Skeletal muscle reprogramming of metabolic, ribosomal and developmental pathways contributes to Roux-en-Y gastric bypass-induced adaptation in women Authors: Chantal A. Pileggi, Dhanuddara Mohottalage, Luke S. Kennedy, Majid Nikpay, Lauren M.K. Hamilton, Martín Roffe, Michel N. Kanaan, Amer Jarrar, Nicole Kolozsvari, Dongdong Wang, Miroslava Cuperlovic-Culf, Tommy Alain, Gregory R. Steinberg, Ruth McPherson, Robert Dent, Mary-Ellen Harper Published: 2026 ABSTRACT SummaryBackgroundUnderstanding how skeletal muscle responds to weight loss is crucial for developing targeted strategies to manage obesity and promote sustained improvement in metabolic health. Here, we investigated the molecular mechanisms underlying skeletal muscle reprogramming of gene expression and metabolic activity following Roux-en-Y gastric bypass (RYGB).MethodsForty-one women were studied before and one year after RYGB surgery. We leveraged multi-omics (DNA methylomics and transcriptomics) and machine learning approaches to complement muscle metabolic analyses and clinical data to identify mechanisms underlying RYGB-induced muscle metabolic reprogramming.FindingsRYGB markedly decreased body weight and fat mass and improved metabolic health. Integrative analysis of vastus lateralis muscle identified 8233 genes with differentially methylated regions and 2173 differentially expressed genes post-RYGB surgery, of which 1197 genes were both differentially methylated and differentially expressed. Promoter hypomethylation was associated with the enhanced expression of transcription factors involved in skeletal muscle development and ribosomal subunits. In contrast, expression of genes encoding mitochondrial proteins decreased despite increases in mitochondrial content and enhanced mitochondrial function in skeletal muscle post-RYGB. Pre-operative muscle OXPHOS capacity, and expression of skeletal muscle hypertrophy and differentiation genes MYOC and EHMT2 were associated with weight loss success.InterpretationRYGB improves systemic metabolic health and induces sustained skeletal muscle bioenergetic reprogramming characterised by enhanced expression of genes involved in myogenesis and protein translation, but decreased expression of genes involved in mitochondrial metabolism, which may reflect improved mitochondrial quality and function. These findings advance our understanding of skeletal muscle metabolic responses to weight loss and of individual variability in metabolic phenotypes.FundingCanadian Institutes of Health Research (CIHR PJT183651-M-EH, 201709FDN-CEBA-116200-GRS), Diabetes Canada Investigator Award grant OG-3-22-5645-GS (GRS), J. Bruce Duncan Endowed Chair in Metabolic Diseases (GRS), Tier 1 Canada Research Chair in Mitochondrial Bioenergetics and Metabolic Health (M-EH), Tier 1 Canada Research Chair in Metabolic Diseases (GRS). Research in contextEvidence before this studyObesity (BMI ≥ 30 kg/m2) is a multifactorial metabolic disease with an elevated risk for all-cause mortality. Roux-en-Y gastric bypass (RYGB) yields durable weight loss with long-term favourable metabolic outcomes in individuals with obesity, yet the mechanisms underlying metabolic improvements are poorly understood. Skeletal muscle plays a central role in glucose homoeostasis and whole-body energy metabolism, however the adaptations in skeletal muscle to RYGB remain unclear.Added value of this studyUsing extensive bioinformatic analyses to analyse the methylome and transcriptome, here we compared skeletal muscle obtained from 41 women with obesity undergoing RYGB and at one year post-surgery. In this uniquely well powered study we demonstrate that certain regions of DNA became less methylated, leading to the upregulation of genes involved in skeletal muscle development and protein synthesis. Interestingly, despite decreases in the expression of several mitochondrial genes, mitochondrial function and content were higher at one-year post-RYGB. Pre surgical vastus lateralis muscle oxidative phosphorylation capacity, expression of skeletal muscle hypertrophy and differentiation genes MYOC and EHMT2 may be predictive of weight loss success.Implications of all the available evidenceRYGB enhances systemic metabolic health and induces durable skeletal muscle bioenergetic reprogramming, mediated by epigenetic and transcriptional remodelling. Furthermore, our findings highlight the predictive potential of baseline biomarkers in weight loss success in response to RYGB. Research in context Evidence before this study Obesity (BMI ≥ 30 kg/m2) is a multifactorial metabolic disease with an elevated risk for all-cause mortality. Roux-en-Y gastric bypass (RYGB) yields durable weight loss with long-term favourable metabolic outcomes in individuals with obesity, yet the mechanisms underlying metabolic improvements are poorly understood. Skeletal muscle plays a central role in glucose homoeostasis and whole-body energy metabolism, however the adaptations in skeletal muscle to RYGB remain unclear. Added value of this study Using extensive bioinformatic analyses to analyse the methylome and transcriptome, here we compared skeletal muscle obtained from 41 women with obesity undergoing RYGB and at one year post-surgery. In this uniquely well powered study we demonstrate that certain regions of DNA became less methylated, leading to the upregulation of genes involved in skeletal muscle development and protein synthesis. Interestingly, despite decreases in the expression of several mitochondrial genes, mitochondrial function and content were higher at one-year post-RYGB. Pre surgical vastus lateralis muscle oxidative phosphorylation capacity, expression of skeletal muscle hypertrophy and differentiation genes MYOC and EHMT2 may be predictive of weight loss success. Implications of all the available evidence RYGB enhances systemic metabolic health and induces durable skeletal muscle bioenergetic reprogramming, mediated by epigenetic and transcriptional remodelling. Furthermore, our findings highlight the predictive potential of baseline biomarkers in weight loss success in response to RYGB. Introduction Obesity is a multifactorial disease and is associated with an increased risk of cardiometabolic diseases and overall mortality.1,2 Despite the recent advances in incretin-based pharmacotherapies, bariatric surgery remains the most effective weight loss intervention and management of obesity-related cardiometabolic complications.3,4 Roux-en-Y gastric bypass (RYGB) is considered one of the best surgical interventions owing to substantial weight loss results and favourable long-term metabolic outcomes, including the remission of type 2 diabetes (T2D).4, 5, 6 However, interindividual responses to obesity treatments are highly variable, and 5–30% of patients fail to lose clinically beneficial weight loss following RYGB surgery.7, 8, 9, 10 As the largest tissue in the body by weight and the primary site of post-prandial glucose uptake, skeletal muscle is a principal regulator of systemic metabolic health and whole-body energy expenditure.11,12 The excess adiposity associated with obesity can lead to ectopic lipid accumulation and inflammation, which impair skeletal muscle insulin sensitivity, mitochondrial function and metabolic flexibility.13 Variability in responses to obesity therapies is impacted by skeletal muscle mass, composition and metabolism,14, 15, 16, 17, 18 highlighting a key role for skeletal muscle in determining weight loss outcomes and mediating the weight loss-independent antidiabetic effects of RYGB. Bariatric surgery induces metabolic reprogramming, whereby the surgical restrictions in food intake and/or nutrient malabsorption drive reciprocal (or bi-directional) changes in metabolism and the epigenome (e.g., DNA methylation, histone modifications, non-coding RNA) which result in sustained adaptations in whole-body metabolism.19, 20, 21 Environmental factors drive modifications to the DNA epigenome and can alter chromatin accessibility, which can modify transcriptional responses and gene expression programs.22 DNA methylation involves the addition of a methyl group to the 5′ position of cytosines located within CpG dinucleotides. Typically, DNA methylation at gene promoters and enhancers is associated with decreased gene expression, whereas DNA methylation in gene bodies is associated with active gene expression.23,24 Epigenome-wide association studies have associated alterations in DNA methylation with body mass index and markers of adiposity (waist circumference etc.) and can predict the future risk of T2D.25 Skeletal muscle DNA methylation patterns can alter the expression of genes involved in mitochondrial function and metabolic disease.26 Therefore, the reversal of obesity-related DNA methylation or demethylation may play a role in restoring skeletal muscle metabolism and weight loss response following bariatric surgery. Here, we aimed to identify the mechanisms underlying RYGB-induced metabolic reprogramming of skeletal muscle metabolism and to identify biomarkers of successful weight loss. We hypothesised that RYGB induces sustained metabolic reprogramming of skeletal muscle through enhanced mitochondrial oxidative capacity and coordinated epigenetic and transcriptional remodelling, and that baseline muscle molecular signatures predict postoperative metabolic outcomes. The objectives were to characterise the functional and molecular remodelling of skeletal muscle following RYGB, with a focus on mitochondrial oxidative capacity and integrated transcriptomic and epigenetic (DNA methylation) changes; to determine how these adaptations relate to improvements in whole-body metabolic health; and to identify preoperative skeletal muscle molecular and metabolic signatures that predict metabolic outcomes. Methods Study design and subject details Forty-one women undergoing RYGB surgery at The Ottawa Hospital Bariatric Centre of Excellence (Ottawa, Ontario, Canada) were recruited for this study following the Ontario Bariatric Network standard inclusion and exclusion criteria. Women account for roughly 80% of this surgical population. There were no exclusions based on ethnicity. Patients were screened prior to enrolment and excluded from the study if they had uncontrolled diabetes (HbA1C > 8.5) or a weight change of ± 5% in the 6 months prior to surgery. Three patients were excluded from the study, as two did not proceed with bariatric surgery and one patient was not eligible for testing during the follow-up period. RNA-seq experiments are powered to detect differences in gene expression of |log2FC| ≥ 0.50 at 80% power with n = 15 patients (Pre-/Post-RYGB) (see “RNA-seq” methods section for details). DNA methylation experiments are estimated to be powered with n = 35 patients to detect mean methylation differences of 5% at 71–89% power and 96–97% for differences of 10% (see “Differential methylation analysis” methods section for details). Anthropometric measurements, blood laboratory tests, and vastus lateralis muscle biopsies were collected 4–6 weeks prior to their scheduled RYGB surgery (before starting a prescribed preoperative hypocaloric diet, if required) and 52 weeks