Functionally diverse heteromeric traps for ligands of the transforming growth factor-β superfamily

Generation of selective heterodimeric ligand traps

We generated ligand-trapping fusion proteins in Chinese hamster ovary (CHO) cells by expression of ECD-Fc polypeptide chains, which become linked covalently by disulfide bonds between the Fc domains to form stable dimers with antibody-like pharmacokinetic properties²⁴. Whereas homodimeric fusion proteins of this type comprise a pair of identical polypeptide chains (Fig. 2A, IIB-Fc:IIB-Fc), heterodimeric constructs require preferential pairing of different ECD chains. We therefore engineered differentially charged amino acid substitutions into the Fc domains of intended polypeptide pairs to produce complementary electrostatic interactions that favor heterodimeric chain pairing over homodimeric pairing^25,26,27. In addition, a hexameric histidine tag was included at the carboxy terminus of the type I receptor chain in each heterodimeric construct to facilitate its purification (Fig. 2A, H6 Tag). In this manner, we produced a set of three heterodimeric fusion proteins (IIB-Fc:ALK3-Fc, IIB-Fc:ALK4-Fc, and IIB-Fc:ALK7-Fc) in which a human ActRIIB ECD-Fc fusion was alternately paired with polypeptide chains incorporating the ECDs from human ALK3 (BMPR1A), human ALK4 (ACVR1B), or human ALK7 (ACVR1C), respectively (Fig. 2A). For comparison, we also produced the analogous homodimers IIB-Fc:IIB-Fc, ALK3-Fc:ALK3-Fc, ALK4-Fc:ALK4-Fc, and ALK7-Fc:ALK7-Fc.

Profiles of ligand dissociation kinetics for heterodimeric ligand traps vary with type I receptor component. (A) Domain schematics depicting variants based on naturally occurring pairings of TGF-β superfamily type I receptor ECDs with ActRIIB ECD. The Fc domains of heterodimeric fusion proteins are modified (Mod) to promote desired pairing of monomers during cellular production. H6, histidine hexamer. (B–D) Ligand sequestration profiles for ActRIIB-Fc-based heterodimeric traps vary markedly with the identity of the partnered type I receptor ECD. Semi-schematic graphs depict the change in off-rate (k_d) of key ligands, as determined by surface plasmon resonance (see Table 1 for exact values), when one ECD in the IIB-Fc:IIB-Fc homodimer is replaced with ALK4 (B), ALK7 (C), or ALK3 (D).

Ligand binding profiles of selective heterodimeric ligand traps in vitro

We used surface plasmon resonance (SPR) analysis to systematically measure the ligand binding kinetics of these homodimeric and heterodimeric fusion proteins based on a panel of 11 homodimeric ligands from the TGF-β superfamily (Table 1, Supplemental Fig. S1, Supplemental Table S1). Ligands selected for analysis included five that activate SMAD2/3 signaling (activin A, activin B, GDF3, GDF8, and GDF11) and six that activate SMAD1/5/8 signaling (BMP2, BMP4, BMP6, BMP7, BMP9, and BMP10)^4,28. The eponymous superfamily ligands TGF-β1, TGF-β2, and TGF-β3 were notably excluded from analysis as they are thought to bind predominantly, if not exclusively, to TGFBRII and ALK5 but not to the receptors tested here^6,10,29.

Characterization of ligand-binding kinetics revealed that the three heterodimeric constructs differed markedly from the IIB-Fc:IIB-Fc homodimer and among themselves. Kinetic parameters for these constructs as well as homodimeric counterparts based on ALK3, ALK4, and ALK7 are shown in Table 1, and Fig. 2 depicts pronounced differences among the heterodimeric constructs regarding the ligand dissociation rate constant (k_d), or off-rate, an important parameter describing duration of ligand sequestration. A useful way to interpret the diagrams in Fig. 2B–D is by the overall pattern of lines depicting change in off-rate for the ligands, and for a given comparison between IIB-Fc:IIB-Fc and heterodimer the number of line intersections provides an indication of the degree to which the off-rate profile is altered by ECD replacement.

