Codon-optimized human Smad7 gene therapy enhances skeletal muscle mass and function in a murine model of Duchenne muscular dystrophy

TL;DR

A codon-optimized human Smad7 gene therapy improved muscle mass and function in a mouse model of Duchenne muscular dystrophy without causing harm.

Contribution

A codon-optimized human Smad7 gene therapy was developed and shown to enhance muscle function in a Duchenne muscular dystrophy model.

Findings

Codon-optimized human Smad7 (AVGN7.2) improved skeletal muscle hypertrophy and isometric torque in mice.

AVGN7.2 enhanced muscle mass and contractile function in Duchenne muscular dystrophy models without causing muscle degeneration.

The therapy's effects were comparable to dystrophin-targeting drugs, suggesting potential for combinatorial treatments.

Abstract

Commercial development of gene therapeutics often requires transitioning to human payload genes as initial proof-of-concept studies in animal models often use taxa-specific orthologs. Such transitions also provide opportunities to address potential secondary structure and immune-related subsequences as with human Smad7 cDNA, which was optimized by removing several repeats, potential hairpins and negative cis elements. Thermodynamic modeling at or above minimal free energy states revealed substantial improvements in secondary structure with fewer hairpins and improved diversity scores. Serotype 6 adeno-associated viral vectors with optimized human Smad7 (AVGN7.2) expression constructs were equally or more effective than those with wild-type mouse Smad7 in stimulating skeletal muscle hypertrophy and enhancing isometric torque of hind-limb dorsiflexor muscles in vivo. In murine models of Duchenne Muscular dystrophy, where deficits in muscle mass and disproportionate declines in force are pathognomonic, AVGN7.2 proportionally increased muscle mass and isometric torque while normalizing contractile kinetics. Such improvements occurred without deleterious impacts on serum creatine kinase, fibrosis or myofiber central nucleation. These data suggest that AVGN7.2 is capable of enhancing dystrophic muscle function without exacerbating muscle degeneration. Although these functional effects were partial, they resembled those of several dystrophin-targeting drugs and suggest that combinatorial approaches may safely yield further benefit.

Figures

Codon optimization of hSmad7.A Optimization of the hSmad7 cDNA sequence variant 1 (GenBank [NM_005904.3](NM_005904.3)) was performed using three independent algorithms. The output sequences were then used to generate the final codon-optimized hSmad7 cDNA consensus sequence. The wild-type (blue) and codon-optimized (orange) sequences were then compared to reference genes expressed in human skeletal muscle to calculate the frequency of conserved codons across the coding sequence. B The percent distribution of codons in computed codon quality groups; higher is better as 100 represents the highest

Wild-type and codon-optimized Smad7 similarly increase muscle mass.Three rAAV6 vectors were generated, each containing either the wild-type mouse (m) or codon-optimized hSmad7 cDNA (h) paired with the CMV promoter or the muscle-specific CK8 promoter/regulatory domain [[17](#CR17)], producing the following vectors: rAAV6:CMV-mSmad7 (CMV-m), rAAV6:CMV-hSmad7 (CMV-h) and rAAV6:CK8-hSmad7 (CK8-h). Tibialis anterior (TA) muscles of 2 m.o. mice were then injected once with 1 × 10^10^ vg of CMV-m, CMV-h or CK8-h. For each mouse, one TA was injected with vector while the contralateral TA was injected

Wild-type and codon-optimized Smad7 similarly increase muscle fiber size.Tibialis anterior (TA) muscles of 2 m.o. mice were injected once intramuscularly with 1 × 10^10^ vg of rAAV6:CMV-mSmad7 (CMV-m), rAAV6:CMV-hSmad7 (CMV-h) or rAAV6:CK8-hSmad7 (CK8-h). For each mouse, one TA was injected with vector while the contralateral TA was injected with the same volume of saline. A Representative cross-sectional images. Nuclei were stained with DAPI (blue) and the sarcolemma with Alexa 647-labeled wheat germ agglutinin (cyan). Colored letters in the panel labels are used to highlight vector differenc

AVGN7.2 (rAAV6:CK8-hSmad7) increases tibialis anterior mass and fiber size in *mdx* mice.Wild-type (WT) and *mdx* neonates were injected R.O. when 1 w.o. with 0, 5 × 10^12^ (5e12), 1.7 × 10^13^ (1.7e13) and 5 × 10^13^ (5e13) vg/kg body mass AVGN7. The mass (A–D) of the indicated muscles were determined after 12 weeks and fiber size (E–G) was quantified histologically by measuring minimal feret diameters (EDL, extensor digitorum longus). E Treatment-induced shift in the fiber size distribution of TA muscles from control and high-dose groups, also reflected in overall mean fiber sizes (F) and in

AVGN7.2 does not exacerbate dystrophic histopathologies.Wild-type (WT) and *mdx* neonates were injected R.O. when 1 w.o. with 0, 5 × 10^12^ (5e12), 1.7 × 10^13^ (1.7e13) and 5 × 10^13^ (5e13) vg/kg body mass AVGN7 and terminated after 12 weeks. A Relative levels (mean ∓ SEM) of serum creatine kinase (units/L) and B muscle fibrosis obtained from Masson’s trichrome stained TA sections. In B, areas of blue stained collagen were normalized to the total area imaged. C Muscle central nucleation (% of fibers with 1 or more central nuclei) as a marker of compensatory muscle regeneration due to enhance

