Multi-omic integrated networks connect DNA methylation and miRNA with skeletal muscle plasticity to chronic exercise in Type 2 diabetic obesity

Abstract

Epigenomic regulation of the transcriptome by DNA methylation and posttranscriptional gene silencing by miRNAs are potential environmental modulators of skeletal muscle plasticity to chronic exercise in healthy and diseased populations. We utilized transcriptome networks to connect exercise-induced differential methylation and miRNA with functional skeletal muscle plasticity. Biopsies of the vastus lateralis were collected from middle-aged Polynesian men and women with morbid obesity (44 kg/m² ± 10) and Type 2 diabetes before and following 16 wk of resistance (n = 9) or endurance training (n = 8). Longitudinal transcriptome, methylome, and microRNA (miRNA) responses were obtained via microarray, filtered by novel effect-size based false discovery rate probe selection preceding bioinformatic interrogation. Metabolic and microvascular transcriptome topology dominated the network landscape following endurance exercise. Lipid and glucose metabolism modules were connected to: microRNA (miR)-29a; promoter region hypomethylation of nuclear receptor factor (NRF1) and fatty acid transporter (SLC27A4), and hypermethylation of fatty acid synthase, and to exon hypomethylation of 6-phosphofructo-2-kinase and Ser/Thr protein kinase. Directional change in the endurance networks was validated by lower intramyocellular lipid, increased capillarity, GLUT4, hexokinase, and mitochondrial enzyme activity and proteome. Resistance training also lowered lipid and increased enzyme activity and caused GLUT4 promoter hypomethylation; however, training was inconsequential to GLUT4, capillarity, and metabolic transcriptome. miR-195 connected to negative regulation of vascular development. To conclude, integrated molecular network modelling revealed differential DNA methylation and miRNA expression changes occur in skeletal muscle in response to chronic exercise training that are most pronounced with endurance training and topographically associated with functional metabolic and microvascular plasticity relevant to diabetes rehabilitation.

morbid obesity and associated metabolic disease and Type 2 diabetes mellitus (T2D) is an increasingly prevalent phenomenon in Westernized countries. A sedentary lifestyle in people with Type 2 diabetes or obesity contributes to skeletal muscle pathophysiology, in which the muscle has lower mitochondrial density (∼30%) and oxidative enzyme activity and is insulin resistant (42, 59, 68). Insulin resistance in the skeletal muscle reduces baseline lipid oxidation rates and increases and intramuscular triacylglycerides (45), lipid metabolites (69), and redox stress (60). Such changes inhibit insulin signaling transduction leading (60), which is mechanistically associated with feedback inhibition of the insulin signal transduction cascade to GLUT4 translocation (83). Other explanations for the impaired muscle glucose observed in obese muscle include pathophysiological changes to the extracellular matrix, such as endomyosial fibrosis (83). Other explanations for impaired muscle glucose handling include extracellular matrix pathology presenting as endomyosial fibrosis, low capillarity, and endothelial cell insulin resistance (21). Moreover, myofibril disorganization, abnormal mitochondria, and low respiratory capacity further impair myofibrillar metabolic substrate handling (42, 43). Fortunately, chronic endurance and resistance exercise training (independently or combined) may improve respiratory capacity and vascular development in T2D skeletal muscle (19, 54) and facilitate improvements in systemic metabolic nutrient disposal (100). However, the global factors regulating exercise training-induced plasticity in obese and insulin resistant (100). However, the global molecular characteristics controlling rehabilitative plasticity in skeletal muscle are largely unknown.

Understanding the molecular regulation of muscle plasticity may help better define the most effective long-term T2D rehabilitation strategies. Recent evidence suggests that methylation of cytosine in CpG dinucleotides of DNA may be involved in transcriptional reprogramming activated by endurance exercise. Cross-sectional evidence suggests that DNA methylation may contribute to metabolic dysfunction in muscle by modifying genes controlling mitochondrial biogenesis and energy homeostasis (9). Dose-dependent changes in mRNA expression and corresponding hypomethylation of gene promoter regions of peroxisome proliferator-activated receptor gamma, coactivator 1α (PGC1α), pyruvate dehydrogenase kinase 4 (PDK4), and peroxisome proliferator-activated receptor delta (PPARδ) have been observed in response to endurance exercise (10). In a familial-linked study, methylation of genes in retinol metabolism and calcium signaling pathways decreased after 6 mo of exercise in T2D adults (65). Retention of defective insulin signaling and metabolism in cultured myotubules derived from T2D (13) suggests epigenetic modifications affect the metabolic memory of skeletal muscle and contribute to the development of insulin resistance (8). Therefore, we hypothesized that chronic exercise training in obese T2D adults will reverse changes in genomic methylation in a direction consistent with attenuation of T2D metabolic pathology in muscle.

Similarly, posttranscriptional mRNA silencing by microRNAs (miRNA, miR) is another potential epigenetic mechanism regulating skeletal muscle plasticity in exercise. MiRNAs are conserved trans-acting nonprotein coding RNA molecules that regulate protein abundance and function by repressing translation of target mRNAs (81). In skeletal muscle, so termed myomiRs (e.g., miR-1, -133a/b, and -206) regulate myogenesis in vitro (30) and mediate processes inherent to exercise adaptation, vascular disease including angiogenesis (102), inflammation (23), and mitochondrial metabolism (16). In vivo, miR-378, -29a, -26a, and -451 were differentially expressed between high and low hypertrophy responders to resistance exercise (22), while 12 wk of endurance training decreased miR-1, -133a, -133b, and -206 in men (64). Gallagher et al. (28) argued that miR-1 and miR-133a may regulate metabolic dysfunction in diabetic muscle. Therefore, our second hypothesis was that longitudinal differential myomiR expression derived from analysis of the miRNA-ome (miRome) would inverse-correlate with functional transcriptome networks altered by exercise; specifically metabolic and angiogenic in response to endurance training, and muscle growth relating to resistance training.

Certain ethnic groups, such as Maori and Pacific Island people, are more vulnerable to T2D (88), making them well positioned to benefit from advances in diabetes rehabilitation intervention. Therefore, the purpose of this paper is to demonstrate a network medicine (7) approach for examining the epigenomic regulation of T2D skeletal muscle affected by 16 wk of endurance or resistance exercise. We hypothesized that by integrating the methylation-affected genomic data and miRNA regulatory elements with the primary bioinformatic-derived functional transcriptome, top-ranking epigenomic regulatory sites governing skeletal muscle plasticity in morbidly obese T2D adults would be identified. With this approach both known and novel differentially methylated genomic regions and regulated miRNAs can be associated with common functional molecular pathways in skeletal muscle and with changes in microvascular and metabolic phenotype markers within the skeletal muscle.

METHODS

Study Design

All methods were reviewed and approved by Polynesian cultural consultants and the Central Regional Ethics Committee, New Zealand (CEN/07/08/054, ACTRN# 12609001085268). Participants were randomly assigned following baseline testing via computer-generated randomization list (http://www.randomization.com), stratified by sex in blocks of four, to receive either resistance training or endurance training for 16 wk. Outcome measures were collected the week prior to and the week following the 16 wk intervention, where postintervention was completed ∼72 h following the final exercise training session to minimize any confounding effect of the acute transcriptome and signaling response to acute exercise (99).

Participants

Full details of participant recruitment, screening, retention are provided elsewhere (88). For the current analysis, 17 (13 female, 5 male) middle-aged (49 yr ± 5) self-identified Maori or Pacific Islanders completed all requirements of the study. All participants had clinical diagnosis of T2D for 3.3 ± 3.3 yr; glycosylated hemoglobin HbA1c was 9.8 ± 2.1% and blood glucose 9.8 mmol/l ± 3.4, and HOMA2-IR index 3.4 ± 2. Visceral obesity was ≥88 cm in women and ≥102 cm in men (class III morbid obesity 43.8 kg/m² ± 9.5), and participants performed no regular exercise during the previous 6 mo.

Interventions

The exercise interventions were described in detail elsewhere (87, 88). Following baseline testing, participants were randomized into endurance or resistance exercise groups comprising supervised progressive-loading exercise sessions 3 ×/wk on nonconsecutive days. The resistance training group (n = 9) performed two or three sets of eight exercises using machine weights (Cybex International, Medway, MA) targeting all major muscle groups for six to eight repetitions to fatigue. The endurance training group (n = 8) performed exercise on a cycle ergometer for 40–60 min (Life Fitness, Schiller Park, IL).

Skeletal Muscle Biopsy

A biopsy was collected from the right vastus lateralis under local anesthesia (1% Xylocaine; Astra Zeneca, Auckland, New Zealand) via a 5 mm Bergstrom needle with applied suction at week 0 and week 17. Muscle samples were immediately cleaned of any connective tissue and fat with pieces transferred untreated into Eppendorf cryotubes and snap-frozen in liquid nitrogen. Another section was oriented longitudinally in TissueTek (Sakura Finetek, Tokyo, Japan) and snap-frozen in liquid nitrogen-cooled isopentane. Samples were then stored at −80°C until analysis.

Analysis Workflow

Nucleic acids and proteins were extracted (detailed below) followed by quantitative DNA methylation and mRNA and miRNA analysis by microarray. The proteome was analyzed by mass spectrometry (MS), glycolytic and mitochondrial enzyme activity by standard enzyme assays and ELISA, tissue capillary and myofiber GLUT4 density by immunohistochemistry, and lipid density from electromyographs. The epigenomic outcomes were integrated within the top-ranked functional transcriptome networks, with protein phenotype and miRNA expression validation indicated in the workflow shown in Fig. 1.

Fig. 1.

Experimental work flow diagram. Shown are the research design and skeletal muscle analysis processes used to construct the integrated molecular networks defining multi-omic regulation of skeletal muscle plasticity in Type 2 diabetes (T2D) rehabilitation.

Multi-omic Analysis

Transcriptome and miRNA profiling.