Based on off-rates obtained by SPR analysis, the heterodimeric construct containing an ALK4 ECD (IIB-Fc:ALK4-Fc) retained relatively strong binding to most SMAD2/3-pathway ligands but showed markedly reduced preference for BMP4, BMP9, BMP10, and GDF3 compared with the parent ActRIIB-Fc homodimer (Fig. 2B). This heterodimeric construct was a dramatically more effective trap than the corresponding ALK4-Fc:ALK4-Fc homodimer (Table 1), consistent with the relatively weak binding by activin-class ligands to their cognate type I receptors without an accompanying type II receptor^30,31.

The heterodimer containing an ALK7 ECD (IIB-Fc:ALK7-Fc) displayed an off-rate profile resembling features of the IIB-Fc:ALK4-Fc profile. Like IIB-Fc:ALK4-Fc, IIB-Fc:ALK7-Fc retained the overall preference for SMAD2/3-pathway ligands of its parent homodimeric construct, IIB-Fc:IIB-Fc. With the exception of activin A, the rank order of ligand off-rates remained largely the same for IIB-Fc:ALK7-Fc as for IIB-Fc:IIB-Fc, but most off-rate values were increased approximately tenfold, thereby lessening the sequestration effectiveness for the weaker-binding ligands in particular. These included BMP4, BMP9, BMP10, and GDF3, but especially activin A, which displayed a disproportionately large increase in off-rate (approximately 100-fold) compared with IIB-Fc:IIB-Fc (Fig. 2C). As expected, the IIB-Fc:ALK7-Fc heterodimeric construct was a much more effective trap than the corresponding type I homodimeric construct, ALK7-Fc:ALK7-Fc (Table 1).

In marked contrast, the heterodimeric construct containing ALK3—a key regulator of SMAD1/5/8 signaling—bound with greatly increased strength to BMP2, BMP4, BMP6, and BMP7 but with lower strength to the cohort of SMAD2/3-pathway ligands (Fig. 2D). For all three heterodimeric ligand traps, BMP9 and BMP10 exhibited faster off-rates (Fig. 2B–D), consistent with the observation that these ligands signal primarily through receptor complexes containing ALK1 as the type I receptor³² and not through ALK4, ALK7, or ALK3. Together, these SPR data demonstrate that each combination of receptor ECDs exhibits a distinct binding profile in vitro, and by replacing one arm of the ActRIIB homodimer with a type I receptor ECD it is possible to modify binding properties to preferentially sequester different subsets of TGF-β superfamily ligands.

Inhibitory potency of selective heterodimeric ligand traps in vitro

We next investigated whether the altered ligand binding profiles observed for the heterodimeric ligand traps in SPR assays translate to novel patterns of signal inhibition in a cell-based reporter gene assay. The three heterodimeric and four homodimeric trap constructs were evaluated for their ability to inhibit signal transduction initiated by four SMAD2/3-pathway ligands in the activin class (activin A, activin B, GDF8, and GDF11) and two SMAD1/5/8-pathway ligands (BMP9 and BMP10). The SMAD2/3-pathway ligands were tested using A204 cells transfected with a SMAD2/3-responsive luciferase reporter plasmid (utilizing a CAGA12 promoter³³), whereas the SMAD1/5/8 ligands were tested using T98G cells transfected with a SMAD1/5/8-responsive reporter (utilizing a BRE promoter³⁴).