after their RYGB surgery. Height was determined using a freestanding stadiometer. Blood samples were collected after an overnight fast, and routine analyses were carried out in the Clinical Biochemistry Laboratory at the Ottawa Hospital (Ottawa, Ontario, Canada). Body composition was measured using bioelectrical impedance analysis (BIA; Tanita Corporation). Skeletal muscle sampling and tissue collection Following an overnight fast (12 h), vastus lateralis biopsies were taken under local anaesthesia with a Bergstrom needle modified for manual suction. The obtained sample was visually cleaned from adipose and subsequently divided for analysis based on priority and tissue availability. Tissue was snap frozen in liquid nitrogen and stored at −80 °C for later analyses. One portion of freshly biopsied tissue was placed in culture media for isolation of satellite cells (if biopsy size permitted). Otherwise the muscle was used for the multi-omics approaches described below, or placed in ice-cold BIOPS buffer for high resolution respirometry analysis of mitochondrial function [BIOPS (in mM): 2.77 CaK2EGTA, 7.23 K2EGTA, 5.77 Na2ATP, 50 MES, 20 imidazole, 20 taurine, 15 PCr, 0.5 DTT, 6.56 MgCL2–6H2O, pH 7.1 at 0 °C]. High-resolution respirometry The freshly biopsied muscle in ice-cold BIOPS buffer was permeabilised with 50 μg/mL saponin for 30 min at 4 °C under gentle agitation. High-resolution respirometry was conducted in duplicate using Oxygraph-2k systems (O2K, Oroboros Instruments, Innsbruck, Austria) in mitochondrial respiration media [MiR05 (in mM): 110 sucrose, 60 K-lactobionate, 20 HEPES, 20 taurine, 10 KH2PO4, 3 MgCl2, 0.5 EGTA, and 1 mg/mL fraction V BSA, pH 7.1 at 37 °C]. The assay protocol consisted of consecutive additions of: 2 mM malate, 5 mM pyruvate, 10 mM glutamate (CI substrates), 5 mM ADP (CI OXPHOS), 10 mM succinate (CI + II OXPHOS), 2.5 μM oligomycin (CI + II Leak), and 2.5 μM antimycin A (AA). Values were adjusted for non-mitochondrial oxygen consumption (i.e., that in the presence of AA) and expressed relative to tissue wet weight. Citrate synthase (CS)-normalised oxygen flux (pmol O2/s/CS) was calculated to account for potential variations in mitochondrial mass. Isolation of primary myoblasts Satellite cells were isolated from freshly biopsied vastus lateralis muscle and immunopurified using CD56 MicroBeads (Miltenyi Biotec). Purified myoblasts were cultured on Matrigel-coated dishes in Ham's F10 media containing 15% bovine growth serum, 1% antibiotic-antimycotic, 2.5 μg/mL gentamycin, 826 nM dexamethasone, 8.3 ng/mL human epidermal growth factor, and 25 pmol insulin. To avoid any batch effects, purified myoblasts from patients pre- and post- RYGB were stored in liquid nitrogen and were later thawed for outcome determinations. To induce differentiation into myotubes, upon confluency, myoblasts were cultured in low glucose DMEM supplemented with 2% horse serum, 1% antibiotic-antimycotic, and 2.5 μg/mL gentamycin. Seahorse analysis of primary myotube bioenergetics Cellular bioenergetics of primary myotubes were measured using the Seahorse XFe96 Analyser (Agilent). Myoblasts were seeded at 15,000 cells/well on Matrigel-coated Seahorse plates and differentiated in low glucose DMEM for 5–7 days. One hour before the experiment, the cells were washed and incubated in HCO3-free DMEM supplemented with 4 mM l-glutamine, 1 mM Na-pyruvate, and 5 mM d-glucose (pH 7.4). Myotubes were treated with sequential injections of oligomycin (2 μg/mL), FCCP (2 μM), antimycin A (1 μM) with rotenone (1 μM), and monensin (20 μM). After the assay, the Seahorse XF medium was removed, and the cells were washed with PBS before being lysed with RIPA Lysis Buffer (0.5 M Tris–HCl, pH 7.4, 1.5 M NaCl, 2.5% deoxycholic acid, 10% NP-40, 10 mM EDTA). Protein content was quantified using the BCA assay method with the Pierce™ BCA Protein Assay Kit (Thermofisher). Bioenergetic capacity and fuel flexibility were analysed by plotting glycolytic and oxidative ATP production as previously described.27 Protein extraction Frozen muscle was homogenised in ice-cold RIPA buffer (Millipore) supplemented with a protease inhibitor cocktail (Sigma) and phosphatase inhibitor cocktail (Thermofisher) using a bead mill homogeniser (Fisherbrand). Protein concentration was determined using the Pierce™ BCA Protein Assay Kit (Thermofisher). Enzyme activities Enzymatic activities for citrate synthase (CS) were conducted as previously described using the BioTek Synergy Mx Microplate Reader (BioTek Instruments, Inc., Winooski, VT).28 Briefly, CS activity was determined by measuring the rate of change of absorbance at 412 nm in 50 mM Tris–HCl (pH 8.0) with 0.2 mM DTNB, 0.1 mM acetyl-CoA. CS enzyme activity was calculated using extinction coefficient of ε = 13.6 mM−1cm−1 for CS. Immunoblotting Proteins were separated by SDS-PAGE under reducing conditions and transferred to PVDF membranes, and incubated overnight with primary antibodies: 4E-BP1 (Cell Signalling Cat# 9644, RRID: AB_2097841), Phospho-4E-BP1 (Thr37/Thr46) (Cell Signalling Cat# 2855, RRID: AB_560835), eIF2α (Cell Signalling Cat# 5324, RRID: AB_10692650), Phospho-eIF2α (Ser51) (Cell Signalling Cat# 3398, RRID: AB_2096481), EIF4E (Proteintech Cat# 11149-1-AP, RRID: AB_2934847), PTRF/Cavin1 (Proteintech Cat# 18892-1-AP, RRID: AB_10596795), RPL13 (Proteintech Cat# 11271-1-AP, RRID: AB_2269735), RPL14 (Proteintech Cat# 14991-1-AP, RRID: AB_2285319), RPS6 (Cell Signalling Cat# 2217, RRID: AB_331355), RPS10 (Proteintech Cat# 14894-1-AP, RRID: AB_2182493), RPS19 (Proteintech Cat# 15085-1-AP, RRID: AB_2180202), Total OXPHOS Human Antibody Cocktail (Abcam Cat# ab110411, RRID: AB_2756818), UBF (Santa Cruz Biotechnology Cat# sc-13125, RRID: AB_671403). GAPDH (Proteintech Cat# 60004-1-Ig, RRID: AB_2107436) was used as a loading control. Protein bands were visualised using the ChemiDoc™ MP Imaging System (Bio-Rad). Densitometry band analyses were performed using ImageJ software (NIH). RNA/DNA isolation Importantly, RNA and DNA were isolated simultaneously from the same 15–20 mg portion of snap-frozen muscle using the AllPrep DNA/RNA/miRNA Universal Kit (Qiagen) following the manufacturer's instructions. DNA methylation array and data processing DNA methylation analysis was performed by The Centre for Applied Genomics (TCAG), SickKids Research Institute (Toronto, Ontario, Canada). Genomic DNA was sodium bisulfite converted using the Epitect Bisulfite Kit (Qiagen). DNA was hybridised using Infinium® MethylationEPIC BeadChips (EPIC-8v2.0, Illumina) and subsequently imaged using the iScan Microarray Scanner (Illumina). Differential methylation analysis DNA methylation data were first genome-annotated according to the IlluminaHumanMethylationEPICv2anno.20a1.hg38 v1.0.0 library in R followed by processing using minfi v1.52.1 for quality control and normalisation29 for a robust and flexible analysis pipeline, as used by others.30, 31, 32 Raw probe values were processed via the “preprocessQuantile” function after removing probes associated with SNPs (via the “dropLociWithSnps” function with default parameters). Subsequently, normalised M- and β-values were obtained for a total of 914,791 CpG sites after excluding the Y chromosome and 458,086 gene-associated sites used for the assessment of gene expression-methylation associations (chromosome locations and gene association identified via the EPIC-8v2.0 array manifest by calling the “getAnnotation” minfi function). Per-sample difference in methylation profiles was computed and visualised through principal component analysis (PCA) computed across all CpG site normalised M-values via the “plotMDS” function of limma v3.62.2.33 Hierarchical clustering of methylation profiles was computed and visualised via the ComplexHeatmap v2.22.034 library, using Euclidean distances and complete linkages across the 10,000 most variable CpG sites (normalised M-values) for determining dendrograms. Overall difference of methylation was quantified by comparing mean M-values across CpG sites per-individual. Individual differentially methylated CpG sites were identified using limma with normalised M-values and paired sample design for patients pre- and post-RYGB. Statistical power for detecting differentially methylated sites were calculated in R via EpipwR v1.0.0.35 Power estimates were calculated (“get_power_cc” function) assuming 100 true different sites (FDR < 0.05 and Welch's t-test) with either 5% or 10% mean methylation differences between conditions out of 915,000 total sites. We report the range of power estimates over available references in EpipwR, as this influences estimates,35 determined with stoppage at a 0.01 margin of error of the 95% confidence interval for average power over a maximum of 10,000 simulated datasets (in our calculations, 1000 or less were required). Differentially methylated regions (DMRs) were obtained using the gold standard36,37 DMRcate v3.2137 library, using a threshold of DMRs containing more than two CpG sites (914,791 sites considered, normalised M-values) and a minimum smoothed FDR < 0.05. Methylation regions were determined with default parameters lambda = 1000 and C = 2, as suggested by the authors.37 Genomic regions of DMRs were determined according to the locations of individual CpG sites constituting each region. RNA-seq RNA quality was assessed using Bioanalyzer (Agilent 2100). RNA library preparation and sequencing were performed by the McMaster University Genomics Facility (Hamilton, Ontario, Canada). Enrichment of purified poly(A) mRNA was achieved using NEBNext® Poly(A) mRNA Magnetic Isolation Module. The recovered poly(A) selected RNA was used for cDNA library preparation using NEBNext® Ultra™ II Directional RNA Library Prep Kit. RNA-seq was performed using the Illumina NextSeq 2000 (P2 Flow cell, 2 × 50 bp configuration) system. Quality control of raw data was performed using FASTQC and MultiQC, and transcript abundance was quantified using Salmon38 and transcript and genome FASTA files from GENCODE GRCh38.p14 for index construction. These index files are available upon request. Transcript counts were then collapsed to gene-level counts by tximeta v1.24.039 and org.Hs.eg.db v3.20.0 in R for downstream analysis. Sample size for RNA-seq experiments were calculated in R via RNASeqPower v1.46.040 using the “rnapower” function for effect sizes |log2FC| = 0.5 at alpha = 0.05. We assumed an average gene count depth of 20 and a coefficient of variation of 0.4, as is expected for human tissue samples.40 Differential expression analysis Transcriptomics data were analysed for differentially expressed genes post-RYGB through DESeq2 v1.46.0.41 Gene expression values were normalised by the median of ratios technique within DESeq2, and differential gene expression was analysed for genes expressed in more than 25% of samples (n = 18), totalling 20,519 genes, with a paired sample design for patients pre- and post-RYGB. PCA of sample gene expression was computed using variance stabilising transformed (VST) values of the 10,000 most variable genes across samples. Hierarchical clustering of specified gene sets was computed and visualised via the