Systemic Delivery of Full-Length Dystrophin in DMD Mice

Abstract Current gene therapy for Duchenne muscular dystrophy (DMD) utilizes adeno-associated virus (AAV) to deliver miniaturized dystrophin (micro-dystrophin or µDys), which does not provide full protection for striated muscles as it lacks many important functional domains within full-length (FL) dystrophin. Here we develop a triple vector system to deliver FL-dystrophin into skeletal and cardiac muscles. We rationally split FL-dystrophin into three fragments (N, M, and C) linked to two orthogonal pairs of split intein, allowing efficient, unidirectional assembly of FL-dystrophin. The three fragments packaged in myotropic AAV (MyoAAV4A) restore FL-dystrophin expression in both skeletal and cardiac muscles in male mdx 4cv mice. Dystrophin-glycoprotein complex components are also restored in the sarcolemma of dystrophic muscles. MyoAAV4A-delivered FL-dystrophin significantly improves muscle histopathology, contractility, and overall strength comparable to µDys, but unlike µDys, it also restores defective ERK signaling in heart. The FL-dystrophin gene therapy therefore promises to offer superior protection for DMD.


INTRODUCTION
Duchenne muscular dystrophy (DMD) is a fatal genetic disease a icting approximately 1 in 3800-6300 live male births and caused by genetic mutations in the DMD gene located on the X chromosome. 1 The dystrophin protein encoded by the DMD gene belongs to the spectrin superfamily of cytoskeletal proteins. 2The major isoform expressed in striated muscles is a 427 kDa protein, containing four main domains: an N-terminal actin-binding domain, a long rod-like domain composed of spectrin-like repeats, a cysteine-rich domain that binds to dystroglycan and other membrane-associated proteins, and a Cterminal domain that interacts with syntrophin and α-dystrobrevin, forming a large dystrophinglycoprotein complex (DGC). 3,46][7] Disrupted expression of dystrophin leads to repeated cycles of muscle damage and repair, which eventually exhaust the regenerative capacity of muscle stem cells and cause brosis and fatty replacement.The progressive loss of muscle mass and function in the skeletal, cardiac, and respiratory muscles eventually leads to impaired mobility, cardiomyopathy, respiratory failure, and early death. 8,9gene replacement therapy using adeno-associated virus (AAV) to deliver a correct copy of the DMD gene could allow the restoration of dystrophin to halt/reverse the deterioration of muscles.10 However, the limited cargo capacity of AAV (~ 4.5 kb) makes it challenging to deliver the FL DMD gene (cDNA is over 11 kb).Early studies found that patients carrying a large internal deletion of the DMD gene developed a mild form of the disease, Becker muscular dystrophy (BMD), owing to the expression of truncated but partially functional dystrophin proteins.11 These ndings led to the development of microdystrophins (µDys) gene therapy, which aims to restore the expression of a truncated version of the dystrophin protein.[12][13][14] These miniaturized forms of dystrophin are only about 1/3 of the FL dystrophin with some of the essential functional domains of the protein retained and can t into one AAV vector for in vivo delivery, thereby restoring muscle integrity and improving muscle in preclinical animal models.13- 15 Recently, the US Food and Drug Administration (FDA) granted accelerated approval to Sarepta's Elevidys (µDys delivered in AAVrh.74) for DMD patients at 4 to 5 years of age.16 However, due to the lack of 2/3 of the dystrophin coding sequence containing critical rod and hinge domains of dystrophin needed to connect with other proteins in the dystrophin complex, these µDys gene therapies are unable to provide full protection of skeletal and heart muscle integrity and function, 15,[17][18][19] highlighting the urgent need to develop novel strategies to deliver FL dystrophin.
Several different approaches have been explored using dual AAV vectors to deliver larger payloads, including fragmented genome assembly, overlapping, trans-splicing, and hybrid approaches.1][22][23] Upon co-delivery, the e ciency of homologous recombination is however a limiting factor for this strategy.Another approach known as trans-splicing is to take advantage of the natural concatamerization ability of AAV to expand the transgene size using dual AAV vectors 24,25 .However, the head-to-tail concatamer formation and the trans-splicing of the pre-mRNA across the ITR junction pose rate-limiting steps.These approaches may be combined as shown for hybrid vector systems. 26Despite the varying success of these dual AAV vector strategies in preclinical studies, the e ciency of expression from the available dual-vector systems is insu cient for many clinical gene therapy applications.It is even more challenging to deliver the FL dystrophin sequence, which is about three times the packaging capacity of AAV, thus requiring at least three vectors to package the entire coding sequence.A previous attempt with triple trans-splicing AAV vectors showed the feasibility of expressing FL dystrophin, albeit with only very low e ciency. 27lit inteins are small polypeptides that self-assemble and undergo a protein trans-splicing (PTS) reaction, resulting in the formation of a mature, fully functional protein in a "traceless manner".9][30] Leveraged on this success, we developed a triple AAV system with orthogonal split inteins to deliver FL dystrophin protein.Packaged into an engineered myotropic AAV capsid 31 (MyoAAV4A), this triple vector combination restored the expression of FL dystrophin and signi cantly improved muscle histopathology and function in a mouse model of DMD.