Table 2 lists mean values for half-maximal inhibitory concentration (IC₅₀) determined in these assays and Fig. 3 depicts representative inhibition curves for each ligand-construct combination for which an IC₅₀ value could be determined. The three heterodimeric constructs inhibited signaling by BMP9 and BMP10 with similarly reduced potency compared with the IIB-Fc:IIB-Fc homodimer, with reductions of approximately 15-to-54-fold for BMP9 and 23-to-67-fold for BMP10, but exhibited diverse signal inhibition patterns for the activin-class ligands. For activin A inhibition, IIB-Fc:ALK4-Fc (IC₅₀ = 0.19 ± 0.04 nM) displayed potency comparable to that of IIB-Fc:IIB-Fc (IC₅₀ = 0.11 ± 0.01 nM), whereas IIB-Fc:ALK7-Fc (IC₅₀ = 2.86 ± 0.60 nM) and IIB-Fc:ALK3-Fc (IC₅₀ = 2.62 ± 0.15 nM) showed reduced inhibitory potencies consistent with their faster ligand off-rate profiles determined by SPR (Fig. 2C). For activin B inhibition, IIB-Fc:ALK3-Fc (IC₅₀ = 1.65 ± 0.02 nM) exhibited the greatest reduction in potency compared with IIB-Fc:IIB-Fc (IC₅₀ = 0.06 ± 0.01 nM), whereas IIB-Fc:ALK4-Fc (IC₅₀ = 0.16 ± 0.01 nM) and IIB-Fc:ALK7-Fc (IC₅₀ = 0.26 ± 0.05 nM) constructs were intermediate in potency. The inhibition profiles of GDF8 and GDF11 resembled each other. In each case, IIB-Fc:IIB-Fc was the most potent inhibitor, IIB-Fc:ALK7-Fc and IIB-Fc:ALK3-Fc were least potent, and IIB-Fc:ALK4-Fc was intermediate in potency. As expected, the homodimeric type I constructs were uniformly poor inhibitors of signaling by these six ligands in the reporter gene assay (Table 2).

The type I receptor component in heterodimeric traps impacts signaling activity in a cellular context in vitro. Representative curves from assays conducted with a pSMAD2/3-responsive luciferase reporter gene in A204 cells (activin A, activin B, GDF8, and GDF11) or a pSMAD1/5/8-responsive luciferase reporter gene in T98G cells (BMP9 and BMP10) are depicted. Data from three independent experiments were used to determine mean values for half-maximal inhibitory concentration (IC₅₀) shown in Table 2. Signaling level of 100% was defined as reporter activity for cells treated with ligand alone and 0% as reporter activity for cells treated with media alone. Values above or below 100% at the lowest ligand trap concentrations therefore reflect experimental variability observed for this representative replicate.

Together, these cell-based data indicate that each heterodimeric combination of receptor ECDs exhibits a novel pattern of signal inhibition when compared among themselves and with their corresponding homodimeric constructs. Furthermore, the directionality of IC₅₀ shifts observed by this complementary assay was consistent with the heterodimer-specific ligand binding profiles obtained by SPR, particularly the increased off-rates observed for the tested ligands.

Pharmacological activity profiles of selective heterodimeric ligand traps in vivo

We characterized the activity of systemically administered heterodimers (IIB-Fc:ALK4-Fc, IIB-Fc:ALK7-Fc, and IIB-Fc:ALK3-Fc) in wild-type C57BL/6 mice in comparison with the IIB-Fc:IIB-Fc homodimer. For mice receiving each individual ligand trap, we measured total body weight, skeletal muscle weight, body fat as a percentage of body weight, and bone mineral density—parameters known to be regulated by TGF-β superfamily signaling^1,2,3. Drug or vehicle control (phosphate buffered saline, PBS) were administered by subcutaneous (s.c.) injection twice weekly for 28 days. To assess the biologic effects of treatment on multiple tissue composition, body fat and bone mineral density were measured by nuclear magnetic resonance (NMR) imaging and dual-energy x-ray absorptiometry (DXA), respectively, at baseline (day -1) and day 27. The gastrocnemius muscle, which responds strongly to manipulation of TGF-β superfamily signaling^8,35, was chosen as a representative muscle and weighed at day 28. Control experiments indicated that replacement of a human Fc domain with its murine counterpart does not affect activity of a IIB-Fc:IIB-Fc homodimeric construct tested in mice for 8 weeks (Supplemental Fig. S2).

