A transcriptomics-based drug repositioning approach to identify drugs with similar activities for the treatment of muscle pathologies in spinal muscular atrophy (SMA) models

Abstract Spinal muscular atrophy (SMA) is a genetic neuromuscular disorder caused by the reduction of survival of motor neuron (SMN) protein levels. Although three SMN-augmentation therapies are clinically approved that significantly slow down disease progression, they are unfortunately not cures. Thus, complementary SMN-independent therapies that can target key SMA pathologies and that can support the clinically approved SMN-dependent drugs are the forefront of therapeutic development. We have previously demonstrated that prednisolone, a synthetic glucocorticoid (GC) improved muscle health and survival in severe Smn−/−;SMN2 and intermediate Smn2B/− SMA mice. However, long-term administration of prednisolone can promote myopathy. We thus wanted to identify genes and pathways targeted by prednisolone in skeletal muscle to discover clinically approved drugs that are predicted to emulate prednisolone’s activities. Using an RNA-sequencing, bioinformatics, and drug repositioning pipeline on skeletal muscle from symptomatic prednisolone-treated and untreated Smn−/−; SMN2 SMA and Smn+/−; SMN2 healthy mice, we identified molecular targets linked to prednisolone’s ameliorative effects and a list of 580 drug candidates with similar predicted activities. Two of these candidates, metformin and oxandrolone, were further investigated in SMA cellular and animal models, which highlighted that these compounds do not have the same ameliorative effects on SMA phenotypes as prednisolone; however, a number of other important drug targets remain. Overall, our work further supports the usefulness of prednisolone’s potential as a second-generation therapy for SMA, identifies a list of potential SMA drug treatments and highlights improvements for future transcriptomic-based drug repositioning studies in SMA.


Introduction
Spinal muscular atrophy (SMA) is a heterogenous autosomal recessive neuromuscular disorder (NMD) characterized by motor neuron degeneration alongside progressive muscle atrophy and weakness [1].Being the leading monogenic cause of infant mortality [2], around 96% of SMA cases are mapped to homozygous loss-of-function and deletion mutations in the survival of motor neuron 1 (SMN1) gene [3,4], which ubiquitously expresses SMN, a protein that current and ongoing research has linked to diverse housekeeping and tissue-specific cellular functions [5][6][7].Although complete SMN loss is embryonic lethal in most organisms [8], humans can overcome the complete loss of the SMN1 gene due to incomplete rescue by the homologous SMN2 gene [9,10].In essence, the presence of a single nucleotide mutation in SMN2 promotes exon 7 alternative splicing that limits full length SMN (FL-SMN) expression in this gene to 10% [11].Consequently, the limited FL-SMN expression makes SMN2 gene copy number an important disease modifier, impacting SMA type and severity [12].
Thus, additional SMN-independent therapies that target affected tissues such as muscle are needed to further enhance and support the benefits of SMN-dependent treatments [21].Indeed, pre-clinical studies and primary patient data have reported innate muscular defects in SMA, which include myogenesis [22], regeneration [23], contraction [24,25], regulation [25], growth [26], and metabolism [27], highlighting skeletal muscle as a primary therapeutic target.Although two novel skeletal muscle-specific SMA therapies, Apitegromab™ [28] (ClinicalTrials.govID: NCT03921528) and Reldesemtiv™ [29] (ClinicalTrials.govID: NCT02644668), are showing progress in clinical trials, the high expenses involved in novel drug research and development (R&D) [30] alongside the costs of the three clinically approved SMN-dependent therapies may lead to elevated prices for combinatorial treatments [31], thus further widening the accessibility gap for SMA patients.
A useful alternative for ensuring accessibility of SMNindependent treatments for all SMA patients would be the identification of cost-effective generic drugs via drug repositioning, a strategy aimed at finding new therapeutic activities for existing pharmacological compounds [32].One such example is prednisolone, a synthetic glucocorticoid (GC) administered to relieve muscle inf lammation in Duchenne muscular dystrophy (DMD) patients [33,34].Interestingly, evidence also emerged that short-term prednisolone treatment additionally conferred ergogenic muscle benefits in DMD patients [35,36], which was also validated in the mdx mouse model of DMD [37,38].
Although GCs are not used for the treatment of SMA patients, prednisolone is administered for a short period (∼30 days, 1 mg/kg) to alleviate the immunological adverse effects of Zolgensma ® [18].However, given prednisolone's potential muscle benefits, we have previously investigated and demonstrated that prednisolone treatment (5 mg/kg, every other day) in severe Smn −/− ;SMN2 and intermediate Smn 2B/− SMA mice improved survival, weight and neuromuscular phenotype [27].Although the study was aimed at investigating prednisolone's activity on the GC-Krüppel-Like-Factor-15 (KLF15) pathway in SMA skeletal muscle, synergistic muscle improvement was also observed in prednisolone-treated Smn −/− ;SMN2 SMA mice overexpressing Klf15 specifically in skeletal muscle [27], suggesting that prednisolone may act on SMA skeletal muscle via numerous effectors and pathways.
Despite the benefits observed, our study did not evaluate prednisolone's long-term effects [18,27].Given that chronic GC usage increases myopathy [35,39], it is unclear whether long-term prednisolone treatments would be similarly detrimental in SMA muscle.Furthermore, the rapid onset and progression of disease in SMA mouse models (1-3 weeks on average) [9,40,41] does not allow sufficient comparison of intermittent vs chronic studies.
Thus, in this study, we used a transcriptomics and drug repositioning pipeline based on our previously published experimental paradigm [42] to uncover the genes and pathways restored by prednisolone in skeletal muscle of SMA mice and identify existing non-GC drugs predicted to have similar activities (Fig. 1).Our study uncovered that prednisolone restored pathways linked to growth, metabolism, and regulation in SMA skeletal muscle and identified 20 leading commercially available non-GC drugs predicted to emulate its action.Based on oral bioavailability and evidence of safety treatment in children, we selected and validated metformin and oxandrolone in SMA cellular and animal models (Fig. 1).Although, both metformin and oxandrolone improved neuromuscular activity in the Caenorhabditis elegans (C.elegans) model for severe SMA, we found that higher metformin doses reduced survival in the Smn 2B/− SMA mouse model.On the other hand, oxandrolone treatment partially improved survival in Smn 2B/− SMA mice, albeit not to the same extent as prednisolone [27].
Nevertheless, our study computationally uncovered new mechanisms behind prednisolone's beneficial activity in SMA muscle, identified numerous potential SMA muscle-specific therapeutic candidates and highlighted the importance of transcriptomic-based drug repositioning for SMN-independent drug discovery.