ComplexHeatmap v2.22.034 library, using Euclidean distances and complete linkages across relevant gene expression profiles (VST values) for determining dendrograms. Integrative analysis of differentially expressed genes and differentially methylated regions with BETA We used BETA (binding and expression target analysis)42 for integrative analysis, a software tool developed for integrating ChIP-seq and transcriptomics (RNA-seq or microarray) datasets to identify potential transcriptional regulation by transcription factor binding. BETA defines regulatory influence as a function of the genomic distance between differentially bound peaks in ChIP-seq data and the transcription start site (TSS) of DEGs in transcriptomics data (See Wang et al. 201342 for details). As BETA has been adapted by others43 to explore epigenetic regulatory potential from DNA methylation by substituting ChIP-seq peaks with differential methylated CpG islands, we applied this to mine our datasets. BETA integration was applied to differential gene expression results from DESeq2 and differentially methylated CpG islands (replacing ChIP-seq peaks). Differentially methylated islands were identified using the GRCh38 CpG island bedfile obtained from the UCSC Genome Browser44 to select islands that overlapped with DMRs identified by DMRcate, identifying 10,268 differentially methylated islands. The BETA Basic tool on the cistrome webserver (http://cistrome.org) was used with differentially expressed genes defined as those with adjusted P < 0.05, leaving the remaining measured genes as background. Potential regulatory scores considered the influence of all identified peaks within 100 kb of a gene transcription start site. Enrichment analysis Functional enrichment of DMRs was performed via missMethyl v1.40.3 for enrichment of KEGG and Gene Ontology terms.45 Promoter DMRs were selected as those that overlapped with TSS200, TSS1500, or exon 1 regions. Functional enrichments for genes were determined via ClusterProfiler v4.1446 and tissue-specific gene enrichment in transcriptomics data were obtained from the TissueEnrich webserver (https://tissueenrich.gdcb.iastate.edu)47 using the Human Protein Atlas dataset and all tissue-specific gene sets. Transcription factor identification For transcription factor identification, the DEGs post-RYGB were submitted to the ChEA3 web-based application using all transcription factor-target gene libraries.48 To identify transcription factors regulated by methylation, the transcription factors identified by ChEA3 were cross-referenced to the list of genes that were both differentially methylated and differentially expressed (meDEG) post-RYGB. Correlation analysis Gene expression and methylation correlations and significance tests were computed using the base R library stats v4.4.3. Since associations between gene expression and DNA methylation are not necessarily linear,49 Spearman correlation values were computed pairwise between 458,086 gene-associated CpG site normalised M-values and normalised counts for DEGs (2173 genes) associated with common gene symbols. Multiple testing corrections were calculated via the Benjamini-Hochberg method with the “p.adjust” function. Classification models for post-RYGB weight loss Classifier models were trained and tested using the nestedcv v0.8.050, caret v7.0-151 and glmnet v4.1-1052 libraries in R. A nested cross-validation schema was applied to our 14 samples of high- and low-weight loss groups defined as top and bottom quintiles of age-adjusted percentage weight loss (PWL) at one-year post-RYGB (n = 7 for each group), with five-fold and leave-one-out cross-validation (LOOCV) for the inner and outer folds, respectively. This framework is commonly applied53, 54, 55 as a rigorous supervised learning approach to minimise model overfitting, which is particularly important in small sample size settings such as ours,56 where the number of possible predictors (e.g., genes) greatly outweighs the number of samples. Either pre-RYGB CpG normalised β methylation or DESeq-normalised RNA-seq counts were used for model features with a standard data scaler and 10 initial model features selected from all CpG sites (914,791) or genes (20,519). Features were selected according to t-test P-value computed across training samples for each outer-fold to identify features potentially relevant to post-RYGB weight loss without leaking information from the unseen test samples for model evaluation. Elastic net logistic regression models were trained at each inner cross-validation fold using the “nestcv.glmnet” function with binomial distribution, an α parameter sweep of [0.1, 0.2, …, 0.9], min_lse = 1 and a log loss metric for training. 100 repetitions of nested cross-validations for these models were applied for stability analysis for feature importance determination. The repeated gene models had average final parameters of α = 0.1 and λ = 0.001148. Feature importance was determined by variable stability (“var_stability” function) to identify genes or CpG sites selected across classifier repetitions, which constitute final model features. All model evaluations were computed on the unseen outer loop LOOCV test samples for each cross-validation fold using the “metrics” and “mcc” functions of nestedcv. Epigenetic age analysis DNA methylome predicted age (Zhang et al. (2019)57) was calculated using the MethylDetectR.58 Age acceleration values for both pre- and post-surgery were calculated by subtracting the chronological age from the estimated DNA methylation age. rDNA copy number Relative ribosomal DNA copy numbers (18S, 5.8S 28S and 5S) were determined through qPCR as previously described.59 Ten μL qPCR reactions were performed in duplicate using 1X SsoAdvanced™ Universal SYBR® Green Supermix (Bio-Rad), qPCR data were compiled using the CFX Maestro software (Bio-Rad) and analysed manually using the 2−ΔΔCT method. As per Gibbons et al. (2015),59 TP53 was used as a reference to assess rDNA dosage. Epigenome and transcriptome association with weight loss As age was predictive of PWL, we used age-adjusted PWL values for subsequent analysis. To identify genes that may associate with weight loss, DESeq2 normalised gene expression data and Minfi normalised DNA β methylation data present in at least 30 baseline samples (914,791 sites and 13,330 genes) were used to compute Pearson correlations with age-adjusted PWL and significance using the “rcor.test” function. For PWL correlations with both gene expression and CpG site methylation, P < 0.01, Pearson's |r| > 0.4 values were used as cut-offs for nominal significance and an FDR < 0.05 for true significance. Pearson correlation was chosen to determine direct, linear relationships between clinical outcomes and pre-processed omics data. Benjamini-Hochberg P-value corrections were computed with the “p.adjust” function. Ethics All participants provided informed oral and written consent. This study was approved by the Ottawa Health Science Network Research Ethics Board (Protocol #2006538-01H). This study was performed in accordance with the 2024 Declaration of Helsinki. Statistical analysis Statistical analysis was performed using Prism (GraphPad Software Inc) and R version 4.4.3.60 The data were first subjected to the Shapiro–Wilk normality test and the method to identify statistical significance of pre-RYGB vs post-RYGB was based on the normality of the data distribution, using either a paired two-tailed Student's t-test for parametric tests or a Wilcoxon matched-pairs signed-rank test for nonparametric tests. Statistical differences were considered significant when the P-value was ≤0.05. For multiple testing, an FDR < 0.05 was defined as significant. Data are presented as means ± SD unless otherwise specified. Standardised statistical methods were utilised in different stages of our analyses, including correlation (Pearson and Spearman), unsupervised projection analysis (i.e., PCA), unsupervised ML analysis (hierarchical clustering) and supervised, classification analysis with elastic nets. For analyses using previously established R libraries, all methodological details are provided in their respective methods sections. Role of the funding source This work was supported by the Canadian Institutes of Health Research. The funder had no influence on the conceptualisation, data collection, analysis, or interpretation of data presented. Results RYGB induces substantial weight loss and improves glucose homoeostasis To determine the mechanisms of skeletal muscle metabolic programming in response to bariatric surgery, we studied a cohort of 41 women 6–8 weeks prior to RYGB surgery and 1 year after RYGB surgery (Fig. 1A). Self-reported ethnicity revealed that 31 patients were of European descent, 2 patients were of African descent, 2 patients were of Asian descent, 1 patient was of Indigenous descent, 1 patient was of Latin-American descent, and 4 patients were multiracial. RYGB surgery induced an average of 32.2% total weight loss and 68.6% excess weight loss (P < 0.001, Table 1). Body composition analysis by bioimpedance (BIA) revealed an average loss of 31.8 kg (48.1%) of fat mass and 8.4 kg (14.2%) of fat-free mass (FFM) (P < 0.001, Table 1). RYGB surgery improved glucose handling as reflected by a >75% reduction in HOMA-IR and a 10.9% decrease in HbA1c (P < 0.001, Table 1). These improvements in glucose homoeostasis were accompanied by reductions in both systolic and diastolic blood pressure, as well as improved plasma lipid profiles (all P < 0.001, Table 1).Fig. 1RYGB induces global skeletal muscle DNA methylation changes in regions associated with muscle development and differentiation. (A) Schematic summary of the time points for sample collection. Briefly, 41 women with obesity were recruited 4–6 weeks prior to RYGB surgery, and one-year following RYGB surgery for collection of blood samples and vastus lateralis muscle biopsies. Diagram created with BioRender.com. (B) Principal component analysis of sample methylation profiles. Lines in plot connect pre- and post-RYGB samples corresponding to the same individual. (C) Mean DNA methylation (M-values) across sites of sample methylation profiles in skeletal muscle pre- and post- RYGB. (D) Volcano plot of differentially methylated regions (DMRs) in vastus lateralis obtained pre- and post-RYGB surgery. Highlighted are DMRs containing ≥2 CpGs, FDR < 0.01, |maxdiff| > 0.025. (E) GO enrichment plot of DMRs analysed and visualised using MissMethyl. (F) Calculated DNA methylation age (EpiAge, Zhang et al. 201957). Age acceleration (EpiAge-true age) decreased at one-year post-RYGB. Comparisons between pre- and post-RYGB were determined using a paired two-tailed Student's t-test. P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001, ∗∗∗∗P < 0.0001. Unless otherwise stated, all values are presented as means ± SD (n = 35).Table 1Clinical characteristics of patients pre- and post-RYGB.VariablePre-RYGBPost-RYGBP-valuen3838Age (years)44.4 ± 9.645.4 ± 9.6Height (cm)163.6 ± 9.6163.0 ± 6.9Body weight (kg)125.7 ± 18.985.8 ± 17.31.1 × 10−24BMI (kg/m2)46.9 ± 5.632.1 ± 5.01.0 × 10−23Waist circumference (cm)129.7 ± 11.7100.8 ± 11.56.2 × 10−19Hip circumference (cm)144.3 ± 12.4116.5 ± 13.91.3 × 10−19Heart rate (BPM)79.4 ± 10.169.7 ± 11.75.8 × 10−5Systolic BP (mmHg)142 ± 15124 ± 135.92 × 10−7Diastolic BP (mmHg)89 ± 878 ± 85.8 × 10−7Body composition by bioelectrical impedance analysis Fat mass (kg)67.0 ± 12.535.3 ± 12.13.0 × 10−25 Fat-free mass (kg)58.7 ± 6.950.3 ± 5.85.7 × 10−20 % Body fat52.9 ± 2.340.0 ± 6.21.68 × 10−17Fasting blood biochemistry Glucose (mmol/L)5.74 ± 0.994.98 ± 0.388.9 × 10−5 HbA1c (%)5.80 ± 0.765.16 ± 0.434.7 × 10−10 Insulin (pmol/L)207.7 ± 154.254.5 ± 18.31.2 × 10−7 HOMA-IR7.9 ± 5.81.8 ± 0.64.8 × 10−7 Creatinine (μmol/L)65.1 ± 12.962.2 ± 10.1NS Triglycerides (mmol/L)1.77 ± 0.691.06 ± 0.355.8 × 10−9 Total cholesterol (mmol/L)4.84 ± 0.944.25 ± 0.631.7 × 10−5 HDL-cholesterol (mmol/L)1.29 ± 0.291.54 ± 0.349.5 × 10−8 LDL cholesterol (mmol/L)2.73 ± 0.792.23 ± 0.572.7 × 10−5 TSH (mlU/L)2.40 ± 1.512.35 ± 1.79NS Free T3 (pmol/L)4.69 ± 0.644.13 ± 0.516.3 × 10−4 Free T4 (pmol/L)14.3 ± 2.215.1 ± 1.83.8 × 10−6BMI, body mass index; BP, blood pressure; HbA1c, glycated haemoglobin A1c; HDL, high density lipoprotein; HOMA-IR, homoeostatic model assessment for insulin resistance; LDL, low density lipoprotein; TSH, thyroid-stimulating hormone; T3, triiodothyronine; T4, thyroxine. Values are means ± SD. Pairwise two-tailed Student's t-test or Wilcoxon's signed-rank test (fasting blood glucose, HbA1c, insulin, HOMA-IR, triglycerides, total cholesterol and TSH). RYGB induces global skeletal muscle DNA methylation changes in regions associated with muscle development and differentiation. (A) Schematic summary of the time points for sample collection. Briefly, 41 women with obesity were recruited 4–6 weeks prior to RYGB surgery, and one-year following RYGB surgery for collection of blood samples and vastus lateralis muscle biopsies. Diagram created with BioRender.com. (B) Principal component analysis of sample methylation profiles. Lines in plot connect pre- and post-RYGB samples corresponding to the same individual. (C) Mean DNA methylation (M-values) across sites of sample methylation profiles in skeletal muscle pre- and post- RYGB. (D) Volcano plot of differentially methylated regions (DMRs) in vastus lateralis obtained pre- and post-RYGB surgery. Highlighted are DMRs containing ≥2 CpGs, FDR < 0.01, |maxdiff| > 0.025. (E) GO enrichment plot of DMRs analysed and visualised using MissMethyl. (F) Calculated DNA methylation age (EpiAge, Zhang et al. 201957). Age acceleration (EpiAge-true age) decreased at one-year post-RYGB. Comparisons between pre- and post-RYGB were determined using a paired two-tailed Student's t-test. P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001, ∗∗∗∗P < 0.0001. Unless otherwise stated, all values are presented as means ± SD (n = 35). Table 1Clinical characteristics of patients pre- and post-RYGB.VariablePre-RYGBPost-RYGBP-valuen3838Age (years)44.4 ± 9.645.4 ± 9.6Height (cm)163.6 ± 9.6163.0 ± 6.9Body weight (kg)125.7 ± 18.985.8 ± 17.31.1 × 10−24BMI (kg/m2)46.9 ± 5.632.1 ± 5.01.0 × 10−23Waist circumference (cm)129.7 ± 11.7100.8 ± 11.56.2 × 10−19Hip circumference (cm)144.3 ± 12.4116.5 ± 13.91.3 × 10−19Heart rate (BPM)79.4 ± 10.169.7 ± 11.75.8 × 10−5Systolic BP (mmHg)142 ± 15124 ± 135.92 × 10−7Diastolic BP (mmHg)89 ± 878 ± 85.8 × 10−7Body composition by bioelectrical impedance analysis Fat mass (kg)67.0 ± 12.535.3 ± 12.13.0 × 10−25 Fat-free mass (kg)58.7 ± 6.950.3 ± 5.85.7 × 10−20 % Body fat52.9 ± 2.340.0 ± 6.21.68 × 10−17Fasting blood biochemistry Glucose (mmol/L)5.74 ± 0.994.98 ± 0.388.9 × 10−5 HbA1c (%)5.80 ± 0.765.16 ± 0.434.7 × 10−10 Insulin (pmol/L)207.7 ± 154.254.5 ± 18.31.2 × 10−7 HOMA-IR7.9 ± 5.81.8 ± 0.64.8 × 10−7 Creatinine (μmol/L)65.1 ± 12.962.2 ± 10.1NS Triglycerides (mmol/L)1.77 ± 0.691.06 ± 0.355.8 × 10−9 Total cholesterol (mmol/L)4.84 ± 0.944.25 ± 0.631.7 × 10−5 HDL-cholesterol (mmol/L)1.29 ± 0.291.54 ± 0.349.5 × 10−8 LDL cholesterol (mmol/L)2.73 ± 0.792.23 ± 0.572.7 × 10−5 TSH (mlU/L)2.40 ± 1.512.35 ± 1.79NS Free T3 (pmol/L)4.69 ± 0.644.13 ± 0.516.3 × 10−4 Free T4 (pmol/L)14.3 ± 2.215.1 ± 1.83.8 × 10−6BMI, body mass index; BP, blood pressure; HbA1c, glycated haemoglobin A1c; HDL, high density lipoprotein; HOMA-IR, homoeostatic model assessment for insulin resistance; LDL, low density lipoprotein; TSH, thyroid-stimulating hormone; T3, triiodothyronine; T4, thyroxine. Values are means ± SD. Pairwise two-tailed Student's t-test or Wilcoxon's signed-rank test (fasting blood glucose, HbA1c, insulin, HOMA-IR, triglycerides, total cholesterol and TSH). Clinical characteristics of patients pre- and post-RYGB. BMI, body mass index; BP, blood pressure; HbA1c, glycated haemoglobin A1c; HDL, high density lipoprotein; HOMA-IR, homoeostatic model assessment for insulin resistance; LDL, low density lipoprotein; TSH, thyroid-stimulating hormone; T3, triiodothyronine; T4, thyroxine. Values are means ± SD. Pairwise two-tailed Student's t-test or Wilcoxon's signed-rank test (fasting blood glucose, HbA1c, insulin, HOMA-IR, triglycerides, total cholesterol and TSH). RYGB alters skeletal muscle DNA methylation profiles To investigate the epigenetic mechanisms underlying the reprogramming of skeletal muscle metabolism, we performed DNA methylation profiling on skeletal muscle DNA from 35 patients both pre- and post-RYGB. When applying |log2 FC| > 0.5 and an unadjusted P < 0.01 as cut-offs, we detected 53 differentially methylated CpG positions (DMPs) between pre- and post-RYGB skeletal muscle (Table S1). Several of these DMPs were associated with genes involved in DNA-binding/transcription regulation (TP53, TOPBP1, ATF5, PAX6), and ribosome biogenesis and mRNA transport (NUP153, RPP38, NUP62, FGF12) (Table S1). Notably, DMPs in the promoter region of genes involved in iron homoeostasis (FTL, GCLM, SLC40A1, TF) were hypomethylated post-RYGB (Table S1). However, none of the individual DMPs passed false discovery rate (FDR) < 0.05 testing. This is not surprising, as with complex phenotypes, small differences may translate into biologically significant changes, and effect sizes are rarely greater than 5%.61 Indeed, multidimensional scaling (MDS) analysis illustrates that the differences between patient pre- and post-RYGB methylation profiles are moderate and vary substantially between patient profiles (Fig. 1B and Fig. S1A). Overall, DNA methylation slightly decreased in skeletal muscle post-RYGB (P < 0.05, Fig. 1C). We next focused on determining DMRs, constituting regions with multiple contiguous CpG sites, which better reflect changes in gene expression compared to individual differentially methylated CpG sites.61,62 We detected 9537 DMRs defined as regions across all CpG sites with more than two contiguous sites and a minimum smoothed FDR < 0.05,63 of which 4132 were hypermethylated post-RYGB and 5405 DMRs were hypomethylated post-RYGB (Fig. 1D, Table S2). Genomic annotations of the 9537 DMRs mapped to 8233 genes and were located primarily in the gene body and promoter regions (Fig. S1B). Genes hypermethylated following RYGB included secretagogin (SCGN), a calcium-binding protein involved in insulin signalling,64 and NAD+ kinase (NADK), which converts NAD+ into NADP+. In contrast, hypomethylated genes included BARX homoeobox 2 (BARX2) and the four-and-a-half LIM domain protein 3 (FHL3), which play roles skeletal muscle metabolism and regeneration65,66 (Table S2). GO term enrichment analysis revealed that genes associated with DMRs, and particularly DMRs located in the promoter regions, were involved in processes related to embryonic organ and tissue development, skeletal muscle development and structure, and cell proliferation (Fig. 1E and Fig. S1C). Of note, DMRs overlap with several obesity-related candidate genes, which are differentially methylated in response to various weight loss interventions,67, 68, 69 including LEP, TNF, GDNF, PTPRE, ACACB, PDK4, and PPARGC1A (Table S2). As DNA methylation patterns of specific CpGs that change with age can be used as a measure of biological age (EpiAge) using epigenetic clocks,57,70 we next tested whether RYGB could reverse the accelerated ageing associated with obesity. While EpiAge did not change post-RYGB (Fig. 1F), EpiAge acceleration, which is the difference between EpiAge compared to chronological age, decreased post-RYGB (P < 0.0001, Fig. 1F). RYGB reprogrammes skeletal muscle gene expression To assess how changes in DNA methylation affected gene transcription, we subsequently performed RNA-seq on RNA simultaneously purified with DNA from the same portion of muscle. PCA showed a modest but distinct shift in overall transcriptomic profiles in post-RYGB skeletal muscle (Fig. 2A). Pairwise comparisons of the muscle transcriptome identified 974 genes with increased expression and 1199 genes with decreased expression post-RYGB (FDR < 0.05) (Fig. 2B, Table S3). Top differentially expressed genes included the increased expression of disco-interacting protein 2 homologue C (DIP2C), bridge-like lipid transfer protein family member 1 (BLTP1), nicotinamide N-methyltransferase (NNMT), and Wnt family member 9A (WNT9A) (all P < 0.0001, Fig. 2B and C), which may play roles in regulating lipid metabolism and muscle satellite cell differentiation.71,72 Top genes downregulated by RYGB included leptin (LEP), myostatin (MSTN/GDF8), a myokine that inhibits skeletal muscle growth, carnosine synthase 1 (CARNS1), and lactate dehydrogenase A (LDHA) (all P < 0.0001, Fig. 2B and C).Fig. 2RYGB remodels skeletal muscle gene expression, enhancing expression of genes related to muscle growth, differentiation, and protein translation while decreasing expression of genes encoding mitochondrial proteins. (A) Principal component analysis of sample gene expression profiles. (B) Volcano plot of differentially expressed genes (DEGs) from vastus lateralis obtained pre- and post-RYGB surgery. Highlighted are DEGs with |log2FC| > 1 and FDR < 0.05. (C) Gene expression of selected top DEGs in vastus lateralis obtained pre- and post-RYGB surgery. (D) Tissue enrichment analysis of DEGs demonstrating that most DEGs were not tissue-specific. (E) Gene expression of selected adipose-specific DEGs in vastus lateralis obtained pre- and post-RYGB surgery. (F and G) GO enrichment analysis of DEGs analysed and visualised using ClusterProfiler. (F) GO enrichment plot of altered pathways in vastus lateralis post-RYG (G) Enrichment of common down-regulated and up-regulated pathways in vastus lateralis post-RYGB. Comparisons between pre- and post-RYGB were determined using a paired two-tailed Student's t-test (DIP2C in panel C and DGAT2 in panel E) or Wilcoxon's signed-rank test (all other targets in panels C and E). P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001, ∗∗∗∗P < 0.0001. Unless otherwise stated, all values are presented as means ± SD (n = 35). RYGB remodels skeletal muscle gene expression, enhancing expression of genes related to muscle growth, differentiation, and protein translation while decreasing expression of genes encoding mitochondrial proteins. (A) Principal component analysis of sample gene expression profiles. (B) Volcano plot of differentially expressed genes (DEGs) from vastus