Engineering and Therapeutic Application of Single-Chain Bivalent TGF-β Family Traps

The TGF-β superfamily includes more than 30 ligands that regulate several physiologic processes, including cell proliferation, migration, and differentiation. Perturbation of their levels and/or signaling gives rise to significant pathologic effects. For instance, TGF-β and activin play critical pathogenic roles in many diseases, including cancer and fibrosis (1–6). Myostatin/GDF8 is a validated therapeutic target for muscle wasting diseases such as muscular dystrophy (7, 8). Bone morphogenetic proteins (BMP) have been implicated in cardiovascular diseases (9) and high levels of BMP2 and BMP4 are expressed in calcified atherosclerotic plaques and diseased aortic valves (10).

One approach to developing therapeutic agents that inhibit TGF-β family function is to block ligand access to its receptors, with both antiligand antibodies and soluble receptor ectodomain-based ligand traps being pursued (11–14). Receptor ectodomain-based traps are a class of therapeutic agents built on structural knowledge of receptor/ligand interactions and can thus be optimized using protein engineering approaches (15–17). Our prior studies on the kinetics of TGF-β ligand–receptor interactions indicated that artificial dimerization of the ectodomain of the TGF-β type II receptor (TβRII-ED) greatly improved its antagonist potency. This increased efficacy results from an avidity effect that promotes more effective ligand sequestration because of decreased ligand dissociation rates (15, 18, 19). Indeed, IgG Fc-dimerized receptor ectodomains targeting various ligands have emerged as successful therapeutics, with some advancing into the clinic (20–22).

In this article, we describe a novel design strategy for generating minimized TGF-β family ligand traps in which bivalency is achieved via covalent linkage of 2 receptor ectodomains (C-terminus of one ectodomain linked to N-terminus of the other) in a single polypeptide, rather than via dimerization domains. Generally, flexible nonnatural polyglycine or glycine–serine sequences are used as linkers to connect the 2 folded binding domains in bivalent fusion proteins (e.g., ref. 23). In contrast, our strategy exploits the intrinsically disordered regions (IDR) that flank the structured ligand-binding domain of TGF-β family receptor ectodomains. We speculated that using IDRs of the cognate human receptor as linkers might bring advantages relative to artificial Gly/Ser-based linkers in terms of reduced immunogenicity. In addition, improved affinity might be achieved because of the IDRs having structural plasticity (ref. 24; and references therein), which could promote bridged bivalent binding.

Specifically, we designed traps for the TGF-βs, BMPs, activins, and myostatin using "natural sequence" flexible linkers by fusing the C-terminal disordered region from one receptor ectodomain to the N-terminal disordered region of the second ectodomain. Using 2 TGF-β traps, (TβRII)2 and (TβRIIb)2, as primary examples, we show that they exhibit improved ligand binding and neutralization potencies compared with the cognate monovalent receptor ectodomain. We have also assessed the in vivo efficacy of the (TβRII)2 trap compared with 1D11 antibody, a well-studied TGF-β–neutralizing antibody with proven efficacy as an anticancer therapeutic (25–29). Using the 4T1 mouse mammary tumor model, we show that the single-chain TGF-β trap, but not 1D11 antibody, is able to reduce the growth of established primary tumors.

TGF-β ligands were provided by Dr. Andrew Hinck (University of Texas Health Science Center, San Antonio, TX). BMP2, TβRII-Fc, and anti-TGFβRII and 1D11 antibodies (R&D Systems), anti Smad2 and phosph-Smad2 antibodies (New England Biolabs) and human serum (Sigma-Aldrich). Mink lung (Mv1Lu) and 4T1 mouse mammary carcinoma cell lines (American Tissue Culture Collection).

Gene constructs of TβRII-ED, (TβRII)2, and (TβRIIb)2 traps are preceded by an N-terminal cassette consisting of the human VEGF signal peptide/8×His tag/Thrombin or TEV cleavage sites. Gene assemblies were cloned in the pTT plasmid vector (30) and expressed in modified human embryonic kidney cells (293-EBNA1 clone 6E) expressing EBNA1 (30, 31). Trap protein was purified on Fractogel-Cobalt column and desalted in PBS (32).

