Participants: Recruitment, analyses, and ethics
Cohorts from two studies were used in this paper as outlined in Fig. 1. The Oslo cohorts of PM women were recruited consecutively through an outpatient medical clinic. The women gave their verbal and written consent as described previously [12]. These cohorts represented healthy women (n = 18) or women with established primary osteoporosis (n = 17), (age, 55–80 years) without cardiovascular, endocrine, or neurological diseases. The osteoporotic women received anti-resorption medication (bisphosphonates), which was withdrawn 3 months prior to starting this study. The number of previous smokers and the level of physical activity were similar between the groups, as were other lifestyle factors and nutrition as previously described [12]. The serum and urine biomarkers of all participants were normal. Clinical evaluation: All participants underwent a clinical examination and completed detailed interview questionnaires on present and previous diseases, nutrition, and lifestyle factors (smoking, alcohol, physical activity). All participants received daily supplements of vitamin D3 (1000 IU) and calcium (1000 mg). The Ball State University (BSU) cohorts have been extensively described previously [9] but are summarized below and in Table 1. BSU participants were excluded based on similar criteria as the Oslo cohort: any acute or chronic illness, cardiac, pulmonary, liver, or kidney abnormalities, uncontrolled hypertension, insulin- or non-insulin-dependent diabetes, abnormal blood, or urine chemistries, arthritis, a history of neuromuscular problems, or if they smoked tobacco. From the BSU cohorts comprising 12 old donors (6 women, 6 men, average age 84 ± 3 years) or 15 young donors (7 women, 8 men, average age 24 ± 4 years), we used transcriptome data based on thigh muscle biopsies taken before the first and before the last exercise in their 12-week training period. The data are available from GEO: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE28422
Fig. 1 Outline of study Full size image
Table 1 Demographic characteristics of muscle biopsy donors Full size table
Training protocols
Oslo Cohorts: The total duration of the intervention was 13 weeks for healthy and 15 weeks for the patients with primary osteoporosis because the first two weeks were used to familiarize them with the training protocol starting with lighter training loads. The training loads were gradually increased to ensure that the 13 weeks of training were conducted with optimal loading to improve muscle strength and muscle mass [12]. The training period was performed as traditional heavy-load strength training: three times per week with 1–3 sets involving all major muscle groups as detailed previously [12]. Briefly, the training protocol consisted of three exercises for the leg muscles (squat, leg press, and standing toe rise), and three exercises for the upper body muscles (chest press, seated rowing, and shoulder press). In addition, the participants performed self-selected exercises for the abdominal and lower back muscles at the end of each session. The strength-training regimen was a mix of linear periodization and daily undulating periodization. The participants started with 8–12 repetition maximum (RM) sets, and ended the 13-week protocol with 4–8 RM sets. In two sessions per week, the sets were run until failure (RM-sets); in the third session, performed between the two maximal sessions, sets were run with a load corresponding to 80–90% of the actual RM load. The total duration of training was about 60 min per session, and the participants exercised in groups of three with a personal instructor present.
The BSU cohorts underwent a 12-week training regimen with progressive resistance training (PRT). The groups performed a smaller, leg focused selection of exercises, but included 36 training sessions (3 days/week) with three sets of 10 bilateral knee extensions at 70–75% of their 1 RM [9]. Thus, training of the vastus lateralis muscle group was similar in the Oslo and BSU studies, although the load used in the BSU cohort was somewhat lower.
Questions may be asked if the transcription repertoire (the transcriptome) in the basal resting state is the same in mechanically loaded skeletal muscle (e.g., thigh) as in unloaded muscle (e.g., iliac/pelvic). If not, their responses to training could have been different, and the results would have been needed a further explanation. Thus, we compared if these muscles, serving vastly different functions (dynamic versus static), had the same transcriptional profile in the basal, rested state selecting muscle biopsies from pelvis of well characterized female donors (12 healthy and 12 with osteoporosis) with similar age and BMI and not part of a training study [13].