Proof-of-concept design of split intein constructs to assemble FL-dystrophin
The cDNA for FL-dystrophin is over 11 kb, about three times of the AAV packaging capacity.Thus, three AAV vectors are required to package the entire coding sequence of FL-dystrophin.We rationally split FLdystrophin cDNA into three fragments based on 1) the fragment size, 2) the protein domain structure, and 3) the junctional sequence compatibility for split intein.We previously utilized Cfa 32 and Gp41-1 33 , two orthogonal inteins with a remarkably fast rate of protein trans-splicing (PTS), to successfully assemble adenine base editor (ABE) 28 .We therefore chose these two orthogonal pairs of inteins for our initial test.The rst split site was chosen in between spectrin repeat (SR) 8 and 9, where a cysteine residue is followed by a bulky tryptophan residue as required for e cient protein splicing by the Cfa intein (Fig. 1a).The second split site was chosen within the end of Hinge 3 (H3) domain where two consecutive serine residues are located, which may facilitate protein splicing by the Gp41-1 intein (Fig. 1a).In addition, this is also where the native Dp140 isoform starts.We named the three fragment constructs of this rst version as Dys-N1, Dys-M1, and Dys-C1, respectively.All expression cassettes utilized a mini-CMV promoter with a muscle creatine kinase enhancer (meCMV).Transfection of Dys-N1/M1/C1 into HEK293 cells resulted in the expression of FL-dystrophin detectable by three different anti-dystrophin antibodies that speci cally recognize the N, M, or C fragments, respectively (Fig. 1b).
However, the FL-dystrophin band was much weaker than the un-assembled or partially assembled dystrophin fragments.Varying the ratio of the three plasmids improved the relative abundance of the FL versus the unassembled or partially assembled dystrophin signals, with the 4:2:1 ratio of N1:M1:C1 plasmids yielding the highest level of FL-dystrophin.The C-terminal fragment band was much more intense than either the FL or MC bands, indicating that the assembly between the M and C fragments mediated by Gp41-1 intein needs to be improved.

Optimization of split intein constructs to improve FLdystrophin assembly
Encouraged by this initial observation, we attempted to optimize the assembly between the M and C fragments.First, we tested a different split site (IGA-SPT) within the H3 domain and mutated the − 1 position from alanine to tyrosine to favor the Gp41-1-mediated PTS (Fig. 2a).These changes (Dys-M2 and Dys-C2) led to a substantial improvement in the FL-dystrophin assembly (Fig. 2b, c) with reduced unassembled (Fig. 2d-f) and partially assembled fragments (Supplementary Fig. 1a-c).In addition, we removed a small intron within the C fragment construct (Dys-C3), which signi cantly increased the FLdystrophin band intensity (Fig. 2b, c).Next, we reasoned that the + 1 to + 3 position on the extein (e.g.Dys-C3) may also affect the PTS e ciency.We thus mutated the SPT sequence to SSS at the junction site on the C fragment construct (Fig. 2g).In addition, we changed the strong Kozac sequence to a weak one considering the abundance of the C fragment.However, these two changes (Dys-C4) led to a substantial reduction of the FL-dystrophin band with an increased accumulation of the NM band and C band (Fig. 2h-l and Supplementary Fig. 1d-f).We reasoned that this may be caused by the use of the weak Kozac sequence, which potentially leads to the translation initiation from a downstream in-frame start codon so that the C fragment was expressed without the intein fusion.Indeed, after we switched the weak Kozac sequence back to a strong one in this version of the C fragment construct (Dys-C5), FLdystrophin expression was restored (Fig. 2h-l and Supplementary Fig. 1d-f).Of note, we also added a synthetic signal for ubiquitin-dependent proteolysis (PB29) 34 in Dys-C5 construct to lower the expression level of the C fragment.Although the FL-dystrophin band signal was reduced when compared to Dys-C3, we observed that the NM fragment was almost undetectable, suggesting that these changes improved NM and C assembly.We further tested a different intein (IMPDH-1), which has a fast PTS rate comparable to Gp41-1 with a native junction sequence of GGG-SIC, 33 similar to the split site of dystrophin (IGA-SPT) (Fig. 2m).We also mutated alanine at the − 1 position of Dys-M2 extein to glycine to further mimic the native junction sequence of IMPDH-1.These changes (Dys-M3 and Dys-C6) signi cantly improved the FL-dystrophin signal by ~ 86% (Fig. 2n, o).Moreover, the unassembled C fragment band was dramatically reduced by ~ 76% as compared to Dys-N1/M2/C5 (Fig. 2n, r), with marginal effects on the unassembled N and M fragments (Fig. 2n, p, q) and partially assembled fragments (Supplementary Fig. 1g-i).Finally, we inserted two different poly-adenylation signal sequences 35 at the upstream of the start codon in order to further lower the expression level of the C fragment (Dys-C7 and Dys-C8) (Supplementary Fig. 2a).These changes signi cantly reduced the Cfragment band intensity while still maintaining a high level of FL-dystrophin expression (Supplementary Fig. 2b, c, f) and similar N/FL (Supplementary Fig. 2d) or MC/FL ratio (Supplementary Fig. 2i).However, we found that lowering C-fragment expression further caused a concomitant accumulation of the M (Supplementary Fig. 2e) and NM (Supplementary Fig. 2g, h) fragments.These data suggest that some excess of the C-fragment favors the assembly of FL-dystrophin.