AVGN7 enhances the function of dystrophic muscle.Wild-type (WT) and *mdx* neonates were injected R.O. when 1 w.o. with 0, 5 × 10^12^ (5e12), 1.7 × 10^13^ (1.7e13) and 5 × 10^13^ (5e13) vg/kg body mass rAAV6:CK8-hSmad7. Plantarflexion force/torque was quantified after 12 weeks using an in vivo whole-animal murine system with increasing stimulation frequencies. A Total force generating capacity (mean ∓ SEM) generated at each frequency. Dashed lines frame the physiologically relevant range of stimulation frequencies and significant differences (ANOVAs) between *mdx* control (0) and *mdx* 5e13 gro

Full text

Introduction

Translational proof-of-concept studies for gene therapeutics often match the payload gene sequence to the preclinical animal model as this avoids immune responses that could compromise drug efficacy. For the same reason, human sequences are eventually incorporated before performing preclinical safety and toxicology studies that assess the final drug product. “Bridging” studies are then needed to demonstrate comparability as regulatory agencies require that any change to the final drug product (e.g., modifying the capsid, replacing the promoter/gene regulator cassette, incorporating the human sequence, etc.) produce similar or superior efficacy if data from earlier studies are incorporated into regulatory filings. This transition to the human sequence also provides an opportunity to improve drug design with codon optimization, which has long been integral to the commercial production of therapeutic proteins and is now being used when designing mRNA and gene therapeutics [1].

Codon optimization is founded upon the fundamental genetic principal of codon usage bias, the preferential use of synonymous codons that varies phylogenetically, among gene groups and between tissues with different tRNA expression profiles [2, 3]. Optimizing a particular transcript’s sequence can eliminate problematic mRNA secondary structure, improve translational efficiency and significantly increase recombinant protein production or, with gene therapy, protein levels of the encoded payload gene. Such improvements can substantially impact clinical use of gene therapeutics by reducing minimal effective doses. This is particularly important when using viral vectors as lowering titers similarly lowers the potential risk for off-target or immune-related toxicities. It also reduces the manufacturing burden for approaches utilizing systemic administration as manufacturing capacity is often a limiting factor [4].

AVGN7 is a serotype 6 adeno-associated viral (AAV6) gene therapeutic being developed for treating muscle wasting diseases, including muscular dystrophies and myopathies [5]. The AAV6 capsid is muscle-trophic, not muscle-specific, yet it displays a much better muscle/non-muscle distribution ratio than the other muscle-trophic capsids (e.g., AAV1, AAV8 & AAV9) that actually demonstrate preference for the liver, especially when compared to AAV6 [6–9]. Even the recently described MyoAAV bioaccumulates in the liver at levels that are 50- to 250-times greater than in any muscle and directs liver transgene expression comparable to AAV9 and AAVrh74 [9]. The AAV6 capsid was, therefore, chosen as high doses of other muscle-trophic AAVs have been associated with liver toxicities (e.g., elevated circulating liver enzymes, cholestatic liver failure, hepatocellular liver failure, thrombocytopenia & complement activation) and death [10, 11].

In mice, AVGN7 effectively prevents muscle wasting with either local or systemic administration and its mechanisms of action have been thoroughly described [12, 13]. By overexpressing Smad7, it attenuates activin receptors (ActRII & ActRIIb) and Smad2/3 signaling regardless of the multiple activating ligands that include myostatin, growth/differentiating factor 11, the activins (ActA, ActB & ActA/B) and others [5, 14, 15]. The functional impact of AVGN7 is independent of muscle twitch type [13] and yields improvements across several established metrics including direct measures of muscle function ex vivo (i.e., maximal tetanic twitch force) and in vivo assessments of grip strength, forced running distance, exercise capacity and cardiac performance [12, 13]. Most importantly, AVGN7 prevents muscle wasting in different murine models of catabolic insult such as diseases with elevated inflammatory cytokines and in experiments with direct cytokine challenge or constitutively active Smad3 overexpression [13]. Such broad applicability is rooted in its ability to increase muscle protein synthesis and to suppress expression of the muscle-specific E3 ubiquitin ligases MuRF1 and MAFbx/atrogin-1 that mediate muscle protein degradation [5, 13]. Its hypertrophic actions are limited, however, to regulating muscle fiber protein synthesis and degradation as satellite (a.k.a. stem) cells are not amenable to AAV-transduction [16].

Studies to date with AVGN7 (rAAV6:CMV-mSmad7) have been performed using the CMV promoter and the wild-type mouse ortholog (mSmad7) as the payload gene. Here we report bridging studies that incorporated a codon-optimized human ortholog (hSmad7) and the muscle-specific CK8 promoter into the drug design, generating AVGN7.2 (rAAV6:CK8-hSmad7). Our results demonstrate effective muscle hypertrophy, a conserved action with local and systemic delivery in wild-type mice. We further show systemic AVGN7.2 to effectively increase muscle mass and function in a murine model of Duchenne muscular dystrophy (DMD), notably without any exacerbation of the disease phenotype. This latter finding supports development of AVGN7.2 as a stand-alone or combinatorial therapeutic for DMD and possibly other muscular dystrophies.

