Mechanisms Underlying Absent Training-Induced Improvement in Insulin Action in Lean, Hyperandrogenic Women With Polycystic Ovary Syndrome

Polycystic ovary syndrome (PCOS) is the most common endocrine abnormality in reproductive-aged women, with a prevalence of 5–13%, depending on the diagnostic criteria (1). PCOS is characterized by the presence of polycystic ovaries, menstrual dysfunction, and hyperandrogenism (2). Women with PCOS have been shown to be less insulin sensitive than healthy control (CON) women and, therefore, at an increased risk of developing type 2 diabetes (3,4). Exercise training has been shown to improve insulin action in healthy individuals (5,6) and to reduce the incidence of diabetes in individuals at risk for diabetes (7). Most training studies conducted in women with PCOS include overweight and obese women, and the improvement in insulin action was often obtained with a concomitant loss of body weight (8). Lean women do not have a similar capability for weight loss as obese women. Notably, aerobic exercise training without loss of body weight has been shown to increase whole-body insulin action in healthy lean (9,10) and obese women (11). To date, only three studies have evaluated the effect of exercise training on whole-body insulin action in lean women with PCOS, but with conflicting results (12–14).

Skeletal muscle accounts for up to ∼85% of insulin-stimulated peripheral glucose uptake (15), and improved insulin action after exercise training in sedentary, healthy women and men has been associated with increased skeletal muscle glucose uptake (10,16). The aim of the current study was to evaluate the effect of exercise training on insulin action and especially on insulin-stimulated glucose uptake in skeletal muscle as well as on hepatic insulin action in lean women with hyperandrogenism and PCOS and healthy CON women matched for age, BMI, and VO_2max. We hypothesized that exercise training without weight loss would improve insulin action in lean women with hyperandrogenism and PCOS.

Furthermore, possible mechanisms underlying training-induced increase in glucose uptake in skeletal muscle in lean women with PCOS are not clear. Therefore, another aim was to clarify molecular mechanisms in skeletal muscle of lean women with PCOS underlying the metabolic adaptations to exercise training and insulin action.

This study was conducted in accordance with the principles of the Declaration of Helsinki and approved by the regional ethics committee of the Capital Region of Denmark (H-1-2013-034). Written informed consent was obtained from all participants before enrollment in the study. The study was registered at ClinicalTrials.gov (NCT02429128).

Women with PCOS and healthy CON women were recruited through newspaper advertisement and social media. Participants with hyperandrogenism and PCOS were included on the basis of 1) clinical (Ferriman-Gallwey score ≥8 [17]) and/or biochemical (free testosterone [free T] >0.034 nmol/L) hyperandrogenism combined with 2) PCO morphology by transvaginal ultrasonography and/or 3) oligomenorrhea (>35 days apart). Inclusion criteria for CON were regular menstrual cycles together with clinical and biochemical normoandrogenism. Additional inclusion criteria for both PCOS and CON were 1) BMI between 19.5 and 27 kg/m², 2) inactive or low physical activity level, 3) low to moderate fitness level determined as Vo_2max <3 L/min, and 4) age between 20 and 40 years. Exclusion criteria for both groups were 1) breastfeeding or use of oral contraceptives 3 months before the study, 2) use of glucose homeostasis–regulating medicine 3 months before study, and 3) smoking.

Many individuals applied for participation (>200). Eleven participants each in PCOS and CON fulfilled the criteria and were accepted to participate in the study. After inclusion, two women from each group were excluded during the intervention because of either injury, lack of training compliance, or weight gain >2 kg. Hence, nine PCOS and nine CON participants, matched for age, BMI, and Vo_2max, completed the study (Table 1).

After enrollment, physical activity level was evaluated by 14 days of step registration, using an accelerometer step counter (Polar, Kempele, Finland). Four days of dietary registration, where all food and beverages were weighed to 1 g of accuracy, were completed, and the macro- and micronutrient distribution and energy intake were calculated (Dankost, Copenhagen, Denmark). Body weight was registered every morning. Pretraining testing included measurements of Vo_2max on a bicycle ergometer, body composition by DEXA scanning, insulin action by an oral glucose tolerance test (OGTT), and a hyperinsulinemic-euglycemic clamp. B-cell function was assessed by OGTT. A 14-week controlled aerobic and strength exercise training intervention was initiated, which included repeated measurements of body weight, nonexercise physical activity level, and dietary frequency questionnaires. After the intervention, posttraining testing was performed, including similar experiments as at pretraining with additional dietary analysis. All experiments were performed during the early follicular phase of the menstrual cycle in CON and whenever feasible in PCOS. Participants were instructed to abstain from vigorous physical activity 48 h before testing.