RNA was isolated from muscle (∼10 mg) using mirVana Kit (Applied Biosystems/Ambion, Austin, TX). Concentrations of RNA were determined by NanoDrop spectrophotometer ND-1000 (NanoDrop Technologies, Wilmington, DE), and quality was assessed with an Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA). A total of 200 ng of RNA used for mRNA and miRNA expression profiling from each sample using Illumina bead array Human-HT12 V4 for mRNA (Illumina, San Diego, CA) and Affymetrix GeneChips 2.0 for miRNA expression (Affymetrix, Santa Clara, CA). For mRNA, reverse transcription (RT), and in vitro transcription amplification incorporating biotin-labeled nucleotides was performed with Illumina TotalPrep -96 RNA Amplification Kit (Ambion). We hybridized 750 ng of the biotin-labeled IVT product (cRNA) to HumanHT-12_v4_BeadChip (Illumina) for 16 h followed by washing, blocking, and streptavidin-Cy3 staining according to the Whole-Genome Gene Expression Direct Hybridization protocol (Illumina). Arrays were scanned with the HiScanSQ System. Decoded images were analyzed by GenomeStudio (Illumina). For miRNA analysis, low-molecular-weight RNA was biotin-labeled with FlashTag Biotin HSR RNA Labeling Kit (Genisphere, Hatfield, PA) and verified (Enzyme Linked Oligosorbent Assay, Genisphere). We hybridized 21.5 μl of high-quality biotin-labeled miRNA sample to Affymetrix Gene-Chip miRNA 2.0 Array for 16 h according to Affymetrix protocol. The arrays were washed and stained on the Affymetrix Fluidics station 400 and scanned with a Hewlett Packard G2500A gene Array Scanner. Affymetrix miRNA QC Tool v 1.1.1.0 was used for the data summarization, normalization, and microarray quality control.

Methylation.

Genomic DNA was extracted from 5–10 mg muscle (Qiagen DNA extraction kit, Germantown, MD) and the dilution (Quant-iTTM PicoGreen dsDNA Reagent, #P7589; Life Technologies, Carlsbad, CA; fluorescence with the Agilent Bioanalyzer) and purity quantified (Nanodrop-1000; Thermofisher, Wilmington, DE). Bisulphite conversion (Illumina) was checked with methylation-specific PCR. We used 4 μl of bisulphite-converted DNA for hybridization on Infinium Human Methylation 450 BeadChip (Illumina). Following washing, single nucleotide extension was performed using hybridized bisulphite-treated DNA as a template. Nucleotides were labeled with biotin (ddCTP and ddGTP) and 2,4-dinitrophenol (DNP) (ddATP and ddTTP). After single base extension, repeated rounds of staining were performed with a combination of antibodies that differentiated DNP and biotin by fixing with different fluorophores. The bead chip was washed prior to scanning (Illumina HiScan SQ). Image intensity was extracted using Genome Studio Methylation module 1.8.5, which also conducted quality control for coverage (fraction of CpGs with detectable intensity values above background) and bisulphite conversion efficiency. The platform assayed the methylation status of >450,000 CpG sites covering all designated RefSeq genes, including promoter, 5′- and 3′-regions, and also CpG islands and shores, CpG sites outside of CpG islands, non-CpG methylated sites, CpG islands outside of coding regions, miRNA promoter regions, and disease-associated regions identified through genome-wide association studies.

miRNA expression real-time QRT-PCR.

miRNA expression magnitude of the top network connected miRNA was determined by RT TaqMan miRNA assay (Life Technologies, Applied Biosystems, Carlsbad, CA) according to manufacturer's protocols on both endogenous control (RNU-48) and miRNAs (endurance training: miR-29a, miR-132, miR-221; resistance training: miR-23a, miR-1207-5p, miR-195). The primer sequences for each miRNA (miRBaseID) were: miR-29a, UAGCACCAUCUGAAAUCGGUUA (hsa-miR-29a-3p); miR-132, UAACAGUCUACAGCCAUGGUCG (hsa-miR-132-3p); miR-221, AGCUACAUUGUCUGCUGGGUUUC (hsa-miR-221-3p); miR-23a, AUCACAUUGCCAGGGAUUUCC (miR-23a-3p); miR-1207, UGGCAGGGAGGCUGGGAGGGG (miR-1207-5p); miR-195, UAGCAGCACAGAAAUAUUGGC (hsa-miR195); RNU48, GATGACCCCAGGTAACTCTGAGTGTGTCGCTGATGCCATCACCGCAGCGCTCTGACC. All samples were run in triplicate on 384-well clear plates using a Life Technologies 7900HT Taqman system. Relative expression was determined by the comparative Ct method (52).

Proteomics.

We cut 50 serial 10 μm muscle biopsy sections from two representative male and two representative female subjects per exercise group by cryostat (Leica Microsystem, Nussloch, Germany), transferred them to a microcentrifuge tube, and suspended them in 100 μl of chilled RIPA buffer (product #89900; Thermo Scientific, Rockford, IL) with complete-mini EDTA free protease inhibitor (Roche #11836170001, Auckland, NZ). The sample was homogenized (Ultraturrax; IKA, Wilmington, NC) for 15 s on ice, spun at 2,000 rpm for 3 min, rehomogenized for 15 s, and then spun 10 min at 13,000 rpm. The supernatant was transferred to a fresh microcentrifuge tube. Protein concentration was determined in each samples using BCA assay (product #23225, Thermo Scientific). Protein aliquots (50 μg) from each sample was dissolved in 4× LDS running buffer and 10× reducing agent and loaded into 4–12% Bis-Tris gel system for separation (Invitrogen, Carlsbad, CA). Gels were fixed for 30 min with a mixture of methanol-water-acetic acid (45:45:10, vol/vol/vol) and stained with Biosafe Coomassie for 1 h at room temperature and destained overnight with water at 4°C. Individual protein bands were cut and digested overnight with 12.5 ng/μl Trypsin (#V5280; Promega, Fitchburg, WI), followed by peptide extraction.

Peptide fractions were injected via an autosampler (6 μl) into a Symmetry C18 trap column for 10 min at 10 μl/min, 100% A, and separated by a C18 reverse-phase column at a flow rate of 250 nl/min by nano-HPLC (Eksigent, Dublin, CA). The mobile phases consisted of water with 0.1% formic acid and 90% acetonitrile. A 65 min linear gradient from 5 to 60% acetonitrile was employed. Eluted peptides were introduced into the mass spectrometer via a 20 μm inner diameter, 10 μm silica tip (New Objective, Ringoes, NJ) adapted to a nano-electrospray source (Thermo Scientific). The spray voltage was set at 1.4 kV and the heated capillary at 200°C. The LTQ (Thermo Scientific) was operated in data-dependent mode with dynamic exclusion in which one cycle of experiments consisted of a full-MS (300–2,000 m/z) survey scan and five subsequent MS/MS scans of the most intense peaks using collision-induced dissociation with the collision gas (helium) and normalized collision energy value set at 35%.

Output files were searched to identify proteins using the Sequest algorithm (Bioworks Browser 3.3.1, ThermoFisher Scientific) against human Uniprot database indexed for fully tryptic peptides, two missed cleavages, and potential modification of oxidized methionine (15.9949 Da). Peptide mass tolerance was set at ±1.5 Da for parent ion and ±1 Da for fragment ions. Search results were loaded into ProteoIQ (NuSep, Bogart, GA) and filtered: XCorr > 1.9, > 2.5 and > 3.5 for singly, doubly, and triply charged ions, respectively, peptides > 6 amino acids, minimum of 2 spectral count per protein, 0.98 peptide probability, and 0.95 protein probability. Scan counts were normalized based on total sample intensity. Protein fold change values were calculated by using the ratio of scan counts for each sample vs. its paired control. Protein annotations were acquired using the Keyword and GO terms in the Uniprot Protein Knowledgebase (http://www.uniprot.org/).

Measures of Tissue Capillary Density and Metabolic Substrate Handling

Immunohistochemistry.

Serial 10 μm transverse muscle biopsy sections were cut by cryostat (Leica Microsystem) and mounted on standard microscope slides treated with Vectabond Reagent (Vector Laboratories, Burlingame, CA). Sections were dried at room temperature for 30 min followed by a 15 min rinse in phosphate buffered saline to remove OCT residue. Slides were treated with 100 μl of 1% bovine serum albumin, incubated for 30 min, and then rinsed with PBS before application of primary antibodies. For each subject, two slides were incubated for 4 h with 100 μl Abcam rabbit anti-GLUT4 antibody (1:1,000) (AB654; Sapphire Bioscience, Waterloo, Australia) and two slides with 100 μl rabbit anti-von Willebrand factor antibody (1:2,000) (AB7356; Chemicon, Billerica, MA). Negative controls were treated with 100 μl of 1% BSA only. After being rinsed with PBS, all slides were treated for 4 h with 100 μl of AlexaFluor 594-conjugated anti-rabbit (1:250) secondary antibodies (Invitrogen). Slides were rinsed and coverslipped with Vecta-Shield anti-fade mounting medium (Vector Laboratories). Sections were examined using a compound fluorescence microscope (Olympus BX-50; Olympus, Tokyo, Japan) with image capture (SPOT-RT Slider cooled CCD camera; Diagnostic Instruments, Sterling Heights, MI). Image quantification was performed on eight-bit TIFF images in software (Adobe Photoshop CS4; Adobe Systems, San Jose, CA). GLUT4 and capillary density were quantified as number of background-adjusted reactive pixels per muscle fiber.

Biochemistry.