Materials and methods

Codon optimization

Optimization of the human (h) Smad7 cDNA sequence variant 1 (GenBank NM_005904.3) was performed using three independent commercial algorithms at GenScript (Piscataway, NJ), Synbio Technologies (Monmouth Junction, NJ) and Biomatik (Wilmington, DE). These first-generation sequences were then aligned using the CLUSTALW algorithm and MacVector in order to generate a consensus sequence. Ambiguities were resolved by majority rule to produce the final codon-optimized hSmad7 cDNA sequence that was then compared to the first-generation sequences for subsequence analysis and, using the GenScript algorithm, to human reference genes. Secondary structure of the hSmad7 mRNA was modeled using computational software on the RNAfold Webserver (Institute for Theoretical Chemistry, University of Vienna). Hairpin motifs were separately identified from minimal free energy (MFE) and thermodynamic predictions, generating dot-bracket notations for each prediction. Secondary structures were then separately modeled from these predictions to produce a single MFE structure for each sequence and a centroid structure generated from multiple sub-optimal predicted foldings using increments above MFE. Nucleotide positions and base pairings were constructed according to probabilities, which were calculated from positional entropies.

Vector synthesis

The hSmad7 cDNA sequence was synthesized by GenScript and subcloned into a proprietary transfer plasmid with AAV2 genomes and inverted terminal repeat sequences. Two plasmids were constructed with different gene promoter/regulatory domains, one with the cytomegalovirus (CMV) promoter and another with the muscle-specific CK8 regulatory domain [17, 18]. These plasmids and an equivalent with the wild-type mSmad7 sequence were then used to manufacture three different recombinant vectors, rAAV6:CMV-mSmad7 (CMV-mSmad7), rAAV6:CMV-hSmad7 (CMV-hSmad7) and rAAV6:CK8-hSmad7 (CK8-hSmad7), in the Research Vector Core of the Children’s Hospital of Philadelphia. Empty capsids were separated from full particles using CsCl density-gradient centrifugation and each vector was further purified using proprietary methods. Quality control assays included proprietary bioburden and endotoxin assays, genome titers were determined using a proprietary qPCR assay and sterile vector was supplied in PBS with 0.001% Pluronic F-68.

Animal experiments

All animal experiments were performed by Myologica LLC at the University of Maryland Baltimore Veterinary Resources. Experiments were conducted in accordance to animal use protocols preapproved by the university’s institutional animal care and use committee and according to National Institutes of Health guidelines. Wild-type C57BL/6 and C57BL/10ScSnJ as well as dystrophic C57BL/10ScSn-Dmd^mdx^*/*J (a.k.a. mdx) mice were obtained from Jackson Laboratory (Bar Harbor, MN). Animals were randomly assigned to treatment groups during the acclimation phase and after a physical exam. They were included or excluded based on veterinary assessment of body condition and health. Technicians performing the studies were blinded to the treatments.

For local vector delivery, mice were anesthetized with isoflurane before making a single IM injection of 1 × 10^9^ or 1 × 10^10^ vg. Vectors were diluted in Hank’s buffered saline solution (HBSS) and directly injected into the tibialis anterior (TA) muscles. Needles were inserted in the caudal end of the muscle, just above the tendon moving up to the cranial. Injections began as the needle was slowly withdrawn, pausing occasionally to allow the solution (20 μl/muscle) to be absorbed. Control-injections of 20 μl carrier were also performed on the contralateral TA of each mouse while mice in the control group received saline injections in only 1 TA. For systemic delivery studies with neonates, different doses of AVGN7 (5 × 10^12^, 1.7 × 10^13^ and 5 × 10^13^ vg/kg) in a 10 μl volume of HBSS was administered retro-orbital (R.O.) in 1 w.o. neonates. Higher doses and dose volumes could not be administered due to the small size of neonates. Muscle function was then assessed after 8 or 12 weeks, as described below, and all mice were eventually killed via thoracotomy under deep isoflurane anesthesia before excising, weighing and processing muscles.

Histology

Muscles in OCT cryoprotectant were rapidly frozen in dry ice-cooled isopentane and subsequently cryosectioned at 10 µm thickness starting at the muscle mid-belly. Sections were fixed with 4% paraformaldehyde then stained with Alexa 647-labeled wheat germ agglutinin (Invitrogen/ThermoFisher Scientific, Waltham, MA) to identify sarcolemmal boundaries and mounted using ProLong Diamond Antifade Mountant with DAPI (Invitrogen/Fisher Scientific, Waltham, MA). Sections were then imaged using a Nikon (Melville, NY) Eclipse Ti2 microscope equipped with a Nikon DS-Qi2 monochrome camera and a Lumencor SpectraX light engine. Muscle fiber boundaries were automatically defined with predictive software and minimal Feret’s diameter and the presence of central nuclei were quantified (Nikon Elements v4.51). All fibers of entire sections were assessed to avoid regional and user bias. To assess muscle fibrosis, sections were stained with a Masson’s trichrome stain kit (Sigma-Aldrich, St. Louis, MO) according to manufacturer’s instructions. Fibrosis was then quantified by normalizing blue-stained connective tissue area to the total area imaged (Nikon Elements v4.51).