Muscle biopsy collection and RNA purification
Biopsy collection and RNA purification from the Oslo and BSU cohorts used similar methods and technology, as previously described [9, 10]. In brief, thigh muscle biopsies were obtained under local anesthesia (xylocaine adrenalin, 10 mg/ml + 5 μg/ml; AstraZeneca, London, UK) from the mid portion of the vastus lateralis before and after the training period using a modified Bergström technique. The muscle samples were obtained at least 2 days after any training or testing, and the second biopsy was obtained approximately 3 cm distal to the previous site. The biopsies were taken from fasting participants of both studies in the morning (07–09 AM) to allow for similar levels of physical activity and dietary intake. A sample (10–20 mg) to be used for RNA extraction was immediately frozen in liquid nitrogen (applied to the cohort of osteoporotic women; muscle biopsies from all other participants were stored in RNAlater (Merck, Darmstadt, Germany) for 1 day at 4 °C before freezing. The samples were then stored at -80 °C until RNA purification using a RNeasy Mini Kit (Qiagen, Oslo, Norway) according to the manufacturer’s instructions.
Microarray and data analysis
The same type of microarray analysis was performed on the BSU and Oslo cohorts employing Affymetrix HG-U133 Plus 2.0 or Affymetrix HuGene-1_0-st-v1 arrays (Thermo Fisher Scientific, Waltham, MA, USA). Robust microarray analysis (RMA) yielding normalized log2 transformed signal intensities was applied for normalization for both array types. (http://bip.weizmann.ac.il/toolbox/overview/Partek_Users_Guide.pdf). Gene transcripts with maximal signal values of < 5 (log2 values) across all arrays were removed to filter for low- and non-expressed genes. Differentially expressed transcripts before vs. after training were identified using two-way analysis of variance (ANOVA) as implemented in Partek Genomics Suite (Partek, St. Louis, MO, USA).
Further bioinformatics analysis on thigh muscle biopsies was conducted on the significant genes to identify functional implications with Ingenuity Pathway Analysis (Ingenuity Systems, Redwood City, CA, USA). The microarray data generated from the Oslo and BSU cohorts have previously been validated using real-time qRT-PCR on selected genes [9, 11].
Comparison of basal iliac and thigh muscle transcriptomes
In a previous study [13] we collected postmenopausal trans-iliac bone biopsies using a Bordier trephine [14]. Iliac muscle attached to the pelvic side of these bone biopsies were separated and snap frozen in liquid nitrogen (n = 24). Subsequent RNA isolation was performed as described for thigh muscle biopsies. However, Affymetrix HG-U133 Plus 2.0 arrays were used for transcriptome profiling, while HuGene-1_0-st-v1 arrays were used for transcriptome profiling of thigh muscle RNA. Use of different array types prevented us from normalizing the two datasets together. However, we conducted a Pearson correlation analysis of the Log 2 transformed signal values using the 17 652 transcripts present in both datasets (93.5% of all thigh muscle transcripts). To visualize the similarity the average signal values for the 17 652 transcripts were plotted against each other (Fig. S1). As a further measure of similarity/difference between the iliac and thigh muscle transcriptomes a paired T-test was performed on the datasets.
Functional enrichment analysis
Functional enrichment analysis identifies trends in large scale biological datasets and determines whether some functions are enriched in our set of differentially expressed genes. We used Ingenuity Pathway Analysis (IPA) (Qiagen, Beverly, MA, USA) for this task. Fisher’s exact test was used by IPA to identify enriched gene sets with all genes on the Human Gene 1.0 ST Array as the background gene list. Further information is found in the Table 2 legend.
Table 2 Over-represented diseases and functions Full size table
Principal component analysis (PCA)
PCA is a frequently used dimensionality-reduction method with the aim to reduce the dimensionality of large data sets by transforming extra sized sets of variables into smaller ones that still contain most of the information. We used the PCA generator as implemented in the software Partek Genomics Suite (Partek, St. Louis, MO, USA).
Volcano plot
The volcano plot was generated using the volcano plot generation module in Partek Genomics Suite 6.6 (Partek).
Analyses of association between transcripts responsive to heavy-load exercise training with muscle eQTLs and loci for handgrip strength in women
We identified SNPs associated with skeletal muscle expression for the genes listed in Table S1 using eQTL databases from the Genotype-Tissue Expression (GTEx) portal (https://gtexportal.org). Subsequently, significant SNPs were tested for association with low handgrip strength in females as summarized in a meta-analysis [11] adopting a nominal p-value < 0.01.