lateralis obtained pre- and post-RYGB surgery. Highlighted are DEGs with |log2FC| > 1 and FDR < 0.05. (C) Gene expression of selected top DEGs in vastus lateralis obtained pre- and post-RYGB surgery. (D) Tissue enrichment analysis of DEGs demonstrating that most DEGs were not tissue-specific. (E) Gene expression of selected adipose-specific DEGs in vastus lateralis obtained pre- and post-RYGB surgery. (F and G) GO enrichment analysis of DEGs analysed and visualised using ClusterProfiler. (F) GO enrichment plot of altered pathways in vastus lateralis post-RYG (G) Enrichment of common down-regulated and up-regulated pathways in vastus lateralis post-RYGB. Comparisons between pre- and post-RYGB were determined using a paired two-tailed Student's t-test (DIP2C in panel C and DGAT2 in panel E) or Wilcoxon's signed-rank test (all other targets in panels C and E). P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001, ∗∗∗∗P < 0.0001. Unless otherwise stated, all values are presented as means ± SD (n = 35). To examine whether changes in gene expression derive from muscle changes or just decreased adipose content we performed tissue enrichment analysis.47 While the majority of DEGs were not tissue specific (Fig. 2D), differences in muscle-specific lipid metabolism genes, including downregulated DGAT2 (P < 0.001), PLIN1 (P < 0.05), FABP4 (P < 0.01), and ADIPOQ (P < 0.001), likely reflect a decrease in intramuscular lipids post-RYGB (all P < 0.05, Fig. 2E). Functional enrichment analysis of the DEGs revealed associations with mitochondrial organisation, metabolic processes, and protein translation (Fig. 2F). Upregulated genes following bariatric surgery primarily related to ribosomal biogenesis and RNA translation (Fig. 2G). In contrast, genes associated with the electron transport chain and oxidative metabolism were downregulated post-RYGB (Fig. 2G), which is consistent with previous findings.73,74 RYGB-induced changes in skeletal muscle DNA methylation partially regulate gene expression To determine if RYGB-induced changes in DNA methylation contribute to the reprogramming of skeletal muscle gene expression, we next compared the differentially methylated and differentially expressed genes (Fig. 3A). Of the 8233 genes that were differentially methylated (9537 DMRs) and the 2173 genes that were differentially expressed post-RYGB surgery, 745 genes were both differentially methylated (with DMRs in the gene body or within 1500 bp of the TSS) and differentially expressed (Fig. 3B and C and Table S4). When also considering potential long-range regulation (≤100 kb from a gene's TSS as predicted by BETA42), 1197 DEGs were identified as regulated by DMRs.42 Associations between these long-range regulatory DMRs and upregulated genes post-RYGB indicate genome-wide activation of skeletal muscle gene expression, supporting the idea of epigenetic reprogramming in response to surgery (Fig. 3D). The 745 short-range regulated DEGs were mainly enriched in pathways related to skeletal muscle development and differentiation (Fig. 3E), but also included several ribosomal subunits (i.e., 16/68 of the ribosomal DEGs also had DMRs) and genes involved in carbohydrate and fatty acid metabolism (Table S4).Fig. 3Skeletal muscle gene expression changes with RYGB are associated with DNA methylation changes. (A) Schematic summary of the integrative data analysis to identify methylation-regulated differentially expressed genes (meDEGs). Diagram created with BioRender.com. (B) Venn diagrams depicting the 745 genes that were both differentially methylated and differentially expressed. The input included 8233 genes that were differentially methylated and 2173 genes that were differentially expressed in vastus lateralis post-RYGB. (C) Quadrant plot of differential methylation vs. differential expression of genes post-RYGB. Plotted values are the change in expression of genes from RNA-seq and the maximum difference in CpG methylation of methylated regions. (D) BETA integration of RNA-seq and methylation data predicting global epigenetic regulation of positive (red) and negative (blue) DEGs post-RYGB relative to unchanged genes (dashed line). P-values for overall up/downregulation significance indicated in legend parentheses. (E) GO enrichment dot plots of 745 genes that were both differentially methylated and differentially expressed. (F) Spearman correlations between CpGs and gene expression. Highlighted are genes with |Spearman Rho| > 0.5 and FDR < 0.05. Orange symbols, positive correlations; purple symbols, negative correlations. Skeletal muscle gene expression changes with RYGB are associated with DNA methylation changes. (A) Schematic summary of the integrative data analysis to identify methylation-regulated differentially expressed genes (meDEGs). Diagram created with BioRender.com. (B) Venn diagrams depicting the 745 genes that were both differentially methylated and differentially expressed. The input included 8233 genes that were differentially methylated and 2173 genes that were differentially expressed in vastus lateralis post-RYGB. (C) Quadrant plot of differential methylation vs. differential expression of genes post-RYGB. Plotted values are the change in expression of genes from RNA-seq and the maximum difference in CpG methylation of methylated regions. (D) BETA integration of RNA-seq and methylation data predicting global epigenetic regulation of positive (red) and negative (blue) DEGs post-RYGB relative to unchanged genes (dashed line). P-values for overall up/downregulation significance indicated in legend parentheses. (E) GO enrichment dot plots of 745 genes that were both differentially methylated and differentially expressed. (F) Spearman correlations between CpGs and gene expression. Highlighted are genes with |Spearman Rho| > 0.5 and FDR < 0.05. Orange symbols, positive correlations; purple symbols, negative correlations. Transcription factor enrichment analysis identified 58 known transcription factors that were both differentially methylated and differentially expressed (Table S5). RYGB-induced promoter hypomethylation was associated with enhanced expression of the SRY-box transcription factor 15 (SOX15), the vitamin D receptor (VDR), and the transcriptional repressor B cell lymphoma 6 (BCL6) (Table S5), indicating that the demethylation in the promoter regions of these genes likely contributes to their activation through epigenetic mechanisms. Notably, both SOX15 and VDR play key roles in skeletal muscle regeneration and maintenance,75,76 and BCL6 may enhance skeletal muscle ribosome biogenesis and assembly and protein translation efficiency.77 Both hypermethylated (1 DMR) and hypomethylated (3 DMRs) regions of oestrogen-related receptor-γ (ESRRG) were associated with increased expression, which may contribute to enhanced mitochondrial content and function.78 To further investigate the epigenetic control of skeletal muscle gene expression following RYGB surgery, correlations were calculated between the gene expression of 2173 DEGs and the methylation of specific CpG sites associated with each gene, which identified 545 CpG sites that correlated with 212 DEGs (Spearman FDR < 0.05) (Table S6). Interestingly, the hindlimb transcription factor,79 PITX1, had 11 positive gene body CpG correlations and 5 negative promoter CpG correlations with PITX1 gene expression. Multiple CpG sites were also highly correlated with the expression of genes involved in skeletal muscle structure and function (ATP2A1, MYO1C, MYOM2, PDLIM7, TPM1), transcriptional regulation and cell cycle progression (CDKN1C, ESRRG, SELENOW, S100A1), and metabolism (ALDH5A1, SLC16A3) (Fig. 3F, Table S6). Taken together, our integrative analysis suggests that RYGB induces the re-programming of skeletal muscle gene expression at least in part through epigenetic mechanisms. RYGB improves skeletal muscle mitochondrial function Skeletal muscle mitochondrial dysfunction in obesity is associated with low mitochondrial content and impaired bioenergetic function. Given that our integrated analyses identified several mitochondria-related changes, including the decreased expression of genes involved in OXPHOS, we next sought to explore the impact of RYGB on skeletal muscle mitochondrial function. Citrate synthase activity, a biomarker of mitochondrial content,80 was higher in muscle post-RYGB (P < 0.05, Fig. 4A, Fig. S2A), in line with previous findings,81 and reveals a dissociation between the observed decrease in expression of genes encoding mitochondrial proteins and mitochondrial content.Fig. 4RYGB-induced weight loss improves skeletal muscle mitochondrial metabolism. (A) Citrate synthase (CS) activity increased in vastus lateralis at one-year post-RYGB (n = 32). See also Fig. S1. (B) High-resolution respiratory flux per mg wet weight (w.w.) of saponin-permeabilised vastus lateralis muscle. RYGB surgery enhanced complex I (CI) leak and oxidative phosphorylation (OXPHOS) and CI + II OXPHOS and leak respiration (n = 29). See also Fig. S2. (C) Immunoblot analyses revealed higher protein expression of OXPHOS complexes CI, CII, CIV, and the ATP synthase (CV) post-RYGB (n = 30). (D) Seahorse analysis demonstrated greater basal, leak, and maximal oxygen consumption in primary myotubes obtained post-RYGB (n = 20). See also Fig. S2. Comparisons between pre- and post-RYGB were determined using a paired two-tailed Student's t-test or Wilcoxon's signed-rank test (CIV protein abundance in panel C). ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001, ∗∗∗∗P < 0.0001. All values are presented as means ± SD. RYGB-induced weight loss improves skeletal muscle mitochondrial metabolism. (A) Citrate synthase (CS) activity increased in vastus lateralis at one-year post-RYGB (n = 32). See also Fig. S1. (B) High-resolution respiratory flux per mg wet weight (w.w.) of saponin-permeabilised vastus lateralis muscle. RYGB surgery enhanced complex I (CI) leak and oxidative phosphorylation (OXPHOS) and CI + II OXPHOS and leak respiration (n = 29). See also Fig. S2. (C) Immunoblot analyses revealed higher protein expression of OXPHOS complexes CI, CII, CIV, and the ATP synthase (CV) post-RYGB (n = 30). (D) Seahorse analysis demonstrated greater basal, leak, and maximal oxygen consumption in primary myotubes obtained post-RYGB (n = 20). See also Fig. S2. Comparisons between pre- and post-RYGB were determined using a paired two-tailed Student's t-test or Wilcoxon's signed-rank test (CIV protein abundance in panel C). ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001, ∗∗∗∗P < 0.0001. All values are presented as means ± SD. High-resolution respirometry conducted on saponin-permeabilised vastus lateralis fibres revealed that mitochondrial complex I (CI) and CI + II OXPHOS and leak respiration were higher in muscle obtained post-RYGB (P < 0.05 and P < 0.001 respectively, Fig. 4B). However, when normalising respiration to CS activity, only complex I OXPHOS was significantly increased post-RYGB (P < 0.05, Fig. S2B), indicating that there are significant metabolic improvements in whole muscle oxidative characteristics, and this is largely driven by increased mitochondrial content. Consistent with this, immunoblotting of key subunits of the OXPHOS complexes revealed that expression of CI, CII, CIV and CV, but not CIII, were higher one-year post-RYGB (P < 0.01, Fig. 4C), further indicating that there may not be a fixed stoichiometry between mitochondrial gene and protein expression levels. The improvement in skeletal muscle mitochondrial oxidative capacity correlated with the decrease in HbA1c post-RYGB (Fig. S2C), consistent with the well-known role of skeletal muscle in glucose disposal and overall glucose homoeostasis. Bioenergetic analyses of primary muscle cells derived from the pre- and post-RYGB skeletal muscle biopsies revealed that the improvements in biopsied muscle mitochondrial metabolism were driven by intrinsic mechanisms within the cells rather than by external factors that could exist in vivo. At one-year post-RYGB, basal, leak, and maximal oxygen consumption rates were higher in primary myotubes (P < 0.05, Fig. 4D and Fig. S2D–S2F). Taken together, these cell-autonomous improvements in skeletal muscle mitochondrial oxidative capacity are indicative of sustained metabolic reprogramming. RYGB induces skeletal muscle ribosomal biogenesis and enhances protein translation Ribosomes are the rate-controlling enzymatic machinery of protein synthesis and of the ∼48 r-proteins that comprise the large 60S subunit (along with 28S, 5.8S, and 5S rRNAs), and the 33 r-proteins that comprise the small 40s subunit (along with 18S rRNA), 34 large r-proteins and 26 small r-proteins were higher post-RYGB (Fig. 5A, Table S3). Given the finding of increased enrichment of ribosomal proteins, we sought to verify if RYGB surgery was associated with enhanced ribosomal biogenesis.Fig. 5RYGB induces skeletal muscle ribosomal biogenesis. (A) Heatmaps with hierarchical clustering (Euclidean distance, complete linkage) showing the change in gene expression of ribosomal subunits post-RYGB (n = 35). (B) Heatmaps with hierarchical clustering (Euclidean distance, complete linkage) showing the change in gene expression of selected factors regulating ribosomal biogenesis post-RYGB (n = 35). (C) Immunoblot analysis demonstrating enhanced expression of upstream binding factor (UBF), the transcription termination factor, polymerase I and transcript release factor (PTRF/Cavin-1) in vastus lateralis post-RYGB (n = 30). (D) Relative rDNA dosage determined via quantitative PCR (qPCR) revealed higher 5s rDNA post-RYGB (n = 34). (E) Protein expression of selected large (RPL13, RPL14) and small (RPS6, RPS10, RPS19) ribosomal subunits tended to increase in vastus lateralis post-RYGB (n = 30). Comparisons between pre- and post-RYGB were determined using a paired two-tailed Student's t-test (UBF in panel C; 5s rDNA in D; RPS19 in E) or Wilcoxon's signed-rank test (PTRF in panel C; all other targets in panels D and E). P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001, ∗∗∗∗P < 0.0001. Heatmaps represent log2 (fold change); all other values are presented as means ± SD (n = 35). RYGB induces skeletal muscle ribosomal biogenesis. (A) Heatmaps with hierarchical clustering (Euclidean distance, complete linkage) showing the change in gene expression of ribosomal subunits post-RYGB (n = 35). (B) Heatmaps with hierarchical clustering (Euclidean distance, complete linkage) showing the change in gene expression of selected factors regulating ribosomal biogenesis post-RYGB (n = 35). (C) Immunoblot analysis demonstrating enhanced expression of upstream binding factor (UBF), the transcription termination factor, polymerase I and transcript release factor (PTRF/Cavin-1) in vastus lateralis post-RYGB (n = 30). (D) Relative rDNA dosage determined via quantitative PCR (qPCR) revealed higher 5s rDNA post-RYGB (n = 34). (E) Protein expression of selected large (RPL13, RPL14) and small (RPS6, RPS10, RPS19) ribosomal subunits tended to increase in vastus lateralis post-RYGB (n = 30). Comparisons between pre- and post-RYGB were determined using a paired two-tailed Student's t-test (UBF in panel C; 5s rDNA in D; RPS19 in E) or Wilcoxon's signed-rank test (PTRF in panel C; all other targets in panels D and E). P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001, ∗∗∗∗P < 0.0001. Heatmaps represent log2 (fold change); all other values are presented as means ± SD (n = 35). De novo synthesis of ribosomes from rDNA by Pol I is facilitated by several protein factors.82 While gene expression of Pol I subunits and components of the pre-initiation complex (PIC) remained relatively unchanged post-RYGB (Fig. 5B), protein expression of upstream binding factor (UBF) and the transcription termination factor, polymerase I and transcript release factor (PTRF/Cavin-1) were moderately higher post-RYGB (P < 0.05 and P < 0.001 respectively, Fig. 5C). Transcription of rDNA by Pol I results in the synthesis of 45S pre-rRNA, which is subsequently processed into the mature 18S, 5.8S, and 28S rRNAs, whereas the 5s rRNA is transcribed by Pol III.82 Quantification of rDNA copy number revealed that 18S, 5.8S, and 28S rDNA were similar pre- and post-RYGB in skeletal muscle, whereas 5s rDNA copy number increased post-RYGB (P < 0.05, Fig. 5D). Furthermore, in line with gene expression, protein expression of ribosomal subunits tended to increase post-RYGB, as evidenced by increased expression of RPS6 and RPS19 post-RYGB (P < 0.01 and P < 0.05 respectively), along with a trend for increased expression of RPL13 (P = 0.096, Fig. 5E). Taken together, these results indicate that ribosomal biogenesis increases in skeletal muscle following bariatric surgery. We next analysed the expression of key proteins involved in initiating translation. Despite increased levels of eIF4E post-RYGB, indicating increased translation initiation, inactivation of 4E-BP1, as indicated by the phospho/total 4E-BP1 ratio, remained unchanged (Fig. S3A and S3B). Expression of eIF2α, involved in the assembly of a ternary complex for ribosome recruitment, was higher post-RYGB (Fig. S3B). Total eIF2α increased expression post-RYGB, although inactivation of eIF2a by ser51 phosphorylation was similar pre- and post-RYGB (Fig. S3B). Differences in translational capacity and protein synthesis are challenging to detect in the absence of an anabolic stimulus (e.g., exercise, dietary amino acids, anabolic hormones).83,84 Nevertheless, these data indicate that there is an overall increase in protein translation machinery in skeletal muscle at one-year post-RYGB. Pre-RYGB skeletal muscle DNA methylation and gene expression associate with weight loss success Interindividual responses to obesity treatments are highly variable, and we have previously shown that weight loss in diet-adherent patients during a behavioural program predicts weight loss following RYGB surgery.15,85 Moreover, patients who lose weight rapidly in response to a hypocaloric diet exhibit several differences in skeletal muscle composition and metabolism, such as higher proportions of oxidative fibre types, enhanced mitochondrial bioenergetics and increased expression of ribosomal proteins,17,86,87 which may enhance their weight loss capacity. As such, one of our primary goals was to identify key factors associated with weight loss success. At one-year post-RYGB, percentage weight loss (PWL) ranged from 17.8 to 46.7% in this cohort of women, representing ∼2.6-fold difference (Fig. 6A).Fig. 6Pre-RYGB skeletal muscle DNA methylation and gene expression predict weight loss success. (A) Waterfall plot showing the variation in percentage weight loss (PWL) at one-year post-RYGB (n = 38). (B) Pearson correlation between age and PWL (n = 38). (C) Pearson correlations between baseline complex I (CI) and CI + CII OXPHOS and PWL (n = 29). (D) Pearson correlations between baseline gene expression of mitochondrial genes and PWL (n = 35). COX17, cytochrome c oxidase copper chaperone; RBFA, ribosome binding factor A; UCP3, uncoupling protein 3; BOK, BCL2 family apoptosis regulator; SLC25A5, adenine nucleotide translocase 2/solute carrier family 25 member 5; REXO2, RNA exonuclease 2. (E) Pearson correlations between baseline gene expression of selected non-mitochondrial genes and PWL (n = 35). SMAD4, SMAD family member 4; MMP2, matrix metalloproteinase-2; SOX6, SRY-box transcription factor 6; SERPINE2, serpin family E member 2; LncRNA LOC105373244; CAMLG, calcium modulating ligand; EHMT2, euchromatic histone lysine methyltransferase 2; JAK2, Janus kinase 2. (F) Pearson correlations between baseline CpG methylation of all genes and PWL. Highlighted are genes with |Pearson Rho| > 0.6 and P ≤ 0.01 (n = 35). Blue symbols, positive correlation; Green symbols, negative correlation. (G) Pearson correlations between baseline CpG methylation of selected genes and PWL (n = 35). Pre-RYGB skeletal muscle DNA methylation and gene expression predict weight loss success. (A) Waterfall plot showing the variation in percentage weight loss (PWL) at one-year post-RYGB (n = 38). (B) Pearson correlation between age and PWL (n = 38). (C) Pearson correlations between baseline complex I (CI) and CI + CII OXPHOS and PWL (n = 29). (D) Pearson correlations between baseline gene expression of mitochondrial genes and PWL (n = 35). COX17, cytochrome c oxidase copper chaperone; RBFA, ribosome binding factor A; UCP3, uncoupling protein 3; BOK, BCL2 family apoptosis regulator; SLC25A5, adenine nucleotide translocase 2/solute carrier family 25 member 5; REXO2, RNA exonuclease 2. (E) Pearson correlations between baseline gene expression of selected non-mitochondrial genes and PWL (n = 35). SMAD4, SMAD family member 4; MMP2, matrix metalloproteinase-2; SOX6, SRY-box transcription factor 6; SERPINE2, serpin family E member 2; LncRNA LOC105373244; CAMLG, calcium modulating ligand; EHMT2, euchromatic histone lysine methyltransferase 2; JAK2, Janus kinase 2. (F) Pearson correlations between baseline CpG methylation of all genes and PWL. Highlighted are genes with |Pearson Rho| > 0.6 and P ≤ 0.01 (n = 35). Blue symbols, positive correlation; Green symbols, negative correlation. (G) Pearson correlations between baseline CpG methylation of selected genes and PWL (n = 35). In line with previous findings,88 younger patients lost more weight compared to older patients (Fig. 6B). Pre-surgical weight, BMI, and body composition did not associate with weight loss capacity (data not shown). Of note, individuals with higher pre-surgery skeletal muscle CI + II OXPHOS respiration tended to lose more weight compared to those with lower CI + II OXPHOS respiration (P < 0.05, Fig. 6C). We next sought to determine whether expression of genes for mitochondrial proteins in skeletal muscle pre-RYGB associate with weight loss outcomes, suggesting possible molecular mechanisms of weight-loss, using PWL values adjusted for age. High pre-RYGB skeletal muscle expression of the complex IV assembly factor, COX17 (P < 0.001), uncoupling protein 3 (UCP3) (P < 0.01), ribosome assembly factor RBFA (P < 0.01), and the mitochondrial folate enzyme ALDH1L2 were moderately correlated with increased weight loss (Fig. 6D, Table S7). In contrast, high baseline expression of the apoptotic factor, BOK, the RNA exonuclease, REXO2, and the