Surface plasmon resonance (SPR) data were generated using a ProteOn XPR36 instrument (BioRad Inc.) at 25°C using HBST (10 mmol/L HEPES, 150 mmol/L NaCl, 3.4 mmol/L EDTA, and 0.05% Tween-20) as running buffer. TGF-β (500 resonance units) was immobilized onto a BioRad GLC sensor chip using standard amine coupling methods. TGF-β trap samples were injected for 120 seconds followed by 600 seconds dissociation at a flow rate of 50 μL/min. Specific trap–TGF-β interaction sensorgrams were double referenced to blank control surface and buffer blank (BioRad Protein Manager Software v2.1) and normalized to a maximum 100 resonance units response (BiaEval v3.2, Biacore Inc.).

SPR data were generated using a Biacore 3000 instrument. A high-density, amine-coupled 1D11 anti-TGF-β antibody surface (∼10,000 resonance units) on a Biacore CM-5 sensor chip was used. Solutions containing TGF-β trap variant or 1D11 were preincubated with 0.3 nmol/L TGF-β at 5°C and then injected over 1D11 sensor chip under mass transport–limiting conditions for 3 minutes with a 30-second dissociation time at a flow rate of 5 μL/min at 25°C. Sensorgrams were aligned to the injection start point and double referenced to control surface and blank injection sensorgrams. The % free TGF-β was calculated based on the binding-rate value, derived from the double referenced sensorgram (BiaEval v3.2; Biacore Inc.) and normalized to the 0.3 nmol/L TGF-β control rate. Binding curves were plotted to determine binding IC₅₀ values (Supplementary Table S3).

TGF-β and BMP reporter assays were carried out using Mv1Lu or C2C12BRA cells having a PAI-1-luciferase (33) or BMP–luciferase reporter gene (34), respectively. Cells were treated for 16 hours at 37°C with trap samples plus 20 pmol/L TGF-β or 1 nmol/L BMP2 in Dulbecco's Modified Eagle's Medium, 1% FBS, 0.1% bovine serum albumin. Cells were then lysed and luciferase activity was measured (Promega Corp.). TGF-β neutralization IC₅₀ values are listed in Supplementary Table S4.

The (TβRII)2 trap was [¹²⁵I]-labeled (Pierce IODO-GEN tube kit; Fisher Sci.). Male Sprague Dawley rats (Charles River) were injected via tail vein with a single dose of trap (10 mg/kg unlabeled trap + 15 μCi [¹²⁵I]-(TβRII)2). Blood and urine samples were collected at various time points postinjection and counted in a WallacWizard 1470 gamma counter (PerkinElmer Life Sci.). Tissues were removed from animals sacrificed at selected time points. Trap levels are expressed as microgram equivalents, as calculated from the cpms.

4T1 mouse mammary tumor cells (4 × 10⁴ in 40 μL saline) were orthotopically implanted in the left #4 inguinal mammary fat pad of 6- to 8-week-old female BALB/c mice. After 24 hours, animals were randomized and treatment was initiated for 2 weeks with daily tail vein injections of (TβRII)2 trap (10 mg/kg) or saline (vehicle control), whereas 1D11 antibody was administered thrice a week (5 mg/kg). Animals were euthanized on day 15 and tumors were weighed and processed for histology or snap-frozen for Q-PCR analysis. In a second study, mice with established 4T1 tumors (∼100 mm³) were randomized and treated daily intraperitoneally with saline or trap (10 mg/kg) or thrice a week with 1D11 (5 mg/kg). P values were calculated using a 2-tailed nonparametric Whitney–Mann test.