Western blotting

Tissues were homogenized in T-Per (ThermoFisher Scientific, Waltham, MA) tissue protein extraction reagent supplemented with EDTA-free Pierce Protease and Phosphatase mini-tablets (ThermoFisher Scientific, Waltham, MA) in a NextAdvance Bullet Blender at maximum speed with 0.5 mm zirconium oxide beads. Samples were clarified by centrifugation at 13,000 × g for 10 min at 4 °C and protein concentrations were determined using a Pierce protein assay kit (ThermoFisher Scientific, Waltham, MA). Protein fractions (40 mg/lane) were separated by SDS-PAGE using pre-cast 4–20% tris-glycine gels (Bio-Rad, Hercules, CA) and electrotransferred onto Immobilon-FL PVDF membranes (MilliporeSigma, Burlington, MA) that were subsequently blocked in Intercept buffer (LI-COR, Lincoln, NE) before incubating with primary antibodies (Abcam, Cambridge, UK) for MADH7/Smad7 (1:500, ab226872) and GAPDH (1:10 K, ab8245). The former polyclonal was generated against the first 50 amino acids of human Smad7, which is 100% similar and 96% identical to the mouse region. Membranes were then washed in TBST before probing with goat-anti-rabbit 800CW and goat-anti-mouse 680RD secondary antibodies (LI-COR). Positive immunodetections were obtained and quantified using an Odyssey DLx immager (LI-COR). When appropriate, Smad7 protein levels were normalized to those of GAPDH, both as optical density units.

Assessment of muscle function

Muscle function was quantified in vivo with a 305C muscle lever system (Aurora Scientific Inc., Aurora, CAN), as previously described [19, 20]. Animals were anesthetized via inhalation (~3% isoflurane, SomnoSuite, Kent Scientific) and placed on a thermostatically controlled table with anesthesia maintained via nose-cone (~2% isoflurane). The knee was stabilized with a pin against the tibial head and the foot firmly fixed to a footplate on the torque motor shaft. For assessment of plantarflexor function, contractions were elicited by percutaneous electrical stimulation of the tibial nerve. For assessment of dorsiflexor function, contractions were elicited by percutaneous electrical stimulation of the peroneal nerve. Optimal isometric twitch torque was determined by increasing the current with a minimum of 30 s between each contraction in order to avoid fatigue. A series of stimulations were then performed at increasing frequencies (0.2 ms pulse, 500 ms train duration) of 1, 20, 40, 60, 80, 100, and 150 Hz. Data were then analyzed using the manufacturer’s software.

Statistical design

Sample size was determined using power calculations with alpha of 0.05 and 1-beta of 0.8. Differences between means were determined by a Student’s t-test and by 1- or 2-way analysis of variance with Tukey’s or Fisher’s Least Significant Difference post hoc tests as appropriate. Correlation analyses were also performed to define relationships between different dependent variables and with all tests, significance was determined by p ≤ 0.05.

Results

Sequence analysis

Optimization significantly changed the overall sequence as wild-type and optimized hSmad7 are only 79% identical. Although the number of subsequence changes were minor (Table 1), several motifs capable of causing secondary structure problems were removed and include a direct repeat (GGAGGAGGAGGA) starting at positions 91 and 151, antiviral motifs and negative cis elements including a splice cite (587-ATCACC-592), a polyT site (651-TTTT-654) and a polyA site (946-AAAA-949). Optimization also improved the codon adaptation index (CAI), a measure of codon usage bias that equates to higher protein levels [3], as well as the frequency of conserved codons throughout (Fig. 1A). In fact, the percent distribution of codons in computed codon quality groups (100 represents the highest codon use) was also higher in the optimized sequence (Fig. 1B).Fig. 1. Codon optimization of hSmad7.A Optimization of the hSmad7 cDNA sequence variant 1 (GenBank NM_005904.3) was performed using three independent algorithms. The output sequences were then used to generate the final codon-optimized hSmad7 cDNA consensus sequence. The wild-type (blue) and codon-optimized (orange) sequences were then compared to reference genes expressed in human skeletal muscle to calculate the frequency of conserved codons across the coding sequence. B The percent distribution of codons in computed codon quality groups; higher is better as 100 represents the highest use frequency while codons below 30 can impair expression [57]. C Relative GC content plotted by position. Large peaks represent opportunities for 2° structure interference in translation. D Alignment of the wild-type and codon-optimized hSmad7 cDNA sequences using CLUSTALW. Sequences are named on the left, nucleotide positions are numbered on both sides and boxed regions reflect areas of 100% sequence identity. E & F Differences in the predicted secondary structure of wild-type (WT, E) and codon-optimized (F) hSmad7 mRNA. Hairpin motifs were separately identified from minimal free energy (MFE) and thermodynamic predictions using RNAfold computational software. Nucleotide positions and base pairings are color-coded according to probabilities, which were calculated from positional entropies. “Cool” colors (blue to green) represent lower probabilities, “warm” colors (yellow to red) represent higher probabilities (i.e., red = approaching 1.0).Table 1. Comparative summary of codon optimization.SequenceCAIGC%CFD%Antiviral motifsNeg cis elementsSTOPIdeal0.8–130–70<3000TAAWT0.7864433TAGOptimized0.9261.7000TAAWT, wild-type. CAI codon adaptation index: 0.8 is sufficient, 1.0 is perfect [57], GC guanine/cytosine, CFD codon frequency distribution of rare tandems, Neg cis direct or inverted elements with the potential to cause secondary structure, STOP translation-terminating stop codon.