adenine nucleotide translocase 2 SLC25A5 were negatively correlated with weight loss post-RYGB (all P < 0.01, Fig. 6D, Table S7). We then examined correlations of genes encoding non-mitochondrial proteins in pre-RYGB skeletal muscle that may explain other cellular mechanisms involved in weight loss outcomes. This identified 37 genes positively correlated with PWL (|Pearson's r| > 0.4 and P < 0.01), including the TGF-β mediator, SMAD4, the extracellular matrix metalloproteinase, MMP2, the cell differentiation factor, SOX6, and the protease inhibitor, SERPINE2 (Fig. 6E, Table S7). In contrast, 44 genes were negatively correlated with PWL, including an X-linked LncRNA LOC105373244, calcium signalling mediator, CAMLG, the histone methyltransferase, EHMT2, and the tyrosine kinase, JAK2 (Fig. 6E, Table S7). These correlations, however, are only nominally significant and do not pass multiple testing correction with FDR < 0.05 and should be considered as exploratory without further validation. Furthermore, as DNA methylation has also been associated with weight loss response to bariatric surgery and other weight loss treatments,67 we also assessed whether CpG methylation was associated with PWL in our cohort. When applying P < 0.01 as a cut-off, 11,171 CpG sites that mapped to 5616 genes were associated with PWL (Table S8). Of the strongly correlated CpG sites (|Pearson's r | > 0.6), INPP5D had multiple CpG sites (Fig. 6F, Table S8). Of note, both pre-RYGB gene expression and pre-RYGB methylation of CpGs associated with MMP2, SOX6, JAK2, RBFA, SERPINE2, and SMAD4 correlated with PWL (Fig. 6G, Table S8). These correlations represent potential reprogramming mechanisms of weight loss but notably do not pass multiple testing correction with FDR < 0.05, and thus only suggest regulatory mechanisms that warrant further validations to determine true associations from these hypotheses. To understand differences in, and identify potential markers of, weight loss outcomes, machine learning classifiers were applied to distinguish individuals according to post-RYGB weight loss by pre-RYGB skeletal muscle gene expression and methylation profiles. Nested cross-validated elastic net logistic regression models were trained to classify individuals in the top and bottom quintiles of weight loss (classified as “high” and “low” PWL groups according to age-adjusted PWL, n = 7 per group) using either pre-RYGB gene expression or CpG methylation features to identify molecular associations with weight loss (independent of known age associations (Fig. 6B)). The nested cross-validation framework avoids overfitting in small sample size settings, allowing for robust classification and identification of relevant features. Using 10 classifier features selected via t-test (determined using training samples for each out cross-validation fold), gene expression but not methylation models could delineate weight loss groups. Methylation models demonstrated virtually no classification accuracy when evaluated on held-out test samples during cross validation (data not shown), whereas gene expression models reached average AUROC = 0.694 (area under receiver operating characteristic curve), AUPRC = 0.677 (area under precision recall curve), F1 score = 0.667, and MCC = 0.289 (Matthew's correlation coefficient). Values of 1 indicate perfect classification for these metrics and 0 (−1 for MCC) is completely incorrect classification. Model feature analysis revealed three genes, EHMT2, Myocilin (MYOC), and Mitotic Arrest Deficient 2 Like 2 (MAD2L2), selected for features in 100% of models (Fig. 7A), suggesting that they are robustly associated markers of weight loss response across our cohort of patients.Fig. 7Pre-RYGB skeletal muscle gene expression of MYOC, EHMT2, and MAD2L2 are associated with weight loss success. (A) Top 20 genes selected across high and low percentage weight loss (PWL) classification model iterations according to t-test and pre-RYGB transcriptomics profiles. Variable importance is the average absolute model coefficient for each gene with error bars as ± 1SD. Dot size and colour indicates the number of occurrences for a gene selected as a model feature (1400 models total). (B) Volcano plot of pre-RYGB DEGs from vastus lateralis in high and low PWL individuals post-RYGB. Highlighted are DEGs with |log2FC| > 0.5 and FDR < 0.05. (C) Pre-RYGB gene expression in high and low PWL individuals for genes identified by weight loss classifier model. (D) Confusion matrix for three gene weight loss classifier model applied to inter-quintile individuals grouped into greater than median PWL (n = 10) and less than or equal to median PWL (n = 11) categories. Predicted groups by the model are defined with a classification threshold of 0.50. Pre-RYGB skeletal muscle gene expression of MYOC, EHMT2, and MAD2L2 are associated with weight loss success. (A) Top 20 genes selected across high and low percentage weight loss (PWL) classification model iterations according to t-test and pre-RYGB transcriptomics profiles. Variable importance is the average absolute model coefficient for each gene with error bars as ± 1SD. Dot size and colour indicates the number of occurrences for a gene selected as a model feature (1400 models total). (B) Volcano plot of pre-RYGB DEGs from vastus lateralis in high and low PWL individuals post-RYGB. Highlighted are DEGs with |log2FC| > 0.5 and FDR < 0.05. (C) Pre-RYGB gene expression in high and low PWL individuals for genes identified by weight loss classifier model. (D) Confusion matrix for three gene weight loss classifier model applied to inter-quintile individuals grouped into greater than median PWL (n = 10) and less than or equal to median PWL (n = 11) categories. Predicted groups by the model are defined with a classification threshold of 0.50. To further investigate these potential markers of weight loss response, transcriptomic analysis for these high and low PWL subgroups revealed 11 differentially expressed genes pre-RYGB (FDR < 0.05, |log2 FC| > 0.5) (Fig. 7B and C and Table S9), including both MYOC and EHMT2 of the genes identified in our classifier models. Repeating the same classification model training and testing framework but using only pre-RYGB gene expression for MYOC, EHMT2, and MAD2L2 identified by feature selection achieves perfect classification of subgroups (i.e., values of 1.0 for all evaluation metrics on held-out samples during cross-validation). Furthermore, both MYOC (r = 0.485, P < 0.01) and EHMT2 (r = −0.565, P < 0.001) were amongst the genes most correlated with PWL across our whole cohort (Table S7). Due to the lack of publicly available transcriptomics data for validation with patient-level weight loss values post-RYGB, we attempted to validate our model with the individuals in our cohort excluded from predictive modelling (i.e., those in the intermediate quintiles for weight loss). These intermediate quintile individuals are also delineated by the three gene model into groups with greater than median age-adjusted PWL (n = 10) and less than or equal to median age-adjusted PWL (n = 11) (Fig. 7D; F1 = 0.556, MCC = 0.270). Together, these findings support that pre-RYGB gene expression of MYOC, EHMT2, and MAD2L2 captures dynamics of post-RYGB weight loss across all cases in our cohort. However, true external validation cohorts will be essential to confirm any associations with weight loss. Discussion Here we demonstrate using integrative multiomic and functional analyses of the skeletal muscle using state of the art RNA sequencing and DNA methylation array platforms that alterations in skeletal muscle development, metabolism, and ribosomal biogenesis contribute to metabolic reprogramming after RYGB-induced weight loss. Unlike most previously studies, these effects are specific to RYGB and not confounded by multiple types of bariatric surgery within our larger cohort of patients. These analyses revealed a broad activation of specific gene programs via DNA methylation related to muscle growth and differentiation in response to RYGB surgery, with 2173 differentially expressed genes in total and 9537 differential methylated regions across the genome potentially contributing to altered expression in 1197 genes. Additionally, alterations in many metabolic gene groups suggest complex metabolic reprogramming of muscle post-RYGB. Relevant to clinical outcomes, we also demonstrate that pre-surgical skeletal muscle oxidative capacity and gene expression of MYOC and EHMT2 may be associated with weight loss success at one-year post-RYGB surgery. Bariatric surgery is highly effective at improving glucose homoeostasis and resolving T2D, particularly in patients who have sufficient pancreatic beta-cell function prior to surgery.3,89 Mechanisms underlying improved glucose homoeostasis include both weight loss-dependent and -independent mechanisms, such as enhanced insulin sensitivity and incretin response (GLP1/GIP), reduced hepatic glucose output, and increases in circulating bile acids.90, 91, 92 In line with previous findings in both permeabilised skeletal muscle fibres and primary myotubes,93, 94, 95 we demonstrate that obesity-induced mitochondrial dysfunction is restored by RYGB surgery and reflects the improvements in glucose homoeostasis. Previous research demonstrates that these RYGB-induced improvements in skeletal muscle mitochondrial function may be related to increases in the proportion of oxidative fibres,96 the improvements in mitochondrial networking,97 and enhanced oxidation of intramuscular lipids,98 including increased oxidation of DAGs and ceramides,69,99 which can impair mitochondrial oxidative capacity.100,101 Indeed, intramuscular lipid content dramatically decreases following bariatric surgery and as expected, correlates with decreases in both body weight and fat mass.102 Importantly, while limited sample availability meant that we could not perform immunohistological analysis of muscle fibre type for this study, previous research has demonstrated that the proportion of oxidative type I fibres is inversely associated with obesity susceptibility and positively with weight loss capacity,14,86,96,103 and the size of types I and II fibres is smaller at 3 and 9 months post-bariatric surgery.103 The dissociation between the decrease in expression of genes encoding mitochondrial proteins and mitochondrial content (e.g., ETC complex protein expression and CS activity) with the increase in mitochondrial function at one year post-RYGB suggests there may not be a fixed stoichiometry between gene and protein expression levels. The assembly and maintenance of the OXPHOS complexes depends on multiple processes, including translation of both mitochondrial and nuclear-encoded mitochondrial proteins, protein import, and quality control.104 These previously published studies are consistent with the idea that enhanced ribosomal abundance contributes to improved mitochondrial oxidative capacity and metabolic processes overall following RYGB, analogous to adaptations observed in response to exercise training.105 Compared with lean controls, patients with obesity have decreased enrichment of ribosome and spliceosome