Tween-20, 250 mM NaCl, 20 mM TrisHCl pH 8.0. For kinetic analysis, all tests were performed in duplicate using a minimum of five concentrations. Binding data were analyzed with BIAevaluation software version 4.1 (GE Healthcare) and fit using a global 1:1 Langmuir binding with mass transfer model.


AB204 is more osteogenic than rhBMP2 in vitro

Activin A and BMP2 were divided into 6 segments, each based on their secondary structure elements, and were combined to create a library of Activin A/BMP2 (AB2) chimeras denoted by the code (BXXXXX) where X is either A (Activin A) or B (BMP2).[27] AB204 (BABBAA) was designed to have the type II receptor binding epitope of Activin A (segments 2, 5, and 6) and the type I receptor binding epitope of BMP2 (segments 1, 3, and 4). The full segmental makeup of AB204 at the amino acid level is shown in Fig. 1A. We previously reported that AB204 has enhanced Smad1/5/8 signaling activity relative to BMP2, as measured using in vitro luciferase assays, and that it is insensitive to Noggin, a soluble, secreted BMP antagonist that blocks BMP signaling but not Activin signaling.[27] We also confirmed our hypothesis that AB204 by retaining BMP2-like type I receptor binding specificity will not signal through Smad 2/3 (Supplemental Fig. S1).

Figure 1.

The structure of AB204 reveals a BMP-like fold and key residues responsible for diminished type I receptor affinity and Noggin insensitivity. (A) Sequence alignment of BMP2 (cyan), Activin-ßA (purple), and AB204 (cyan and purple). Six segments are depicted to show AB204's composition (BABBAA), and the segments containing type I and type II epitopes are indicated. Key residue involved in binding the type I receptor, BMPRIa, is boxed. (B) Superposition of backbone Cα traces for AB204 and BMP2. BMP2 is depicted with a thin cyan line. AB204 is depicted with a thicker line with its BMP2 segments in cyan and its Activin segments in purple. N- and C-termini are marked. (C) Key residues predicted to be responsible for the lower binding affinity of AB204 for the type I receptor BMPRIa relative to BMP2. Top image: top view of the BMP2 (cyan)/BMPRIa (gray) complex. The key region of the BMP2/BMPRIa interface is boxed. Bottom left: close-up of this region showing critical interactions between BMP2's Tyr103 residue (cyan) and BMPRIa's Asp84 and Phe85 residues (gray). Bottom right: same region, with BMP2 replaced by AB204 (purple) showing the substitution of the interacting BMP2 Tyr103 residue with the noninteracting Ile103 residue from AB204. (D) Concentration-dependent luciferase reporter assay comparing Smad 1/5/8 activity induced by AB204 and AB204(I103Y). (E) Von Kossa staining measuring the effects of ligand-induced MC3T3-E1 cell differentiation to osteoblasts. Black or deep brown staining is proportional to the area of mineralized cellular matrix reflective of osteoblast differentiation. (F) Quantification of the intensity of staining was done using Motic Image 2.0 and ImageJ software based on microphotographs. Data are shown as the mean ± SD. The p values were obtained using one-way ANOVA followed by Dunnett's multiple comparion test. *p < 0.05.

Here, we tested the osteogenic activity of AB204 relative to that of BMP2 by comparing the ability of these ligands to increase mineralized calcium nodule formation and alkaline phosphatase activity. We also tested if the osteogenic activity of AB204 is inhibited by Noggin. We treated pre-osteoblast MC3T3-E1 cells cultured in osteogenic media with AB204 or BMP2 and quantified calcium nodule formation, an important and pivotal step in bone apatite formation. We found that AB204 is superior to BMP2 in promoting calcium nodule formation in this assay (Fig. 2A, B). AB204 also induced alkaline phosphatase (ALP) activity to a greater extent than BMP2 (Fig. 2D), another measure of its superior osteogenic potential. We also find that Noggin supplementation using a 1:5 ligand to Noggin molar ratio results in inhibition of extracellular mineralization and alkaline phosphatase activity in the BMP2-treated cells but has no effect on the AB204-treated cells (Fig. 2C, E).

Figure 2.