Tissue protein homogenate derived from ∼15–25 mg of wet muscle was used to determine citrate synthase (CS), beta-hydroxy acyl-CoA-dehydrogenase (BHAD), and cytochrome c oxidase (COX) enzyme activity. Protein extraction was performed in a biocontainment hood on thawed muscle using 1:25 ratio of muscle to extraction buffer (10 mM HEPES pH 7.4, 70 mM sucrose, 1 mM EDTA, 220 mM mannitol) with 0.3% Brij-35 and complete-mini EDTA free protease inhibitor (cat. #11836170001; Roche, Auckland, NZ) added immediately before homogenization with an IKA ultra-turrax with S10N-5G dispersing element (Rose Scientific, Edmonton, Alberta, Canada). Muscle was blended for 20 s repeating three times with 10 s pauses between blends. The homogenate was then vortexed for 1 h at 4°C and then further homogenized by passing through a clean 25-gauge needle using a 1 ml syringe 20 times. Homogenate was then centrifuged 600 g for 15 min at 4°C and removal of supernatant. To extract residual activity discovered in assay development, 200 μl of extraction buffer was then added to pellets and passed through a clean 25-gauge needle, followed by removal of supernatant and addition to the first supernatant aspirate extraction. Protein concentration of supernatant was measured using the BCA Protein assay (Merck, Darmstadt, Germany). Assay conditions were adopted from spectrophotometric assays described elsewhere for CS (63), COX (29), and BHAD activity (44) and adapted for 96-well microplate reader (Benchmark Plus; Bio-Rad, Hercules, CA). Samples were run in triplicate, and the intra-assay coefficients of variability (CVs) were 3.1, 2.1, and 2.0%, respectively, and interassay CVs were 3.1, 1.4, 3.7%, respectively. Hexokinase (HK) activity was assayed as described previously (80), and glycogen synthase kinase-3 beta (GSK-3β) protein content was detected and quantified using GSK-3β ELISA kit (Invitrogen).

Electron microscopy.

Intramyocellular lipid (IMCL) density and inspection of mitochondrial morphology were estimated by direct visualization from electron microscopy using an adaptation of the validated method for transmission electron microscope (TEM) (89). We fixed 10–25 mg of muscle in half strength Karnovsky's fixative (0.1 M cacodylate buffer, 2% paraformaldehyde, 2% glutaraldehyde, 1 mM calcium chloride, 20 mM sucrose). Tissue was dehydrated in ethanol, embedded in Epon type resin, and then sectioned (70 nm). Sections were viewed at ×6,500 using a TEM (Philips CM100, Eindhoven, Netherlands). Eight to 15 images/sample were obtained under light-standardized conditions using a film picture camera (Kodak 4489, Rochester, NY) from two randomly selected fibers, with ⅓ of images from the subsarcolemmal region near the nucleus, ⅓ from the subsarcolemmal region away from the nucleus, and ⅓ in the middle of the fiber. Plates were digitized as 1,200 dpi Tiff images.

IMCL density was determined in Photoshop CS5.5 and Fovea Pro (Reindeer Graphics, Asheville, NC). Image areas were cropped to exclude nonmyocellular space. Image pixels were standardized to 1 μm grid reference image taken per sample batch. Lipid droplets were identified to the software by manual selection and quantified by filter feature function, with threshold filtering to exclude artefacts. IMCL features were recoded black. The nonlipid background was then removed from the image (converted to white space) followed by manual deletion of any nonlipid features, permitting subsequent quantification of IMCL droplet number and IMCL area relative to total image area. Outcomes were reported as the muscle fraction (μm²) occupied by IMCL droplets, which yields values similar to grid-point counting estimates (89).

Statistical Analyses

Histology, immunohistochemistry, enzyme activity, physical function.

A mixed-model ANOVA (SAS 9.1; SAS, Cary, NC) with subject as the random effect was used to determine the within-group effect of exercise on outcomes. A sex fixed effect was not considered due to insufficient sample size. Intramyocellular lipid, GLUT4, and capillary density data were log-transformed prior to analysis to account for heteroscadasticity (Table 1). The standardized within-subject baseline score was included as a covariate. Inference to effect size was by standardized difference (i.e., Cohen's d) and clinical likelihood thresholds derived from the confidence interval (36).

Table 1.

Statistical summary for the effect of endurance and resistance training on intramyocellular lipid, mitochondrial, and glucose-handling associated enzyme activity, GLUT4, and capillary density

Parameter	Endurance		Resistance
	Standardized Differencea ±95% CLb	Inferencec	Standardized Differencea ±95% CL	Inference
IMCL density	−0.9 ± 0.5	moderate‡	−0.7 ± 0.5	moderate†
Capillary density	1.1 ± 0.9	moderate†	−0.1 ± 3.5	unclear
Mitochondrial enzyme activity
BHAD	2.1 ± 1.3	v. large‡	4.6 ± 2.1	ex. large‡
CS	0.4 ± 1.0	unclear	1.7 ± 1.2	large†
COX	0.9 ± 1.1	moderate*	3.1 ± 1.6	v. large‡
Glucose handling
HK	2.2 ± 1.4	v. large‡	0.6 ± 1.5	unclear
HK/CS	1.8 ± 1.2	large‡	0.3 ± 1.5	unclear
GSK-3B	0.1 ± 0.8	unclear	−0.5 ± 0.8	unclear
GLUT4	2.1 ± 1.3	v. large†	−0.0 ± 0.8	unclear

^aData are the within-group standardized effect (Cohen's d effect size) of exercise training.

^b±95% confidence limits (CL): add and subtract this number to the mean to obtain the 95% CL for the true difference.

^cEffect size (standardized difference) qualifiers: 0–0.2 trivial, 0.2–0.6 small, 0.6–1.2 moderate (mod.), 1.2–2.0 large, 2.0–4.0 very (v.) large, >4.0 extremely (ex.) large.

Threshold for the probability of a standardized effect that is > +0.2 (increase) or < −0.2 (decrease): 75–95% likely (*), 95-99.5% very likely (†), >99.5% most certain (‡). An effect is unclear if its confidence interval overlaps a standardized difference of both +0.2 and −0.2 (36). IMCL, intramyocellular lipid.

Microarray.

Signal values for mRNA data were generated in GenomeStudio Gene Expression Module (Illumina, San Diego, CA), background adjusted, and quantile normalized, with quality control according to Illumina guidelines. Unsupervised clustering was used for outlier discovery. We excluded lowest 10% of the probes with the expression values that were below or close to background. No technical outliers were detected in the datasets, but one outlier was detected in the methylation dataset by principle components analysis for endurance training and excluded. Affymetric CEL files for miRNA interrogation were normalized using robust multichip analysis-detection above background in Expression Console (Affymetrix). miRNA probes were included only if called present (P < 0.05) in at least 10% of the samples. Probes with the lowest expression values (lowest 10%) were also excluded. Methylation data were normalized using the subset-quantile within array normalization (SWAN) for Illumina Infinium HumanMethylation450 BeadChips (55) method of normalization and was processed as described in the minfi package in Bioconductor (http://www.bioconductor.org/packages/release/bioc/html/minfi.html). All average log2 transformed microarray signal data were analyzed by paired t-test (i.e., within-subject contrasts) (Partek 6.6; Partek, St. Louis, MO).

Inflation of error arising from low sample size and multiple testing was managed by application of a novel procedure for post hoc selection of microarray probes. The procedure is based on uncertainty in the standardized magnitude of a change (36, 84) rather than approaches based on null-hypothesis testing (11, 94). The training change (post- minus pre-exercise) (11, 94). The post-pre exercise training change in microarray probe signal was standardized by dividing the mean log-transformed signal by the standard deviation of the log-transformed baseline value. The resulting standardized change, the degrees of freedom, and the corresponding P value obtained from a paired t-test were entered in an adapted spreadsheet (35) to calculate the standardized false discovery rejection probability (sFDR). The sFDR is the probability that the change in the probe was either not substantially positive (smallest Cohen d effect size: <0.20 standardized units) for probes showing a substantial increase or not substantially negative (>−0.20) for probes showing a substantial decrease following chronic exercise training (see Data Supplement 1, Endurance_mRNA worksheet); probes selected where the mean effect was trivial were omitted.¹ Assuming independent effects, we selected probes to ensure an overall false discovery rate (FDR) of <5% as follows: probes were rank-ordered by ascending sFDR probability; starting with the probe with lowest probability, probes were then selected sequentially until including another probe would result in the sum of their probabilities exceeding 5%. [For independent effects with false-positive probabilities P₁, P₂, P₃…, the cumulative probability of at least one false positive is 1 − (1 − P₁)(1 − P₂)(1 − P₃)…, which for low P values is almost exactly P₁ + P₂ + P₃…, and hence the selection of probes based on keeping this sum <5%.]

Bioinformatics.

Genes downstream of methylated loci differentially affected by exercise within the gene body, 5′- or 3′-untranslated region, transcription start site, or 1st exon were selected for bioinformatics interrogation using Ingenuity Pathway Analysis (IPA) software (Feb. 2013, Ingenuity Systems, http://www.ingenuity.com). Methylation probe alignments were determined from the UCSC Genome Browser database (assembly Feb. 2009, GRCh37/hg19; http://genome.ucsc.edu/cgi-bin/hgGateway); probes with undefined genomic locations or inside and within 10 bp of single nucleotide polymorphisms adjacent to CpG loci were removed from the analysis. Core analyses filtered for species were also run on the transcriptome probe selections. The resulting networks, molecular and physiological function annotations, and upstream regulators were ranked by enrichment score. The top-ranked networks containing biological functions common to both the methylome and the transcriptome were then selected for subsequent analysis using the comparisons analysis utility in IPA. The resulting top-ranked functions common to both the methylome and transcriptome were utilized following the principles of network medicine (7) to construct the primary epigenomic-associated integrated networks defining the underlying molecular biology responding to exercise. Integration of miRNA within the transcriptome networks was obtained from regulatory binding activity analysis in IPA (miRNA Target filter). This analysis provides literature validated and predicted associations from three databases: Targetscan (http://www.targetscan.org), mirecords (http://c1.accurascience.com/miRecords/), and mirbase (http://www.mirbase.org). The subsequent inverse-correlated miRNA associations were connected to the top-ranked functional transcriptome network modules.

RESULTS

Multi-omic Data Description

The experimental raw microarray data were deposited in the Gene Expression Omnibus (GSE58250). The microarray probe selections and gene abbreviations, the overrepresented molecular functions, and the top-ranked IPA categories are in Data Supplement 1, 2, and 3, respectively. The top-ranked miRNA and proteome are presented in Tables 2 and 3, respectively. Discovery genomic-site methylation characteristics are summarized (Fig. 2, A and C), and the differential methylation clustering is visualized by heat maps (Fig. 2, B and D). The top-ranked methylome and transcriptome pathways were first organized by biological function and then connected with the computed miRNA binding targets to create integrated multi-omic networks representative of skeletal muscle response to endurance and resistance training (Fig. 3).