proteins.74 Multi-omics analyses of skeletal muscle obtained from monozygotic twin pairs discordant for BMI demonstrated that pathways related to ribosome biogenesis and protein translation are downregulated in the twin having a higher BMI,106 indicating that reduced ribosome capacity is a consequence of obesity. The finding that gene expression of ribosomal proteins was higher at one-year post-RYGB is consistent with transcriptomic and proteomic analyses previously reported in muscle biopsies collected at 3 months post-RYGB surgery.74 The sustained increase in ribosomal gene expression demonstrates a prolonged reprogramming of skeletal muscle ribosomes. Importantly, alterations in DNA methylation were demonstrated in ∼23% (16/68) of the differentially expressed ribosome genes, indicating that DNA methylation partially contributes to the reprogramming of ribosomes. RYGB-induced hypomethylation of promoter regions of key transcription factors like BCL6 likely contributes to the activation of a transcriptional program that enhances skeletal muscle ribosomal biogenesis and protein translation efficiency.77 Similarly, differential methylation is associated with several differentially expressed genes in skeletal muscle metabolic pathways and may also be regulated by methylation-induced increases in the expression of upstream transcription factors such as ESRRG, and VDR. These observations of RYGB-induced altered methylation in key genes involved in metabolism (PDK4, FTO, PPARGC1A) are consistent with previous observations,68,69 further supporting the idea that DNA methylation contributes to the broad metabolic reprogramming that occurs following RYGB-induced weight loss. However, these and all previous papers reporting on effects of bariatric surgery on muscle DNA methylation have at best used arrays with 200,000 fewer probes, giving us much greater analytical coverage, especially in gene enhancer regions, open chromatin and CTCF-binding domains. Other studies have demonstrated that lipid exposure elicits differential responses in gene expression and DNA methylation in primary human skeletal muscle cells from women with obesity.107 The present finding of altered methylation of genes related to skeletal muscle morphogenesis and muscle differentiation is in line with mechanistic evidence of enhanced expression of myogenic markers in primary muscle cells isolated from patients at 9 months post-RYGB, indicating improved proliferation and cell commitment potential.95 Changes in skeletal muscle DNA methylation following RYGB are not immediate, as relatively few alterations in DNA methylation have been observed in biopsies obtained 2 weeks following RYGB surgery.69 Furthermore, while the deceleration of epigenetic age observed in post-RYGB skeletal muscle is not causative of improved life expectancy, it highlights the improvements in overall health that occur with RYGB-induced weight loss, which include increased life expectancy by up to 3 years.108 Although RYGB results in better weight loss success beyond 10 years compared to most other surgical treatments, interindividual responses to RYGB surgery remain highly variable, and a subset of patients fail to achieve and maintain clinically significant weight loss.7, 8, 9, 10,109 The response to RYGB is highly heritable, as first-degree relatives exhibit similar responses to surgery compared to unrelated individuals.110 Behavioural and psychosocial factors, combined with demographic factors such as sex, ethnicity, and age, are also well-established predictors of weight loss responses to RYGB.111, 112, 113, 114 Variability in RYGB response is also influenced by skeletal muscle differences in skeletal muscle composition. For example, an increased proportion of type I fibres and decreased intramuscular fat infiltration in rectus abdominus associate with greater weight loss following RYGB surgery.96,115 Here, we demonstrate that baseline skeletal muscle mitochondrial metabolism is also associated with RYGB-induced weight loss success. Furthermore, we identify both mitochondrial and non-mitochondrial genes in pre-RYGB skeletal muscle that may correlate with weight loss success at one-year post-surgery. The observed association between skeletal muscle UCP3 expression and RYGB-induced weight loss is consistent with our previous findings in diet-adherent women who have an enhanced capacity for diet-induced weight loss.87 While the mitochondrial inner membrane carrier protein UCP3 is expressed at low levels in skeletal muscle, it can mediate mitochondrial proton leak,116 and UCP3 expression is associated with increased skeletal muscle fatty acid oxidation and improved redox status.117, 118, 119 Therefore, increased UCP3 expression may indicate enhanced mitochondrial bioenergetics including improved fatty acid oxidation capacity and redox, which can facilitate weight loss success and metabolic health. Interestingly, both gene expression and methylation of CpGs associated with SMAD4, MMP2, SOX6, JAK2, RBFA, and SERPINE2 were nominally correlated with weight loss success, indicating that epigenetic mechanisms may contribute to weight loss capacity through the reprogramming of skeletal muscle gene expression. The mitochondrial ribosome binding factor A (RBFA) plays a role in mitoribosome formation of the 28S small subunit.120,121 The positive correlation between weight loss and skeletal muscle RBFA expression may indicate an increase in mitochondrial protein translation and thus, mitochondrial function in patients with a higher weight loss capacity. However, mechanistic studies have demonstrated that RBFA depletion results in growth defects without any apparent effects on mitochondrial protein translation of OXPHOS capacity in both HeLa and HEK293 cells.120 Matrix metalloproteinases (MMPs) play central roles in tissue repair and remodelling, and MMP2 negatively correlates with BMI and waist circumference in individuals with obesity.122 MMP2 is highly expressed in type II fibres and may facilitate differentiation by degrading ECM components.123,124 Bariatric surgery reduces MMP2 levels in serum,125 and thus the observed positive correlation between MMP2 expression and weight loss at one-year post-RYGB may reflect an improved capacity to maintain type II muscle fibres. These potential associations will require further validations to identify any true relationships. The pre-surgical levels of MYOC, EHMT2, and MAD2L2 associate with degree of weight loss one-year post-RYGB, most evident within the extremes of weight loss in our cohort. This apparent effect may be explained by the roles of these genes in cell differentiation and development (EHMT2,126, 127, 128 MAD2L2129,130) and muscle hypertrophy (MYOC131,132), suggesting potential mechanisms through which skeletal muscle responds and adapts to weight loss. However, it is important to note that these findings are limited without validations on similar datasets with individual patient-level weight loss data, which will be essential to confirm any associations as datasets become available. Additionally, mechanistic studies of the roles of these genes in skeletal muscle are needed. Altogether, these findings suggest that DNA (de)methylation influences skeletal muscle gene expression to promote weight loss capacity by enhancing gene expression programs related to maintaining skeletal muscle mass and metabolism. The inclusion of only female patients is a limitation of our study but reflects the majority of the population undergoing bariatric surgery, as females account for ∼80% of bariatric surgery patients.133 Moreover, although our findings are consistent with established biological mechanisms, the cross-sectional design of our study precludes the ability to infer causal relationships between methylation, gene expression, and weight loss. Future studies will be essential to determine whether these epigenetic and transcriptional alterations exert causal regulatory effects on weight loss, potentially through experimental approaches such as targeted methylation editing. In conclusion, the present study both investigates factors driving RYGB-induced metabolic improvements in skeletal muscle and identifies putative biomarkers that may indicate successful weight loss outcomes. RYGB-induced weight loss and improvements in whole-body metabolism were accompanied by increases in myofibre mitochondrial oxidative capacity, which was also observed in primary muscle cells, indicative of sustained metabolic reprogramming at the cellular level. Moreover, RYGB is associated with the reprogramming of genes involved in metabolism and ribosomal biogenesis, partially resulting from alterations in DNA methylation. Altogether, these findings provide molecular insights into the factors, including mitochondrial oxidative capacity, muscle growth and differentiation, and ribosomal biogenesis, that mediate improvements in skeletal muscle metabolism and that influence interindividual variability in weight loss following RYGB surgery. Contributors Conceptualisation: CAP, DM, LSK, GRS, RD, RM, M-EH. Methodology: CAP, DM, LSK, MN, MR, MNK, MCC, TA, GRS, RD, RM, M-EH. Software: LSK, MN, MCC, DW. Investigation: CAP, DM, LSK, LMKH, MN, MR, MNK, DW. Accessed and Verified the Underlying Data: CAP, DM, LSK, LMKH, MN, MCC, DW, M-EH. Resources: MCC, TA, GRS, RD, RM, M-EH. Writing–Original Draft: CAP, DM, LSK, M-EH. Writing–Review & Editing: CAP, DM, LSK, LMKH, MN, MR, MNK, AJ, NK, DW, MCC, TA, GRS, RD, RM, M-EH. Visualisation: CAP, DM, LSK. Supervision: MCC, GRS, RD, RM, M-EH. Funding acquisition: M-EH, GRS. All authors read and approved the final manuscript. Data sharing statement Clinical data used for analysis in this study are held at the Ottawa Hospital and the University of Ottawa. Data may be shared with the use of an institutional data sharing agreement. Further information and requests for reagents and resources should be directed towards the corresponding author, Mary-Ellen Harper (mharper@uottawa.ca). RNA-Seq and methylation data are deposited at the NCBI Gene Expression Omnibus (GEO). All GEO datasets and code developed for DNA methylation, RNA-seq analysis, and bioinformatics analysis are available at https://github.com/lkenn012/bariatric_epigenetics. Original Western blot data are available in the Source Data file. Declaration of interests DW received consulting fees from Aureka. RD received speaker honoraria from Novo Nordisk, and Eli Lilly, and consulting fees from Boehringer Ingelheim. GRS received grants or contracts from Espervita Therapeutics, Merck, Novo Nordisk, Catalym Therapeutics, Poxel Pharmaceuticals, and Nestle; consulting fees from Merck, Alveus, Poxel, Novo Nordisk, Versant Ventures, Curie Bio, Korro Bio, Keros Therapeutics, Nuances Health, BioXcel, Madrigal, and Betagenon; honoraria from Eli Lilly and Novo Nordisk. GRS has patents planned, issued or pending for GDF15 as a biomarker for metformin, an ACLY inhibitor for fibrosis and a 5HT inhibitor for obesity. GRS holds stock or stock options in Espervita Therapeutics. M-EH received speaker honoraria from Boehringer Ingelheim and Novartis. All other authors declare no conflicts of interest.