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

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.


ABSTRACT 21
Spinal muscular atrophy (SMA) is a genetic neuromuscular disorder caused by the reduction of 22 However, long-term administration of prednisolone can promote myopathy. We thus wanted to 29 identify genes and pathways targeted by prednisolone in skeletal muscle to discover clinically 30 approved drugs that are predicted to emulate prednisolone's activities. Using an RNA-sequencing, 31 bioinformatics and drug repositioning pipeline on skeletal muscle from symptomatic prednisolone-32 treated and untreated Smn -/-;SMN2 SMA and Smn +/-;SMN2 healthy mice, we identified molecular 33 targets linked to prednisolone's ameliorative effects and a list of 580 drug candidates with similar 34 predicted activities. Two of these candidates, metformin and oxandrolone, were further 35 investigated in SMA cellular and animal models, which highlighted that these compounds do not 36 have the same ameliorative effects on SMA phenotypes as prednisolone; however, a number of 37 other important drug targets remain. Overall, our work further supports the usefulness of 38 prednisolone's potential as a second-generation therapy for SMA, identifies a list of potential SMA 39 Spinal muscular atrophy (SMA) is a heterogenous autosomal recessive neuromuscular disorder 45 (NMD) characterized by motor neuron degeneration alongside progressive muscle atrophy and 46 weakness 1 . Being the leading monogenic cause of infant mortality 2 , around 96% of SMA cases 47 are mapped to homozygous loss-of-function and deletion mutations in the survival of motor neuron 48 1 (SMN1) gene 3,4 , which ubiquitously expresses SMN, a protein that current and ongoing research 49 has linked to diverse housekeeping and tissue-specific cellular functions 5-7 . Although complete 50 SMN loss is embryonic lethal in most organisms 8 , humans can overcome the complete loss of the 51 SMN1 gene due to incomplete rescue by the homologous SMN2 gene 9,10 . In essence, the presence 52 of a single nucleotide mutation in SMN2 promotes exon 7 alternative splicing that limits full length 53 SMN (FL-SMN) expression in this gene to 10% 11 . Consequently, the limited FL-SMN expression 54 makes SMN2 gene copy number an important disease modifier, impacting SMA type and severity 55

Statistical Analyses 285
Statistical analyses were carried out using the most up to date GraphPad PRISM software. Prior to 286 any analyses, outliers were identified via Grubb's test (GraphPad) and subsequently removed. 287 Appropriate statistical tests include unpaired t-test, one-way analysis of variance (ANOVA), and 288 two-way ANOVA. Each post-hoc analyses used is noted in the respective figure legend. Kaplan-289 Meier survival curves were analysed with a log-rank test. Statistical significance was considered 290 at p < 0.05, described in graphs as *p < 0.05, **p < 0.01, ***p < 0.001 and ****p < 0.0001.

RESULTS 297
Prednisolone restores the expression of a large subset of genes involved in canonical skeletal 298 muscle pathways in SMA mice. 299 As described in an earlier study, we have previously demonstrated that treating SMA mice with 300 prednisolone significantly improved several disease phenotypes, including survival, weight, and 301 muscle health 27 . To have a more in depth understanding of the impact of prednisolone on SMA 302 skeletal muscle at a molecular level, we performed bulk RNA-Seq on skeletal muscle of untreated 303 and prednisolone-treated Smn -/-;SMN2 SMA and Smn +/-;SMN2 healthy mice. Specifically, we 304 administered prednisolone (5 mg/kg, gavage, every 2 days) starting from P0 until P7 to Smn -/-305 ;SMN2 SMA and Smn +/-;SMN2 healthy mice 27 . Triceps were harvested from P7 prednisolone-306 treated and untreated mice for RNA-Seq via Illumina NextSeq550 and a HISAT2-FeatureCounts-307 DESeq2 pipeline against a Mus Musculus mm10 genome for parameters of "condition" and 308 "treatment" (Figures S1).  Tables S3-5). 