In contrast to the canonical inteins, atypical inteins with small intein N and large intein C have also been discovered. 36,37To test whether the atypical inteins could also enable the assembly of FL-dystrophin, we modi ed the N, M and C constructs using the atypical inteins Cat 38 and VidaL 39 to mediate the N-M and M-C splicing, respectively (Supplementary Fig. 3a).Western blotting analysis showed that these atypical inteins can also confer the assembly of FL-dystrophin, but the e ciency was much lower than Dys-N1/M3/C6 (Supplementary Fig. 3b).We chose the best optimized version Dys-N1/M3/C6 for further in vivo studies.
Restoration of FL-dystrophin expression in mdx 4cv mice following systemic MyoAAV delivery of Dys-N1/M3/C6 We rst replaced the generic promoter meCMV with a synthetic muscle-speci c promoter Spc5-12 40 (for the N and M fragment construct) or Spc2-26 40 (for the C fragment construct) for AAV packaging.We chose a recently engineered myotropic AAV capsid, MyoAAV4A, for packaging because of its superior muscle and heart transduction in mice and monkeys following systemic delivery 31 .A total dose of 2E + 14 vg/kg AAV vectors consisting of N1, M3 and C6 at 2:1:1 molar ratio was delivered into a cohort of mdx 4cv mice (N = 13) at the age of 3-4 weeks via retro-orbital injection (Fig. 3a).
Immuno uorescence staining was performed using the aforementioned anti-dystrophin antibodies (Fig. 3b).Dystrophin expression could be detected at the sarcolemma of wild-type (WT) gastrocnemius (GA) muscle with all these three antibodies, while GA muscle from mdx 4cv mice was negative for any of these antibody stains (Fig. 3b).AAV administration restored dystrophin expression that can be detected by all these three antibodies (Fig. 3b).Importantly, dystrophin signals were correctly localized at the sarcolemma without noticeable accumulation in the cytoplasm.On average, dystrophin was detected in 83.4 ± 2.7% muscle bers (Fig. 3c).Dystrophin expression was also robustly rescued in the cardiac muscles of mdx 4cv mice with 78.3 ± 2.7% cardiomyocytes being dystrophin positive following AAV administration (Supplementary Fig. 4a, c).However, dystrophin + bers in diaphragm muscles (8.6 ± 2.6%) were much lower than those in the GA muscles (Supplementary Fig. 4b, d).
Western blotting was also performed to substantiate these observations.Again, FL-dystrophin was readily detectable using the three different N-, M-or C-recognizing antibodies in GA muscles from mdx 4cv mice treated with AAV-N1/M3/C6 (Fig. 3d).Interestingly, both the N-and M-recognizing antibodies detected mostly the FL-dystrophin with weak partially assembled dystrophin fragments, while the unassembled N-or M-fragment was hardly discernable (Fig. 3d).The C-recognizing antibody detected FL-dystrophin and unassembled C-fragment with roughly equal intensities, while the partially assembled MC fragment was almost undetectable (Fig. 3d).FL-dystrophin expression was also readily detectable in the heart muscles of AAV-treated mdx 4cv mice (Supplementary Fig. 4e), but the diaphragm muscle showed a much weaker expression of FL-dystrophin and the C fragment following AAV treatment (Supplementary Fig. 4f), likely re ecting weak activity of the Spc2-26 promoter and/or MyoAAV4A capsid in diaphragm muscle.