TGF-β superfamily growth factors adopt the cysteine knot fold and form covalent disulfide-linked symmetric homodimers (reviewed in refs. 35, 36). The ectodomains of the TGF-β superfamily receptors are also structurally conserved and contain a single structured domain belonging to the snake toxin family (37). An additional feature is that their ectodomains typically contain IDRs flanking the structured ligand-binding domain (Fig. 1A, Supplementary Table S1). Structural elucidation of several TGF-β superfamily ligand–receptor complexes confirmed that 2 receptor ectodomains bind simultaneously to one homodimeric ligand molecule (38–44).

To generate minimized TGF-β family ligand traps, we used a single-chain approach in which 2 ectodomains are fused covalently in a head-to-tail manner. Examination of the crystal structures of TGF-β superfamily ligand–receptor complexes showed that the linear distance between the 2 receptor ectodomains, when bound to ligand, varies between 45 to 80 Å (Supplementary Table S2). Thus linkers of different lengths are required for each TGF-β family member single-chain trap. A comparison of the minimum number of residues required for linkage versus the number of residues in natural linkers formed by fusing the C-terminal IDR of the first receptor ectodomain with the N-terminal IDR of the second is shown in Supplementary Table S2. Using the single-chain TGF-β trap termed (TβRII)2 as an example, this table shows that the linker assembled by fusion of the IDRs has more than sufficient length (10 plus 25 = 35 a.a. vs. the minimal 32 a.a. required length) to span over the TGF-β ligand and enable bivalent binding. The existence of a splicing variant, termed TβRIIb, which features a 25-residue insertion in its N-terminal IDR, allowed us to examine the effect of a longer 60 a.a. “natural” linker. The resultant traps are shown schematically in Fig. 1B. Molecular mechanics based 3 dimensional models of the TGF-β–bound (TβRII)2 trap further confirmed that the length of the natural linker is sufficient to circumvent the ligand (Fig. 1C). A similar approach was used to design traps for additional TGF-β superfamily ligands, for example, activin and BMPs (Supplementary Table S2 and Fig. S1).

In the case of the (TβRII)2 trap, the feasibility of appropriate bivalent ligand binding was further assessed by molecular dynamics simulation in explicit solvent. As seen in Supplementary Fig. S2A, the TGF-β3–bound (TβRII)2 trap attained a stable molecular dynamic solution structure that preserves the mode of simultaneous binding observed experimentally for the 2 unlinked ligand-binding domains (38, 39). This molecular dynamic analysis also indicated that the linker of (TβRII)2 becomes relatively rigid, with only 6 residues experiencing greater mobility than those of the structured ligand-binding domains of the trap (Supplementary Fig. S2B). In addition, this analysis indicated that the trap can establish a favorable interaction with TGF-β3, with a highly favorable solvated interaction energy (45) of −25.4 kcal/mol (Supplementary Fig. S2C). This further substantiated the feasibility of using natural IDR-based linkers as there were no significant unfavorable steric and electrostatic contacts predicted between the linker and the ligand. On the basis of this molecular dynamic analysis of the (TβRII)2 trap, we assumed that the 25-residue longer linker of the (TβRIIb)2 trap will also allow unobstructed binding of TGF-β.

To investigate the ability of these ligand traps to bind TGF-β, his-tagged versions of (TβRII)2 and (TβRIIb)2 traps and TβRII-ED monomer were expressed in modified HEK-293 cells and purified using cobalt-affinity columns. The (TβRII)2 and (TβRIIb)2 proteins migrated over a 50- to 60-kDa range because of glycosylation, as assessed by SDS-PAGE (Fig. 2A). The binding of (TβRII)2, (TβRIIb)2, TβRII-ED, and 1D11 (a pan-TGF-β–neutralizing antibody) to immobilized TGF-β isoforms was compared by SPR (Figs. 2B–D). The sensorgrams representing binding of these proteins to TGF-β1 and TGF-β3 were essentially identical. The kinetics of TβRII-ED binding to TGF-β1 and TGF-β3 was characteristic of the monomeric receptor (i.e., fast on- and off-rates; ref. 18). The binding of TβRII-ED to TGF-β2 was undetectable, as expected, because TβRII has a low affinity for TGF-β2 (15, 46). In contrast to the monovalent TβRII-ED, both (TβRII)2 and (TβRIIb)2 dissociated from TGF-β1 and TGF-β3 with slow off-rates, indicating that the linkage of 2 TβRII ectodomains resulted in improved ligand binding. This is also supported by the observation that, in contrast to TβRII-ED, (TβRII)2 and (TβRIIb)2 were both able to bind TGF-β2 (Fig. 2D). Nonetheless, the fast off-rate component in their TGF-β2–binding sensorgrams suggests that their affinity for TGF-β2 is much weaker than for TGF-β1/3. The 1D11 antibody exhibited similar binding to TGF-β1 and TGF-β3 as (TβRII)2 and (TβRIIb)2. However, the slow off-rate with TGF-β2 indicated stronger binding of 1D11 to this isoform.