Table 2.

Discovered miRNA and the inverse expression binding mRNA target detected as altered in response to 16 wk of endurance or resistance training in skeletal muscle of morbidly obese patients with Type 2 diabetes

miRNAa (IPA node; seed sequence)	Affymetrix Array Standardized Difference ±99% CLb	Standardized Difference ±99% CLb of miRNA Expression via Taqman qPCR ±95% CLb	miRNA Expression Confirmation via Literaturec	Inverse Expression Predicted Binding Target mRNAd	Gene Functions of Top Ranked Target mRNAd
Endurance Training
let-7e (let-7a-5p; GAGGUAG)	1.3 ± 1.2		(25)	ANKRD28, BCL7A, GOLT1B, RBBP4, STK40, SEC14L5	DNA methylation and transcriptional repression signaling (RBBP4); NF-kappaB cascade (GOLT1B)
miR-132 (miR-132-3p; AACAGUC)	2.2 ± 2.2	0.7 ± 0.6		MYO18B, NBN, PGM5, PDE7A, SREBF1	deposition of triglyceride, lipid metabolism (SREBF1); DNA stability (NBN); glucose and glucose-1-phosphate degradation (PGM5); signalling (PDE7A)
miR-221 (miR-221b-3p; GCUACAU)	2.3 ± 1.7		(6, 15)	MYOZ2, RREB1	transcription regulation (RREB1); calcineurin signalling (MYOZ2)
miR-4312 (miR-4312; GCCUUGU)	−0.8 ± 0.7			UQCRB	respiration (UQCRB)
miR-548c-5p (miR-548 h-5p; AAAGUAA)	2.1 ± 1.1			FLRT3, G3BP1
miR-664* (miR-4794; CUGGCUA)	0.5 ± 0.5			C17orf101, SIK2, SORBS1, TMEM108, WDPCP	glucose transport, cellular response to insulin, focal adhesion assembly, muscle contraction, actin cytoskeletal binding (SORBS1); oxidoreductase activity (C17orf101); kinase activity, insulin receptor signalling pathway (SIK2)
miR-29a (miR-29b-3p; AGCACCA)	−1.4 ± 1.2	−1.6 ± 0.5	(32, 104)	ATP5G1, ANTXR2, CAV2, COL4A1, FOXO3, IFITM3, OSBPL3, NKTR, PHACTR2, SLC39A9	formation of basement membrane (ANTXR2, COL4A1); proliferation of endothelial cells (CAV2, COL4A1, FOXO3); proliferation of satellite cells, immune complex-mediated inflammation (FOXO3)
Resistance Training
miR-1207-5p (miR-1207-5p; GGCAGGG)	−1.0 ± 0.9	−1.1 ± 0.8		AIF1L, C9orf16, CAPN8, COL4A2, ECM1, FSCN1, RBM3	development of blood vessel (COL4A2, ECM1); gene expression (COL4A2, ECM1)
miR-195 (miR-16-5p; AGCAGCA)	0.69 ± 0.67	1.2 ± 1.2		CRYZL1, JPH1, MAP2K4, RAD23B, RAF1, USP9X, ZFHX3	gene expression (MAP2K4, ZFHX3); force generation of skeletal muscle (JPH1); development of blood vessel (RAD23B, RAF1)
miR-193b (miR-193a-3p; ACUGGCC)	−0.87 ± 0.69			PDE1A, SH3RF3
miR-23a (miR-23a-3p; UCACAUU)	1.7 ± 1.7		(78)	GREM1, HMGB2, MIS18A, PALD1, PKIA, RAD23B, TMEM101, ZFHX3	development of blood vessel (HMGB2, RAD23B); gene expression (HMGB2, PKIA, GREM1); organization of collagen fibrils (GREM1); cell cycle (ZFHX3)
miR-3178 (miR-3178; GGGCGCG)	0.73 ± -.73			CPEB1, SLC5A5, SPPL2B	gene expression (CPEB1)
miR-483-5p (miR-483-5p; AGACGGG)	2.2 ± 2.1		(50)	KIAA1191, RAD23B	development of blood vessel (RAD23B)
miR-487 (miR-487b-3p; AUCGUAC)	0.69 ± 0.89			MAP2K4	gene expression (MAP2K4)

^amiR IPA miRNA group node name. miR relationships from inverse-expression binding target mRNA were experimentally observed as recorded by mirbase, mirecords, or had high predicted binding likelihood in Targetscan.

^b99% CL.

^cmiRNA expression confirmed in mammalian skeletal muscle via literature search; Empty cell denotes no available literature.

^dGene functions identified by the IPA curated knowledge database. Gene abbreviations in Data Supplement 1.

Table 3.

Proteins detected as altered in response to 16 wk of endurance or resistance training in skeletal muscle of morbidly obese patients with Type 2 diabetes

Endurance Training	Δ		Δ	Resistance Training	Δ
Cell Membrane		Mitochondria		Cell Membrane
Cadherin-13	⇑	10 kDa heat shock protein	⇑	band 3 anion transport protein	⇓
Integrin beta-1	⇑	2,4-dienoyl-CoA reductase	⇑	Ca2+/calmodulin-dependent kinase type II SU α	⇓
L-xylulose reductase	⇑	28S ribosomal protein S36	⇑	Nucleus	⇓
Moesin	⇓	ATP synthase SU b, d, f, g	⇑	probable C->U-editing enzyme APOBEC-2	⇓
Polymerase I and transcript release factor	⇓	ATP synthase SU gamma	⇑	Cytoplasmic
Protein ptase 1 regulatory SU 3A	⇓	citrate synthase	⇑	Cytoskeleton
Radixin	⇓	coiled-coil-helix domain-containing protein 3	⇑	myosin-9	⇑
Sodium/potassium-transporting ATPase SU α-1	⇓	cytochrome b-c1 complex SU 7	⇑	myosin-Ic	⇑
Nucleus		cytochrome b-c1 complex SU Rieske	⇑	troponin I	⇓
Exportin-2	⇑	cytochrome c	⇑	Cytosol
GTP-binding nuclear protein Ran	⇑	cytochrome c oxidase SU 2, 5A, 5B, and 6C	⇑	ankyrin repeat domain-containing protein 2	⇑
Heterogeneous nuclear ribonucleoprotein K	⇓	delta(3,5)-Delta(2,4)-dienoyl-CoA isomerase	⇑	glutathione S-transferase omega-1	⇓
Matrin-3	⇓	electron transfer flavoprotein SU alpha	⇑	creatine kinase M-type	⇓
Nucleosome assembly protein 1-like 4	⇓	ES1 protein homolog	⇑	hemoglobin SU delta	⇓
Ser/thr-protein ptase 2B catalytic SU α	⇓	glutathione peroxidase 1	⇑	phosphoglycerate kinase 2	⇓
Ser/thr-protein ptase 2B catalytic SU β	⇓	GTP:AMP phosphotransferase	⇑	alpha-enolase	⇓
UV excision repair protein RAD23 homolog B	⇓	heat shock protein 75 kDa	⇑	gamma-enolase	⇓
Cytoplasmic		mitochondrial 2-oxoglutarate/malate carrier	⇑	hemoglobin SU alpha	⇓
Cytoskeleton		NADH dhrase 1 α subc. SU 12, 2, 5	⇑	Lysosome	⇓
Ankyrin repeat domain-containing protein 2	⇑	NADH dhrase 1 β subc. SU 10, 8	⇑	Ras-related protein Rab-7a	⇓
Keratin	⇑	NADH dhrase flavoprotein 2	⇑	Mitochondria
Kinesin-1 heavy chain	⇑	NADH dhrase iron-sulfur protein 3, 4, 7, 8	⇑	10 kDa heat shock protein	⇑
Myosin-14	⇑		⇑	3-hydroxyisobutyrate dhrase	⇑
Myosin-9	⇑	prohibitin	⇑	3-ketoacyl-CoA thiolase	⇑
Programmed cell death 6-interacting protein	⇑	pyruvate dhrase E1 component SU α	⇑	ATP synthase SU e	⇑
Dynactin SU 1	⇓	short-chain specific acyl-CoA dhrase	⇑	cytochrome c oxidase SU 5A	⇑
FH1/FH2 domain-containing protein 1	⇓	succinate dhrase iron-sulfur SU	⇓	dihydrolipoyllysine-residue acetyltransferase, PDH	⇑
Gelsolin	⇓	succinyl-CoA ligase [ADP-forming] SU beta	⇓	cytochrome b-c1 complex SU 1	⇓
Keratin	⇓	superoxide dismutase [Cu-Zn]	⇓	isocitrate dhrase [NADP]	⇓
Myosin light chain kinase 2	⇓	voltage-dependent anion-selective channel prot. 2		medium-chain specific acyl-CoA dhrase	⇓
Myosin-binding protein C	⇓	3-ketoacyl-CoA thiolase	⇓	PDH E1 component SU α	⇓
Synaptopodin	⇓	adenylate kinase 2	⇓	short-chain specific acyl-CoA dhrase	⇓
Synaptopodin 2-like protein	⇓	complement component 1 Q subcomponent-bp		Proteosome	⇓
Cytosol		mitochondrial inner membrane protein		proteasome SU alpha type-7	⇓
14 kDa phosphohistidine ptase	⇑	trifunctional enzyme SU beta	⇑	Scaroplasmic reticulum
14-3-3 protein beta/alpha	⇑	Proteosome	⇑	cytosolic 5′-nucleotidase 3	⇑
14-3-3 protein zeta/delta	⇑	proteasome SU alpha type-3	⇓	alpha-1-antitrypsin	⇓
Calpain small SU 1	⇑	ubiquitin carboxyl-terminal hydrolase isozyme L3	⇓	calnexin	⇓
Cytosolic 5′-nucleotidase 3	⇑	proteasome SU alpha type-1, type-6		reticulon-4	⇓
Fructose-bisphosphate aldolase C	⇑	ubiquitin carboxyl-terminal hydrolase 14	⇑	Cytoplasmic - other
Glycogen (starch) synthase	⇑	40S ribosomal protein S3		aldo-keto reductase family 1 member C2	⇑
L-lactate dhrase B chain	⇑	Sarcoplasmic reticulum	⇑	cytoplasmic aconitate hydratase	⇑
Microtubule-associated protein tau	⇑	acylphosphatase-2	⇑	dukaryotic translation initiation factor 3 SU C	⇑
PreB-cell leukemia TF-interacting protein	⇑	transmembrane emp24 domain-containing prot.10	⇑	glutathione S-transferase Mu 1	⇑
SH3 domain-binding glutamic acid-rich protein	⇑	calnexin	⇓	peptidyl-prolyl cis-trans isomerase A	⇑
Adenine phosphoribosyltransferase	⇓	calreticulin	⇓	E3 ubiquitin-protein ligase RNF123	⇓
Eukaryotic translation initiation factor 3 SU A	⇓	protein niban	⇓	low mol. wt. phosphotyrosine protein ptase	⇓
Fatty acid-binding protein	⇓	reticulon-4	⇓	myomesin-3	⇓
General vesicular transport factor p115	⇓	Cytoplasmic - other		transgelin	⇓
Glycerol-3-phosphate dhrase [NAD+]	⇓	cystatin-B	⇑	Obg-like ATPase 1	⇓
Glycogenin-1	⇓	cytoplasmic aconitate hydratase	⇑	Extracellular space	⇓
Hemoglobin SU gamma-1	⇓	glutaredoxin-1	⇑	apolipoprotein A-I	⇓
NADP-dependent malic enzyme	⇓	heat shock protein beta-3	⇑	Unknown	⇓
Phosphoglycerate kinase 2	⇓	peptidyl-prolyl cis-trans isomerase A	⇑	leucine-rich repeat-containing protein 20	⇓
Rab GDP dissociation inhibitor alpha	⇓	(protein ADP-ribosylarginine) hydrolase-like prot. 1	⇓
Rab GDP dissociation inhibitor beta	⇓	5-oxoprolinase	⇓
Extracellular Matrix	⇓	beta-taxilin	⇓
Galectin-1	⇓	ribonuclease inhibitor	⇓
Golgi		Unknown
Quinone oxidoreductase	⇑	putative triosephosphate isomerase-like protein	⇑
Ras-related protein Rab-1B	⇑	T-complex protein 1 SU eta	⇓