To examine if AAV-N1/M3/C6 treatment improves the histopathology of mdx 4cv mice, we performed Hematoxylin and Eosin (H&E) staining of skeletal muscle sections from the animals.While WT GA muscle sections showed a normal musculature, mdx 4cv mice displayed a typical muscular dystrophy phenotype as evidenced by the presence of central nucleated muscle bers (CNFs), muscle necrosis and regeneration.These pathologies were substantially ameliorated by AAV-N1/M3/C6 administration (Fig. 4c).To further quantify the percentages of CNFs, we performed immuno uorescence staining of the muscle sections with anti-laminin α2 and 4',6-diamidino-2-phenylindole (DAPI) (Fig. 4c).The CNFs in the GA muscles of mdx 4cv mice were reduced from 58.2 ± 1.7% to 25.1 ± 1.5% by AAV-N1/M3/C6 treatment (Fig. 4d).Owing to the repeated cycles of degeneration and regeneration, the distribution of muscle ber size in mdx 4cv GA shifted to lower sizes as compared to WT (Fig. 4e), whereas AAV-N1/M3/C6 treatment shifted the ber size distribution towards those of the WT muscles (Fig. 4e).It appears that AAV-N1/M3/C6 treatment also improved the histopathology of mdx 4cv diaphragm muscles (Supplementary Fig. 5a), but we did not observe signi cant changes in CNFs in the diaphragm muscles (Supplementary Fig. 5b), consistent with the low dystrophin restoration in AAV-treated mdx 4cv diaphragm muscles.To examine the impact of AAV-N1/M3/C6 treatment on brosis, we performed Masson's Trichrome staining on muscle sections, which showed that the brosis in both GA (Fig. 4c, f) and diaphragm (Supplementary Fig. 5c, d) muscles of mdx 4cv mice was greatly attenuated by AAV-N1/M3/C6 treatment.

Effects of promoter and dose on FL-dystrophin expression in skeletal and heart muscles of mdx 4cv mice
We reasoned that the Spc2-26 promoter may not work e ciently in diaphragm muscle, thus yielding low restoration of FL-dystrophin in this tissue.To test this, we changed the promoter in the C construct to Spc5-12, and designated AAV-N1/M3/C6-Spc5-12 as AAV-FL-v2 and the original AAV-N1/M3/C6 as AAV-FL-v1.Both AAV mixtures (a total dose of 2E + 14 vg/kg at 2:1:1 ratio for N, M and C vectors) were systemically administered into a cohort of mdx 4cv mice via retro-orbital injection.At 6 weeks post AAV injection, the animals were sacri ced for examination.
Western blotting analysis showed that the expression of FL-dystrophin was comparable between AAV-FL-v1 and AAV-FL-v2 in GA muscles (Fig. 5a-d).To estimate the relative amount of rescued FL-dystrophin compared to endogenous dystrophin in healthy skeletal muscle, control human skeletal muscle lysate was loaded at 50%, 25% and 10% on the gel.Three different antibodies recognizing the N, M or C fragment reported 32.6%-49.5% and 36.9%-43.4% of FL-dystrophin restoration following AAV-FL-v1 and AAV-FL-v2 treatment, respectively (Fig. 5b-d).In diaphragm muscles, AAV-FL-v2 treatment signi cantly increased FL-dystrophin expression when compared to AAV-FL-v1 as detected by anti-M and C antibodies (Fig. 5e-h).Similar improvement of FL-dystrophin rescue was observed in heart muscles (Fig. 5i-l), indicating that the Spc5-12 promoter works more e ciently in diaphragm and heart muscles than Spc2-26.It is of note that we used human skeletal muscle lysate to estimate the relative amount of FL-dystrophin rescue for both diaphragm and heart muscles, as we do not have access to human diaphragm and heart muscle samples.
As the total dosage of MyoAAV that we used was relatively high (2E + 14 vg/kg), we wondered whether a lower total dosage (8E + 13 vg/kg at 2:1:1 of N, M and C) could yield e cient FL-dystrophin restoration in mdx 4cv mice.As shown in Fig. 5a-d, FL-dystrophin was readily detectable in GA muscles treated with the lower dose, albeit at a lower level as compared to the high dose group.The differences between the low and high dose groups were less evident in diaphragm muscle (Fig. 5e-h) and heart (Fig. 5i-l).
