Muscle metabolic remodelling patterns in Duchenne muscular dystrophy revealed by ultra-high-resolution mass spectrometry imaging

Duchenne muscular dystrophy (DMD) is a common and severe X-linked myopathy, characterized by muscle degeneration due to altered or absent dystrophin. DMD has no effective cure, and the underlying molecular mechanisms remain incompletely understood. The aim of this study is to investigate the metabolic changes in DMD using mass spectrometry-based imaging. Nine human muscle biopsies from DMD patients and nine muscle biopsies from control individuals were subjected to untargeted MSI using matrix-assisted laser desorption/ionization Fourier-transform ion cyclotron resonance mass spectrometry. Both univariate and pattern recognition techniques have been used for data analysis. This study revealed significant changes in 34 keys metabolites. Seven metabolites were decreased in the Duchenne biopsies compared to control biopsies including adenosine triphosphate, and glycerophosphocholine. The other 27 metabolites were increased in the Duchenne biopsies, including sphingomyelin, phosphatidylcholines, phosphatidic acids and phosphatidylserines. Most of these dysregulated metabolites are tightly related to energy and phospholipid metabolism. This study revealed a deep metabolic remodelling in phospholipids and energy metabolism in DMD. This systems-based approach enabled exploring the metabolism in DMD in an unprecedented holistic and unbiased manner with hypothesis-free strategies.


Results
Histological findings. Morphological lesions meeting DMD diagnostic criteria were observed in muscle biopsies including fiber necrosis with inflammatory response, diffuse variation in fiber size with rounded and hypercontracted fibers, basophilic regenerative fibers, and endomysial and perimysial fibrosis. Immunohistochemical studies revealed that six patients lacked DYS1, and three had severely decreased and irregular DYS1 immunoreactivity, all patients but one having no DYS2 and DYS3 (Fig. 1, Supplementary Tables 1 and 2). These expression patterns were confirmed by western blot analyses. Immunohistochemical or western blot techniques showed that absent or decreased DYS was associated with absent or decreased proportions of α-, β-, γ-and δ sarcoglycans (Fig. 1, Supplementary Tables 1 and 2). Molecular analyses allowed for identifying different variants consisting of large deletions in the DMD gene sequence in 8 patients and a small frameshift deletion in one patient (Supplementary Tables 1 and 2).
Mass spectrometry-based metabolomics imaging. This work explored the differential metabolic patterns between Duchenne and control biopsies by analyzing the spectral fingerprints extracted in MSI experiments from the analyzed muscle biopsies. These fingerprints were formed from ions generated by the mass spectrometry analysis used for data analysis and metabolite identification. Average spectra are presented in Supplementary Fig. 1 The first statistical analysis yielded 52 discriminant ions with significant differences between the Duchenne and control biopsies. Further annotation steps allowed filtering and cleaning this list to include only unambiguously identified metabolites. In positive-ion mode, the combination of high mass accuracy and MS/ MS experiments highlighted phosphatidylcholine (PC) and sphingomyelin (SM) lipids due to their fragmentation patterns. PC and SM lipids can be detected in positive-ion mode owing to their quaternary amine groups. The fragmentation ions observed upon collisional activation of both [ To explore the distribution of this signature and its expression across samples, we performed a clustering analysis using Euclidean distance, a similarity metric, between samples. Figure 2A shows two distinct clusters between the DMD and control biopsies due to the differential expression of the above-mentioned metabolites across samples. We also explored the covariation of these metabolites using Spearman correlation analysis. Figure 2B shows two main co-expression clusters that include upregulated and downregulated metabolites. Intraclass subclusters are observed such as a module that includes phosphatidylcholines, phosphatidic acids and phosphatidylserines. Another module includes SMs and lysophosphatidylcholines. Figure 2C presents the directional changes of these metabolites in the DMD samples along with their statistical significance. This figure reports adjusted p-values and estimate which indicates metabolite change direction. Figure 3 presents boxplots of the glycerophosphocholine, lysophosphatidylcholine, phosphatidylcholine, phosphatidic acid, SM, phosphatidylserine, along with their tissue distributions in the DMD and control biopsies with their related   www.nature.com/scientificreports/ adjusted p-values. Even though the figures don't show a cellular, sub-cellular or fiber level resolution, they highlight the heterogenous distribution of intensities across the tissue section that mirrors the distribution of the related metabolites.