AB204 induces higher levels of mineralized calcium nodule formation, alkaline phosphatase activity, osteogenic marker gene expression, and Smad1/5/8 phosphorylation than BMP2. (A) Alizarin red staining of MC3T3-E1 cells cultured in OM supplemented with three different concentrations of BMP2 (top) or AB204 (bottom) after 7 days. (B) Close-up of cells cultured with and without 30 ng of BMP2 and AB204 after 7 days and spectrophotometric measurements of deposited mineralized calcium for the same cell cultures. (C) Alizarin red staining of MC3T3-E1 cells cultured in OM supplemented with 30 ng/mL of BMP2 (top) or AB204 (bottom) and treated with three different ligand:Noggin ratios after 7 days. (D) Assessment of alkaline phosphatase activity in MC3T3-E1 cells cultured in OM media supplemented with either 30 ng/mL of BMP2 or AB204 for 3 days. (E) Alkaline phosphatase activity comparison between cells cultured in Noggin-treated media supplemented with BMP2 and AB204. (F) Time-dependent induction levels relative to Glyceraldehyde 3-phosphate dehydrogenase standard of osteogenic markers: alkaline phosphatase (ALP), collagen Iα (Col1α), osteocalcin (OCN), and osteopontin (OPN) in MC3T3-E1 cells supplemented with BMP2 or AB204. (G) Anti-phospho-Smad1/5/8 Western blot showing time-dependent Smad1/5/8 phosphorylation levels over 12 hours in MC3T3-E1 cells cultured in OM supplemented with either BMP2 or AB204. Data (B, D, F) are shown as the mean ± SD. Significant differences compared with the control were determined by Dunnett's multiple comparison test (*p < 0.05, **p < 0.01, ***p < 0.001).

Next, we compared the abilities of AB204 and BMP2 to trigger expression levels of osteogenic marker genes. We performed quantitative RT-PCR after AB204 or BMP2 treatment to measure expression levels of several markers associated with bone differentiation to osteocytes and bone matrix production. The results show that the expression level of ALP, osteocalcin (OCN), osteopontin (OPN), and collagen Iα (ColIα) genes are significantly higher in cells treated with AB204 when compared with those treated with BMP2 (Fig. 2F).

We further tested AB204-induced Smad1/5/8 phosphorylation relative to that of BMP2. We compared the level and duration of Smad1/5/8 phosphorylation after treatment of MC3T3-E1 cells with AB204 and BMP2 and found that AB204 once again outperforms BMP2 (Fig. 2G). Western blotting using phospho-Smad1/5/8 antibody shows that the level of BMP2-induced Smad1/5/8 phosphorylation peaks after 30 minutes and then decreases to undetectable levels after 3 hours. By contrast, AB204 treatment results in Smad1/5/8 phosphorylation that reaches a maximum 180 minutes after stimulation and that is still detectable after 12 hours (Fig. 2G). The maximum level of phosphorylation was ∼6 times higher in AB204-treated cells than in cells treated with BMP2.

AB204 is superior to rhBMP2 in facilitating new bone formation during bone repair in vivo

In initial studies, we tested the relative abilities of AB204 and rhBMP2 to heal CSD in the murine tibia. We created osteoperiosteal segmental defects on the midshaft of diaphyses in 6-month-old male C3H mice using two different methods of supporting the defect and fixed the collagen sponge (CS) carrying rhBMP2 or AB204 into the tibial defect sites. The results show that only AB204 facilitates full healing of CSD in mice tibia (Fig. 3A, C). Furthermore, analysis of the ratio of the new bone volume to the volume of the segmental defect indicates that the volume of the new bone induced by AB204 is greater than that induced by rhBMP2 and, furthermore, that AB204 can cause the bone to overgrow the volume of the defect at the doses used (Fig. 3B). Similar results were also achieved from a partial tibial defect model (Supplemental Fig. S2).

Figure 3.