In wild-type mice, the IIB-Fc:ALK4-Fc heterodimer displayed an activity profile similar to that of IIB-Fc:IIB-Fc, which is known to alter body composition³⁶. Specifically, treatment with IIB-Fc:ALK4-Fc produced a dose-related increase in body weight (at 10 mg/kg: 41.55 ± 2.01, P < 0.001) comparable in magnitude to the effect of the IIB-Fc:IIB-Fc homodimer (38.47 ± 1.49, P < 0.001) (Fig. 4A). Similarly, IIB-Fc:ALK4-Fc treatment caused a significant increase in gastrocnemius weight (at 10 mg/kg: 234.4 ± 9, P < 0.001) and a significant reduction in body fat (at 10 mg/kg: − 3.34 ± 0.45, P < 0.001) comparable in magnitude to changes caused by IIB-Fc:IIB-Fc (gastrocnemius weight: 230.6 ± 8.66, P < 0.001; body fat: − 3.51 ± 0.33, P < 0.001) (Fig. 4B,C), consistent with previous observations^35,37. These tissue alterations in the presence of IIB-Fc:ALK4-Fc are likely achieved, at least in part, through hypertrophy of individual muscle fibers, and have been reported to accompany enhanced muscle twitch force³⁵. Interestingly, the selectively diminished binding to BMP9 and BMP10 by IIB-Fc:ALK4-Fc (Fig. 2B, Table 1) did not appreciably alter its effects on these endpoints. This result could argue against major roles of these BMPs on homeostasis of skeletal muscle or adipose tissue under the conditions tested, although BMP9 has been associated with hepatic function, insulin resistance, obesity, and regulation of energy balance^{38,39,40,41,42}. An important distinction between the two fusion proteins emerged when comparing bone mineral density, as treatment with IIB-Fc:ALK4-Fc did not produce a significant increase in this parameter when compared with vehicle (at 10 mg/kg: 11.61 ± 0.94, P = 0.32), whereas IIB-Fc:IIB-Fc did (14.29 ± 1.69, P = 0.02) (Fig. 4D). Overall, the similar effects of IIB-Fc:ALK4-Fc heterodimer and IIB-Fc:IIB-Fc homodimer on body weight and body composition accord well with the shared high affinities of these agents for the SMAD2/3-pathway ligands activin A, activin B, GDF8, and GDF11 (Fig. 2B, Table 1).

Activity of IIB-Fc:ALK4-Fc heterodimer in wild-type mice. Eight-week-old wild-type C57BL/6 mice (n = 9 per group) were injected s.c. with IIB-Fc:IIB-Fc (10 mg/kg), IIB-Fc:ALK4-Fc (either 3 mg/kg or 10 mg/kg), or vehicle control (PBS) twice weekly for 28 days. (A) Percentage change in mouse total body weight from day -1 to day 28. (B) Weight of the gastrocnemius muscle isolated by dissection on day 28. (C) Percentage change in total fat mass normalized to body weight from day -1 to day 27 as assessed by NMR. (D) Percentage change in bone mineral density from day -1 to day 27 as assessed by DXA. Data are means ± SEM. See Supplemental Table S2 for associated statistics. Veh vehicle; mg/kg milligrams per kilogram; NS not significant. *P < 0.05, ***P < 0.001 vs. vehicle.