Prednisolone restores the expression of a large subset of genes involved in canonical skeletal muscle pathways in SMA mice
As described in an earlier study, we have previously demonstrated that treating SMA mice with prednisolone significantly improved several disease phenotypes, including survival, weight, and muscle health [27].To have a more in depth understanding of the impact of prednisolone on SMA skeletal muscle at a molecular level, we performed bulk RNA-sequencing (RNA-Seq) on skeletal muscle of untreated and prednisolone-treated Smn −/− ;SMN2 SMA and Smn +/− ;SMN2 healthy mice.Specifically, we administered prednisolone (5 mg/kg, gavage, every 2 days) starting from postnatal day (P) 0 until P7 to Smn −/− ;SMN2 SMA and Smn +/− ;SMN2 healthy mice [27].Triceps were harvested from P7 prednisolonetreated and untreated mice for RNA-Seq via Illumina NextSeq550 and a HISAT2-FeatureCounts-DESeq2 pipeline against a Mus Musculus mm10 genome for parameters of "condition" and "treatment" (Fig. S1).

Figure 1.
Graphical representation of the study aimed at using a transcriptomics-based drug repositioning strategy to predict and validate drug candidates that emulate prednisolone's activity in SMA skeletal muscle.(1) 5 mg/kg prednisolone was administered was administered to Smn −/− ;SMN2 SMA mice and Smn +/− ;SMN2 healthy control animals every other day from post-natal day (P) 0 to P7. (2) Total RNA was extracted from the triceps of prednisolone-and untreated SMA mice and Smn +/− ;SMN2 healthy control animals and underwent library preparation for next generation sequencing against Mus musculus mm10 genome using Illumina NextSeq 550.(3) The RNA-sequencing (RNA-Seq) reads initially underwent QC analysis via FASTQC prior to a transcriptomics pipeline of mm10 M. musculus genome alignment (HISAT2), read count quantification (FeatureCounts) and differential gene expression analysis (DESeq2) to generate differentially expressed genes.(4)The differential gene expression patterns between prednisolone-treated vs untreated Smn −/− ;SMN2 SMA triceps was used to identify significant pathways targeted by prednisolone (iPathwayGuide).Both differentially expressed genes and significant pathways were used to predict clinically approved drug candidates from online databases (iPathwayGuide, KEGG drugs and drug gene interaction database (DGIdb)) that emulate prednisolone's activity in skeletal muscle of SMA mice.(5) The activity and safety of predicted drug candidates of interest and their target genes were then assessed in both in vivo (Smn 2B/− SMA mice and Caenorhabditis elegans smn-1 (ok355)) and in vitro (C2C12 muscle-like cells) models.This figure was generated in BioRender.
Combined, our transcriptomics and pathway analyses suggest that prednisolone treatment attenuated muscle pathologies in SMA mice [27] by targeting key muscle metabolism, atrophy and regulatory pathways.

Drug repositioning algorithms identify novel pharmacological compounds predicted to emulate prednisolone's activity in skeletal muscle of SMA mice
As mentioned above, while prednisolone treatment significantly improves muscle health and overall disease progression in SMA mice, chronic use of prednisolone can negatively impact skeletal muscle [35,39].As such, we used the DEGs and associated KEGG pathways identified in prednisolone-treated Smn −/− ;SMN2 SMA mice to discover alternative drugs predicted to mimic prednisolone's molecular effects in SMA skeletal muscle.Initially, we utilized the in-built integration of KEGG drugs database in iPathwayGuide [50] and the DGIdb v3.0 [51] database to initially reveal a total of 580 compounds (Tables S7-S10).To filter down our list, we focused on the drug compounds 1) that targeted > 5 prednisolone-targeted pathways or linked to upstream regulators, 2) were clinically approved and 3) were not associated with promotion of muscle-wasting (e.g.primary anti-cancer drugs [52]), leaving a total of 20 potential candidates (Tables 3-4).
Interestingly, our combined in silico drug repositioning approach revealed a subset of candidates previously investigated in SMA such as celecoxib [53] (ClinicalTrials.govID: NCT02876094) and colforsin [54].To further validate our bioinformatics strategy, we chose to continue our study with drugs not yet assessed for SMA, focusing on those previously used safely in young patients and orally bioavailable.With these criteria, we narrowed down our selection to metformin, a generic asymmetric dimethyl-biguanide type 2 diabetes mellitus (T2DM) drug [55] and oxandrolone, a synthetic anabolic steroid with a higher ratio of anabolic: androgynous effects for further study [56].
Thus, using a transcriptomics-based in silico drug repositioning platform, we were able to generate a list of clinically approved pharmacological compounds that are predicted to emulate prednisolone's activity in skeletal muscle.

Molecular docking simulations support the predicted drug-target interactions identified by the transcriptomics pipeline
Having selected to move forward with metformin and oxandrolone based on a combined bioinformatics and drug repurposing strategy, we performed molecular docking simulations to confirm that our drugs of interest indeed interacted with their predicted molecular targets (Prkag3 for metformin and androgen receptor (AR) for oxandrolone).In the case of metformin, no crystallographic data was available for Prkag3 (AMPK-γ 3).We thus studied the interactions of metformin with Prkag1 (AMPK-γ 1) (Fig. 3a), as both encode the AMPK-γ subunit, share 64% sequence homology and can form an AMPK trimeric complex with the same AMPK α2 and β2 subunits in human and mouse skeletal muscle [57].Nevertheless, both metformin and oxandrolone demonstrated strong binding affinity to their targets of interest with metformin binding at Prkag1 key residues such as Ala 204, Ala 226 and Asp316 (Fig. 3a) and oxandrolone anchoring to the AR active sites via hydrogen bonds with Asn 705, Arg 752 and Thr 877 (Fig. 3b).Furthermore, docking studies across our larger list of orally bioavailable drug candidates (Tables 3-4) show that most of the drug molecules were predicted to fit in their respective target active sites, aligning well with known co-crystallized ligands (Fig. S2, Table S11).Thus, these molecular docking simulations support further investigating the therapeutic potential of the drug candidates identified via a transcriptomics and drug repurposing approach.