Comparison of MyoAAV delivered FL-and micro-dystrophin in mdx 4cv mice Finally, to benchmark against micro-dystrophin gene therapy, we performed a comparative study for MyoAAV-delivered FL-and micro-dystrophin gene delivery in mdx 4cv mice.Two different microdystrophin constructs were tested with one like the P zer version but containing the fusion of spectrin repeat (SR) 2 and 22 (designated as µ-v1), and the other originally developed in Duan's laboratory (designated as µ-v2) (Fig. 6a).All the dystrophin constructs were under the control of the Spc5-12 promoter.AAV-µ-v1 and µ-v2 were delivered at 8E + 13 vg/kg, while the total dose of AAV-FL-v2 were 2E + 14 or 8E + 13 vg/kg (an effective dose of 5E + 13 or 2E + 13 vg/kg determined by the lowest dose among the three fragments).Consistent with the data shown in Fig. 5, FL-dystrophin was restored to 46.3% of normal human skeletal muscle level in GA muscles after AAV-FL-v2 delivery at the high dose group (Fig. 6b, c), however, the micro-dystrophin proteins were overexpressed at 3.7 and 2.6 folds following AAV-µ-v1 and µ-v2 delivery, respectively (Fig. 6b, c).Similarly, AAV-FL-v2 delivery led to the restoration of FL-dystrophin to 10.0% in diaphragm and 69.3% in heart using human skeletal muscle as a reference, while micro-dystrophin was overexpressed at 2.6-7.8folds in diaphragm and heart following AAV-µ-v1 and µ-v2 delivery (Supplementary Fig. 6a-d).
To examine the expression of FL-dystrophin and micro-dystrophin proteins in muscle bers, we performed co-immuno uorescence staining with anti-N (10F9, which recognizes the hinge1 region) and anti-M (8773R, which recognizes the SR16/17 of human dystrophin only) on one section, and coimmuno uorescence staining with anti-N and anti-C (Ab15277, which recognizes the C terminus) on a consecutive section of GA muscles.As shown in Fig. 6d, WT skeletal muscle was positive for both N and C antibodies but not for M antibody as expected, and mdx 4cv skeletal muscle showed only background signal for either of these antibodies.AAV-µ-v1 yielded positive staining for only N antibody while AAV-µ-v2 led to positive staining for both N and M antibodies.The mdx 4cv mice treated with AAV-FL-v2 were positive for all three antibodies, and the signals were correctly localized at the sarcolemma.Interestingly, visual examination of the uorescence images could rarely nd any muscle bers positively for only one or two antibodies (e.g. if a muscle ber was positive for one antibody, it was also positive for the other two antibodies) in mdx 4cv mice treated with AAV-FL-v2.To further illustrate this high degree of correlation among the three antibody signals, we performed line pro le analyses on these images for AAV-FL-v2 treated mice.As shown in Supplementary Fig. 7, the N and M immuno uorescence signals showed almost identical patterns at three arbitrarily selected lines.Similarly, the N and C immuno uorescence signals also showed highly overlapping peaks along the lines.These results suggest that muscle bers expressing only the unassembled or partially assembled dystrophin fragments are very rare if any.
The serum CK levels were signi cantly dropped in all AAV treated groups with the µ-v1 and µ-v2 groups approaching the WT levels (Fig. 6e).Muscle contractility was signi cantly increased in all treatment groups compared to control mdx 4cv mice, and there was no signi cant difference among the treatment groups (Fig. 6f).We further performed a wire hanging test to evaluate the overall muscle strength in these mice.The latency to when the animal falls was recorded and compared for the animals in each group.On average, WT mice stayed on the wire mesh for over 555.1 ± 30.4 s (n = 8), while mdx 4cv mice held only for 215.4 ± 51.6 s (n = 4; p = 0.0002, Fig. 6g).Remarkably, all AAV treatments completely normalized the hanging time on the wire mesh (AAV-FL-v2: 542.1 ± 35.2 s, n = 5; AAV-FL-v2-L: 466.0 ± 60.8 s, n = 5; AAV-µ-v1: 538.0 ± 56.2 s, n = 4; AAV-µ-v2: 476.6 ± 52.6 s, n = 5; Fig. 6g).
Similar histopathological improvement was observed in diaphragm (Supplementary Fig. 8b).Histological examination of heart sections showed no signi cant differences among all groups (Supplementary Fig. 8c).