VO_2max was determined using an incremental exercise test to exhaustion, with a stepwise increase of 15 W every minute on an ergometer bicycle (Monark Ergomedic 839E; Monark, Sweden) using a respiratory gas exchange system (Masterscreen CPX; IntraMedic, Gentofte, Denmark). Heart rate (HR) was continuously recorded (SC400; Polar, Kempele, Finland). The participants were familiarized with the equipment and test procedure before the initial pretraining test.

Evaluation of body composition was performed by DEXA (DPX-IQ; Lunar Corporation, Madison, WI) after an overnight fast (12 h).

Participants arrived at the laboratory at 8:00 a.m. by bus or car after an overnight fast (12 h). A catheter was inserted into an antecubital vein, and three basal fasting blood samples were drawn at 5-min intervals. A glucose load (75 g of glucose in 250 mL water) then was ingested within 5 min, and blood samples were obtained at 15, 30, 45, 60, 90, 120, and 180 min postingestion while participants rested in a sitting or supine position.

At least 3 days apart from the OGTT, participants arrived at the laboratory at 8:00 a.m. by bus or car after an overnight fast (12 h). A catheter was inserted into an antecubital vein, and a fasting blood sample was collected. Catheters were inserted into a femoral vein and a superficial dorsal hand vein, the latter being heated by a heating pad, for arterialized blood measurements. Participants rested in a supine position, and a primed (priming dose 2.6 mg/kg), constant [6,6-²H₂]glucose tracer infusion (0.044 mg/kg/min) was initiated. After 2-h tracer infusion, a 120-min hyperinsulinemic-euglycemic clamp was initiated. The insulin infusion rate during the clamp was 1 mU insulin/kg/min administered after an initial insulin bolus (9.0 mU/kg) (Actrapid; Novo Nordisk, Bagsværd, Denmark). During the clamp, glucose was infused (20% glucose solution enriched with 1.9% [6,6-²H₂]glucose) at a rate ensuring euglycemia, matching the fasting arterialized blood glucose level on the preintervention experimental day. Femoral arterial blood flow was measured by the ultrasound Doppler technique (Philips iU22; ViCare Medical, Birkerød, Denmark) every 20 min before arteriovenous blood was sampled. Biopsy samples from the vastus lateralis muscle were obtained before and at the end of the clamp under local anesthesia using a Bergström needle with suction. One part was frozen in liquid nitrogen and stored at −80°C for subsequent biochemical analysis. Another was mounted in embedding medium and frozen in precooled isopentane and then stored at −80°C for subsequent histochemical analysis.

Femoral venous catheterization and a muscle biopsy sample were not obtained from one participant with PCOS because of withdrawal of consent for these procedures. Therefore, leg balance data and muscle analysis for the PCOS includes seven participants. Histochemical analysis was only performed on a subgroup (n = 4–9) of both CON and PCOS because of lack of muscle material (specified in Table 4).

Participants underwent 14 weeks of controlled and supervised exercise training consisting of two interval-based, high-intensity aerobic exercise sessions on an indoor bicycle (45 min) with integrated watt recording (Body Bike, Frederikshavn, Denmark) and one whole-body strength training session (45 min) per week. For the cycling sessions, the average intensity was 60–65% of watt maximum, with periods of higher and lower watts for a minimum of 2 min/period. The participants were instructed to maintain a cadence of ∼70 rpm. Continuous HR registration (SC400; Polar) was applied and evaluated by the computer program Polar Protrainer 5 (Polar). Tests of VO_2max were performed frequently to adjust exercise workload to progression in aerobic capacity. Strength training comprised nine different whole-body exercises with three sets of 8–12 repetitions. Progression was achieved by increasing resistance (weight) and difficulty level throughout the training period. All training sessions started with an ∼10-min warmup and were instructed and supervised.