Results are from analysis of the proteome in 4 participants (2 men, 2 women) for each of the 2 exercise modes. Inclusion criteria were for the post/pre change in protein content in all 4 participants to return a change of at least 2-peptide increase or decrease from baseline. bp, Binding protein; dhdrase, dehydrogenase; D, delta score for post/pre training effect; ptase, phosphatase; prot., protein; SU, subunit; subc., subcomplex; TF, transcription factor; ⇑, increase; ⇓, decrease.

Fig. 2.

Effect of endurance and resistance training on change in global DNA methylation. Shown is the frequency of hypomethylation and hypermethylation according to specified genomic region (UCSC Genome Browser database) for the standardized false discovery rejection probability (sFDR) probe selection following 16 wk of endurance training (n = 6) (A) and resistance training (n = 9) (C) in morbidly obese Polynesians with T2D. Heat maps and hierarchical clustering for DNA methylation relating to the sFDR selection before (Pre) and 16-wk following (Post) endurance (B) and resistance training (D). Maps were constructed using Partek software (Partek 6.6, Partek, St. Louis, MO). Each column represents a skeletal muscle sample from an individual, and each row represents the magnitude of the standardized M value intensity for a single probe. The x-axis signal intensity is the values are the range of M value intensities. Individual relatively hypermethylated and hyporegulated genomic sites are indicated by red and green signals, respectively. The values to the side of the clustering are the Euclidean distance representing the strength of M value signal relationship between probes. Note, clear separation in 4 of the 7 participants after endurance training (B); the hierarchical outliers, participant 026, 019, and 060 completed 52, 45, and 98% (group mean 76%) of the prescribed exercise sessions, respectively.

Fig. 3.

Multiomic integrated epigenomic networks associated with skeletal muscle plasticity in response to 16 wk' endurance (cluster A) or resistance (cluster B) exercise training in morbidly obese Polynesians with T2D. The global network topology was built from the top-ranked (enrichment score) molecular and physiological functions between the methylome and the transcriptome (colored boxes at top). Highlighted are intersecting and overlapping network modules containing the top-ranked functionally common transcripts and methylated genomic regions differentially affected by exercise mode. Diamond shapes represent genomic regions containing both CpG and non-CpG enriched methylated sites, upstream or within the specified genomic region (UCSC Genome Browser database) that were differentially methylated by exercise (methylation site is located in Data Supplement 1). Circle shapes are mRNA transcripts. Triangles are miRNA inverse-expression correlated validated (solid line connections) and predicted (dotted line connections) miRNA-mRNA target relationships derived from IPA miRNA Target Filter analysis. The miRNA confirmed by qRT-PCR analysis are surrounded by thick bold blue parameter lines. Changes in expression magnitude are expressed as standardized difference indicated by the color legend bars to the right. The IPA functional category Cardiovascular Development defines capillary microvascular development in skeletal muscle. Abbreviations for the 279 network gene regions were excluded from the figure capture due to space but can be found in Data Supplement 1, tab: Gene Abbreviations for Fig. 3.

Changes in DNA Methylation

Both resistance and endurance exercise training induced hypomethylation. Post- relative to pretraining median (P25, P75) β-values were: −0.021 (−0.046, 0.007) for endurance and −0.019 (−0.052, 0.006) for resistance training. In response to endurance training the most overrepresented functional gene networks (Data Supplement 2) and categories (Data Supplement 3) represented by differential methylation patterns were related to lipid metabolism, carbohydrate metabolism, metabolic disease, cell death and survival, cardiovascular system development and function, and hematological system development and function. In contrast, the highest ranked functional networks and categories responding to resistance training were cellular assembly and organization, cellular development, tissue morphology, and cardiovascular system development and function. Together these data suggest that endurance and resistant training induce epigenetic changes in different molecular pathways.

miRNA Regulation

We discovered 25 and 23 miRNA probes differentially expressed in response to endurance and resistance training, respectively. Seven probes from each returned inverse-expression correlationships with the transcriptome. Analysis of the miRNA-mRNA predicted binding associations within the networks suggested the myomiRs regulate genes involved in transcriptional regulation, lipid and glucose metabolism, myofibril and connective tissue development with endurance training, and gene expression and blood vessel development with resistance training (Table 2). Top network-connected miRNA analyzed by RT-PCR were exposed to sFDR to limit false discovery probability to <5%. The resulting confirmed selections for endurance training were miR-29a and miR-132, and for resistance training miR-1207-5p and miR-195 (Fig. 3, Table 2). Nonconfirmed were miR-221 (standardized difference ± 95% CL 0.85 ± 1.01, P = 0.16), miR-23a (−0.82 ± 1.19, P = 0.14). In addition, the IPA upstream regulator analysis confirmed miR-29-3p (common miR nodal seed sequence with miR-29a; Z-score −3.7, P value 2.0E-10) and miR-16-5p (common miR nodal seed sequence with miR-195; Z-scores 2.4 for directional consistency; P values for transcript overrepresentation 0.02) and as a predicted regulators of the endurance and resistance exercise transcriptome, respectively.

Transcriptome Integrated Epigenetic Functional Networks

We used the IPA network and function analysis tools to determine the most overrepresented constituent gene categories within the transcriptome and methylated genomic regions to address the question of what functional epigenomic relationships provide the basis for exercise specificity of molecular plasticity in T2D skeletal muscle. The network analysis revealed common categorical molecular and physiological functions between the methylome and transcriptome within exercise mode, although no gene-specific association between differential methylation state and differential mRNA regulation was present. The most overrepresented categories were also similar between endurance and resistance training (Fig. 3, Data Supplement 3), but the constituent gene and gene functions were largely unique between exercise modes, suggesting that endurance and resistance exercise triggers skeletal-muscle tissue remodeling via distinctive molecular systems.

Network Associations With Protein Phenotype Imply Increased Capacity for Skeletal Muscle Respiration, Lipid, and Glucose Turnover After Endurance Training

We provide the first description of differentially methylated genomic regions and miRNA within the skeletal muscle of adults with T2D in response to 16 wk of exercise training. The analysis generated evidence to support our hypothesis that altered DNA methylation and miRNA expression are operational in metabolic plasticity of muscle in response to endurance training. The evidence was that the metabolic module: 1) was central to the integrated network topology dominating mRNA expression (66% of network mRNA) (Fig. 3, Data Supplement 3); 2) was connected to epigenetic regulation by an overrepresented metabolic methylome and downregulated miR-29a (Fig. 3, Table 2); and 3) included upregulated mRNA encoding for mitochondrial ATPase transcripts, which was confirmed within the qualitative proteomic analysis (Table 3) and by increased BHAD and mitochondrial COX activity (Fig. 4, A and B; Table 1). Notable differences that were observed between endurance and resistance trained subjects at the protein level included increased CS, cytochrome c, short chain-specific acyl-CoA dehydrogenase, mitochondrial 2-oxoglutarate/malate carrier and GTP:AMP phosphotransferase in endurance relative to resistance training and decreased E3 ubiquitin-protein ligase RNF123 in the resistance relative to endurance trained group (Table 3).

Fig. 4.