Flowing the traps and 1D11 antibody over immobilized TGF-β1, TGF-β2, and TGF-β3 surfaces allows for a relative ranking of binding activity, as presented above. However, in this assay orientation, the trap may be interacting not only intramolecularly with a 1:1, trap:ligand stoichiometry as designed but also intermolecularly with the 2 TβRII-binding domains within the trap contacting neighboring immobilized TGF-β molecules. To determine whether the traps function in solution as designed, we carried out competitive SPR-binding experiments in which the TGF-β trap was bound to ligand in solution, thereby removing potential ligand-immobilization artifacts. In this assay, the trap protein was first allowed to bind to a fixed amount of TGF-β in solution, after which this mixture was flown over immobilized 1D11 antibody to quantify the amount of ligand left unbound by the trap (TβRII and 1D11 compete for binding TGF-β, i.e., have overlapping epitopes). The binding IC₅₀ values for the various traps, 1D11, and TβRII-ED monomer to TGF-β2 and TGF-β3 were determined by plotting the percent free TGF-β over a range of trap concentrations (Fig. 3A and B, Supplementary Table S3). Single-chain (TβRIIb)2 and (TβRII)2 traps were compared with IgG Fc-dimerized TβRII (TβRII-Fc), monomeric TβRII-ED, and 1D11 antibody. As expected, TβRII-ED was least efficient for binding TGF-β2 and TGF-β3. The 2 single-chain traps, TβRII-Fc and 1D11, all bound TGF-β3 with similar efficacies, exhibiting average IC₅₀ values ranging from 0.1 to 0.4 nmol/L. In contrast, the single-chain traps and TβRII-Fc bound weakly to TGF-β2 (IC₅₀ values >100 nmol/L), whereas 1D11 bound efficiently to this isoform (IC₅₀ = 1.8 nmol/L). When these results were compared with the direct binding data in Fig. 2B–D, it was apparent that slow dissociation rates define the interactions that result in efficient binding/sequestration of ligand in solution, that is, TGF-β2 with 1D11 and TGF-β1/3 with 1D11, (TβRII)2, and (TβRIIb)2.

Neutralization of TGF-β1, TGF-β2, and TGF-β3 by (TβRII)2 and (TβRIIb)2 traps was compared with TβRII-Fc, TβRII-ED, and 1D11 using a cell-based TGF-β reporter assay (Fig. 3C and D, Supplementary Table S4). The average IC₅₀ values of (TβRII)2 for TGF-β3 and TGF-β1 (∼0.3 and ∼1.4 nmol/L, respectively) were similar to those for TβRII-Fc (∼0.3 and ∼0.5 nmol/L for TGF-β3 and TGF-β1, respectively). The IC₅₀ values for (TβRIIb)2 were somewhat lower (∼0.05 and ∼0.1 nmol/L for TGF-β3 and TGF-β1, respectively) and similar to 1D11. In contrast, TβRII-ED was much less potent in neutralizing TGF-β1/3 (IC₅₀ values >100 nmol/L). Finally, TGF-β2 was neutralized by 1D11 but not by the receptor-based traps (Fig. 3D). Our results showed that linkage of 2 TβRII ectodomains with natural linker sequences enhances not only binding to TGF-β1 and TGF-β3 but also neutralization potency. However, the improvement in TGF-β2 binding (Fig. 2D) was not sufficient to enable neutralization of this TGF-β isoform.