Serum concentrations of total testosterone and androstenedione were analyzed by liquid chromatography tandem mass spectrometry. Serum sex hormone–binding globulin (SHBG), luteinizing hormone (LH), and follicle-stimulating hormone (FSH) were determined by immunofluorometric assays. Free T was calculated from the measurement of total testosterone and SHBG (18). Plasma glucose concentration was measured on an ABL 615 (Radiometer Medical A/S, Brønshøj, Denmark), and plasma enrichment of the glucose isotope was measured using liquid chromatography mass spectrometry (Finnigan AQA; ThermoQuest, Austin, TX) as described (19). Plasma insulin and C-peptide concentrations were measured by ELISA kits (ALPCO, Salem, NH). Plasma fatty acids (FAs) (NEFA C Kit; Wako Chemicals, Neuss, Germany); triacylglycerol (TG) (GPO-PAP Kit; Roche Diagnostics, Mannheim, Germany); and total cholesterol (TC), HDL cholesterol (HDL-C), and LDL cholesterol (LDL-C) were measured using colorimetric methods on an autoanalyzer (Hitachi 912; Boehringer Ingelheim, Ingelheim am Rhein, Germany).

Muscle samples were freeze dried and dissected free of all visible adipose tissue, connective tissue, and blood under a dissection microscope. Muscle biopsy samples were homogenized as previously described (4), and lysates were recovered by centrifuging the homogenates (20 min, 18,320g, 4°C). Homogenate and lysate protein content were determined by the bicinchoninic acid method (Pierce Biotechnology, Waltham, MA).

To measure protein expression and phosphorylation, samples were separated on self-cast gels using SDS-PAGE followed by semidry transfer of proteins on polyvinylidene fluoride membranes. Membranes were blocked for 5 min in 2% skim milk in TBS containing 0.05% Tween-20 (Tris-buffered saline with Tween buffer) followed by overnight incubation at 4°C in primary antibodies (Supplementary Table 1). Membranes were incubated with horseradish peroxidase–conjugated secondary antibodies (Jackson ImmunoResearch, Ely, U.K.) for 45 min at room temperature before visualized on a Bio-Rad ChemiDoc MP Imaging System. Signals were quantified using Image Lab 4.0 software. Membranes were Coomassie stained to verify loading consistency.

Muscle glycogen content was measured by a fluorometric method as glycosyl units after acid hydrolysis on 150 μg of homogenates and related to the protein concentration of the homogenate.

RNA was isolated from 20 mg w/w muscle tissue by an acid guanidinium thiocyanate-phenol-chloroform extraction method adapted from Chomczynski and Sacchi (20), except that the tissue was homogenized for 3 min at 30 Hz in a TissueLyser II (QIAGEN, Venlo, the Netherlands). RNA concentration and purity were tested by NanoDrop (NanoDrop 1000; Thermo Fisher Scientific, Waltham, MA). Real-time and quantitative PCR were performed as described previously (21). Sequences used to amplify a fragment of hexokinase II (HKII) were forward primer: 5′ TTGTCCGTAACATTCTCATCGATT 3′; reverse primer: 5′ TGTCTTGAGCCGCTCTGAGAT 3′; and TaqMan probe: 5′ ACCAAGCGTGGACTGCTCTTCCGA 3′. GLUT4 and cyclophilin A (Cyc.A) probes were predeveloped assay reagents from Applied Biosystems (Foster City, CA) (GLUT4: Hs00168966_m1 and Cyc.A: 4310883E). GLUT4 and HKII gene expressions were related to Cyc.A mRNA, which was similar between groups and not affected by exercise training.

Serial cross-sections (10 μm) were cut and stained for myofibrillar ATPase to identify type I, IIA, and IIX muscle fibers (22). Muscle fiber type composition were analyzed using TEMA image analysis software (ChechVision, Støvring, Denmark). In each biopsy, a total of 157 ± 10 fibers were analyzed for fiber type composition and 152 ± 11 fibers for fiber size.