Metformin's primary predicted target gene
Prkag3 is dysregulated in skeletal muscle of both severe Smn −/− ;SMN2 and intermediate Smn 2B/− SMA mice As previously mentioned, metformin is an orally administered T2DM drug that we selected as one of the candidates to validate our bioinformatics-based drug repositioning approach.Importantly, metformin has over 60 years of clinical use with a wellknown safety profile [55] and recorded administration in younger patients [58].Furthermore, it has been previously repositioned and conferred ergogenic activities in muscular disorders such as DMD [59] and congenital muscular dystrophy type 1 A (CMDT1A) [60], highlighting its potential as a skeletal muscle therapy.
We thus measured the mRNA expression levels of Prkag3 and FoxO isoforms in the triceps of both symptomatic P7 severe Smn −/− ;SMN2 and P19 milder Smn 2B/− SMA mice alongside their respective healthy controls.We indeed observed that Prkag3 levels were significantly downregulated in skeletal muscle of both SMA mouse models (Fig. 4a), supporting the bioinformatics data.However, none of the FoxO isoforms were significantly different between SMA mice and their healthy controls (Fig. 4b-e).Previous research also reported no significant upregulation of FoxO isoforms in P7 severe Smn −/− ;SMN2 SMA mice via qPCR [43].Although another study observed a significant upregulation of FoxO isoforms in the hindlimb muscles of Smn 2B/− SMA mice [43], the wide variation observed in our experimental cohort could be due to differential vulnerabilities to denervation-induced muscle atrophy between hindlimb and triceps muscle [63] that could impact FoxO expression levels.However, the fact that our qPCR data did not ref lect the bioinformatics predictions may also be due to variability in our experimental cohorts, the sequencing depth coverage not being sufficiently conservative and/or intrinsic differences between RNA-Seq and primer-based qPCR approaches [64].
Overall, our qPCR experiments revealed that the primary metformin target Prkag3 matched its bioinformatics prediction of being downregulated in both Smn −/− ;SMN2 and Smn 2B/− SMA mice, suggesting that this gene may be involved in both severe and milder SMA pathologies and an appropriate therapeutic molecular target in SMA muscle.

The predicted target genes for metformin are mostly Smn-independent in an SMA muscle cellular model
We next wanted to better understand if the aberrant expression of the metformin target genes was dependent on SMN expression and/or muscle atrophy.Thus, we firstly generated small interfering RNA (siRNA)-mediated Smn-depleted C2C12 myoblast-like cells, a useful and previously successful in vitro model [65].We confirmed by qPCR that Smn mRNA levels were significantly reduced by up to 90% in C2C12 myoblasts and D8 C2C12 myotubes compared to scrambled siRNA and untreated controls (Fig. S4).We next investigated the effects of Smn knockdown on the expression of the predicted metformin target genes.In C2C12 myoblasts, we identified a significant upregulation of only the FoxO3 gene in Smn-depleted C2C12 myoblasts compared to controls (Fig. 5a), which ref lects previous microarray analyses of specific FoxO isoforms upregulated in quadriceps femoralis muscle biopsies from Type 1 SMA patients [66].However, in C2C12 myotubes we found that Smn knockdown (KD) had no effect on the expression of predicted metformin target genes (Fig. 5b), suggesting that for the most part, the expression of the predicted metformin genes is Smn-independent, thus representing ideal targets for SMNindependent therapies.
We next investigated if the expression of the predicted metformin target genes is affected in vitro by muscle atrophy.However, one difficulty in mimicking SMA muscle atrophy in vitro is establishing denervation.Thus, based on evidence of shared pathway similarities from different pro-atrophy factors such as starvation and denervation [67], we used a validated method of 24-h serum-starvation in C2C12 myotubes to induce canonical atrophy, as confirmed by myotube loss and upregulation of proatrophic atrogin-1 levels (Fig. 5c).Next, we evaluated the expression of the predicted metformin target genes and observed a significant upregulation of FoxO3 and FoxO4 isoforms (Fig. 5d), ref lecting their established roles in atrophy-dependent ubiquitinproteasome pathways [43].
We then evaluated whether metformin could attenuate muscle atrophy in C2C12 myotubes.Based on initial gene-dose response experiments in both control C2C12 myoblasts and D8 myotubes, we treated our cells with physiological (60 μM) and supraphysiological (2 mM) metformin concentrations for 24 h (Fig. S5) [68].The 30 μM physiological metformin concentration for 24 h did not attenuate muscle atrophy or impact the expression of the target genes in the serum starved C2C12 myotubes (Fig. 5e and f).However, for the supraphysiological 2 mM metformin concentration [68], we observed an upregulation of Atrogin-1 levels (Fig. 5e), suggesting an exacerbation of muscle atrophy.Further analysis of the predicted metformin target genes revealed no significant impact on their expression patterns either (Fig. 5f), suggesting that exacerbation of atrophy in C2C12 myotubes by supraphysiological metformin concentrations involves factors mostly outside of our predicted targets.However, it should be noted that metformin may have different effects in SMA muscle as there are still differences between distinct pro-atrophic factors [67].
Overall, our in vitro studies revealed that although most of our predicted metformin target genes are SMN-independent with some linked to muscle atrophy, they were mostly not linked to metformin's inf luence on canonical atrophy in C2C12 myotubes.
We initially administered a 200 mg/kg daily dose (diluted in 0.9% saline) starting from P5 until humane endpoint in Smn 2B/− SMA mice and Smn 2B/+ healthy control littermates, based on previous success in the muscle disorder CMDT1A, where metformin concentrations ranged from 100-250 mg/kg [60].We observed no significant improvement in survival of Smn 2B/− SMA mice treated with 200 mg/kg/day compared to untreated (Fig. 6a) and vehicletreated animals (Fig. S6).We also found a significant reduction in the body weight of 200 mg/kg/day metformin-treated Smn 2B/− SMA mice compared to untreated SMA animals, beginning 4 days after initial treatment at P9 (Fig. 6b).However, we did not observe any effects on weight in the 200 mg/kg/day metformin treated Smn 2B/+ healthy control mice (Fig. S6b), indicating a disease specific effect of metformin.In terms of motor function, there was no significant difference in the righting ref lex between untreated and 200 mg/kg/day metformin-treated Smn 2B/− SMA mice (Fig. 6c) and Smn 2B/+ healthy control animals (Fig. S7a-c).
Since our initial 200 mg/kg/day metformin dosage did not improve disease onset or disease progression in Smn 2B/− SMA mice, we conducted pilot studies with a later treatment start point (P8) and a lower dose (100 mg/kg/day), which demonstrated similar effects to our initial dosing regimen (data not shown).We therefore tried a higher daily dosage of 400 mg/kg/day, starting at P5. Surprisingly, the higher 400 mg/kg/day dose significantly reduced survival in Smn 2B/− SMA pups by 5 days (Fig. 6d), while having no significant impact on weight or righting ref lex (Fig. 6e and f).Interestingly, 400 mg/kg/day metformin had no adverse effects in the healthy Smn 2B/+ control littermates (Fig. S7d-f), suggesting that the adverse effects of the higher dose of metformin is disease specific.
Thus, our in vivo experiments demonstrated that metformin did not emulate prednisolone's beneficial effects on SMA disease progression in SMA mice and is in fact not an ideal therapy candidate, due to dose-and disease-dependent adverse effects in SMA.