314 Next, we determined the biological pathways associated with DEGs in prednisolone-treated Smn -315 /-;SMN2 SMA mice compared to untreated Smn -/-;SMN2 SMA mice. Using iPathwayGuide, we 316 identified that 3056 significant DEGs (Log2FC > 0.6, FDR < 0.05) (Table S4) were targeted by 317 prednisolone in the skeletal muscle of Smn -/-;SMN2 SMA mice when compared to untreated Smn -318 /-;SMN2 SMA mice and associated with 28 significant KEGG pathways (p <0.05) ( Table 1). 319 Interestingly, these prednisolone-targeted pathways are closely associated with important skeletal 320 muscle processes such as metabolism, atrophy and regulatory function, alongside previous 321 associations with SMA-related pathways such as FoxO signalling 66 , p53 signalling 67 , AMPK 322 signalling 68 , mitophagy 69 , circadian rhythm 70 , PPAR signalling 71 and autophagy 72 (Table 1). 323 An additional GO analysis also revealed similar skeletal muscle biological processes associated 324 with the DEGs in prednisolone-treated Smn -/-;SMN2 SMA mice such as myotube differentiation, 325 fatty acid oxidation, protein ubiquitination, sarcomere regulation, gluconeogenesis, and circadian 326 rhythm (Table 2; Tables S6-8). 327 Combined, our transcriptomics and pathway analyses suggest that prednisolone treatment 328 attenuated muscle pathologies in SMA mice 27 by targeting key muscle metabolism, atrophy and 329 regulatory pathways. 330 331 Drug repositioning algorithms identify novel pharmacological compounds predicted to 332 emulate prednisolone's activity in skeletal muscle of SMA mice. 333 As mentioned above, while prednisolone treatment significantly improves muscle health and 334 overall disease progression in SMA mice, chronic use of prednisolone can negatively impact 335 skeletal muscle 35,39 . As such, we used the DEGs and associated KEGG pathways identified in 336 prednisolone-treated Smn -/-;SMN2 SMA mice to discover alternative drugs predicted to mimic 337 prednisolone's molecular effects in SMA skeletal muscle. Initially, we utilized the in-built 338 integration of KEGG drugs database in iPathwayGuide 53 and the DGIdb v3.0 58 database to 339 initially reveal a total of 580 compounds (Tables S9-12). To filter down our list, we focused on 340 the drug compounds 1) that targeted > 5 prednisolone-targeted pathways or linked to upstream 341 regulators, 2) were clinically approved and 3) were not associated with promotion of muscle-342 wasting (e.g., primary anti-cancer drugs 73 ), leaving a total of 20 potential candidates (Tables 3-4). 343 Interestingly, our combined in silico drug repositioning approach revealed a subset of candidates 344 previously investigated in SMA such as celecoxib 74 (ClinicalTrials.gov ID: NCT02876094) and 345 colforsin 75 . To further validate our bioinformatics strategy, we chose to continue our study with 346 drugs not yet assessed for SMA, focusing on those previously used safely in young patients and 347 orally bioavailable. With these criteria, we narrowed down our selection to metformin, a generic 348 asymmetric dimethyl-biguanide type 2 diabetes mellitus (T2DM) drug 76 and oxandrolone, a 349 synthetic anabolic steroid with a higher ratio of anabolic: androgynous effects for further study 77 . 350 Thus, using a transcriptomics-based in silico drug repositioning platform, we were able to generate 351 a list of clinically approved pharmacological compounds that are predicted to emulate 352 prednisolone's activity in skeletal muscle. As previously mentioned, metformin is an orally administered T2DM drug that we selected as one 357 of the candidates to validate our bioinformatics-based drug repositioning approach. Importantly, 358 metformin has over 60 years of clinical use with a well-known safety profile 76 and recorded 359 administration in younger patients 78 . Furthermore, it has been previously repositioned and 360 conferred ergogenic activities in muscular disorders such as DMD 79 and congenital muscular 361 dystrophy type 1 A (CMDT1A) 80 , highlighting its potential as a skeletal muscle therapy. 