Our molecular dynamic simulation of the (TβRII)2 trap indicated that the natural linker becomes rigid and may establish molecular interactions with the ligand (Supplementary Fig. S2A), which could contribute to an increased affinity and neutralizing potency of this trap. To test the effect of the linker sequence, we assessed a (TβRII)2 trap in which the 35 a.a. natural linker was replaced by a 35-a.a. artificial linker (composed of [8×GGGS]GGG). We observed an approximately 10-fold decrease of the IC₅₀ value for TGF-β1 neutralization for the artificial versus natural linker trap (13.3 and 1.4 nmol/L, respectively, Supplementary Fig. S3). This suggested that the natural linker in (TβRII)2 has properties that enhance trap affinity and efficacy.

For our in vivo studies, we focused on the (TβRII)2 trap as its sequence mimics the predominant TβRII isoform found in tissues, and it showed TGF-β–binding characteristics similar to (TβRIIb)2 (Figs. 2B–D, 3A). First, we assessed the susceptibility of (TβRII)2 to proteolytic degradation in serum. As shown in Fig. 4A, the (TβRII)2 trap showed no evidence of degradation and retained full neutralization potency after incubation in human serum at 37°C for up to 7 days. Although stable in serum, the size of the (TβRII)2 trap with glycosylation is slightly under the 60-kDa (approximately) molecular cut-off for kidney filtration, hence we anticipated a short circulation time in vivo. This was confirmed by pharmacokinetic studies that showed that (TβRII)2 has a half-life of less than 1 hour, predominantly because of kidney clearance and elimination in the urine (Fig. 4B–D). This is considerably shorter than antibodies, including 1D11 and the humanized equivalent GC1008 (fresolimumab), which have half-lives of several days in vivo (47).

We next tested the in vivo efficacy of the (TβRII)2 trap versus 1D11 antibody using the syngeneic 4T1 mammary tumor model in BALB/c mice. The tumor-inhibiting action of 1D11 has been well documented using this model (25, 26). 4T1 cells were implanted orthotopically into the mammary fat pad, and treatment was then initiated for a period of 2 weeks. To partially compensate for its short half-life, we increased the dose (10 mg/kg) and frequency (daily) of injection of the (TβRII)2 trap compared with 1D11 (5 mg/kg, thrice a week). Figure 5A shows that, compared with the saline controls, treatment with 1D11 and the trap slowed tumor growth by approximately 50%.

We also compared 1D11 and (TβRII)2 trap efficacies in mice with established 4T1 tumors and found that the trap, but not 1D11, significantly reduced further tumor growth (Fig. 6A). This difference may be due to better penetration into tumors of the smaller trap compared with 1D11 antibody. We also assessed lung metastases in these mice and found that both the trap and 1D11 seemed to reduce the number of lung nodules relative to controls (Fig. 6B). This may be due to the trap and 1D11 having access to disseminating 4T1 tumor cells, either within the blood or at the secondary site. Taken together, these studies showed primarily that, despite its rapid elimination in vivo, the (TβRII)2 trap reduces the growth of newly implanted or established primary tumors, indicating a remarkable potency for this trap.