Higher 400 mg/kg/day metformin dosage is associated with hypoglycaemia in non-fasted Smn 2B/− SMA mice
We next investigated the potential causes behind metformin's adverse effects in SMA mice.With metformin being a glucose lowering agent for T2DM, we initially assessed blood glucose levels in P14 non-fasted, untreated, 200-and 400-mg/kg/day metformin treated Smn 2B/− SMA and Smn 2B/+ healthy mice 2 h after the final treatment.This time point was chosen to account for the reduced median survival in the higher dose SMA cohort.We observed that neither 200-or 400-mg/kg/day metformin treatments lowered blood glucose levels in Smn 2B/+ healthy mice (Fig. 7a).However, we reported a significant reduction in blood glucose levels in 400 mg/kg/day metformin-treated Smn 2B/− SMA mice compared to untreated SMA animals (Fig. 7a).Our results suggest that hypoglycaemic shock could have been one of the possible causes behind the premature death in the 400 mg/kg/day SMA cohort, further exacerbating the previously reported hypoglycaemia in SMA models [70] and patients [73,74].

Metformin did impact the expression of the predicted target genes, but not muscle pathology markers in Smn 2B/− SMA mice
We next evaluated the effect of metformin on the expression of the predicted target genes in TAs from P14 untreated, 200-and 400-mg/kg/day metformin-treated Smn 2B/− SMA and Smn 2B/+ healthy mice 2 h after final treatment.Contrary to the drug repositioning prediction that metformin could reverse Prkag3 downregulation in SMA muscle (Fig. S3), we instead discovered that both 200-and 400-mg/kg/day metformin doses exacerbated Prkag3 downregulation in Smn 2B/− SMA muscle (Fig. 7b).Furthermore, although metformin had no impact on FoxO1, FoxO3 and FoxO4 isoforms in both Smn 2B/+ healthy and Smn 2B/− SMA muscle (Fig. 7c-e), for the 200 mg/kg/day Smn 2B/−  SMA cohort, metformin significantly reduced FoxO6 expression (Fig. 7f), which again contrasted our bioinformatics prediction that metformin would upregulate the expression of this isoform in SMA muscle.Thus, the drug-gene response for metformintreated Smn 2B/− SMA skeletal muscle reveals a contrasting pattern that does not match the bioinformatic predictions.
We next investigated metformin's effects on the expression of dysregulated molecular markers associated with muscle atrophy (Atrogin-1 and MuRF-1) and glucose metabolism (Glut4 and Hk2) [75,76].We however observed that neither atrophy (Fig. 7g and h) or glucose metabolism markers (Fig. 7i and j) were affected by 200-and 400-mg/kg/day metformin treatments in the Smn 2B/− SMA mice when compared to untreated animals.
We also investigated markers associated with mitochondrial biogenesis and function in muscle (Pgc1-α, Tfam, Ndufs1, and Nrf1), as previous research has established that these features are impaired in SMA skeletal muscle [46] and a common mechanism of action for metformin is mild inhibition of mitochondrial electron transport complex 1 (or NADH:ubiquinone oxidoreductase) [77].We found that that neither the 200-or 400-mg/kg/day dose of metformin impacted the expression of mitochondrial genes in the skeletal muscle of Smn 2B/− SMA mice (Fig. 7k-n), although we do observe a significant downregulation of Pgc1-α, Tfam, and Ndufs1 in the skeletal muscle of 200 mg/kg/day metformin-treated Smn 2B/+ healthy counterparts (Fig. 7k-n).
Overall, our data highlights that metformin did not have a direct impact on the predicted target genes in skeletal muscle of SMA mice.Furthermore, the absence of direct impact on muscle atrophy, glucose metabolism, and mitochondrial function markers following metformin treatment in Smn 2B/− SMA muscle, suggests that the adverse effects associated with the 400 mg/kg/day dosage may not have been linked to muscle-intrinsic effects.