362 Our iPathwayGuide analysis predicted that metformin could emulate prednisolone's targeting of 363 the KEGG: 04068 FoxO signalling pathway ( Figure S2.a). In particular, metformin was predicted 364 to mimic prednisolone's upregulation of Prkag3, which encodes for the AMPK-3 subunit of the 365 predominant skeletal muscle AMPK-223 isoform complex 81 ( Figure S2). Furthermore, 366 Prkag3 upregulation was predicted to coherently downregulate the expression of FoxO1, FoxO3 367 and Foxo4 isoforms, whilst upregulating FoxO6 ( Figure S2) supporting previous literature 368 associating these FoxO isoforms with promotion of muscle atrophy 66,82 . Importantly, the 369 expression pattern of these genes in the prednisolone-treated Smn -/-;SMN2 SMA mice were 370 normalized to healthy Smn +/-;SMN2 levels ( Figure S2.c), supporting the usefulness of investigating 371 metformin and these targets in SMA skeletal muscle. We next wanted to better understand if the aberrant expression of the metformin target genes was 390 dependent on SMN expression and/or muscle atrophy. Thus, we firstly generated siRNA-mediated 391 Smn-depleted C2C12 myoblast-like cells, a useful and previously successful in vitro model 84 . We 392 confirmed by qPCR that Smn mRNA levels were significantly reduced by up to 90% in C2C12 393 myoblasts and D8 C2C12 myotubes compared to scrambled siRNA and untreated controls ( Figure  394 S3). We next investigated the effects of Smn knockdown on the expression of the predicted 395 metformin target genes. In C2C12 myoblasts, we identified a significant upregulation of only the 396 FoxO3 gene in Smn-depleted C2C12 myoblasts compared to controls (Figure 3.a), which reflects 397 previous microarray analyses of specific FoxO isoforms upregulated in quadriceps femoralis 398 muscle biopsies from type 1 SMA patients 85 . However, in C2C12 myotubes we found that Smn 399 knockdown (KD) had no effect on the expression of predicted metformin target genes (Figure 3.b), 400 suggesting that for the most part, the expression of the predicted metformin genes is Smn-401 independent, thus representing ideal targets for SMN-independent therapies. 402 We next investigated if the expression of the predicted metformin target genes is affected in vitro 403 by muscle atrophy. However, one difficulty in mimicking SMA muscle atrophy in vitro is 404 establishing denervation. Thus, based on evidence of shared pathway similarities from different 405 pro-atrophy factors such as starvation and denervation 86 , we used a validated method of 24-hour 406 serum-starvation in C2C12 myotubes to induce canonical atrophy, as confirmed by myotube loss 407 and upregulation of pro-atrophic atrogin-1 levels (Figure 3.c). Next, we evaluated the expression 408 of the predicted metformin target genes and observed a significant upregulation of FoxO3 and 409   Overall, our data highlights that metformin did not have a direct impact on the predicted target 501 genes in skeletal muscle of SMA mice. Furthermore, the absence of direct impact on muscle 502 atrophy, glucose metabolism, and mitochondrial function markers following metformin treatment 503 in Smn 2B/-SMA muscle, suggests that the adverse effects associated with the 400 mg/kg/day dosage 504 may not have been linked to muscle-intrinsic effects. 505 506 A higher dose of metformin is associated with dysregulation of mitochondrial regulatory 507 genes in the spinal cord of Smn 2B/-SMA mice. 508 We next investigated the effects of metformin on the spinal cord given that metformin is 509 systemically distributed 97 , has the ability to cross the blood-brain-barrier (BBB) 98  6.c), whilst Tfam was not affected by metformin in either cohort (Figure 6.d). 524 Our results demonstrating that the higher dose of metformin (400 mg/kg/day) appears to 525 specifically dysregulate certain mitochondrial genes in the spinal cord of Smn 2B/-SMA mice is 526 supported by recent evidence of tissue-dependent differences in conserved cellular processes 527 between SMA motor neurons and skeletal muscle 72 . Thus, although further in-depth investigations 528 would be needed, our results on mitochondrial health markers suggest that metformin's adverse 529 effects in SMA mice could be linked to the exacerbation of neuronal mitochondrial dysfunction. Our second drug candidate that we selected to mimic prednisolone activities was oxandrolone, a 534 synthetic orally bioavailable anabolic steroid that confers minimal androgynous effects 77 . 