Discussion
Defective dystrophin is known to be the main alteration in the DMD etiology, however, metabolic impairment has been reported in several tissues such as the skeletal and cardiac muscles, liver and brain 27 . Here, we explored the differential metabolic patterns of muscle biopsies from DMD patients and control samples using a hypothesisfree strategy based on MSI. Our results showed profound metabolic pathway remodelling in the muscles of DMD patients compared with the controls grouped in two distinct clusters corresponding to the upregulated and downregulated metabolites. The most discriminative metabolites were primarily phospholipids and energy metabolites, suggesting that both these metabolic pathways are important players in muscle pathology and possibly by extension cognitive/behavioral disabilities. Phospholipids (PLs) are major components of all cell membranes, and glycerophospholipids (GLs) are the most abundant membrane PLs. GLs are composed of a glycerol backbone linked to a phosphate group in position 3. Acyl groups may be attached at positions 1 and 2 of the glycerol. Different moieties, such as choline, inositol, ethanolamine and serine, are linked to the phosphate group, contributing to GL diversity (e.g., PC, phosphatidylinositol [PI], phosphatidylethanolamine (PE), PS, and phosphatidylglycerol). PCs are the most abundant GL in mammalian cells. Other lipid classes, such as cholesterol and glycosphingolipids, play structural roles in cell membranes 28,29 . Disruption of PL metabolism has been previously reported in DMD 10,11,13,20,21,23,30,31 . We showed that several compounds belonging to PC, lysophosphatidylcholine (LPC), phosphatidic acid (PA), PS and SM classes, as well as triacylglycerols, are increased, while glycerophosphocholine (GPC) is decreased in DMD muscles compared with control muscles (Fig. 3). This metabolic remodelling could be attributed to a putative decrease in lysophospholipase activity (Fig. 4). Lysophospholipase catalyzes lysophosphatidylcholine deacylation and mediates GPC production 32 . Sharma et al. considered DMD to be mainly a GPC deficiency 33 . Consequently, this impairment could lead to an accumulation in phosphatidylcholine, which is redirected toward other metabolic pathways with an increase in PA, SM, phosphatidylserine and triacylglycerol (Fig. 4). It has   www.nature.com/scientificreports/ been shown that lysophospholipase activity in erythrocyte membranes of DMD patients was lower than that of healthy age-matched subjects 32 . Chalovich et al. compared the 31 P-NMR spectra of muscles from DMD patients, Werdnig-Hoffman (Spinal Muscular Atrophy Type 1) patients and controls. These authors observed that GPC levels were lower or near absent in DMD muscles compared with control muscles and dramatically increased (15-fold) in one Werdnig-Hoffman patient 30 . Increased triacylglycerol and SM in DMD mouse muscles have been interpreted as resulting from increased phospholipase C activity 13 . High phospholipid levels and high phospholipid-to-cholesterol ratios have been documented using 1 H-NMR spectroscopy, which helps distinguish DMD patients from healthy subjects 34 . Phospholipase D activity is reported to be increased in DMD muscles. Calcium homeostasis has been shown to be altered in DMD patients, with calcium concentration being increased in dystrophin-deficient muscle. Phospholipase D activity is enhanced under high concentrations of free calcium, resulting in increased PA levels 35,36 . In summary, profound phospholipid metabolism remodelling is associated with DYS deficiency; thus, DMD has been considered a glycerophosphocholine deficiency 33,37 . Morphological alterations of the mitochondria have been reported in DMD muscles with higher rates of swollen mitochondria compared with that of normal muscle 38 . PL alterations in DMD and a dysregulation of Ca 2+ homeostasis may induce mitochondrial membrane fragility and associated morphological changes with subsequent energy and oxidative metabolism disruptions [39][40][41][42] . Energy metabolism impairments in DMD are major contributors to DYS-deficient muscle degeneration 43 as illustrated by mitochondrial function improvement upon the partial restoration of DYS expression 44 . Accordingly, we showed a significant increase in ADP and decreased ATP levels in DMD muscles compared with those of the control muscles ( Supplementary Fig. 3). Several preclinical and clinical studies aimed at promoting energy metabolism are ongoing to treat DMD 45 . The need for studies with higher cellular or fiber level resolution would unveil deeper metabolic insights and further explore these results. Van Pelt et al. performed a multi-omics study in a Duchenne mouse model. They reported the impairment of glycolytic metabolism and phospholipids 46 . Two metabolites overlap with our results adenosine diphosphate and PC(36:3) (Supplementary Table 5). Exploring the potential release of the metabolites in less invasive biological fluids is relevant. Spitali et al. reported serum-based metabolomics results in patients affected by several multiple forms of muscular dystrophy. Fifteen metabolites have been reported belonging to energy, amino acid and testosterone metabolisms 47 . Lindsay et al. performed a urine-based metabolomics study in a Duchenne mouse model. They reported that five of seven detected Krebs cycle metabolites were depleted in these mice consistent with an impaired energy metabolism 22 . No overlap has been observed between our results and those reported in these studies (Supplementary Table 5).
In conclusion, although metabolic impairments have been reported in DMD, this work describes for the first time the use of ultra-high-resolution MSI in DMD, which enabled more systematically and integratively exploring on-tissue metabolic disturbances. This powerful technique allowed investigating the different metabolic components in a single experiment. The present results highlight the potential use of MSI technology coupled with systems biology approaches to holistically explore metabolic impairments in DMD. This work lays the foundation for more mechanistic investigations of DMD and other metabolic diseases.

Methods
The overall workflow is presented in Supplementary Fig. 11.
Patients. Nine patients with molecularly proven DMD over the last 10 years were selected for the study. All patients were referred to our neurology department, and biopsies were performed as a part of the diagnostic workup when the disease was suspected. Samples were collected before initiating treatment. Medical charts were reviewed for age at disease onset, age at diagnosis, age at walking unaided, age at walking loss, and age at last visit. Symptoms at disease onset were recorded as calf hypertrophy, muscle weakness, contractures, behavioral problems, speech delay, respiratory distress and cardiac arrhythmia. Disease activity parameters consisted of maximal motor function, maximal CPK, age of initial joint contractures and current joint contractures, surgical therapy and long-term physical therapy. Other parameters consisted of scoliosis occurrence or any other spinal deformities, along with their surgical and nonsurgical therapies, occurrence of bone fractures and other orthopedic surgeries, and mobility at last visit. Evaluated complications were respiratory insufficiency (e.g., IPPB, ventilation, tracheotomy), existing cardiac complications (e.g., rhythm abnormalities or heart abnormalities on ultrasound), digestive complications (e.g., nutritional problems, gastrostomy) and iatrogenic complications. Inclusion in therapeutic protocols, hospitalization frequency, cognitive delays and behavioral problems, including autism and school attendance, were also recorded.