A higher dose of metformin is associated with dysregulation of mitochondrial regulatory genes in the spinal cord of Smn 2B/− SMA mice
We next investigated the effects of metformin on the spinal cord given that metformin is systemically distributed [78], has the ability to cross the blood-brain-barrier (BBB) [79] and can impact the mitochondria in the spinal cord [80].We thus evaluated whether metformin altered the expression of mitochondrial markers (Pgc1α, Tfam, Nrf1 and Ndufs1) in the spinal cord of P14 untreated, 200-and 400-mg/kg/day metformin treated Smn 2B/− SMA mice compared to Smn 2B/+ healthy mice, 2 h after final treatment.
For Pgc1-α, a master regulator of mitochondrial biogenesis and function, we observed that although both 200-and 400-mg/kg/day metformin concentrations significantly reduced its expression  levels in Smn 2B/+ healthy spinal cords (Fig. 8a), it was only the higher concentration that significantly reduced Pgc1-α expression in Smn 2B/− SMA spinal cords (Fig. 8a).Similarly, 400 mg/kg/day metformin significantly reduced Ndufs1 levels in both Smn 2B/+ healthy and Smn 2B/− SMA spinal cords (Fig. 8b), suggesting that for these mitochondrial health markers, the higher metformin dose negatively affected their expression independent of disease status.On the other hand, although not affected by metformin in Smn 2B/− SMA spinal cords, Nrf1 gene expression was significantly reduced by both metformin doses in the spinal cord of Smn 2B/+ healthy mice (Fig. 8c), while Tfam was not affected by metformin in either cohort (Fig. 8d).Our results demonstrating that the higher dose of metformin (400 mg/kg/day) appears to specifically dysregulate certain mitochondrial genes in the spinal cord of Smn 2B/− SMA mice is supported by recent evidence of tissue-dependent differences in conserved cellular processes between SMA motor neurons and skeletal muscle [49].Thus, although further in-depth investigations would be needed, our results on mitochondrial health markers suggest that metformin's adverse effects in SMA mice could be linked to the exacerbation of neuronal mitochondrial dysfunction.
Oxandrolone's predicted target gene, Ddit4, is dysregulated in the skeletal muscle of severe Smn −/− ;SMN2 SMA mice Our second drug candidate that we selected to mimic prednisolone activities was oxandrolone, a synthetic orally bioavailable anabolic steroid that confers minimal androgynous effects [56].Importantly for SMA, oxandrolone has been successful in the promotion of muscle growth for DMD [81] and mixed gender burn injury patients [82].
Oxandrolone was predicted to upregulate the Ar gene in SMA muscle (Fig. S8).The upregulation of Ar was predicted to directly upregulate downstream effectors Igfbp5 and myogenin (or MyoG) (Fig. S8), which both regulate muscle differentiation, regeneration and myofiber growth [83,84].Furthermore, Ar was predicted to indirectly upregulate Dok5, a signalling protein linked to insulin and IGF-1 activity [85] and Akap6, which is involved in the modulation of muscle differentiation and regeneration [86] (Fig. S8).
In addition to these factors, we also decided to investigate Ddit4 as an oxandrolone target based on its direct relation with Ar [87] and being one of the top 20 downregulated DEG targets of prednisolone in Smn −/− ;SMN2 SMA skeletal muscle (Table S2; Fig. S8).
Similar to our metformin strategy above, we initially wanted to evaluate the mRNA expression levels of these target genes in the triceps of both symptomatic P7 severe Smn −/− ;SMN2 and P19 milder Smn 2B/− SMA mice alongside their respective healthy controls.Overall, we identified no significant dysregulated expression of the target genes Ar, Akap6, Igfbp5, Dok5 and MyoG in both severe Smn −/− ;SMN2 and milder Smn 2B/− SMA mice (Fig. 9a-e).However, for Ddit4, we did identify a significant upregulation only in Smn −/− ;SMN2 SMA mice (Fig. 9f), supporting both our bioinformatics data for this gene and its known pro-atrophic role [87], indicating that it may play an important role in SMA muscle pathologies.In summary, the majority of the predicted oxandrolone target genes did not significantly ref lect their bioinformatic predictions.

In vitro oxandrolone treatment prevents canonical atrophy in C2C12 myotubes independently of the predicted Smn-independent targets
Similar to our metformin in vitro studies, we wanted to evaluate whether reduced SMN levels or atrophy inf luenced the expression of the predicted oxandrolone target genes in SMA skeletal muscle.Although none of the target genes were affected in the Smndepleted C2C12 myoblasts (Fig. 10a), we found that Smn KD triggered a significant upregulation of Dok5 only in C2C12 myotubes (Fig. 10b), suggesting that the expression of this gene may be Smndependent.Nevertheless, the expression of the majority of the predicted oxandrolone target genes was Smn-independent.
We next wanted to evaluate the ability of oxandrolone to attenuate canonical atrophy in serum-deprived C2C12 myotubes [88].However, in this case we performed the treatments in D5 C2C12 myotubes instead of D8, as although oxandrolone was non-toxic (Fig. S9), it elicited a greater androgen Ar response at the earlier differentiation stage (Fig. S10).Following confirmation of muscle atrophy in D5 C2C12 myotubes via significant Atrogin-1 and MuRF1 upregulation (Fig. 10c), we observed that the expression of the predicted oxandrolone target genes Akap6, Igfbp5, MyoG and Ddit4 was significantly downregulated in serum-deprived D5 C2C12 myotubes (Fig. 10d).
Interestingly, we found that 24-h treatment with 1 μM oxandrolone attenuated canonical muscle atrophy in these serumstarved D5 C2C12 myotubes as shown by significant downregulation of Atrogin-1 (Fig. 10e).However, we observed that Igfbp5, MyoG and Ddit4 were further downregulated by the 1 μM oxandrolone treatment (Fig. 10f), suggesting that oxandrolone's effects on atrophy are linked to effectors independent of the predicted target genes.
Overall, our in vitro studies have shown that although the expression of the predicted oxandrolone target genes is Smnindependent, they are not involved in oxandrolone's ameliorative effects on canonical atrophy in C2C12 myotubes.