Whole-body insulin action was calculated as the average glucose infusion rate (GIR) for the last 20 min of the clamp expressed relative to body mass. Leg glucose uptake was calculated as femoral arteriovenous blood glucose difference × blood flow and expressed relative to leg mass. The blood flow data shown in Fig. 1K was obtained during the last 20 min of the clamp. Endogenous glucose production (EGP) was calculated during the last 20 min of the basal and clamp period using Steele steady-state calculations (23). During the clamp period, EGP was calculated by subtracting GIR from glucose R_a. HOMA insulin resistance (HOMA-IR) index was calculated as fasting insulin concentration × fasting plasma glucose concentration / 22.5 from the mean of three fasting blood samples obtained before the OGTT. Incremental areas under the curve (AUCs) for plasma insulin and glucose concentrations during OGTT were calculated using the trapezoidal method (24). Evaluation of insulin action from OGTT was done using the Matsuda index (IS_OGTT) (25). Insulin clearance was calculated as insulin infusion rate / (insulin_clamp − [C-peptide_clamp × insulin_fasting/C-peptide_fasting]) (26). B-cell function was evaluated as OGTT AUC_insulin/OGTT AUC_glucose (27). Disposition index was calculated as OGTT AUC_insulin/OGTT AUC_glucose × GIR (28).

All data are expressed as mean ± SEM and analyzed using SigmaPlot Version 13.0 software (SYSTAT Software, San Jose, CA). Participant characteristics and matching at pretraining were evaluated by unpaired t test. The effect of exercise training was evaluated using two-way ANOVA with repeated measures (ANOVA-RM) for groups and training. For variables including insulin stimulation, a two-way ANOVA-RM within fasting and insulin-stimulated values were applied for group and intervention effects and within pre- and postintervention for group and insulin effects. Data were transformed when appropriate to ensure normal distribution and equal variance. Tukey test was used as post hoc test. The significance threshold was P < 0.05.

The data set generated and analyzed in this study is available from the corresponding author upon reasonable request.

All PCOS participants were clinical and/or biochemical hyperandrogenic, and the group was thus characterized by elevated serum concentrations of total testosterone (approximately twofold, P < 0.01), free T (approximately threefold, P < 0.001), and ∼37% lower serum SHBG concentrations compared with the CON group (P < 0.01) (Table 1). Clinical hyperandrogenism was present in 70% of the participants with PCOS (P < 0.01). PCOS had elevated serum LH levels (P < 0.01) and a twofold higher serum LH/FSH ratio (P < 0.05) than CON (Table 1). At pretraining, body fat percentage did not differ between groups (Table 2), neither did VO_2max nor habitual physical activity level (registered steps/day) or habitual daily energy intake and diet composition. Whole-body insulin action, insulin-stimulated leg glucose uptake, glucose tolerance, β-cell function, disposition index (Fig. 1), and HOMA-IR (Table 3) were not different between groups at pretraining. Fasting plasma concentrations of TG, TC, HDL-C, and LDL-C were similar between groups (Table 3).

The number of completed training sessions was >90% in both aerobic cycling (CON 96 ± 3.3%, PCOS 95 ± 2.4%, not significant [NS]) and strength training (CON 99 ± 0.7%, PCOS 99 ± 0.9%, NS). The aerobic cycling sessions were completed with the same average intensity in both groups (CON 86.3 ± 0.5% HR_max, PCOS 84.4 ± 1.0% of HR_max, NS). Exercise training induced an ∼17% increase in VO_2max in both groups (P < 0.001) (Table 2). Dietary macronutrient composition did not change during the intervention period. In accordance with the study design, participants in both groups remained weight stable (Table 2). The total body fat percentage was overall decreased by 1.5–1.7% (P < 0.001), and lean body mass was increased by ∼1 kg (P < 0.001) after exercise training, with no differences between groups (Table 2). In PCOS, but not CON, the fat percentage in the abdominal compartment was decreased by ∼9% (P < 0.01) (Table 2).

Fasting serum SHBG concentration increased by ∼12% (P < 0.05) (Table 3) in both groups after exercise training. In PCOS, reductions, but not normalizations, were observed in serum concentration of total testosterone (−25%, P < 0.05) (Table 3) and androstenedione (−26%, P < 0.01) (Table 3). As a result of the lowering of total serum testosterone and increased SHBG, free T was decreased by ∼30%, but not normalized, in PCOS (P < 0.01) (Table 3). These improvements in PCOS hyperandrogenism were not accompanied by changes in serum LH levels or LH/FSH ratio (Table 3). Exercise training increased plasma HDL-C concentration by ∼14% in both groups (P < 0.05) (Table 3).