These changes had a substantial impact on the projected secondary structure that often reflects pockets of high GC content. When compared by position, the relative GC content varied between sequences as large peaks in the first 450 bp, representing opportunities for secondary structures capable of reducing translational efficiency, were removed with optimization, although the overall GC content was only slightly reduced (Fig. 1C, D). Converting the optimized sequence to dot-bracket notation (data not shown) identified several hairpin motifs that were either removed or shortened. Modeling the MFE state identified 28 and 17 hairpins in the wild-type and codon-optimized sequences, respectively, for a net reduction of 11 hairpins with optimization. Thermodynamic modeling above MFE identified fewer hairpins as expected with a net reduction of 6 with optimization. Both models also computed secondary structures and revealed highly compact structures for wild-type hSmad7 with regions of low base-pairing probabilities (Fig. 1E, F). By contrast, the codon-optimized structures were less compact and contained mostly regions of high base-pairing probabilities. Such differences are reflected in diversity scores generated with thermodynamic modeling above MFE as the optimized score was 315.44 versus 439.34 for the wild-type sequence.

Comparable bioactivity of codon-optimized hSmad7

Intramuscular injections were performed using a contralateral control system and used to demonstrate comparable bioactivity of wild-type mSmad7 and optimized hSmad7. Three vectors were injected IM and include rAAV6:CMV-mSmad7, the previously developed therapeutic containing a wild-type mouse Smad7 cDNA, and rAAV6:CMV-hSmad7 containing codon-optimized hSmad7 cDNA, both of which utilized the CMV promoter. The third vector tested, rAAV6:CK8-hSmad7, also contained a codon-optimized hSmad7 cDNA as well as a promoter with high specificity and activity for striated muscle [17, 18].

Although each vector increased muscle mass after just 4 weeks, statistical significance was greatest with rAAV6:CK8-hSmad7, which produced the largest difference and was therefore the most effective (Fig. 2A, B). Suboptimal doses, based on studies using AAV6 capsids at much higher doses [21, 22], were used to avoid maximizing efficacy and to help distinguish differences between vectors. Notwithstanding, Smad7 protein levels were generally higher in muscles injected with rAAV6:CK8-hSmad7 (Fig. 2C) likely due to the use of the muscle-specific CK8 promoter while overall, the relative difference in mass between control and treated muscles was positively correlated to hSmad7 protein levels (Fig. 2D). These data together indicate that the combined use of a muscle-specific promoter and a codon-optimized hSmad7 cDNA sequence (i.e., rAAV6:CK8-hSmad7 or AVGN7.2) is more effective than using the ubiquitous CMV promoter and a non-optimized sequence (i.e., rAAV6:CMV-mSmad7).Fig. 2. Wild-type and codon-optimized Smad7 similarly increase muscle mass.Three rAAV6 vectors were generated, each containing either the wild-type mouse (m) or codon-optimized hSmad7 cDNA (h) paired with the CMV promoter or the muscle-specific CK8 promoter/regulatory domain [17], producing the following vectors: rAAV6:CMV-mSmad7 (CMV-m), rAAV6:CMV-hSmad7 (CMV-h) and rAAV6:CK8-hSmad7 (CK8-h). Tibialis anterior (TA) muscles of 2 m.o. mice were then injected once with 1 × 10^10^ vg of CMV-m, CMV-h or CK8-h. For each mouse, one TA was injected with vector while the contralateral TA was injected with the same volume of saline. After 4 weeks, the absolute mass of each muscle (A) as well as the relative difference between treated and control limbs (B) were calculated. hSmad7:GAPDH ratios were determined by western blotting (C) and correlated to differences in TA mass (D, r^2^ = 0.5124, p = 0.0008). Representative western results are inset of D (∓ = treatment). E–H TA muscles were injected with saline, 1 × 10^9^ (1E9) or 1 × 10^10^ (1E10) vg CK8-h (a.k.a. AVGN7.2). Control mice were injected with saline in only one TA while the other was not injected. Treated mice received either saline or AVGN7.2 using the contralateral control system. The mass of each muscle (E) and dorsiflexion force/torque (F) were then measured after 8 weeks. A–C, E–G Box plots display range, mean and individual values (n = 5–6). A–C Significant differences between control and treated groups are indicated by p levels (t-tests) and between any two groups in multiple group comparisons by different letters (ANOVAs, p < 0.05), whereas the same or shared letters indicate no significant difference. E–G Letters are again used to indicate differences between groups (ANOVAs) and asterisks (t-tests) from contralateral controls. H Force and mass relationship among treated muscles were correlated, regression analysis results are inset for each dose.

Muscle function was subsequently assessed in vivo using a footplate assay that quantifies dorsiflexion force/torque. After 8 weeks, both doses of rAAV6:CK8-hSmad7 increased muscle mass significantly with the higher dose producing the greatest effect (Fig. 2E). Both doses also increased dorsiflexion force over that of the contralateral control limbs and over the saline-injected limb of control mice (Fig. 2F). Note that treating with rAAV6:CK8-hSmad7 overcame the negative effects of injecting TAs as the forces of treated limbs were similar to those of the non-injected controls. The relative change compared to the contralateral limb of treated mice was ~40% with both doses, suggesting that the minimal effective dose with IM administration lies either below 1E9 vg or, given the higher variability with this dose, somewhere between 1E9 and 1E10 vg (Fig. 2G). This is supported by similarly positive force-by-mass relationships for both groups that were not different at lower doses (Fig. 2H).