To probe the mechanisms underlying the reduction of 4T1 tumor growth, we analyzed the effect of the (TβRII)2 trap on 4T1 cells in vitro and on tumors from the animals in which treatment was initiated the day after tumor cell implantation. In 4T1 cells, the (TβRII)2 trap effectively antagonized TGF-β1 and TGF-β3 signaling and partially antagonized TGF-β2 signaling, as measured by expression of the TGF-β response marker PAI-1 (Supplementary Fig. S4B). Similarly, the (TβRII)2 trap inhibited TGF-β1- and TGF-β3–mediated Smad2 phosphorylation in 4T1 cells (Supplementary Fig. S4C). In contrast, we were unable to detect a significant decrease in PAI-1 levels in 4T1 tumors from mice treated with trap or 1D11 compared with the saline controls (Fig. 5B). Furthermore, Ki67 staining of the tumors did not reveal any differences in cell proliferation (Supplementary Fig. S5). This was expected because 4T1 cells are not growth inhibited in vitro by any of the TGF-β isoforms (Supplementary Fig. S4A). These results therefore suggested that the observed tumor growth reduction is not due to inhibition of TGF-β signaling within tumor cells or direct effects on tumor cell proliferation, but may be due to TGF-β neutralization effects on other cell types in the tumor microenvironment. In support of this, we observed a significant increase in infiltrating T cells (as detected by CD3 staining) in tumors of both trap- and 1D11-treated mice (Fig. 5C). This correlated with increased granzyme B (a cytotoxic T-cell effector molecule) and Rae-1γ (an immune cell activator) mRNA levels, particularly in tumors from trap-treated mice (Fig. 5B). In addition, trap- and 1D11-treated tumors exhibited a marked reduction in blood vessel density (Fig. 5D). Together these results indicated that, similar to 1D11, the (TβRII)2 trap inhibits tumor growth by reversing TGF-β–mediated immune suppression and angiogenesis. Overall, these results showed the promising potential of the (TβRII)2 trap to act as a therapeutic addressing TGF-β–driven diseases.

Here we present a novel design strategy for the generation of single-chain, bivalent receptor ectodomain-based traps that neutralize members of the TGF-β superfamily of ligands. Our approach exploits the presence of IDRs that flank either side of the structured ligand-binding domain, a feature shared by the receptor ectodomains from this family. Fusion of the C-terminal IDR of one ectodomain with the N-terminal IDR of the second ectodomain, thus, forms a bivalent trap that has 2 structured binding domains connected by a natural flexible linker.

This design strategy was supported by empirical estimates based on known atomic distances between 2 receptor domains when simultaneously bound to ligand and which indicated that natural flexible linkers formed by joining receptor IDRs should have sufficient length to span over the ligand dimer (Supplementary Table S2). In the simplest case, affinity enhancement can be achieved if the linkers act purely as flexible polypeptides that constrain the distance between 2 receptor-binding domains, thereby increasing the effective concentration (C_eff) of the second binding domain (48, 49). However, our results show that, when linkers are composed of natural IDR sequences, they may contribute to enhancement of the affinity of the trap and neutralization efficacy through additional effects beyond simply acting as a leash. The first evidence alluding to this is provided by the molecular dynamic simulation of the (TβRII)2 trap, which indicated that, upon binding TGF-β, the natural linker does not remain free in space but rather becomes rigid, establishing potential molecular interactions with the ligand (Supplementary Fig. S2A). This notion is supported by our data showing that the replacement of the natural linker of the (TβRII)2 trap by an artificial linker caused a 10-fold decrease in TGF-β1 neutralization efficacy (Supplementary Fig. S3). These results suggest that IDR-based linkers may take on a folded conformation that is adapted to interact with its target, a well-recognized capability of intrinsically disordered proteins (50). These findings highlight a select advantage of using a linker composed of natural IDR sequences.

As a demonstration of the general applicability of this IDR-linker design strategy, in addition to TGF-β, we designed single-chain traps for other TGF-β superfamily members, for example, BMP and activin (Supplementary Fig. S1). To confirm affinity enhancement, we produced a BMPR1a receptor-based trap, termed (BMPR1a)2, and determined that it neutralized BMP-2 more effectively than monovalent BMPR1a-ED (Supplementary Fig. S6). Taken together, our data indicate that a receptor trap design strategy that uses linkers composed of natural IDRs provides a generally applicable engineering strategy for effective traps for the TGF-β superfamily of ligands.