At preintervention, insulin infusion resulted in similar GIR in both CON and PCOS (Fig. 1A). Whole-body insulin action was increased after exercise training in CON by ∼26% (P < 0.01) but remained unchanged in PCOS (Fig. 1A). Plasma glucose was 5.2 ± 0.1 mmol/L during the clamp in both groups at pre- and posttraining (Table 3). Plasma FA concentration was suppressed during the clamp to a greater extent at posttraining compared with pretraining in both groups, with no group differences (Table 3). Preintervention fasting respiratory exchange ratio (RER) levels did not differ between groups (CON 0.79 ± 0.01, PCOS 0.76 ± 0.01, NS). Both fasting RER and the insulin-induced increase in RER during the clamp (Fig. 1B) remained unchanged in both groups with training. This indicated that the increased glucose disposal in CON after training was not related to increased glucose oxidation but, rather, to increased nonoxidative disposal.

Exercise training resulted in a lowering of the incremental area under the OGTT curve for plasma glucose (P < 0.05) and insulin (P < 0.001) in CON but not in PCOS (Fig. 1B and C). Thus, the Matsuda index, calculated from the OGTT, increased by ∼29% in CON (P < 0.01) but not in PCOS (Fig. 1D). HOMA-IR decreased by ∼30% after the intervention in CON (P < 0.05) because of decreased fasting plasma insulin concentration in this group only (P < 0.01) (Table 3). Calculated β-cell function in response to oral glucose challenge was lowered in CON by ∼26% (P < 0.05) but remained unchanged in PCOS after exercise training (Fig. 1E). When β-cell function was related to GIR during the clamp, as evaluated by the disposition index, this remained unchanged in both groups (Fig. 1F).

Fasting and insulin-stimulated EGP, together with the relative suppression of EGP with insulin, were similar between groups at pretraining and unaffected by exercise training (Fig. 1G). Insulin clearance rate during the clamp remained unchanged with exercise training in both groups (Fig. 1H).

At pretraining, there was no difference in the insulin-stimulated leg glucose uptake between groups (Fig. 1I). Exercise training resulted in an ∼53% increase in insulin-stimulated leg glucose uptake in CON (P < 0.01), whereas no change was observed in PCOS. At pretraining, femoral blood flow was similar at basal and increased to a similar extent in both groups during insulin-stimulated conditions (55 ± 11 and 47 ± 13 mL/min in CON and PCOS, respectively). At posttraining, the insulin-induced increase in femoral blood flow was greater in CON compared with PCOS (113 ± 20 and 64 ± 22 mL/min, P < 0.05), being 29% greater in CON compared with pretraining (Fig. 1J), while no change was evident in PCOS. In CON, insulin-stimulated femoral arteriovenous blood glucose difference, measured during the last 20 min of the clamp, was 33% increased at posttraining (P = 0.06). This indicates that an improved insulin-mediated increase in leg blood flow is the main contributor to the increased insulin action at posttraining in CON.

The increased glucose disposal in CON, but not in PCOS, was observed concomitantly with data showing that muscle homogenate glycogen content was similarly increased after exercise training in both groups (CON 166 ± 43 and PCOS 345 ± 117 nmol/mg protein) (Fig. 1K). Muscle glycogen content was, however, greater in PCOS than in CON both at preintervention (38%) and at postintervention (57%).

There were no differences between groups in GLUT4 and HKII gene and protein expressions in muscle at pretraining (Fig. 2). In CON, the elevated leg glucose uptake was accompanied by an increase in skeletal muscle GLUT4 gene (∼58%, P < 0.05) and protein (∼30%, P < 0.01) expressions, while both remained unchanged in PCOS with exercise training (Fig. 2A and B). Likewise, increased HKII gene (∼270%, P < 0.001) and protein (∼68%, P < 0.01) expressions were observed in CON in response to exercise training, but not in PCOS (Fig. 2C and D). Citrate synthase (CS) did not differ between groups at pretraining and increased by ∼65% in both groups with training (P < 0.05) (Fig. 2E).