535 Importantly for SMA, oxandrolone has been successful in the promotion of muscle growth for 536 DMD 100 and mixed gender burn injury patients 101 . 537 Oxandrolone was predicted to upregulate the Ar gene in SMA muscle ( Figure S7). The 538 upregulation of Ar was predicted to directly upregulate downstream effectors Igfbp5 and myogenin 539 (or MyoG) ( Figure S7), which both regulate muscle differentiation, regeneration and myofiber 540 growth 102,103 . Furthermore, Ar was predicted to indirectly upregulate Dok5, a signalling protein 541 linked to insulin and IGF-1 activity 104 and Akap6, which is involved in the modulation of muscle 542 differentiation and regeneration 105 ( Figure S7). In addition to these factors, we also decided to 543 investigate Ddit4 as an oxandrolone target based on its direct relation with Ar 106 and being one of 544 the top 20 downregulated DEG targets of prednisolone in Smn -/-;SMN2 SMA skeletal muscle 545 (Table S4; Figure S7). 546 Similar to our metformin strategy above, we initially wanted to evaluate the mRNA expression 547 levels of these target genes in the triceps of both symptomatic P7 severe Smn -/-;SMN2 and P19 548 Nevertheless, the expression of the majority of the predicted oxandrolone target genes was Smn-564 independent. 565 We next wanted to evaluate the ability of oxandrolone to attenuate canonical atrophy in serum-566 deprived C2C12 myotubes 87 . However, in this case we performed the treatments in D5 C2C12 567 myotubes instead of D8, as although oxandrolone was non-toxic (Figures S8), it elicited a greater 568 androgen Ar response at the earlier differentiation stage ( Figure S9). Following confirmation of 569 muscle atrophy in D5 C2C12 myotubes via significant Atrogin-1 and MuRF1 upregulation ( Figure  570 8.c), we observed that the expression of the predicted oxandrolone target genes Akap6, Igfbp5, 571 MyoG and Ddit4 was significantly downregulated in serum-deprived D5 C2C12 myotubes ( Figure  572 8.d). 573 Interestingly, we found that 24-hour treatment with 1 M oxandrolone attenuated canonical 574 muscle atrophy in these serum-starved D5 C2C12 myotubes as shown by significant We next assessed the impact of oxandrolone in SMA mice. We initially tested preliminary 584 treatment regimens of 1 -8 mg/kg/day starting from P5 or P8 in Smn 2B/-SMA and Smn 2B/+ healthy 585 mice (data not shown), based on previous studies in models of spinal cord injury (SCI) 107 and burn 586 injury 108 . We also stopped oxandrolone treatments at P21 as previous research has shown that 587 shorter oxandrolone treatments are more effective 107 . These pilot studies allowed us to identify 588 the optimal dosing regimen of 4 mg/kg/day oxandrolone treatment from P8 to P21, which 589 significantly improved the median survival of Smn 2B/-SMA mice (Figure 9.a). 590 However, we found that the body weight of 4 mg/kg/day oxandrolone-treated Smn 2B/-SMA mice 591 was significantly lower compared to their untreated counterparts (Figure 9.b), which is most likely 592 due to the intrinsic smaller sizes of the randomly assigned litters, as demonstrated by the difference 593 in weight starting 4 days prior to initial treatment (Figure 9.b). In terms of motor function, we 594 observed no significant difference in righting reflex between untreated and oxandrolone-treated 595 SMA animals (Figure 9.c). Furthermore, we identified no impact of vehicle treatment on survival, 596 weight, and righting reflex in Smn 2B/-SMA mice (Figures S10). 597 In the Smn 2B/+ healthy mice, although 4 mg/kg/day oxandrolone had no effect on survival or motor 598 function in treated animals (Figures S11.a-b), we did observe a significant decrease in bodyweight 599 starting from P9, one day after initial treatment ( Figure S11.c), suggesting that oxandrolone may 600 Nevertheless, our results demonstrated that although 4 mg/kg/day oxandrolone treatment 602 improved survival in Smn 2B/-SMA mice, its effect on survival was still minor compared to 603 prednisolone 27 , suggesting that oxandrolone is not a suitable substitute as an SMA skeletal muscle 604 therapy. 605 606 Oxandrolone did not impact the expression of the predicted target genes or muscle pathology 607

markers. 608