When comparing efficacy of the different vectors, differences in muscle mass (Fig. 2A, B) were reflected in increased muscle fiber size as demonstrated qualitatively by sarcolemma and nuclear staining of muscle fibers (Fig. 3A). Differences were also noted when quantifying the total size of all fibers (Feret diameter, Fig. 3B), although a statistical difference was only noted with rAAV6:CK8-hSmad7. These results are meaningful as detecting statistical differences in global fiber size is quite difficult due to the normal disparity in range of fiber sizes. A more detailed analysis of fiber size distributions also revealed differences between control and treated muscles as well as between vectors. In fact, a rightward shift (fewer small fibers and/or more larger fibers) was detected with each treatment group, although differences in both size groups were only detected in mice expressing the codon-optimized hSmad7 (Fig. 3C–E). Direct comparisons of fiber size distributions between vectors (Fig. 3F, G) failed to detect significant differences yet the trend of fewer small fibers and more larger fibers was again apparent in mice injected with rAAV6:CK8-hSmad7. Future studies are needed to determine potential changes in fiber type distribution, although we previously documented AVGN7 to increase the size but not numbers of type IIa and IIx/b fibers in the TA [13].Fig. 3. Wild-type and codon-optimized Smad7 similarly increase muscle fiber size.Tibialis anterior (TA) muscles of 2 m.o. mice were injected once intramuscularly with 1 × 10^10^ vg of rAAV6:CMV-mSmad7 (CMV-m), rAAV6:CMV-hSmad7 (CMV-h) or rAAV6:CK8-hSmad7 (CK8-h). For each mouse, one TA was injected with vector while the contralateral TA was injected with the same volume of saline. A Representative cross-sectional images. Nuclei were stained with DAPI (blue) and the sarcolemma with Alexa 647-labeled wheat germ agglutinin (cyan). Colored letters in the panel labels are used to highlight vector differences. B Mean (±SEM, n = 3/group) muscle fiber size in control and treated muscles, determined by measuring minimum feret diameter. P levels (t-tests) for comparisons between control and treated means for each vector are indicated. C–E Comparisons of small (<20 μm), medium (20–70 μm) and large (>70 μm) fibers in control and treated muscles. Significant differences (t-tests, p < 0.05) within each binned size range (means ∓ SEM) are represented by asterisks. F Relative differences between control and treated muscles in the overall distribution of fiber sizes (minimal feret diameter ranges) for each vector. G The difference in fiber sizes between treated muscles and their respective controls. Mean (±SEM) values are plotted for each vector.

Enhancing dystrophic muscle mass

Male mdx mice were treated at 1 week of age with 5e12, 1.7e13, and 5e13 vg/kg rAAV6:CK8-hSmad7 and evaluated after 12 weeks. The two higher doses increased the mass of different muscle groups independent of twitch type as the slow twitch soleus and fast twitch EDL, TA and gastrocnemius muscles were all significantly larger with treatment (Fig. 4A–D). Even the lowest dose (5e12 vg/kg) was capable of significantly increasing soleus mass.Fig. 4AVGN7.2 (rAAV6:CK8-hSmad7) increases tibialis anterior mass and fiber size in mdx mice.Wild-type (WT) and mdx neonates were injected R.O. when 1 w.o. with 0, 5 × 10^12^ (5e12), 1.7 × 10^13^ (1.7e13) and 5 × 10^13^ (5e13) vg/kg body mass AVGN7. The mass (A–D) of the indicated muscles were determined after 12 weeks and fiber size (E–G) was quantified histologically by measuring minimal feret diameters (EDL, extensor digitorum longus). E Treatment-induced shift in the fiber size distribution of TA muscles from control and high-dose groups, also reflected in overall mean fiber sizes (F) and in comparisons of binned fiber sizes (G; small, <30 mm, medium, 30–60 mm; large, >60 mm; Min, minimal). In each panel, means ∓ SEM are plotted and significant differences (p < 0.05) between groups (A–D, F) and within each size range (G) are indicated by different letters, shared letters signify no difference. In F letters indicating no difference were generated from a 2-way ANOVA whereas the p levels shown signify differences from respective controls using t-tests.

The hypertrophic response appears driven in part by increased muscle fiber size (Fig. 4E–G). This is reflected in a shifted size distribution pattern for both wild-type and dystrophic muscles (Fig. 4E) that is reflected in overall mean fiber sizes (Fig. 4F) and in the fact that there were fewer small and more large fibers in treated muscles (Fig. 4G). Note that fiber size variability appeared somewhat higher in untreated mdx muscles, at least regionally, consistent with the early onset hypertrophy that develops in mdx mice and with DMD [23]. This suggests that rAAV6:CK8-hSmad7 (AVGN7.2) effectively increased muscle mass and fiber size despite evidence of pre-existing disease-induced hypertrophy.