Based on SPR analysis, IIB-Fc:ALK7-Fc resembles IIB-Fc:ALK4-Fc in its ligand sequestration profile (Fig. 2C) but notably shows reduced sequestration of activin A, a selective change which could lead to differences in activity between the two agents in vivo. When evaluated in wild-type mice, IIB-Fc:ALK7-Fc produced a dose-dependent increase in body weight (at 10 mg/kg: 27.22 ± 1.42, P < 0.001) and gastrocnemius weight (at 10 mg/kg: 215.5 ± 7.03, P < 0.001) closely resembling effects of the IIB-Fc:IIB-Fc homodimer (body weight: 30.24 ± 1.07, P < 0.001; gastrocnemius weight: 247.4 ± 4.05, P < 0.001) (Fig. 5A,B). Notably, IIB-Fc:ALK7-Fc did not alter body fat percentage relative to vehicle (at 10 mg/kg: 0.48 ± 0.36, P = 0.89) (Fig. 5C), unlike either the IIB-Fc:IIB-Fc homodimer (− 1.7 ± 0.15, P < 0.01) (Fig. 5C) or the IIB-Fc:ALK4-Fc heterodimer (Fig. 4C), both of which reduced fat while increasing muscle weight. This result suggests that one or more ligands with altered affinity for IIB-Fc:ALK7-Fc compared with IIB-Fc:IIB-Fc play a role in promoting maintenance of fat mass under these conditions, which is intriguing given that ALK7 is implicated in adipose tissue homeostasis^43,44. Bone mineral density was not affected significantly by IIB-Fc:ALK7-Fc (at 10 mg/kg: 9.85 ± 2.03, P = 0.23) or IIB-Fc:IIB-Fc treatment (7.2 ± 1.25, P = 0.74) under these conditions, perhaps due to high intragroup variability (Fig. 5D). Together, these results indicate that treatment with IIB-Fc:ALK7-Fc evokes only a subset of the effects observed after treatment with the IIB-Fc:IIB-Fc homodimer, consistent with its ligand binding profile, and that a reduction in body fat percentage need not accompany increased body weight and muscle weight under the conditions examined.

Activity of IIB-Fc:ALK7-Fc and IIB-Fc:ALK3-Fc heterodimers in wild-type mice. Twelve-week-old wild-type C57BL/6 mice were injected s.c. with IIB-Fc:IIB-Fc (n = 8, 10 mg/kg), IIB-Fc:ALK7-Fc (n = 8, either 3 mg/kg or 10 mg/kg), IIB-Fc:ALK3-Fc (n = 7, 3 mg/kg or n = 8, 10 mg/kg), or vehicle control (n = 7, PBS) twice weekly for 28 days. (A) Percentage change in mouse total body weight from day -1 to day 28. (B) Weight of the gastrocnemius muscle isolated by dissection on day 28. (C) Percentage change in total fat mass normalized to body weight from day -1 to day 27 as assessed by NMR. (D) Percentage change in bone mineral density from day -1 to day 27 as assessed by DXA. Data are means ± SEM. See Supplemental Table S3 for associated statistics. Veh vehicle; mg/kg milligrams per kilogram; NS not significant. *P < 0.05, **P < 0.01, ***P < 0.001 vs. vehicle.

The IIB-Fc:ALK3-Fc heterodimer exhibits a ligand-binding profile markedly different from that of the IIB-Fc:IIB-Fc homodimer, including exceptionally strong binding to BMP2 and BMP4, increased binding to BMP6 and BMP7, and reduced binding to BMP9 and BMP10 (Fig. 2D). When evaluated in wild-type mice, IIB-Fc:ALK3-Fc produced a distinctive biologic activity profile as well. Specifically, IIB-Fc:ALK3-Fc caused a dose-dependent increase in total body weight (at 10 mg/kg: 18.17 ± 0.82, P < 0.001) that was approximately half the magnitude of the weight gain caused by IIB-Fc:IIB-Fc at the same dose level (30.24 ± 1.07, P < 0.001) (Fig. 5A). However, in striking contrast with the homodimer, IIB-Fc:ALK3-Fc increased rather than decreased body fat percentage at the 10 mg/kg dose (7.9 ± 0.74, P < 0.001) (Fig. 5C) without altering gastrocnemius weight (181.8 ± 3.36, P = 0.38) (Fig. 5B) or bone mineral density (10.03 ± 3.08, P = 0.21) (Fig. 5D). Interestingly, the activity profile of IIB-Fc:ALK3-Fc differed also from the ALK3-Fc homodimer, ALK3-Fc:ALK3-Fc, which altered neither body weight nor body fat percentage when tested under equivalent conditions at the same dose level (Supplemental Fig. S3). The heterodimeric construct IIB-Fc:ALK3-Fc thus possesses an activity profile in wild-type mice distinct from either of its corresponding homodimeric traps. These data furthermore suggest that IIB-Fc:ALK3-Fc increases body weight mainly through increased fat mass and not increased muscle mass, and imply that signaling by BMP2 or BMP4 could regulate adipose tissue homeostasis under the conditions tested.