Although the histopathological and functional assays showed similar improvement for AAV-FL-v2 and micro-dystrophins, previous studies showed that micro-dystrophins failed to correct cavin-associated ERK signaling defects in dystrophic mouse heart due to the lack of dystrophin's C terminal domain. 43In agreement with that, our data showed that the membrane localization of cavin-4 is disrupted in cardiomyocytes of mdx 4cv mice and AAV-delivered micro-dystrophins did not correct such mislocalization of cavin-4 (Fig. 7a).However, AAV-delivered FL-dystrophin greatly restored the membrane localization of cavin-4 in mdx 4cv cardiomyocytes (Fig. 7a).In response to cardiac stress/damage, membrane-associated cavin-4 recruits the signaling molecule ERK to caveolae to activate key cardioprotective responses.Western blot analysis showed that ERK phosphorylation was inhibited in mdx4cv mouse heart, which was not affected by micro-dystrophin gene delivery, but was signi cantly improved by FL-dystrophin gene delivery (Fig. 7b-d).Taken together, these data suggest that FL-dystrophin gene therapy is superior to micro-dystrophin gene therapy.

DISCUSSION
The small packaging capacity of AAV vectors limits gene replacement therapy for DMD and many other diseases.Until now, delivering full-length dystrophin to body-wide muscles has been unsuccessful.Here we harnessed the split intein-mediated PTS to develop a triple AAV system for e cient assembly and in vivo delivery of full-length dystrophin.Using the newly engineered myotropic AAV capsid, we achieved systemic full-length dystrophin rescue in both skeletal muscle and heart, which led to signi cant functional, histopathological, biochemical signaling improvement in mdx 4cv mice.
Using two orthogonal split inteins (Cfa and Gp41-1 or IMPDH-1) with a very fast rate of PTS, we demonstrate the feasibility of generating FL dystrophin protein in vitro and in vivo.After a series of optimizations for the construct design (e.g. the choice of split inteins, the split sites, the addition of a protease-degradation signal, mutation of the junctional amino acids, etc.), we gradually improved the assembly e ciency of assembly.The best version (N1/M3/C6) can achieve ~ 60% of FL-dystrophin protein expression (relative to a plasmid carrying the entire dystrophin cDNA) with very low levels of unassembled or partially assembled products in vitro.Earlier studies with Ssp DnaB split intein to assemble a 6.3-kb Becker-form dystrophin 44 and factor VIII 45 were ine cient.These results showed that the selection of split inteins and split sites on dystrophin (and thus the junctional amino acids) are particularly important for e cient FL-dystrophin assembly.Future efforts could systemically screen different orthogonal pairs of split inteins to further increase the assembly e ciency, expression, and purity of fully assembled FL-dystrophin protein.
The split site for N and M constructs is in proximity to the natural initiation site of the Dp260 isoform whereas the split site for M and C constructs overlaps with the start site of the Dp140 isoform.The selection of these split sites may help minimize the potential negative impacts of unassembled C and partially assembled MC fragments.A previous study showed that the Dp260 (MC fragment) restored a stable association between costameric actin and the sarcolemma, assembled the DGC, and signi cantly slowed the progression of muscular dystrophy in mdx mice. 46Metzger and colleagues showed that the NtermDys fragment (corresponding to our NM fragment) did not compete with dystrophin and had no pathological effect, but the 2A protease-cleaved CtermDys (corresponding to our unassembled C fragment) was su cient to cause dystrophic cardiomyopathy in transgenic mice when overexpressed. 47owever, in their CtermDys transgenic mice, the CtermDys was overexpressed by more than 10 folds relevant to the FL-dystrophin (estimated from the Fig. 1B in the reference 47 ), while in our AAV delivered mice, the unassembled C fragment was about equal to the FL-dystrophin product (Fig. 5i).Thus, it is very unlikely that the unassembled C fragment would cause cardiomyopathy in our triple vector delivery.In agreement with this, we did not observe increased cardiac brosis in mdx 4cv mice following the triple vector delivery (Supplementary Fig. 8c).Nevertheless, it is important to carefully investigate the longterm impact of these unassembled and partially assembled products in the future.
We optimized the molar ratios for each vector carrying the N, M, and C fragments by semi-quantitative assessment of each fragment and the resulting assembly products by Western blotting initially in vitro.The N and M fragments appeared to be less stable than the C fragment.We observed that the expression level of the N fragment fused with Cfa N was evidently lower than that fused with the smaller Cat N , indicating that the larger Cfa N fused at the C-terminus of the N construct contributed to the instability of the N fragment fusion protein. 48Although our efforts in this work were primarily centered on optimizing the inteins and split sites as well as tuning down the C fragment expression, future studies can be targeted to improve the stability of the N and M fragments, which may improve the expression level of FL-dystrophin.A few strategies can be explored.First, codon optimization could potentially help to boost the N and M transgene expression.Second, adding an intein "cage" at the end may help to stabilize the disordered intein N . 48Third, stronger enhancer sequences can be tested to drive the N and M expression.