Markers of dystrophic disease include elevated levels of serum creatine kinase, which is a metric of persistent muscle fiber damage, and muscle fibrosis, which is an accumulated response to persistent inflammation. Here we showed that although creatine kinase and fibrosis was elevated in mdx mice, neither were altered with treatment (Fig. 5A, B) nor was central nucleation of muscle fibers (Fig. 5C). The structural differences were readily apparent histologically (Fig. 5D, E) as were differences in fiber size. These results conservatively suggest that codon-optimized hSmad7 promoted muscle fiber hypertrophy without exacerbating the pathobiology of dystrophic muscle. Moreover, hypertrophy occurred without changes in serum CK nor fibrosis, but with an improved central nucleation profile. This suggests that hSmad7 promoted hypertrophy while diminishing pathways that promote muscle fiber damage.Fig. 5AVGN7.2 does not exacerbate dystrophic histopathologies.Wild-type (WT) and mdx neonates were injected R.O. when 1 w.o. with 0, 5 × 10^12^ (5e12), 1.7 × 10^13^ (1.7e13) and 5 × 10^13^ (5e13) vg/kg body mass AVGN7 and terminated after 12 weeks. A Relative levels (mean ∓ SEM) of serum creatine kinase (units/L) and B muscle fibrosis obtained from Masson’s trichrome stained TA sections. In B, areas of blue stained collagen were normalized to the total area imaged. C Muscle central nucleation (% of fibers with 1 or more central nuclei) as a marker of compensatory muscle regeneration due to enhance degeneration. A–C Significant differences (2-way ANOVAs) are indicated by different letters whereas the same letters indicate no difference. D Representative images of TA sections stained with Masson’s trichrome. Sarcolemmal membranes are stained reddish purple, cytoplasm pink and collogen blue. E Representative images of TA sections stained with wheat germ agglutinin Alexa Fluor 488 (reddish purple, sarcolemmal membranes) and DAPI (blue, nuclei), the procedure used for determining fiber size and central nucleation.

Enhancing dystrophic muscle function

Muscle function was assessed in vivo using an established plantarflexor force/torque assay with direct sciatic nerve stimulation spanning the normal in vivo physiological frequency range of 50–120 Hz and were additionally stimulated at 150 Hz to assure that maximal or near-maximal activation was reached within the physiological range. Codon-optimized hSmad7 increased peak isometric force in a dose-dependent manner (Fig. 6A). Dose-dependency was best demonstrated at 80 Hz where the force generated by mdx mice treated with 5e13 vg/kg was greater than that generated by the mdx controls, but not greater than the forces from mice receiving lower doses (Fig. 6B). This difference represents a ~30% increase (140 vs. 185 mN), which is ~40% toward the healthy wild-type levels (246 mN). Although mass/force relationships are often difficult to demonstrate and are discordant in dystrophic animals and patients where muscles become hypertrophied from enhanced muscle regeneration, fibrosis and inflammation, correlation analysis indicates a positive mass/force relationship with treatment (Fig. 6C). Quantifying contracile kinetics, we observed an almost identical pattern in contraction rate while relaxation rate was completely restored by the highest dose. Moreover, the mass/contraction rate relationship mirrored the mass/force relationship (Fig. 6D–H), although the mass/relaxation rate relationship was not correlated (data not shown).Fig. 6AVGN7 enhances the function of dystrophic muscle.Wild-type (WT) and mdx neonates were injected R.O. when 1 w.o. with 0, 5 × 10^12^ (5e12), 1.7 × 10^13^ (1.7e13) and 5 × 10^13^ (5e13) vg/kg body mass rAAV6:CK8-hSmad7. Plantarflexion force/torque was quantified after 12 weeks using an in vivo whole-animal murine system with increasing stimulation frequencies. A Total force generating capacity (mean ∓ SEM) generated at each frequency. Dashed lines frame the physiologically relevant range of stimulation frequencies and significant differences (ANOVAs) between mdx control (0) and mdx 5e13 groups are indicated by the probability levels shown. B Pairwise comparisons of total force generated at 80 Hz, which produced maximal to near-maximal responses for all contractile metrics. Significant differences are indicated by different letters, the same letters indicate no difference. C Total force of all mdx mice correlated to their corresponding plantarflexor muscle mass. The Pearson correlation coefficient (r) and two-tailed probability level (p) for the relationship are inset. D–H Contraction and relaxation rate measures as described for (A–C). I Maximum force at 80 Hz normalized to plantarflexor mass. In all histograms, significant differences between any two means are indicated by different letters (2-way ANOVAs), the same letters indicate no difference.

We further evaluated muscle specific force, which is force nomalized to muscle mass. Deficits in muscle specific force arise post-nataly and are underscored by increased muscle mass due to regenerating muscle fibers [24, 25]. Here we found muscle specific force unchanged with treatment (Fig. 6I). Given that deficits in muscle specific force and twitch kinetics are not intrinisc to dystrophin’s absence but arise early in post-natal growth, these results suggest an intrinsic change linked to dystropic pathology persists on the background of hSmad7’s positive impact on muscle mass.

Discussion

Optimizing hSmad7 cDNA improved codon quality and removed predicted hairpins, GC-rich regions and other negative motifs/elements. These changes were reflected in predicted secondary structures that were less compact and consistent with improvements in translational efficiency. Optimized hSmad7 was equally potent as wild-type mSmad7 in stimulating muscle hypertrophy in mice, ruling out possible negative consequences of optimization, and although similarities in action were expected due to 98.1% identities in amino acid sequences, additional studies are needed to determine if optimization improves translational efficiency in human muscle. Nevertheless, the pairing of optimized hSmad7 with the CK8 promoter appeared superior to using CMV with either mSmad7 or hSmad7, likely reflecting the former’s enhanced activity in skeletal muscle [17, 18, 26, 27]. As a drug product (i.e., AVGN7 vs. AVGN7.2), this pairing was similarly effective with local and systemic routes of administration, effecting multiple muscle types in wild-type and especially mdx mice where muscle mass and function were both enhanced without exacerbating dystrophic pathophysiological markers. Moreover, the pairing of CK8 with a muscle-trophic capsid (AAV6) maintains muscle-specificity and prevents off-targeting. This is indicated in a companion report of AVGN7.2 toxicology and biodistribution [28] that also assessed systemic safety and performed immunogenicity, physiology, ophthalmoscopic, hematological and serum chemistry examinations. Muscle-specificity is key to preventing off-target effects and is critically important as although Smad7 perturbs growth of some tumor types, it is functionally linked to the growth of others [29, 30].