At pretraining, muscle total protein expression of IR, Akt2, and TBC1D4 as well as phosphorylation of Akt2 and its downstream target Akt substrate of 160 kDa (TBC1D4) did not differ between groups (Fig. 3A–D). Exercise training did in both groups increase Akt2 protein expression by ∼30% (P < 0.05) (Fig. 3B), while TBC1D4 protein remained unchanged (Fig. 3C). Increased insulin-stimulated Akt serine 473 (Ser⁴⁷³) (∼37%) and threonine 308 (Thr³⁰⁸) (∼43%) phosphorylation were observed after training in CON (P < 0.01) but not in PCOS (Fig. 3D and E). Basal and insulin-stimulated TBC1D4 Ser⁵⁸⁸ and Ser³¹⁸ phosphorylation did not change with exercise training (Fig. 3F).

AMPKα2 protein expression in muscle was similar between groups at pretraining and remained unchanged with training (Fig. 4A). Basal AMPK Thr¹⁷² phosphorylation was also similar between groups at pretraining but increased in response to training by ∼31% and ∼37% in CON and PCOS, respectively (P < 0.05) (Fig. 4B). Acetyl-CoA carboxylase (ACC) protein expression increased with exercise training in both groups by ∼30%, and basal ACC Ser²²¹ phosphorylation was ∼54% and ∼100% higher after training in CON and PCOS, respectively (P < 0.05) (Fig. 4C and D). Together, this indicates a training-induced increase in AMPK activity as described before (29).

Skeletal muscle protein expression of NADPH oxidase 4 (NOX4), superoxide dismutase 2 (SOD2), and thioredoxin 2 (TRX2) were similar between groups at pretraining and unaffected by training (Fig. 4E–G).

In a subgroup (n = 4–9) (Table 4), muscle fiber type distribution was evaluated. At pretraining, no group differences in fiber type percentage were obtained, but the area of type I, IIa, and IIx fibers were 16–29% lower in PCOS than in CON (P < 0.05) (Table 4). Exercise training resulted in an increased percentage of type IIa fibers (P < 0.01) and decreased percentage of type IIx fibers (P < 0.05) in CON, whereas no changes were observed in PCOS (Table 4). The area of type I and type IIA fibers increased in both groups with training (P < 0.05) (Table 4).

We studied the impact of exercise training without body weight changes on insulin action in lean women with PCOS and especially the effect on insulin-stimulated glucose uptake and regulatory mechanisms in skeletal muscle. Initially, whole-body, hepatic, and peripheral insulin action in the lean women with PCOS were similar to age-, BMI-, and VO_2max-matched CON, a finding which is similar to a previous study that reported similar insulin-stimulated forearm glucose uptake in lean women with hyperandrogenism and PCOS as in age- and BMI-matched control women (30). Fourteen weeks of supervised aerobic and strength training elicited in both groups comparable increases in VO_2max (∼17%) and in CS protein expression (∼65%), a classical marker of training adaptations (31). While CON improved whole-body insulin action with training, as judged by a ∼26% increase in clamp GIR, ∼29% increase in the Matsuda index, and ∼30% decrease in the HOMA-IR, these parameters did not change with training in PCOS. Similar to the lack of training-induced improvement in whole-body insulin action in PCOS, insulin-stimulated glucose uptake in the leg remained unchanged with training in PCOS but increased by ∼53% in CON, which is in line with previous findings in healthy men and women (10,16,32).

The training-induced increase in leg glucose disposal was, in CON, related to a greater insulin-mediated increase in femoral blood flow. A training-induced increase in insulin-stimulated leg blood flow is in line with previous studies in males when insulin was increased during an OGTT (33) or hyperinsulinemic-euglycemic clamp (34). The reason for the failure to improve insulin-stimulated leg blood flow with training in PCOS is not known. In the endothelium, insulin stimulates the production of the vasodilator nitric oxide and the vasoconstrictor endothelin-1 (35). An imbalance in production of nitric oxide and endothelin-1 blunted insulin-mediated increase in blood flow in individuals with obesity and insulin resistance (36,37). It could thus be speculated that a lack of training-induced improvement in production of these blood flow–regulating compounds could contribute to the lack of increased insulin-stimulated blood flow and, in turn, glucose uptake in PCOS. In PCOS, the resting glycogen content in skeletal muscle was higher than in CON. Still, training induced a similar increase in resting glycogen content, and postintervention muscle glycogen was thus remarkably greater (by 57%) in PCOS than in CON. It could therefore be considered that the high postintervention glycogen concentration in PCOS might contribute to impeding insulin-stimulated glucose uptake (38,39).