Effect of 16 wk of endurance or resistance exercise training on capillary density and metabolic adaptation in the skeletal muscle of Polynesian adults with morbid obesity and T2D. A: representative within-subject transmission electron micrographs of the tissue before and following training; noteworthy are intramyocellular lipid vesicles, apparent increase in glycogen granulation, improved sarcomere structure, and resolution of abnormal mitochondria following training in both exercise groups. Group means and SD for lipid density, glycolytic and mitochondrial enzymes (B), and for capillary density and GLUT4 (D). The statistical likelihood (certainty) of a standardized effect that is >+0.2 (increase) or −0.2 (decrease) is noted with asterisk (*) next to the relevant post-pre change: 75–95% likely (*), 95–99.5% very likely (**), >99.5% most certain (***). An unclear effect has no asterisk (34). Representative within-subject immunohistochemistry for the effect of exercise training on capillary density (C) and GLUT4 (E); to improve quality of figure presentation, brightness of the GLUT4 label within images was enhanced by factor of 2.

Activation of metabolic molecular networks was also associated with the IMCL-lowering phenotype, but in both exercise modes. Upstream regulatory analysis was headed with the insulin receptor and leptin with mechanistic network connectivity to insulin receptor substrate 2, sirtuin 1, and signal transducer and activator of transcription 3 (Fig. 5, A and B). Additionally, other known regulators of metabolic adaptation in skeletal muscle: insulin-like growth factor 1 receptor, peroxisome proliferator activator receptor-α (PPARα), phosphoprotein p53, and estrogen receptor-related-α returned evidence for activation (directional Z-score range 1.7 to 2.3; P value range for overrepresentation 4.7E-03 to 1.1E-04).

Fig. 5.

Predicted activity status of regulatory upstream mechanistic networks in response to 16 wk of endurance training comprising insulin receptor (INSR; Z-scores for directional consistency 3.0, P value for transcript overlap 5.2E-06) (A), leptin (LEP; Z-score 2.4, P value 8.0E-03) (B), and interferon-γ (IFNG; Z-score 2.9, P value 4.3E-02) networks (C), and in response to resistance training, transforming growth factor beta 1 (TGFB1; Z-score 2.0, P value 5.6E-03) (D). Other noteworthy highly ranked transcriptional regulators include mitogen-activated protein kinase 1 (MAPK1; Z-score 2.8, P value 1.3E-03) and interleukin-1β (IL1β; Z-score −1.2, P value 3.64E-03). The mechanistic networks were resolved from maps obtained from the IPA database via the unbiased analysis of the respective transcriptome mRNA selections. The activated transcriptome is shown in E for INSR, LEP, PPARA, which supports altered regulation of lipid metabolism, and in F for TGFB1, which may control extracellular matrix and cytoskeletal remodeling, and myogenesis (14); clearly some expression responses may be causal, some may be purely correlative, while others may represent compensatory events. Solid lines represent direct regulatory relationships, and dashed lines indirect regulatory relationships identified by IPA knowledge database. Analyses and graphs from IPA software.

Promoter region hypomethylation of nuclear respiratory factor 1 (NRF1), hypermethylation of fatty-acid synthase (FASN) (Fig. 3) were consistent with activation of a metabolic and physiological lipid lowering phenotype (4, 49). Hypomethylation within the transcription start site of fatty-acid transporter (SLC27A4) and in the gene bodies of cytochrome P450 (CYP26C1) may be functional [involved catabolism of 9-cis and all-trans isomers of retinoic acid implicit in the response to both resistance and endurance training (70)]. Meanwhile, hypomethylation in the gene bodies of 6-phosphofructo-2-kinase (PFKFB3), histone deacetylase (HDAC4), and the gene encoding the multifunctional Ser/Thr protein kinase (GSK3A), may regulate glycolytic flux (Fig. 3). Methylation site alterations cross-associated with very-large increases in the enzyme actively of HK and in total skeletal muscle GLUT4 content, following endurance training only (Fig. 4, D and E; Table 1). By association, up-regulated mitochondrial enzyme activity, HK and GLUT4, and the relative (vs resistance training) predominance of differentially regulated cytosolic glycolytic regulatory units and upregulated mitochondrial peptide content within the endurance-trained proteome all validated the prevailing metabolic network landscape (Table 3).

Resistance training, lowered IMCL density, and increased mitochondrial oxidative and respiratory enzyme activity (Fig. 4, A and B; Table 1), but unlike endurance training, directional change in mitochondrial protein peptide profile was not conclusive (Table 3). Nonetheless, hypomethylation of the upstream CPG island of SLC2A4 (GLUT4) and the genomic regions of several key cellular components responsible for fatty-acid transport and metabolism [Acyl-CoA synthetase long-chain 1, ACSL1; low density lipoprotein receptor-related protein 1 and 10 LRP1, LRP10; solute carrier family 27 (fatty acid transporter), member 1, SLC27A1] suggest possible involvement of DNA methylation in metabolic plasticity of skeletal muscle triggered by resistance exercise.

Epigenomic Network-specific Microvascular Plasticity

Extracellular matrix and vascular remodeling is characteristic of skeletal muscle adaptation to exercise (92). Such adaptations likely contribute to increased tissue perfusion in response to training likely contributing to increased tissue perfusion in T2D (54). Endurance training increased capillary density, but resistance training provided no effect (Fig. 4, C and D; Table 2). The capillarity phenotype was corroborated qualitatively and quantitatively within the epigenomic-transcriptome landscape: cardiovascular system development and function occupied ⅓ of the network topology (Fig. 3), and by overrepresentation of gene functions involved in blood vessel development and vasculogenesis (P value for overrepresentation 3.5E-03 to 7.4E-04) following both resistance and endurance training that was directionally consistent with capillarity phenotype change (i.e., increased with endurance, decreased with resistance; Data Supplement 3). Changes in the expression of several mRNAs consistent with vascular-remodeling functions included insulin-like growth factor (IGF2) (76), cluster of differentiation 34 (CD34) (93), mitogen-activated protein kinase (MAP2K3) (56), and vascular endothelial growth factor B (VEGFB) (67). Responding to endurance exercise, 12 genomic regions were hypomethylated under morphology of cardiovascular system (P value for overrepresentation 9.8E-03) (Fig. 3). Downregulated miR-29a also connected to the vascular development transcriptome network (Fig. 3). In response to resistance training, upregulation of miR-23a and miR-195 had inverse-expression binding association with genes involved in blood vessel development (Table 2): gremlin 1 (GREM1) and high mobility group box 2 (HMGB2), and v-raf-1 murine leukemia viral oncogene homolog 1 (RAF1), respectively; while RAD23 protein (RAD23B) expression was also a target for miR-483-5p (Fig. 3). The remaining associations between methylation and miRNAs with vascular genes were limited.

Alterations in skeletal muscle tissue inflammatory status is connected with remodeling in health and disease (14, 91). Evidence for a decreased inflammatory transcriptome after endurance training was provided by decreased activation of gene functions causing immune cell trafficking (Data Supplement 3). Activity of an inflammatory-cytokine responsive transcriptome was corroborated with predicted activation of IFNG and reduced cellular response to stress indicated by inhibition of MAPK1 (Fig. 5C). Meanwhile, resistance training only activated a TGFB1 network (Fig. 5D). There was little other evidence for exercise-induced alteration in the state of the inflammatory transcriptome after resistance exercise, but hypomethylated sites were noted in 5 of 6 gene body regions associated with the gene function category chemotaxis of phagocytes (directional Z-score −2.2; P value range for overrepresentation 3.0E-02) (Data Supplement 3). Finally, cellular assembly and organization, which contained the overrepresented function organization of the cytoskeleton was the top-ranked overrepresented function induced by resistance training (directional Z-score 3.0; P value range for overrepresentation 1.2E-02).

DISCUSSION

We constructed integrated epigenomic-transcriptome networks from multi-omic microarray analysis of skeletal muscle tissue from morbidly obese Polynesian adults with T2D in response to 16 wk of endurance or resistance exercise training. From our analysis we were able to define: 1) a set of functional genetic categories common to both the transcriptome and methylome network landscapes that were overrepresented irrespective of exercise training mode, 2) the quantitative characteristics (up- or down-regulated, hypo- or hyper-methylated) and miRNA associations of the functional methyl-transcriptome networks unique to exercise mode, 3) the exercise-mode specific regulated epigenomic-regulated transcriptome landscape validated against change in the tissue metabolic and microvascular phenotype. Through observation of differential multi-omic changes in response to endurance and resistant training, our data demonstrates the utility of using multi-omic network approach to understand and target therapeutic approaches for T2D. Our findings warrant the future investigation of multi-omic screening to uncover novel molecular networks in T2D. Such techniques may be utilized in the future to accelerate translational diabetes rehabilitation research.

Epigenomic Features Connect With Metabolic and Microvascular Plasticity in Response to Endurance Training

The current network analysis revealed that several epigenetic modifications were connected to molecular reprogramming of lipid and glucose handling networks in obese T2D skeletal muscle in response to chronic endurance training. Hypomethylation of the promoter region for NRF1 was integrated with network changes in the metabolic-mitochondrial transcriptome and enzyme activity. NRF1 activates the expression of several key genes regulating cell growth, mitochondrial respiratory proteins, mitochondrial DNA transcription and replication (4), and GLUT4 (74). PGC1α controls NRF1 transcription (79). Indeed, the very large increase on PPARGC1A mRNA expression was congruent with the NRF1 methylation response, and was noteworthy considering the previous training session 72 h prior to biopsy sample suggests that basal metabolic-mitochondrial remodeling occurred in response to the endurance training program. PGC1α is a node regulator of mitochondrial biogenesis, also activating the transcription factors the peroxisome proliferator activator receptors (PPARγ, PPARα), and the estrogen-related receptor alpha (ERRα) co-activator of PPARγ (26), evidenced in the regular analysis (Fig. 5). Barres et al. (8) also investigated epigenetic regulation of metabolic gene expression in skeletal muscle after bariatric surgery in obese women. Reduced clinical metabolic dysfunction phenotype was associated with promoter region hypomethylation of PPARGC1A and PDK4 back to control levels observed in the normal-weight, healthy subjects. Collectively, these findings warrant additional investigation to determine if epigenetic changes in NRF1 promoter and other metabolic genes related to PGC1α-regulated mitochondrial biogenesis occurs in response to endurance training. In addition, future work on the impact of exercise training duration and intensity, targeted nutrition or drug interventions on methylation events regulating skeletal muscle mitochondrial function and other diseases associated with low muscle metabolic function maybe rewarding in developing highly targeted therapeutic strategies.