Oxandrolone treatment improves survival in Smn 2B/− SMA mice
We next assessed the impact of oxandrolone in SMA mice.We initially tested preliminary treatment regimens of 1-8 mg/kg/day starting from P5 or P8 in Smn 2B/− SMA and Smn 2B/+ healthy mice (data not shown), based on previous studies in models of spinal cord injury (SCI) [89] and burn injury [90].We also stopped oxandrolone treatments at P21 as previous research has shown that shorter oxandrolone treatments are more effective [89].These pilot studies allowed us to identify the optimal dosing regimen of 4 mg/kg/day oxandrolone treatment from P8 to P21, which significantly improved the median survival of Smn 2B/− SMA mice (Fig. 11a).
However, we found that the body weight of 4 mg/kg/day oxandrolone-treated Smn 2B/− SMA mice was significantly lower compared to their untreated counterparts (Fig. 11b), which is most likely due to the intrinsic smaller sizes of the randomly assigned litters, as demonstrated by the difference in weight starting 4 days prior to initial treatment (Fig. 11b).In terms of motor function, we observed no significant difference in righting ref lex between untreated and oxandrolone-treated SMA animals (Fig. 11c).Furthermore, we identified no impact of vehicle treatment on survival, weight, and righting ref lex in Smn 2B/− SMA mice (Fig. S11).
In the Smn 2B/+ healthy mice, although 4 mg/kg/day oxandrolone had no effect on survival or motor function in treated animals (Fig. S12a and b), we did observe a significant decrease in bodyweight starting from P9, one day after initial treatment (Fig. S12c), suggesting that oxandrolone may have impacted growth.
Nevertheless, our results demonstrated that although 4 mg/kg/day oxandrolone treatment improved survival in Smn 2B/− SMA mice, its effect on survival was still minor compared to prednisolone [27], suggesting that oxandrolone is not a suitable substitute as an SMA skeletal muscle therapy.

Oxandrolone did not impact the expression of the predicted target genes or muscle pathology markers
The improved 3-day survival in 4 mg/kg/day oxandrolone-treated Smn 2B/− SMA mice led us to evaluate whether this beneficial impact was related to targeting muscle pathologies.Thus, we evaluated oxandrolone's effects on the expression of dysregulated molecular markers associated with the SMA hallmark pathology of muscle atrophy (Atrogin-1 and MuRF-1) in the triceps from P19 late symptomatic, untreated and 4 mg/kg/day oxandrolonetreated Smn 2B/− SMA and Smn 2B/+ healthy mice, 2 h after final treatment.We observed no significant reduction in elevated Atrogin-1 or MuRF-1 gene expression levels by oxandrolone in the Smn 2B/− SMA cohort (Fig. 11d and e), suggesting that oxandrolone did not attenuate muscle atrophy.
We next evaluated the effect of oxandrolone on the expression of the predicted target genes in the same P19 untreated and 4 mg/kg/day oxandrolone-treated Smn 2B/− SMA and Smn 2B/+ healthy mice.We found that oxandrolone did not significantly impact the predicated target genes in the triceps from the Smn 2B/− SMA mice (Fig. 11f-k).However, we did observe that 4 mg/kg/day oxandrolone treatment significantly upregulated Dok5 expression in the Smn 2B/+ healthy mice (Fig. 11i).Nevertheless, the pattern observed suggests that oxandrolone did not impact any of the predicted genes in the muscle from Smn 2B/− SMA mice.
Overall, our data shows that oxandrolone did not have an efficient effect on the predicted target genes.Furthermore, its inability to ameliorate muscle atrophy marker dysregulation in SMA skeletal muscle, suggests that improved survival in the Smn 2B/− SMA mice by oxandrolone may be independent of targeting skeletal muscle pathologies.

Both metformin and oxandrolone drug candidates attenuate neuromuscular phenotypes in the C. Elegans severe SMA model
We next wanted to investigate our drug candidates in a separate SMA model to assess whether they could attenuate neuromuscular dysfunctions in a distinct species.For this purpose, we used the C. elegans smn-1(ok355) invertebrate model [91], based on shared conservation of the SMN protein with vertebrate species [92] and the well described phenotypic defects of larval lethality (reduced survival) and impaired neuromuscular function in pharyngeal pumping for feeding and locomotion [91,93].For metformin, administration at higher doses of 50 mM partially ameliorated multiple phenotypes only in C. elegans smn-1(ok355), including pharyngeal pumping defects and the locomotory defect of number of reversals (Fig. 12a and b), however, only the lower dose of 1 mM metformin significantly ameliorated paralysis times (Fig. 12c).On the other hand, oxandrolone across 1-50 μM doses significantly ameliorated pharyngeal pumping defects only (Fig. 12d), with no significant effect on locomotory defects of reversal and paralysis times (Fig. 12e and f), suggesting an improvement in muscular activity in the pharynx region.Overall, given SMN conservation across species, oxandrolone could improve neuromuscular defects across vertebrate and invertebrate models of SMA.