Correlation analyses established a link between hSmad7 protein levels and muscle mass, and between muscle mass and muscle force, in both WT and mdx muscle. This suggests that muscle hSmad7 protein level could function as a preclinical and possible clinical biomarker for improvements in muscle function, although more detailed studies are needed to thoroughly establish the dose relationship. These results may also suggest that the full potential for AVGN7.2 in treating subjects with DMD or other neuromuscular diseases may be in combinatorial therapy. Indeed, restoring a functional dystrophin to near healthy levels has never restored muscle function in clinical trials regardless of the approach. Sarepta’s micro-dystrophin gene therapy (SRP-9001), for example, failed to meet functional endpoints in a recent phase 2 clinical trial that restored dystrophin immunoreactivity to near normal levels [31–33]. This latter point is based on an earlier phase 1/2a trial using the same dose (2e14 vg/kg) to express micro-dystrophin in 81% of fibers and at levels that were 96% of normal (after adjusting for fat and fibrosis) [34]. The inability of SRP-9001 to restore or even reliably enhance muscle function in clinical trials is typical for the field as other dystrophin-targeting genetic strategies (i.e., exon-skipping) have similarly struggled to meet functional outcomes or to significantly enhance muscle function [35, 36].

Often overlooked is the fact that DMD is a progressively developing degenerative disease where muscle pathologies accumulate over time. Restoring muscle function, therefore, must overcome the deleterious effects from many different dystrophic pathologies (e.g., muscle wasting, necrosis, fibrosis, fiber branching, fragmentation of neuromuscular junctions, etc.) that directly impair myofiber contractility and muscle force production [24, 25, 37]. Dystrophin-replacing or -correcting approaches will presumably prevent future damage and degeneration of the muscle fiber, but cannot compensate for the accumulated pathologies within the fiber or the muscle. Thus, a combinatorial strategy to increase muscle mass and strength is likely needed to realize a significant restoration of muscle function particularly in older subjects with more advanced stages of disease.

Several drugs targeting myostatin and other ActRII ligands have demonstrated the ability to increase muscle mass and function in dystrophic animals [38–43]. Results from these studies compare favorably with the mdx data reported herein, although the approaches differ significantly. All previous drugs in the space attempted to prevent ligand binding to activin receptors, either by sequestering the ligands or blocking their binding sites. These include soluble ActRIIb ligand traps [41–44], myostatin and activin binding proteins (i.e., follistatin & its derivatives) [39, 45, 46] as well as myostatin antibodies [38, 40, 47]. In fact, co-delivering a micro-dystrophin gene therapeutic that prevents muscle degeneration with a follistatin or insulin-like growth factor 1 gene therapeutic that enhances muscle mass proved vastly superior to dystrophin replacement alone [48–51]. Combining ActRIIb attenuation, via shRNA interference, with exon skipping to restore partial dystrophin was also vastly superior in restoring muscle function in mdx mice as the change in tetanic and specific force was more than double that with the individual treatments [52]. In fact, specific force was unaffected by ActRIIb knockdown or dystrophin restoration, but increased 30% with the combinatorial. A similar approach featuring an anti-sense morpholino to restore spinal motor neuron 1 expression and a gene therapeutic to express a myostatin binding protein, the myostatin latent associated peptide [5], enhanced motor function in mice with spinal muscle atrophy and was again, far superior to morpholino treatment alone [53].

These studies together validate combinatorial approaches for neuromuscular disease that feature an ActRII-attenuator. They also specifically demonstrate the potential to improve the dystrophic condition, although notable disadvantages and safety concerns of the previously developed attenuators have led to failed clinical trials and suspended programs [5]. Targeting ActRII ligands, which include the activins (Act-A, Act-B & Act-A/B), myostatin, GDF11 and others, presents opportunities for serious off-target effects as these ligands regulate a variety of systems (e.g., reproduction, neurogenesis, angiogenesis, osteogenesis, etc.) and recognize multiple receptors in addition to ActRII [5]. Indeed, a clinical trial of an ActRIIb ligand-trap administered to boys with DMD was terminated prematurely due to serious off-target effects that compromised respiratory epithelium [54] while a monoclonal antibody recognizing the ActRII ligand binding domain was demonstrated to alter pituitary function [55]. This latter effect is consistent with the very well-known effects of activins in this tissue and with the recently described effects of myostatin on pituitary development and function [56]. By contrast, AVGN7.2 was designed to avoid such off-target effects by combining the muscle-trophic AAV6 capsid with the muscle-specific CK8e promoter [5]. It appears, therefore, to be the only drug in the space that attenuates actions of multiple ActRII ligands, in a muscle-specific manner and with the durability of a gene therapeutic, which we believe is well suited for combinatorial approaches to treat DMD and possibly other neuromuscular diseases.