In this proof-of-principle study, we initially utilized a weaker promoter Spc2-26 to drive the C fragment expression and a stronger Spc5-12 to drive the N and M fragment expression.However, we found that the expression of the C fragment and the resulting FL-dystrophin in the diaphragm muscle was very weak.The weak activity of Spc2-26 likely contributed to the ine cient restoration of FL-dystrophin in diaphragm.Indeed, when using Spc5-12 to drive the C fragment expression, we observed some improvement in FL-dystrophin expression in diaphragm and cardiac muscles.Further optimization is required to increase the FL-dystrophin expression in the diaphragm muscle.This could be done by testing different muscle-speci c regulatory cassettes such as those from DES, MYL11, SERCA1, CKM, and FLNC to drive each fragment expression and their stoichiometry in the AAV cocktail.E cient gene replacement therapy for DMD requires a potent AAV capsid with strong muscle tropism.New myotropic capsids have recently been engineered by the directed evolution of AAV capsid libraries containing seven random amino acid insertions into the variable region VIII of AAV9 and in vivo. 31,49,50hese newly engineered myotropic AAV capsids share a common "RGD" motif, which is known to bind to several integrin heterodimers. 51In this study, we took advantage of the high muscle tropism of MyoAAV4A, which allowed simultaneous administration of three AAV vectors at a total dosage that is currently used in clinical trials for neuromuscular diseases.Notably, MyoAAV4A and its related variants were tested in both mouse models and non-human primates, 31 highlighting the promise of their clinical translation.
Finally, we performed a comparative study for FL-dystrophin with the triple vector approach vs the microdystrophins in mdx 4cv mice using the same MyoAAV4A capsid and the same Spc5-12 promoter.Our data showed that delivery of FL-dystrophin and micro-dystrophin at the same total dose (8E + 13 vg/kg) achieved similar levels of functional and histological improvements except for the serum CK measurements.The serum CK levels were normalized to about WT levels in mdx 4cv mice treated with micro-dystrophin, but less pronounced in mice treated with FL-dystrophin.The lower dystrophin positive muscle bers in diaphragm muscles following FL-dystrophin gene delivery can obviously contribute to the higher CK readings.Interestingly, on WB examination, FL-dystrophin was expressed at ~ 30-50% of normal level following the triple vector injection, while micro-dystrophins were expressed ~ 2.6-3.7-fold of normal skeletal muscle (Fig. 6b, c).Overall, these data showed that even a lower level of FL-dystrophin expression can provide a similar level of protection in dystrophic skeletal muscle as much overexpressed micro-dystrophins.Moreover, FL-dystrophin expression can restore the membrane localization of cavin-4 and the defective ERK signaling, which could not be corrected by micro-dystrophin gene delivery in the heart (Fig. 7).We envision that future optimization of the promoter, capsid and intein splits of triple dystrophin vectors can bring FL-dystrophin expression to body-wide skeletal muscle and heart to offer maximal protection.
One potential concern with FL-dystrophin gene therapy is the host immune response towards the nonself epitopes from FL-dystrophin and its fusion products with inteins, particularly in patients who carry large deletions.A recent study reported that ve DMD patients from 4 different clinical trials by Sarepta, Roche (using Sarepta's vector), P zer and Genethon receiving 3 different gene therapy products differing in AAV serotype, promoter, and dose, showed strikingly similar severe adverse advents that suggested a cytotoxic T-cell immune response against micro-dystrophin proteins 52 .All ve patients had similar large overlapping deletions (exon 8 to exon 21), which was present in micro-dystrophins.This highlights the urgent need for speci c interventions to prevent immune responses that can limit the e cacy of gene therapy and cause irreparable harm.Such interventions could also be applicable for many other therapeutic approaches under development such as gene editing therapies (in which, the bacteriaderived Cas9 protein is delivered).Although beyond the scope of this study, our future efforts will be concentrated on studying the potential immune responses and the approaches to mitigate them.Encouragingly, a recent study from clinical trials showed that it is possible to prevent antibody response to AAV gene therapy by a cocktail of immune modulators ( for example, rituximab plus sirolimus in addition to steroids) to prevent anti-AAV antibody formation. 53In addition, immune tolerance could be induced by hepatic gene transfer as supported by coagulation factor VIII studies in both animals and a patient with hemophilia A. 54 These types of immune modulation could be tested in the future to see if they can also induce long-term immune tolerance towards the full-length dystrophin gene therapy.
Collectively, our work addresses a key de ciency of current AAV-based gene replacement therapy for DMD by delivering functional FL-dystrophin.The novel split intein-mediated assembly coupled with potent myotropic AAV capsids enables FL-dystrophin restoration as well as functional, histopathological and signaling improvement in dystrophic mice.Future studies will be warranted to improve the expression levels and investigate the host immune responses towards FL-dystrophin gene therapy.