Discussion
The goal of this study was to use a transcriptomics-based drug repositioning strategy to identify clinically approved drug candidates that could emulate prednisolone's beneficial effects in SMA skeletal muscle and life expectancy [27], without the risks associated with long-term GC exposure [39].
Our major finding was the observation that prednisolone treatment restored specific gene sets associated with key pathological SMA pathways such as FoxO signalling [43], p53 signalling [44], AMPK signalling [45], mitophagy [46], circadian rhythm [47], PPAR signalling [48], and autophagy [49] in Smn −/− ;SMN2 SMA skeletal muscle.Although these pathway results highlight prednisolone's efficacy in improving skeletal muscle health, it should be noted that our transcriptomic data cannot distinguish whether these restorations are targeted directly by prednisolone or a consequence of improved muscle health.Nevertheless, our transcriptomic data complemented the prior phenotypic data of prednisolone's potential as a second-generation SMA therapy [27].
Despite a multitude of promising compounds identified that could be investigated in future studies, one of the findings of our study was that neither of our chosen orally bioavailable drug candidates, metformin and oxandrolone, reproduced prednisolone's previously reported effect on muscle growth and survival in SMA mice [27].For metformin, we observed that both 200-and 400mg/kg/day doses counterintuitively exacerbated Prkag3 downregulation in Smn 2B/− SMA muscle instead of the predicted upregulation, which could have negative consequences since Prkag3 −/− null mice presented metabolic and mitochondrial dysregulation [94,95] similar to those reported in SMA patients [46,74] and models [70,71].More surprising was the reduced survival of Smn 2B/− SMA mice treated with 400 mg/kg/day metformin, which we potentially linked to possible hypoglycaemic shock and/or dysregulation of neuronal mitochondrial components.On the other hand, although various metformin doses ameliorated neuromuscular dysfunction in our SMA C. elegans model, the negative effects observed in Smn 2B/− SMA mice could be to due vertebrate vs invertebrate systems.Curiously, previous research into AMPK agonism via 500 mg/kg AICAR treatment in the SmnΔ7 SMA mice contrasts our metformin data, as they showed improved skeletal muscle health and no exacerbations in neuronal dysfunction [45].One explanation for metformin and AICAR's conf licting results in SMA mice could be related to differential pharmacological activities.Indeed, AICAR is an AMP analog with a low BBB penetrability [96] that directly binds to the γ -AMPK isoform [97,98], while metformin can rapidly penetrate the BBB [79] and has been associated with direct and indirect AMPK activation [78], an example of the latter involving mitochondrial respiratory complex 1 inhibition [77].With emerging evidence of naturally low mitochondrial respiratory complex 1 activity in SMA motor neurons [99] and SMA pathway profiles being tissue-dependent [49], one theory could be that the 400 mg/kg/day metformin dose exacerbated mitochondrial respiratory complex 1 inhibition in SMA motor neurons.However, future studies would be needed to verify this proposed model.Nonetheless, our findings could be important for clinical drug safety, as with reported co-morbidities of diabetes in certain SMA patients [100,101], our pre-clinical data suggests lower metformin doses or non-biguanide drugs may be important to manage diabetes and not risk primary pathologies in SMA patients.
For oxandrolone, our mouse data showed that 4 mg/kg/day treatment from P8 partially improved survival in Smn 2B/− SMA mice, although not to the same extent observed with prednisolone [27].In addition, we identified that the lower dose of 1 μM oxandrolone in vitro attenuated canonical atrophy in C2C12 myotubes, while in vivo oxandrolone attenuated neuromuscular dysfunction in severe SMA C. elegans model, suggesting that in both our SMA vertebrate and invertebrate species oxandrolone may be a beneficial SMN-independent treatment option.
An additional factor that could contribute to the different pharmacological response in observed in our mouse and C. elegans models is sex-specific differences.Indeed, emerging clinical evidence collected from 3838 SMA patients across 18 countries shows that severe neuromuscular pathologies are more frequent in SMA types 2 and 3b male patients compared to females [102].Furthermore, both metformin and oxandrolone have demonstrated sex-specific effects on muscle pathology.Indeed, metformin treatment significantly improved skeletal muscle function only in the female mouse model for CMDT1A [60].On the other hand, studies with Ar knock out (KO) rodent models have demonstrated that Ar absence does not have the same impact on female muscle size compared to males [103].In addition, it is suggested that intramuscular Ar content may have a stronger inf luence on hypertrophy than peripheral androgen levels [104], suggesting that oxandrolone may be more beneficial in males.Thus, future studies need to take into consideration the sex of the SMA models when evaluating drug efficacy.
Despite our study's limitations, it highlighted refinements for future in silico SMA drug repositioning studies.Compared to a previous study that successfully identified and validated harmine's therapeutic potential in SMA muscle [42], ours did not include proteomics.The absence of proteomics can be a caveat for drug studies as transcript levels alone do not proportionally ref lect drug-protein interactions, abundance and translational modifications [105,106].However, a limitation of both transcriptomics and proteomics approaches is that they cannot bridge drug-pathway interactions with disease phenotypes, as demonstrated by a recent proteomics analysis of Spinraza ® -treated Type 2 and 3 SMA patients that could not correlate protein profiles with functional improvements [107].Thus, implementation of metabolomics may be beneficial for linking pathway perturbations with metabolites associated with disease and stages of muscle atrophy [108].To the best of our knowledge, this three-pronged multi-omics approach has not previously been used in SMA drug repositioning, however it has been successful in the identification of 200 biomarker candidates for SMA [109].Importantly, and as demonstrated in our previous work [42], the power of multi-omics lies in the ability of further defining the molecular and signalling changes that occur in cells and tissues following treatment, which can allow for the refinement of therapeutics targets and repurposed drugs as well as for the investigation of potential mechanisms behind responsive and non-responsive cohorts.
Another consideration is systemic effects of the drugs as seen with the enhanced lethality of the 400 mg/kg/day metformin's dose in Smn 2B/− SMA mice being linked to hypoglycaemia and mitochondrial dysfunction in neuronal tissue.Indeed, adverse systemic risks were also found with a previous multi-omics drug repositioning study for SMA muscle that identified the development of tremors in harmine-treated Smn 2B/− SMA mice [42].With tissue-dependent differences in conserved pathways in SMA [49], future omics studies could integrate data from both neuronal and skeletal muscle to minimize systemic adverse risks.
Even with these refinements, future drug repositioning studies for SMA skeletal muscle pathology may need to consider replacing bulk RNA-Seq.Skeletal muscle fibres are comprised of a myriad of different muscle fibre and cell types, alongside non-muscular interconnecting tissues such as neurons, tendons, adipose, immune cells and capillaries [110,111], which altogether does not truly represent the transcriptomic profiles for distinct skeletal muscle cells.Indeed, our significant KEGG pathway results included those associated exclusively with neuronal, immune, and capillary cells such as glioma, atherosclerosis and Th17 cell differentiation.With alterations in fibre type composition [72,112] and satellite cell dysregulation evidenced in SMA muscle [23], an emerging alternative to predict drug candidates that target dysregulated SMA pathways in these muscle types would be single-cell (scRNA-Seq) and/or single nuclear RNA-Seq (snRNA-Seq) [111], which have already been useful tools in other muscular disorders such as DMD [113,114].In addition, snRNA-Seq could have further benefits to narrow in on muscle regions such as nuclei located near the NMJ, since myonuclei display spatial and temporal expression pattern heterogeneity in multi-nucleated myofibers [111].
Although the drug candidates metformin and oxandrolone did not emulate prednisolone's beneficial effects in SMA to the extent previously reported, our transcriptomics-drug repositioning approach did better define prednisolone's activity in SMA muscle and provided a list of potential candidates for future preclinical SMA drug repositioning studies.Furthermore, our study highlights important refinements for future SMA drug repositioning studies.
The Smn 2B/− SMA mouse model [69] was generated by breeding Smn 2B/2B mice (generously provided by Dr Rashmi Kothary (University of Ottawa), Dr Lyndsay Murray (University of Edinburgh) and Professor Matthew Wood (University of Oxford) before being sent to Charles River for rederivation) with Smn +/− mice (B6.Cg-Smn1/J, stock #007963, Jackson Labs).All live procedures on wild type (WT) (C57BL/6 background), Smn 2B/− SMA and Smn 2B/+ healthy littermates were performed in the Keele University BSU, in accordance with the UK Home Office authorisation (Animals Scientific Procedures Act (1986), UK Home Office Project Licence P99AB3B95).
For all behavioural experiments, body weights and righting ref lex [115] (up to 30 s) were assessed daily from birth until animals reached their humane endpoint, defined in our UK Home Office Project Licence (P99AB3B95) as the time at which the animal displays hindlimb paralysis, immobility, inability to right (greater than 30 s), 4 consecutive days of weight loss and/or up to 20% body weight loss.