AB204 excels in inducing orthotopic bone formation in critical size defects of both murine tibia and calvaria. (A) Radiographs of mice tibia with modeled CSD supported by two sets of pins made 1 month after treatment. From left to right: tibia implanted with collagen sponge (CS), tibia implanted with collagen sponge containing 1 µg of BMP2 (CS/BMP2), and tibia implanted with collagen sponge containing 1 µg of AB204 (CS/AB204). (B) Corresponding new bone formation levels. (C) Radiographs of mice tibia with modeled CSD supported by intramedullary pin imaged 1 month after treatment. From left to right: tibia implanted with collagen sponge (CS), tibia implanted with collagen sponge containing 1 µg of BMP2 (CS/BMP2), and tibia implanted with collagen sponge containing 1 µg of AB204 (CS/AB204). (D) Coronal (top) and axial (bottom) views of modeled calvarial CSD (indicated by an arrow) after 1 month of treatment. From left to right: sham control, calvaria implanted with collagen sponge (CS), and calvaria implanted with collagen sponges containing 1 µg of BMP2, 0.1 µg of AB204, and 1 µg of AB204 (F = frontal; O = occipital lobes). (E) Naive and ligand-induced calvarial bone mineral density (BMD) in g/cm2. (F) Axial view of calvaria slice with Villanueva staining for naive calvaria, calvaria implanted with collagen sponges containing 1 µg of BMP2, and calvaria implanted with collagen sponges containing 0.1 µg of AB204 after 1 month of treatment. Red arrows indicate the implant site. (G) Close-up of regions 1, 2 of (F); black arrowheads indicate osteocells and white arrowheads indicate residual CS matrix. OB marks old bone and NB marks new bone region. Data (B, E) are shown as the mean ± SD. Significant differences compared with the control were determined by Dunnett's multiple comparison test (*p < 0.05, **p < 0.01, ***p < 0.001).

We further tested AB204's ability to heal calvarial CSDs. To model a calvarial CSD, a full-thickness standardized trephine defect was made in the calvaria of 6-month-old male C3H mice, and the CS carrying BMP2 or AB204 was implanted into the defect site. CT scans and histological analyses show that only AB204 is able to efficiently heal the defect. This can be found in both the coronal and the axial views in Fig. 3D. The histological analysis reveals that AB204-induced bone is completely fused with the old bone without a visible boundary line, whereas the new bone induced by BMP2 is not fused well, with the boundary line of the old bone clearly visible (Fig. 3F, G). Notably, we observe osteoid cells and mineralized bone in the Villanueva-stained tissue slide, which confirms the fidelity of the new bone. We have measured bone mineral density (BMD) by µCT scan and confirmed that the new bone density of BMP2/AB204-treated mice is not statistically different from that of naive mice (Fig. 3E). Histological examination also revealed that there was no uncontrollable penetration of the newly formed bone into unexpected regions of the brain.

The AB204 structure resembles that of BMP2 and suggests a point mutation that increases type I receptor affinity and further enhances AB204 potency

We solved the crystal structure of AB204 to better understand the mechanistic basis of the improved signaling and functionality of AB204 relative to BMP2. Interestingly, the overall architecture of AB204 is similar to that of BMP family ligands (Fig. 1B), despite the fact that it shares half of its amino acid sequence with Activin A (Fig. 1A). Unlike BMPs, Activin A displays a high level of plasticity with flexible wings and a disordered type I receptor-binding epitope.[23] However, the Activin A sequence in AB204 comprises the most rigid parts of Activin A, ie, segments of finger 1 and finger 2, which form two β-sheets that are conserved among TGF-β superfamily ligands. This provides an explanation for the overall rigid conformation of AB204 and the fact that the spatial arrangements of AB204 in complex with its signaling receptors are the same as for BMP2. Nonetheless, the entire type II receptor binding epitope of AB204, which is positioned near the fingertips of the ligand, consists of sequence derived from Activin A. Therefore, it is not surprising that AB204 and Activin A have similar high affinities for ActRII with Kd's of 0.38 and 0.20 nM, respectively.[27]

Although the type I receptor binding epitope of AB204 is mostly derived from BMP2, its binding affinity for Alk3 (BMPRIa) is much weaker than that of BMP2 with the Kd's for AB204 and BMP2 being 170 nM and 5.0 nM, respectively (Supplemental Fig. S3). Based on the BMP2/BMPRIa/ActRII complex structure,[26] it is evident that there is one important contact between finger 2 of BMP2, which consists of Activin A sequence in AB204, and BMPRIa. Tyr103 of BMP2 creates a strong hydrogen bond to Asp84 of BMPRIa that is aided by additional Pi stacking interactions between Tyr103 of BMP2 and Phe85 of BMPRIa (Fig. 1C). Because Tyr103 is replaced by Activin A-derived Ile103 in the sequence of AB204, this stable interaction at the AB204/BMPRIa interface is absent, possibly explaining the observed reduction in AB204's affinity for its type I receptor relative to BMP2.