The mean patient age at disease onset was 4 ± 1.5 years, at last visit was 10.5 ± 2.2 years, and at time of biopsy was 5.5 ± 1.9 years. All patients are presently alive, and six (P1, P5, P6, P7, P8, and P9) were ambulant at the last visit. One non-ambulant patient (P2) lost the ability to walk at age 7; he was not given steroids, nor included in any therapeutic protocol. The other two patients (P3, P4) lost the ability to walk at 11 and 13 years of age. CK levels ranged from 5418-52,000. Symptoms at disease onset were muscle weakness with difficulty walking and climbing stairs and frequent falls. Eight patients had calf hypertrophy. Six patients presented no motor features at onset but had speech delays (P2, P5, P8) and/or behavioral/autistic disturbances (P5, P8, P9) that had likely led to delayed diagnosis. Four patients had cognitive delays (P2, P5, P8, P9). Two patients (P2, P5) developed behavioral abnormalities. Six patients had learning difficulties affecting speech, writing, attention and memory (P1, P2, P5, P7, P8, P9). Three patients (P4, P5, P9) had nutritional and feeding problems, with anorexia in two and obesity in one (P9). All but two patients were given steroids (P1, P2); one of these two patients were lost to follow up one year after diagnosis, and the other received no treatment owing to poor family compliance. Four patients (P1, P5, P7, P9) could run, the others could walk without help. Two patients had mild respiratory insufficiency that did not require ventilation. Eight patients had no significant cardiac arrhythmia or functional www.nature.com/scientificreports/ abnormalities. Patient P4 had rhythm abnormalities consisting of supraventricular extrasystoles 3.9% of the time, which did not require specific treatment. All but one patient received a prophylactic antiarrhythmic drug (perindopril), and one (P1) was included in a perindopril protocol (2-year double-blind treatment protocol). Three patients (P4, P5, P7) were included in exon-skipping protocols, and one (P6) was included in a givinostat (histone deacetylase inhibitor) protocol. Clinical data overview is presented in Fig. 5 and detailed clinical data are listed in Supplementary Tables 1 and 2.
As age-matched control muscles were not available, nine male adults aged 28-58 years were selected as controls for comparative analyses. These patients presented myalgias, muscle weakness, and/or muscle fatigability. In one case, a polymyositis was suspected; Buschke scleroderma was suspected in another. In all control patients, routine biological tests, CK levels, EMGs, cardiac ultrasonography, pulmonary testing and muscle MRI were considered normal. Their muscle biopsies were also normal, and they were concluded to have probable fibromyalgia. One obvious potential limit of this work that needs to be highlighted consists in absence of age-matched control muscles could hamper interpretation of the results, as biochemical properties of muscles change over the lifetime. Even though paediatric muscle samples could be obtained during orthopaedic surgery, the location of the biopsy may change and may not entirely match with that of Duchenne patients.  . Data size was set at 2 million points for a transient length of 0.87 s, and spectra were acquired with a 97% data file reduction. A single MSI measurement has been performed by specimen. Images were captured using FTMS control and FlexImaging (v 5.0, Bruker) software. Images were processed with SCiLS Lab Pro software (Bruker Daltonics, Bremen, Germany). The total ion current method was used for normalization, and m/z intervals were automatically set at ± 1 ppm. Images were viewed using both FlexImaging and SCiLS Lab software (Bruker Daltonics, Bremen, Germany).

Data analyses.
Ion intensities have been log-transformed. Univariate analyses were performed using t-tests to identify discriminatory ionic species between the assessed groups. Age has been taken into account by adding it as covariate. Spearman correlation analysis was performed using R software. Euclidean distance was used as a similarity measure in the clustering analysis. False discovery rates were corrected using the Benjamini-Hochberg-Yekutieli method, and p < 0.05 was considered statistically significant.
Metabolite annotation and identification. Preliminary assignments based on accurate mass measurements were performed using the mass spectrometry databases, METLIN 50 and HMDB 51 , using a threshold of ± 2 ppm. For some metabolites, the precise raw formula led to one hit. Others were identified via "on-tissue" tandem mass spectrometry experiments using MALDI tandem MS/MS. Ions of interest were first isolated using a window of ± 1 Da, then fragmented by collision-induced dissociation with energy levels between 10 and 40 eV. For each MS/MS analysis, 50 scans were accumulated for better sensitivity. Spectra were reprocessed using Data www.nature.com/scientificreports/