Blood-glucose measurement
Blood was collected from non-fasted Smn 2B/− SMA and Smn 2B/+ healthy mice and glucose levels were immediately measured (mmol/l) via True Metrix ® GO Self-Monitoring Blood Glucose Meter (Trividia Health™) and True Metrix ® Test Strips (Trividia Health™).

RNA-sequencing (RNA-Seq)
Total RNA was extracted from triceps of symptomatic P7 untreated and prednisolone-treated Taiwanese Smn −/− ;SMN2 SMA and Smn +/− ;SMN2 healthy mice (Table S12).The triceps were immediately placed in RNALater (#AM7030, ThermoFisher) following dissection and stored at −20 • C under further analysis.For mRNA isolation, 500 ng of total RNA was used as input for the NEBNext ® Poly(A) mRNA Magnetic Isolation Module' (#E7490L, New England Biolabs (NEB)) in accordance with the manufacturer's standard instruction.Library preparation was carried out using the NEBNext ® Ultra Directional RNA Library Prep Kit for Illumina (#E7420L, NEB).Barcoded libraries from each experimental sample were combined in equimolar concentrations of 1.5 pM prior to sequencing at 75 bp × 1 (singleend) read metric on a NextSeq 550 (Illumina) system.

Molecular docking
Ligands were modelled and optimized using LigPrep (2021-4, Schrödinger LLC, New York, NY) and MacroModel (2021-4, Schrödinger LLC, New York, NY) according to the OPLS4 (2021-4, Schrödinger LLC, New York, NY) forcefield parameters [128].Protein structures were downloaded from the RCSB databank [129] and prepared using the Protein Preparation Wizard (2021-4, Schrödinger LLC, New York, NY) [130], where redundant molecules are removed, hydrogen and missing heavy atoms were added, bond orders were assigned, water orientations were sampled, ionization and tautomeric states of certain residues are set.Restrained force field minimization was applied by converging heavy atoms to 3.0 Å and receptor grids were generated for the active sites.Molecular docking was performed using Glide (2021-4, Schrödinger LLC, New York, NY) at extra precision mode at 100 runs per each ligand-receptor complex [131].
Methodology was validated by a redocking of the co-crystallized ligand for each protein structure.

RNA extraction and quantitative polymerase chain reaction (qPCR)
RNA extraction for C2C12 cells was performed with the ISO-LATE II RNA Mini Kit (#BIO-52073, Meridian BIOSCIENCE) as per manufacturer's instructions.Skeletal muscle (triceps and Tibialis anterior (TA)) and spinal cord tissue samples underwent homogenization with 7 mm stainless steel beads (#69990, Qiagen) in a Tissue Lyser LT (#85600, Qiagen) set at 60 oscillations/second for 2 min followed by microcentrifugation at > 10 000 RCF (MSE Sanyo Hawk 15/05) for 1 min.RNA extractions from harvested skeletal muscle was performed with the RNeasy Fibrous Tissue Kit (#74704, Qiagen) and all other harvested tissues with ISO-LATE II RNA Mini Kit (#BIO-52073, Meridian BIOSCIENCE) as per manufacturer's instructions.RNA concentrations (ng/μl) were measured with the NanoDrop 1000 spectrophotometer (Thermo-Scientific) before reverse transcription was performed using the qPCRBIO cDNA Synthesis Kit (#PB30.11-10,PCR Biosystems) as per manufacturer's instructions.

C. elegans neuromuscular assays
Neuromuscular assays were performed on day 3 animals raised on plates containing the pertinent solvent or drug.The pharyngeal pumping assay was performed as previously described [93].Brief ly, animals were filmed with a 150× objective using a Axio-Cam ICc5 camera at 175 frames/10 s on a Discovery.V8 SteREO microscope.Pumps were manually counted using the Zen Pro software v2.3.A pumping event was defined as a grinder movement in any axis.For locomotion assays animals were filmed with a 63× objective using a AxioCam ICc5 camara at 15 frames/sec on a Discovery.V8 SteREO microscope.Reversals, and paralysis time for 5 min (±SEM) were quantified using WormLab 1.1 software (MBF Bioscience).Final data represents three independent trials (n ≥ 25 animals in total per genotype).

Statistical analyses
Statistical analyses were carried out using the most up to date GraphPad PRISM software.Prior to any analyses, outliers were identified via Grubb's test (GraphPad) and subsequently removed.Appropriate statistical tests include unpaired t-test, one-way analysis of variance (ANOVA), and two-way ANOVA.Each posthoc analyses used is noted in the respective figure legend.Kaplan-Meier survival curves were analysed with a log-rank test.Statistical significance was considered at P < 0.05, described in graphs as * P < 0.05, * * P < 0.01, * * * P < 0.001 and * * * * P < 0.0001.

Figure 3 .
Figure 3. Molecular docking predicted binding of metformin and oxandrolone to their targets of interest.(a) Binding of the co-crystallized Prkag1 activator and the predicted alternative binding modes of metformin in Prkag1 active site.(b) Binding of the co-crystallized testosterone and the predicted binding of oxandrolone in the androgen receptor (AR) active site.Ligands are shown in colour ball-and-stick representation with molecular surface rendered, amino acid residues as sticks and binding interactions as colour dashed-lines.
a Over-representation only.