Low mitochondrial content and basal fatty-acid oxidation is thought to contribute to dysregulated glucose and lipid metabolism in obese and type 2 diabetes (45, 59, 68, 75). Accumulation of intracellular lipids and lipid metabolites (e.g., ceramides, diacylglycerol) and reactive oxygen species are thought to inhibit insulin signaling and thereby decrease glucose transport (3, 20, 33). In this study, evidence for increased capacity to manage oxidative stress with endurance, but not resistance training, was provided from higher endogenous anti-oxidant protein content (glutathione peroxidase 1). Such changes may suggest a greater efficiency of mitochondrial function in response to endurance, and to a lesser extent resistance training. More comprehensive was evidence for the molecular reorganization of skeletal muscle glucose handling with endurance training was provided from observations of wide-spread changes to the metabolic-mitochondrial network landscape and the very large increases in HK activity and GLUT4 content. Activation of the transcriptome downstream of the insulin receptor pathway suggests insulin signaling through phosphatidylinositol-3 kinase (PI3K) pathway was increased in skeletal muscle tissue in response to endurance training, which could promote glucose uptake via increased GLUT4 translocation and expression in the myofiber (24). Activation of the same pathway in endothelial cells could also improve skeletal-muscle capillary recruitment via an eNOS mechanism, leading to improved tissue perfusion and access of glucose and insulin to myofibers (61). In addition, to change in the protein content of several enzymes central to cytosolic glucose metabolism (e.g., glycogen synthase, lactate dehydrogenase, glycogenin-1, phosphoglycerate kinase; Table 3), we discovered hypomethylation at a CpG shelf within the intronic region of the gene body of 6-phosphofructo-2-kinase/fructose-2,6-biphosphatase isoform 3 (PFKFB3) and within an exon region of glycogen synthase kinase 3α (GSKA). These enzymes determine the glycolytic rate via the biosynthesis and degradation of fructose 2,6-bisphosphate and the rate of glycogen synthesis, respectively, and therefore contribute toward muscle capacity for glucose uptake, storage, and utilization. However, gene body methylation may be neutral or even facilitate gene expression (40). Further research is required to determine the magnitude of functional relevance of these endurance-training induced PFKFB3 and GSKA hypomethylation events in skeletal muscle glucose metabolism.

Changes in intracellular lipids in response to endurance training may have also positively affected networks related to glucose homeostasis. We observed hypermethylation of the promoter region of FASN following endurance training. Such changes may contribute to the observed decrease in IMCL by lower FASN activity leading to reduced fatty-acid synthesis and lipid accumulation. In addition, we observed decreased mRNA expression of the transcription factor sterol regulatory element-1 (SREBP1c), which regulates FASN activity in insulin-resistant skeletal muscle in humans (31) and a reduction in fatty-acid binding protein (Table 3). Together these changes in intramuscular lipid stores in T2D may reflect an increased skeletal muscle glucose handling (58).

Accumulating evidence points to a role for miRNAs in posttranslational regulation of gene expression in skeletal muscle response to exercise (22, 64, 102). Our analysis linked miR-29a (and miR-29b-3p by seed sequence homology) with skeletal muscle metabolic and microvascular plasticity in response to endurance exercise in T2D. Previously, miR-29b-3p regulated myofibril development and connective tissue remodeling (103). Upregulation of miR-29a/b/c in diabetic rat skeletal muscle induced insulin resistance in vitro in adipocytes (32), while miR-29a expression in skeletal muscle responds to exercise training (22). In the current study, miR-29a inverse expression-binding mRNA targets regulated endothelial and satellite cell proliferation and immune complex-mediated inflammation (Table 2). Caveolin 2 (CAV2) was one of three upregulated targets connected to miR-29 and is involved in lipid metabolism, cellular growth, assembly, and organization, and apoptosis. CAV1 mRNA was also upregulated, and both caveolins are involved in GLUT4 translocation (66). Connectivity of miR-29a to genes involved in microvascular function, FOXO3 (72) and COL4A1 (86), was also observed. Transcription factors Fox03 and FoxO1 specifically regulate a set of angiogenesis and vascular remodeling-related genes and downregulate endothelial nitric oxide synthase (eNOS) expression (72). The latter was supported by enrichment of the eNOS canonical pathway with endurance training (not shown). Nitric oxide (NO) control of capillary recruitment (of blood flow) is downstream of insulin signaling (61). Abnormalities in vascular NO production are thought to contribute to atherosclerosis and hypertension (61). Therefore, regulation of the transcriptome downstream of miR-29a suggests a novel epigenetic mechanism controlling therapeutically functional vascular plasticity in skeletal muscle to endurance training that is worthy of further investigation in animal antagomiR models and in pathways associated with therapeutic interventions for vascular disease.

Low capillary density, fibrosis, and microvascular dysfunction are emerging mechanisms in the etiology of insulin resistance in skeletal muscle. Poor perfusion reduces nutrient delivery to the limiting metabolic tissue capacity contributing to systemic dyslipidemia and hyperglycemia (12, 21, 61). We have no measure of capillary function; however, the substantial cardiovascular system development network modules, directional gene expression in favor of vasculogenesis, and increased capillary density suggest that endurance training relative to resistance training led to increased perfusion surface area, capacity for nutrient delivery, and lower peripheral resistance. Further evidence for exercise mode-specific functional microvascular plasticity was evident from reduced systolic and diastolic pressures (post-pre training −16 and −5 mmHg, respectively) with endurance exercise, whereas there was no reduction in blood pressure after resistance training (88). In contrast, the vascular remodeling transcriptome responding to resistance training contained more directionally downregulated mRNA, which may account for arrested blood vessel development.

Microvascular plasticity may also involve functional remodeling of the extracellular matrix central to endurance training adaptation (92). Chronic inflammation is linked to deposition of collagen fibers leading to endomysium thickening (fibrosis), which was associated with insulin resistance in obesity and T2D (12, 21). Altered immune-cell trafficking may also be a component of skeletal muscle regeneration from trauma (91) and adaptive remodeling (71). The endurance exercise-altered transcriptome provided some evidence consistent with reduced accumulation of leukocytes, suggesting a moderation of inflammatory pathology. The predicted interferon-γ (INFG) activation suggested a transcriptome favoring active regeneration and remodeling over fibrosis. INFG is expressed by macrophages, T cells, natural killer cells, and myoblasts following acute injury, and knockdown impairs muscle healing associated with impaired macrophage function and the development of fibrosis (18). Inhibition of MAPK1 (ERK2) and the obesity-associated inflammatory cytokine IL1β (85) was also connected to the INFG mechanisms network (Fig. 5). MAPK signaling pathways are stimulated by exercise, promoting improvements in fuel homeostasis and growth and differentiation (77). However, chronic activation of MAPK pathways are also implicated in diabetes (98), and diabetes alters the contraction-mediated MAPK-signaling response (41). Therefore, a reduced resting MAPK (ERK2) transcriptome suggests lower cell stress and cell death following chronic endurance training in T2D skeletal muscle. Unlike in other resistance training studies (90), a MAPK1-activated transcriptome was not seen in the current muscle after resistance training, although the sample time 72 h past last exercise may have been responsible for the absence.

Altered Phenotype but Unremarkable Metabolic Networks With Resistance Training

The extensive metabolic molecular reprogramming evident in response to endurance training was conspicuously absent in the integrated networks and proteome responding to resistance training; this inconsistency was despite large increases in oxidative enzyme activity and lowered IMCL accumulation. The integrative data, therefore, suggest that endurance exercise training promotes quantitatively and qualitatively a more robust stimuli of the two training modes for metabolic-mitochondrial plasticity in the vastus lateralis of morbidly obese T2D individuals. Nevertheless, hypomethylation at TS200 for SLC2A4 suggests epigenetic regulation of GLUT4 expression may be operational in response to resistance training and could help to explain, in part, other reports of improved glycemic control [decreased HbA1c (5, 47)] and insulin sensitivity following chronic resistance training in T2D (5, 34, 39). However, in this study neither skeletal muscle HK activity nor GLUT4 content increased with resistance training. No change in the content of the glucose handling protein and the absence of extensive metabolic molecular reprogramming could be explained simply by the many-fold more contractions initiated by endurance vs. resistance training. Prolonged repetitive contractile activity is necessary for substantial metabolic-mitochondrial adaptations and GLUT4 gene expression through higher activity of NRF1-MEF2 (74), AMPK (57), PGC1α (73), and other signaling activity (53) leading to the beginning of the reversal of dysregulated skeletal muscle glucose handling imposed by many years of inadequate physical activity and positive energy balance.

Increased cytoskeletal plasticity was implicated as the top-ranked transcriptome function responding to resistance training. Recently, alterations in skeletal muscle proteins (e.g., actinin-2, desmin, proteasomes, and chaperones) have been observed in insulin-resistant muscle that may be involved in structure, function, and mechanosignal transduction leading to reduced mitochondrial content and ensuring lipid accumulation and insulin resistance (21). Therefore, we examined the multi-omic dataset to see if altered cytoskeletal plasticity, proteasome, or chaperone expression, or protein content, could account for the increased oxidative enzyme activity and reduced lipid content in the resistance-trained muscle. None of α-actinin-2, cytosolic protein chaperones, intermediate filaments (desmin) proteins associated with insulin resistance in cross-sectional association studies (38), were seen to be altered leaving an explanation for altered lipid and enzyme activity unresolved.