With respect to their ability to neutralize TGF-β1 and TGF-β3 in cell-based assays, the (TβRII)2 and (TβRIIb)2 traps behaved similarly to 1D11 antibody (Fig. 3C and Supplementary Table S4). However, in contrast to 1D11, these traps were unable to neutralize TGF-β2 (Fig. 3D). This distinction is consistent with the fact that these traps display faster TGF-β2 dissociation rates (Fig. 2D) and lower potencies in a competitive binding assay (Fig. 3B), as compared with 1D11.

In vivo we observed that the (TβRII)2 trap and 1D11 reduced the growth of 4T1 tumors to a similar extent when treatment was initiated the day following tumor cell implantation (Fig. 5A). This implies that the TGF-β1 and/or TGF-β3 isoforms play a greater role in tumor growth than TGF-β2 in this tumor model. Importantly, these results show that the (TβRII)2 trap can reduce tumor growth as effectively as 1D11, despite its short circulating half-life (<1 hour). Augmentation of the half-life of the trap, for example, by PEGylation, may improve its in vivo potency further. On the other hand, rapid clearance of the (TβRII)2 trap from the circulation may prove advantageous for disease indications where short-term TGF-β neutralization is preferred.

Our analysis of the mechanisms underlying in vivo efficacy indicate that the (TβRII)2 trap enhanced a T-cell–mediated immune response in the tumor microenvironment and reduced blood vessel formation, which concurs with the study of Nam and colleagues (25) using 1D11. Notably, although the effect of the (TβRII)2 trap and 1D11 on the growth of newly implanted 4T1 tumors was not detectably different, we observed that the T-cell activity markers granzyme B and Rae-1γ were augmented more significantly in tumors from trap-treated compared with 1D11-treated mice (Fig. 5B). As further evidence of the higher efficacy of the trap in animals with established 4T1 tumors, we show that the (TβRII)2 trap, but not 1D11 antibody, reduced primary tumor growth (Fig. 6A). One explanation is that the smaller (TβRII)2 trap may better penetrate established tumors compared with 1D11 antibody, providing a potential therapeutic advantage.

In summary, we present a novel approach for the design of single-chain bivalent receptor traps for TGF-β family ligands. In particular, we show the neutralization potency, serum stability, and in vivo efficacy of selected TGF-β traps. Several characteristics make these traps attractive candidates for therapeutic or diagnostic applications. These advantages include (i) neutralization potencies in the nmol/L to sub-mol/L range, (ii) a single chain nature for ease of production and further engineering (i.e., fusion, PEGylation), and (iii) natural sequence compositions reducing potential immunogenicity.

No potential conflicts of interest were disclosed.

Conception and design: J.C. Zwaagstra, T. Sulea, C. Collins, M.D. O'Connor-McCourt

Development of methodology: J.C. Zwaagstra, J. Baardsnes, R. Tom, Y. Durocher

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.C. Zwaagstra, A.E.G. Lenferink, C. Collins, C. Cantin, S. Grothe, S. Hossain, L.-P. Richer, B. Cass

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.C. Zwaagstra, T. Sulea, J. Baardsnes, A.E.G. Lenferink, C. Cantin, S. Hossain, L.-P. Richer, B. Cass, M.D. O'Connor-McCourt

Writing, review, and/or revision of the manuscript: J.C. Zwaagstra, T. Sulea, J. Baardsnes, A.E.G. Lenferink, M.D. O'Connor-McCourt

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.C. Zwaagstra, A.E.G. Lenferink, C. Collins, B. Paul-Roc, D. L'Abbé, R. Tom, M.D. O'Connor-McCourt

Study supervision: J.C. Zwaagstra, A.E.G. Lenferink, S. Hossain, Y. Durocher, M.D. O'Connor-McCourt

The authors thank Alina Burlacu for help in trap production and Mario Mercier and Cynthia Hélie for their assistance in animal studies as well as Genzyme Inc. (Boston, MA) for supplying the 1D11 antibody used in animal studies.

Research funds were provided by the Genome Heath Initiative Canada.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.