The improved 3-day survival in 4 mg/kg/day oxandrolone-treated Smn 2B/-SMA mice led us to 609 evaluate whether this beneficial impact was related to targeting muscle pathologies. Thus, we 610 evaluated oxandrolone's effects on the expression of dysregulated molecular markers associated 611 with the SMA hallmark pathology of muscle atrophy (Atrogin-1 and MuRF-1) in the triceps from 612 P19 late symptomatic, untreated and 4 mg/kg/day oxandrolone-treated Smn 2B/-SMA and Smn 2B/+ 613 healthy mice, 2 hours after final treatment. We observed no significant reduction in elevated 614 Atrogin-1 or MuRF-1 gene expression levels by oxandrolone in the Smn 2B/-SMA cohort ( Figures  615   9.d-e), suggesting that oxandrolone did not attenuate muscle atrophy. 616 We next evaluated the effect of oxandrolone on the expression of the predicted target genes in the 617 same P19 untreated and 4 mg/kg/day oxandrolone-treated Smn 2B/-SMA and Smn 2B/+ healthy mice. 618 We found that oxandrolone did not significantly impact the predicated target genes in the triceps 619 from the Smn 2B/-SMA mice (Figures 9.f-k). However, we did observe that 4 mg/kg/day 620 oxandrolone treatment significantly upregulated Dok5 expression in the Smn 2B/+ healthy mice 621 (Figure 9.i). Nevertheless, the pattern observed suggests that oxandrolone did not impact any of 622 the predicted genes in the muscle from Smn 2B/-SMA mice. 623 Overall, our data shows that oxandrolone did not have an efficient effect on the predicted target 624 genes. Furthermore, its inability to ameliorate muscle atrophy marker dysregulation in SMA 625 skeletal muscle, suggests that improved survival in the Smn 2B/-SMA mice by oxandrolone may be 626 independent of targeting skeletal muscle pathologies. The goal of this study was to use a transcriptomics-based drug repositioning strategy to identify 649 clinically approved drug candidates that could emulate prednisolone's beneficial effects in SMA 650 skeletal muscle and life expectancy 27 , without the risks associated with long-term GC exposure 651

. 652
Our major finding was the observation that prednisolone treatment restored specific gene sets it is suggested that intramuscular Ar content may have a stronger influence on hypertrophy than 697 peripheral androgen levels 119 . Thus, we cannot conclude whether oxandrolone's treatment 698 efficacy compared to prednisolone was due to gender-specific differences. 699 Despite our study's limitations, it highlighted refinements for future in silico SMA drug 700 repositioning studies. Compared to a previous study that successfully identified and validated 701 harmine's therapeutic potential in SMA muscle 42 , ours did not include proteomics. The absence 702 of proteomics can be a caveat for drug studies as transcript levels alone do not proportionally 703 reflect drug-protein interactions, abundance and translational modifications 120,121 . However, a 704 limitation of both transcriptomics and proteomics approaches is that they cannot bridge drug-705 pathway interactions with disease phenotypes, as demonstrated by a recent proteomics analysis of 706 Spinraza® treated type 2 and 3 SMA patients that could not correlate protein profiles with 707 functional improvements 122 . Thus, implementation of metabolomics may be beneficial for linking 708 pathway perturbations with metabolites associated with disease and stages of muscle atrophy 123 . 709 To the best of our knowledge, this three-pronged multi-omics approach has not previously been 710 have further benefits to narrow in on muscle regions such as nuclei located near the NMJ, since 731 myonuclei display spatial and temporal expression pattern heterogeneity in multi-nucleated 732 myofibers 126 . 733 Although the drug candidate's metformin and oxandrolone did not emulate prednisolone's 734 beneficial effects in SMA to the extent previously reported, our transcriptomics-drug repositioning 735 approach did better define prednisolone's activity in SMA muscle and provided a list of potential 736 candidates for future pre-clinical SMA drug repositioning studies. Furthermore, our study 737 highlights important refinements for future SMA drug repositioning studies.       c. d. e. f.