Examining the modeled AB204/BMPRIa interface led us to hypothesize that the I103Y mutation will restore AB204/BMPRIa affinity to that of BMP2/BMPRIa and thereby further increase the signaling potency of AB204. Indeed, surface plasmon resonance studies demonstrated that the I103Y mutation dramatically increased the Kd of AB204 for BMPRIa from 170 nM to 6.6 nM, an affinity comparable to that of BMP2 (Kd = 5 nM) (Supplemental Fig. S3). Furthermore, the comparison of the signaling potency of AB204 and its I103Y mutant in BMP-responsive luciferase assay revealed that the potency of the mutant is significantly higher than that of AB204 and their respective EC50s are 2 nM and 9 nM (Fig. 1D). Finally, we compared the abilities of BMP2, AB204, and AB204 (I103Y) to promote differentiation of MC3T3-E1 cells into osteoblasts in vitro and found that AB204 (I103Y) possesses osteogenic potency greater than that of BMP2 and AB204 (Fig. 1E, F).


Structure-guided design of biologics holds great promise for the development of novel targeted therapeutics. Here, we highlight this approach by showing that the chimeric TGF-β ligand AB204 has osteogenic and bone-healing properties that are superior to those of BMP2. We hypothesize that the enhanced performance of AB204 in comparison to BMP2 in healing tibia and calvaria CSDs most likely stems from its greater ability to stimulate Smad1/5/8 signaling. We previously showed that AB204 has greater signaling potency than BMP2,[27] and our results here demonstrate that AB204-induced Smad1/5/8 signaling is also greatly prolonged relative to that of BMP2. In addition, both the signaling[27] and osteogenic properties of AB204 are Noggin insensitive. Together, these observations demonstrate a direct correlation between both the strength and duration of ligand-induced Smad1/5/8 signaling and the osteogenic activity of the ligand. On the other hand, the enhanced properties of AB204-induced Smad1/5/8 signaling are most likely the result of the almost 100-fold increase in its affinity to type II receptor (ActRII) relative to BMP2. Indirect evidence supporting this inference comes from the fact that increasing the affinity of AB204 for its type I receptor via the I103Y point mutation led to a corresponding increase in Smad1/5/8 signaling and osteogenic potency.

High doses of rhBMP2 are frequently required to achieve a therapeutic effect, resulting in side effects that can limit medical applications. These include complications such as cyst-like bone formation and abnormal soft tissue swelling.[35] Soft tissue swelling can lead to dysphagia, an ailment that has been observed to be more frequent and severe in patients receiving rhBMP2 as part of their treatment for cervical discectomy and fusion.[36] In this regard, AB204 provides an opportunity to provide effective bone healing at a dose that is at least an order of magnitude lower than rhBMP2. Thus, as we have shown here, even at a dose of 1 µg, which is 10-fold higher than the dose of AB204 that is required to achieve complete healing of CSD, rhBMP2 was unable to completely heal the CSD. Therefore, even if side effects are inherently associated with activation of the BMP2 signaling pathway, the temporal and spatial dynamics related to the lower dosage of AB204 should result in fewer unwanted side effects. Thus, assuming the same diffusion and degradation rate for both BMP2 and AB204, the smaller amount of AB204 needed to achieve therapeutic effects will impact a smaller area at the site of administration and will be active for less time before being cleared from the circulation than a higher dose of BMP2. Additionally, we have shown that AB204 lacks toxicity on osteoblasts (Supplemental Fig. S4), adding to its attractiveness as a therapeutic molecule with the potential to address unmet medical needs in bone injury treatment.

AB204 was originally designed using structural information, and we find that subsequent structure-guided mutagenesis (ie, the I103Y mutation) can increase the type I receptor binding affinity of AB204 and provide further gains in its osteogenic potency. Given the superiority of AB204 relative to BMP2, we predict that the AB204 I103Y mutant will similarly exhibit bone-healing activity exceeding that of AB204 and future studies will test this directly. Overall, our results demonstrate the potential clinical benefit of designer TGF-β chimeras with enhanced signaling properties.


All authors state that they have no conflicts of interest.


This work was supported by a Grant-in-Aid from the Korean Health Industry Development Institute (#A121836) and Incheon Free Economy Zone of Korea (jCB). LE was supported by the Jesse and Caryl Phillips Foundation Graduate Student Endowment. PG was supported by the Clayton Medical Research Foundation, Inc. and the Cancer Center Core Grant (P30 CA014195).

Authors' roles: SC, BY, and WK designed the experiments. BY conducted the in vivo studies. BY, PG, and SY conducted the in vitro studies. CA conducted the affinity studies and mutant design. WK and LE conducted the structural studies. SC, WK, BY, and PG wrote the manuscript. All authors discussed the results and commented on the manuscript.