Inflammation With Resistance Training

Finally, a TGFB1-regulated transcriptome associated with extracellular matrix components and remodeling factors was activated in response to resistance training (Fig. 5). The weightlifting included an eccentric exercise component, which is a potent inducer of muscle remodeling and has caused TGFB1 pathways in skeletal muscle remodeling to become an increasing focus of translational research in muscle disease, including approaches to therapy. TGFB1 pathways appear required for normal muscle regeneration, yet constitutive activation are associated with pathological fibrosis and tissue failure (as is the case with many nonmuscle tissues as well) (62). For example, recent data suggest that genetic polymorphisms in TGFB1 pathways modulate the severity of muscular dystrophy, and blocking the TGFB1 pathway improves muscle disease phenotypes (27, 37). The weightlifting included eccentric exercise causing trauma, inflammation, and a connective-tissue protein gene expression impulse normally part of successful regeneration and remodeling (91). However, with the permanent low-grade inflammation of obesity and T2D, chronically elevated TGFB1 may negatively affect skeletal muscle regeneration by inhibiting satellite cell proliferation, myofiber fusion, and expression of some muscle-specific genes (2). Furthermore, TGFβ has been demonstrated to promote transformation of myogenic cells into fibrotic cells after myofiber injury (51, 82).

Within the integrated molecular model of resistance training presented here, the inflammatory response may have been mitigated by downregulation of target mRNA Gremlin (GREM1) (Fig. 3B). GREM1 is a proinflammatory bone morphogenic protein antagonist expressed by endothelial cells during tissue remodeling, and its downregulation would be expected to exert a proinflammatory effect through release of GREM1 inhibition of TGFB1 signaling. Furthermore, the bioinformatic analysis revealed there was no directional evidence to support a change in inflammatory status following resistance training. Therefore, the available evidence suggests the current resistance training in the morbidly obese diabetic Polynesian cohort may have complemented the inflammatory pathology and activated a transcriptome program of minimal functional benefit to metabolism and extracellular matrix and perfusion remodeling.

Innovation and Limitations

Heterogeneous skeletal muscle cells types.

The current analysis from biopsy material provides tissue-level insight to epigenomic-regulated plasticity derived from the heterogeneous cellular composition of skeletal muscle. MiRNA activity, mRNA expression patterns, and DNA methylation is known to be skeletal-muscle-cell type specific (1, 17, 101). On the other hand, some miRNAs demonstrate paracrine (mircrine) action (95), raising the validity of functional network connectivity for mechanistic inference. Nevertheless, cell-type specificity should be considered in future work validating epigenetic-regulated muscle plasticity.

Managing inflation of error.

In this study we applied a novel procedure for post hoc microarray probe selection based on standardized treatment effect size and limiting the probability of false discovery to 5%. Our motivation evolved from prior familiarity of magnitude-based inference in clinical and exercise science and from the ease and improved sensitivity [⅓ lower sample size requirement (36)] for constructing molecular network-protein phenotype alignment within the low sample-size experiment. In our initial analytic runs, we failed to draw any biologically sensible discovery using a range of null hypothesis-based FDR procedures (11), where in some cases fewer than two probes were selected in the mRNA datasets. Nor could we draw meaningful inference from arbitrary fold-change methods; these initial runs failed to select large numbers of probes with low-level expression magnitude but revealing high signal-to-noise ratio, a problem noted elsewhere (94). To illustrate by example from the current response to endurance exercise: the post-pre fold-change for PPARGC1A was 1.2, too low a level of expression for consideration by many molecular biologists. However, the standardized difference (3.7) was qualified as very large. Validation by network and tissue phenotype analysis and prior knowledge of the central role of PGC1α in metabolic adaptation to endurance exercise (59) supports the utility of the magnitude-based selection procedure.

Our approach amounts to defining discovery on the basis of magnitude of change relative to sampling uncertainty; it can also be considered as a way of selecting genes with the highest signal-to-noise ratios of the change scores for transcript expression or methylation-site intensity. The total false discovery probability was the straightforward calculated sum of the false-discovery probabilities of the traditional smallest substantial Cohen d effect size of 0.2× between-subject SD. We used the same conservative independence assumptions as other researchers have used in various methods to constrain the FDR to null (11), the only difference being our false-discovery probability was estimated for the smallest substantial effect, not null; this approach recognizes biological triviality, which is undefined in all null-hypothesis testing approaches. A limitation with our and other FDR approaches is that a degree of interdependence must exist between the genes, which will result in an increase in the failed-discovery rate. This limitation could be overcome by using a sequential bootstrapping procedure to estimate the overall FDR, but larger sample sizes than in the present study are required for trustworthy bootstrapping.

The current selection threshold based on standardized difference was similar to the approach taken by Tusher et al. (94) in the popular Significance Analysis of Microarrays (SAM) procedure. In SAM, the relative difference [d(i)] for a single probe is derived from the change in experimental intensity divided by the composite SD for the repeated measures plus a constant to attempt to adjust for independence. Probe selection is subsequently based on a user-defined fold-ratio of d(i)/d(e), where d(e) is the expected relative difference. The FDR is estimated from the number of genes that exceed the fold-ratio cut-off compared with the distribution of d(i) following random permutation of the sample. For each d(i), a proportion of all genes in the permutation set (control set) will be found to be discovered by chance, and this parameter is then used to calculate the FDR. The SAM procedure also had greater sensitivity than null hypothesis-based FDR procedures (94) but can be subject to user bias within the arbitrary selection of the d(i)/d(e) fold-change ratio leading to substantial differences in post hoc analysis interpretations (48). Similarly, choosing different standardization thresholds and sFDR percentage may also subject the current approach to different post hoc biological interpretations, but defining the extent and nature is beyond the intention and scope of the current manuscript.

Network functional resolution.

Our analysis resolved common systems-level network topology between the methylome and the transcriptome but did not determine any instances of gene-specific site methylation inversely correlated with mRNA expression. The phenomenon of low negative correlations between directional methylation and mRNA expression is not uncommon in Illumina 450K analyses characterized by high intersubject variability in DNA methylation (46). Indeed, hierarchical clustering analysis revealed higher intersubject variability, but also a relationship between endurance exercise training and the global magnitude of differential methylation (Fig. 2B). The ability of the network approach to resolve functional (categorical) molecular associations with phenotype, however, demonstrates a sensitivity advantage of systems-based approaches in low-sample-size studies in challenging cohorts over reductionist models. Certainly, improved statistical precision and therefore analysis resolution may have been possible with a larger sample size or single-sex sample, but this was not possible in the unique free-living grade III obese cohort we were able to access. Muscle sampling temporally closer to completion of exercise may also have improved site methylation-transcriptome alignment. Barrès et al. (10) reported dynamic functional non-CpG dynamic methylation 3 h after endurance exercise. Other technical issues around inference from 450k methylation arrays and other bisulphite-related procedures is the inability to distinguish between 5mC and 5-hydroxmethylcytosine, which has recently been found in DNA. Therefore, any candidate regulatory methylation sites require validation with functional cellular epigenetic-regulated phenotype models (40).

Finally, integrated network construction utilized evidential support provided by the IPA operator-curated knowledge base. A downside of the knowledge base, however, is gene function may be obtained from cell culture, animal or nontissue-specific relationships, meaning inference maybe disparate in skeletal muscle biology requiring validation in cell culture or animal models. Other limitations to biological inference include the assumption of linear response interaction between components, and statistical assumptions, such as normal distribution and the independence of variables (7), which is suspect, but challenging to resolve analytically (e.g., bootstrapping and sample sizes in within-subject designs of n > 20). As the existing interactome knowledge base improves, so will the precision of the network approach to inference and discovery. Additionally, advances in software to manage nonnormal distribution and modeling approaches integrating known biological relationships (96, 97) may advance the precision of inference possible from integrative systems network modelling.

Conclusion

Integrated molecular system network models suggests that the early phase of skeletal muscle plasticity to chronic endurance and resistance training in morbidly obese T2D adults induces exercise-mode specific changes, promoter or gene body methylation, and miRNA connectivity that align with directional alterations in the metabolic and microvascular transcriptome and functional protein phenotype. Extensive metabolic molecular reprogramming was evident in response to endurance training. However, despite some evidence for improved skeletal muscle lipid turnover, evidence for extensive molecular reprogramming of metabolic plasticity was conspicuously absent in the integrated networks responding to resistance training. Therefore, endurance training appeared to be the superior exercise mode driving plasticity for glucose and lipid handling and microvascular remodeling.

Modular integrated network construction based on standardized magnitude-based dataset selections helped to consolidate the multi-omic complexity and enabled disease and functional modules to be connected quantitatively and qualitatively. This alignment has provided new information on skeletal muscle plasticity responding to endurance or resistance training associated with metabolic diseased tissue. We put forward the multi-omic network approach as an accessible and adaptable approach to discovery and robust new mechanistic hypothesis generation in molecular systems biology in chronic disease rehabilitation.

GRANTS

Funding from National Center for Medical Rehabilitation Research Grant 2R24HD050846-06, Massey University Research Fund, Wellington Medical Research Foundation Inc.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author(s).

AUTHOR CONTRIBUTIONS

Author contributions: D.S.R., R.A.P., W.R.S., B.S.C., B.B., J.M.D., J.K., and E.A.H. conception and design of research; D.S.R., W.R.S., I.H., B.S.C., I.L., and M.J.L. performed experiments; D.S.R., W.R.S., M.G., S.D.G., I.H., P.W.S., S.J.W., Y.H., K.B., R.M., F.E.O.-S., J.M.D., B.L., and E.A.H. analyzed data; D.S.R., W.R.S., M.G., I.H., P.W.S., Y.H., K.B., R.M., F.E.O.-S., J.M.D., G.M., and E.A.H. interpreted results of experiments; D.S.R., P.W.S., and S.J.W. prepared figures; D.S.R., S.D.G., K.B., G.M., and W.G.H. drafted manuscript; D.S.R., R.A.P., I.H., B.S.C., P.W.S., B.B., J.M.D., G.M., and W.G.H. edited and revised manuscript; D.S.R., R.A.P., W.R.S., M.G., S.D.G., I.H., B.S.C., I.L., M.J.L., P.W.S., S.J.W., B.B., Y.H., K.B., R.M., F.E.O.-S., J.M.D., B.L., G.M., J.K., W.G.H., and E.A.H. approved final version of manuscript.