The current model for insulin action in skeletal muscle is activation of Akt2 leading to phosphorylation of the Rab GTPase-activating protein TBC1D4, thereby facilitating GLUT4 trafficking to the plasma membrane for glucose uptake (40). The level of Akt2 protein was increased after exercise training in both groups. The phosphorylation by insulin on both Akt phosphorylation sites (Thr³⁰⁸ and Ser⁴⁷³) was increased with training in CON, supporting a previous study in men (41,42), while the absolute improvement in TBC1D4 insulin-stimulated phosphorylation in CON did not reach statistical significance (although, for the TBC1D4 Ser⁵⁸⁸ site, P = 0.05 was obtained for effect of intervention in CON). These signaling data suggest a training-induced increased stimulus to translocate GLUT4 to the sarcolemma in CON only, congruent with the lack of a training-induced increased muscle glucose disposal in PCOS. It is possible that the 57% greater glycogen content in PCOS muscle contributed to the lack of increased proximal insulin signaling following training. Together with an unchanged GLUT4 mRNA and protein expression in PCOS with training, this supports that the capacity of insulin-stimulated glucose transport in skeletal muscle in women with PCOS was not modified by training. HKII plays an important role in the transarcolemmal glucose gradient by phosphorylation of glucose to glucose-6-phosphate. In accordance with previous studies (10,16,41), the protein expression of HKII was increased with exercise training by ∼68% in CON, an effect not obtained in PCOS. Together, these data imply that the lack of increase in whole-body insulin action in lean women with PCOS with training is due to an impaired ability to upregulate glucose uptake in skeletal muscle because the regulation of hepatic glucose production remained unchanged in response to training.

AMPK has been linked to the regulation of training-induced adaptations in skeletal muscle gene expression of GLUT4 and HKII (43). In the current study, both AMPKα Thr¹⁷² phosphorylation and ACC phosphorylation, the downstream target of AMPK activity, were similarly increased with exercise training in CON and PCOS. Some studies in humans suggest that an increased antioxidant capacity might impair training adaptations in insulin action and reactive oxygen species–sensitive transcriptional regulators, including PPARγ, PGC1α, and PGC1β (44–46). These and other exercise-induced transcriptional regulators are known to be redox regulated (47). In the current study, expression of the antioxidant defense proteins SOD2 and Trx2 in muscle did not differ between groups or were affected by training, and neither were expression changes observed for the superoxide-producing NOX4 enzyme. The lack of a training-induced increase in GLUT4 and HKII mRNA and protein expressions in lean women with PCOS is puzzling and important to pursue. In a subgroup, we observed a training-induced shift only in CON toward a more oxidative fiber distribution, a well-known training adaption (48,49). Because it has been reported that GLUT4 abundancy in human muscle (50,51) and insulin-stimulated glucose uptake in rat muscle (52) increase when the fibers become more oxidative, a training effect on fiber type remodeling may contribute to an increase GLUT4 protein expression and insulin-stimulated leg glucose uptake, which could support the present findings.

The mechanisms explaining the lack of skeletal muscle training adaptations in PCOS are unclear. In a previous study, participants with insulin resistance who did not achieve a response in insulin sensitivity after 8 weeks of endurance training (despite preintervention parameters being similar to the responders) were characterized by a lack of increase in muscle gene expression of GLUT4 (as in the current study) together with a nonresponse in other genes related to glucose transport and metabolism (53). Upstream regulators specifically activated in the nonresponder group included transforming growth factor-β1 (TGF-β1) (53). Elevated muscle TGF-β1 expression has also been shown in rats to be a focal regulator of training adaptations by hyperphosphorylating an important exercise-induced site on Sma/Mad-related protein 2 (54). TGF-β1 activation in PCOS could thus be speculated to contribute to the blunted upregulation of GLUT4 and HKII expression.