Moreover, this approach could potentially be implemented for other diseases with large genetic payloads exceeding the packaging capacity of AAV to be delivered.

Ethics statement
The C57BL/6J and mdx 4cv (B6Ros.Cg-Dmd mdx4cv/J) mice purchased from the Jackson Laboratory and housed at Indiana University Laboratory Animal Resource Center following animal use guidelines.All the experimental procedures were approved by the Institutional Animal Care and Use Committee of Indiana University.All mice were maintained under standard conditions of constant temperature (72 ± 4°F) and humidity (relative, 30-70%), in a speci c pathogen-free facility and exposed to a 12-h light/dark cycle.

Plasmid construction
The plasmid expressing FL dystrophin (p37-2iDMD-LR) was a gift from Michele Calos (Addgene # 88892).The dystrophin fragments, Cfa, Gp41-1, IMPDH-1, Cat and VidaL inteins were synthesized at IDTDNA Technologies and ampli ed by fusion PCR, restriction digested and ligated into pAAV vectors harboring a mini-CMV promoter with a muscle creatine kinase enhancer (meCMV) and a mini-polyA signal.The plasmids were con rmed by Sanger sequencing and/or whole plasmid sequencing.Detailed information on the plasmids is provided in Supplementary Table S1.
Cell culture and transfection AD293 cells were cultured in Dulbecco's Modi ed Eagle's Medium (DMEM) (Corning, Manassas, VA) supplemented with 10% fetal bovine serum (FBS) and 1% 100x penicillin-streptomycin solution (10,000 U/ml, Invitrogen).Cells were plated in 6-well plates and incubated overnight at 37°C.When the cultures reached 80% con uence, they were transfected with plasmids expressing FL or dystrophin fragments using PEI (Polysciences Inc., Pennsylvania, USA).At 72 hours after transfection, cells were collected for protein extraction.

Measurement of serum biomarkers
Blood samples were collected at various time points after retro-orbital injection.The blood samples were allowed to clot for 15 min to 30 min and centrifuged at 2300g for 10 min at room temperature.The supernatant was collected as serum and stored at -80°C in small aliquots for the biochemical assays.
Measurement of CK (326 − 10, SEKISUI Diagnostics LLC) was performed according to the manufacturer's instruction.

Muscle contractility
At 8-12 weeks of age, muscle contractility was measured weekly using an in vivo muscle test system (Aurora Scienti c Inc) as described previously 28,41,42 .Mice were anesthetized with 3% (w/v) iso urane and anesthesia was maintained by 1.5% iso urane (w/v) during muscle contractility measurement.Maximum plantar exion tetanic torque was measured during a train of supramaximal electric stimulations of the tibial nerve (pulse frequency 150 Hz, pulse duration 0.2 ms) using the DMA v5.501 (Aurora Scienti c Inc).

Wire hanging assay
The animal was placed on a custom-made wire mesh, then inverted and suspended above a soft cushion.The latency to when the animal falls is recorded.The mouse will be trained 2-3 times one week before the test.This test is performed three days per week with 2-3 trials per session.The average performance for each session is presented as the average of the trials, and the average for three days is used as the average of the mouse.

Statistical Analysis
The data were expressed as mean ± the standard error of the mean (SEM) and nal gures were assembled with Adobe Photoshop 24.7.0.Statistical differences were determined by two-tailed unpaired Student's t-test for two groups and one-way ANOVA with Tukey's posttests for multiple group comparisons using GraphPad Prism 10.1.0(GraphPad Software, La Jolla, California) with the assumption of Gaussian distribution of residuals.A p value less than 0.05 was considered to be signi cant.HEK293 cell lysate transfected with a FL-dystrophin construct was used as a positive control (FL Ctrl) and the GAPDH was used as a loading control.c-f, i-l, o-r, Densitometry quanti cation of the FL dystrophin band intensity (c, i, o), the ratio of unassembled N versus FL (d, j, p), the ratio of unassembled M versus FL (e, k, q) and the ratio of unassembled C versus FL (f, l, r).One-way ANOVA with Tukey's multiple comparisons test for three groups and two-tailed unpaired Student's t test for two groups.g,m, Diagrams showing the M and C constructs at the split sites with the -1 to -3 and +1 to +3 residues in the dystrophin exteins labeled.The mutated amino acids are labeled in purple (g) and yellow (m).Data are mean ± SEM.Source data are provided as a Source Data le. Figure 6

Figure 1 Design
Figure 1