The most prominent feature in PCOS is hyperandrogenism (2), and lean women with hyperandrogenism and PCOS have previously been suggested to be more insulin resistant than women with PCOS and normal androgen levels (55,56). However, despite including only women with hyperandrogenism and PCOS in the current study, insulin sensitivity was initially comparable to CON. However, some cross-sectional studies did not obtain different insulin sensitivity between lean participants with PCOS and healthy control participants (30,57,58). Thus, inconsistency exists regarding the role of androgens in the development of insulin resistance. This is supported by findings in transsexual individuals, where androgen treatment induces insulin resistance in some (59), but not all (60), studies. Collectively, this suggests that the presence of other factors than hyperandrogenism contribute to regulation of insulin resistance. The women with PCOS in the current study were carefully matched with CON women with regard to age, BMI, and cardiorespiratory fitness. Differences in matching and inclusion criteria for lean women with PCOS between the present and previous studies could potentially explain the divergent observations. A possible limitation in the current study may be the limited sample size of women with PCOS, considering the heterogeneity of this disorder.

Despite the absence of a training effect on insulin action in this group of women with PCOS, VO_2max increased to a similar extent as in CON. Furthermore, training markedly improved plasma androgen and SHBG levels in PCOS in line with some (13,61,62), but not all (14,63,64), training studies in lean and obese women with PCOS. The mechanisms for improvement in circulating androgenic markers remain unknown. Women with PCOS display persistently rapid LH pulsatility and increased amplitude, which further augment ovarian androgen production (65). We did not observe changes in serum LH levels or LH/FSH ratio by training in women with PCOS; however, a training effect on pulsatility or amplitude of LH cannot be excluded.

Regular aerobic exercise is known to be antiatherogenic by increasing plasma HDL-C concentration (66,67). In the current study, plasma HDL-C levels were increased by ∼14% in women with PCOS and in CON women. Previous exercise training studies (yoga, short-term aerobic step exercise ≥20 min/session) in lean women with PCOS have shown modest increases in HDL-C concentrations (2–3%) probably because of training modality (12,14).

In conclusion, a marked training-induced increase in whole-body and muscle insulin action was obtained in healthy sedentary females, whereas insulin action was not improved in lean sedentary women with PCOS. This was likely caused by the lack of training-induced adaptations in muscle fiber type composition and glucose metabolic proteins, greater glycogen content in muscle of PCOS, and lack of an increase in insulin-stimulated blood flow compared with CON. However, other training-induced benefits were demonstrated. Thus, the similar training effort in both groups resulted in a similar increase in VO_2max, upregulation of the mitochondrial marker CS and AMPK signaling in muscle, and improved plasma HDL-C concentration in both lean women with PCOS and healthy CON women. Importantly, in women with PCOS, serum androgen levels were decreased and serum SHBG concentrations were increased by the exercise training program independent of weight loss.

Acknowledgments. The authors acknowledge Irene Bech Nielsen, Betina Bolmgren, and Nicoline Resen (University of Copenhagen) for excellent assistance in the human trials and laboratory analysis. The authors are grateful for the help from bachelor student Sophia Hjorth (Department of Nutrition, Exercise and Sports, University of Copenhagen). The authors appreciate the help from Mette Petri (Rigshospitalet, Copenhagen) in the recruitment process. Furthermore, the authors give a special thanks to all women participating in the study.

Funding. The study was supported by the Novo Nordisk Research Foundation and the Københavns Universitet Excellence Program for Interdisciplinary Research (2016) “Physical Activity and Nutrition for Improvement of Health.” A.-M.L. was supported by a research grant from the Danish Diabetes Academy, which is funded by Novo Nordisk Fonden grant NNF17SA0031406. Postdoctoral research by K.A.S. was supported Det Frie Forskningsråd grant 4092-00309.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. S.L.H., K.N.B.-M., L.N., S.M., J.F.P.W., E.A.R., and B.K. designed the study. S.L.H., K.N.B.-M., L.N., K.A.S., J.F.P.W., E.A.R., and B.K. carried out the clinical experiments. S.L.H., A.-M.L., F.L.H., J.R.H., A.K.S., C.H.O., and B.K. analyzed the data. S.L.H. and A.-M.L. performed the statistics. S.L.H., A.-M.L., and B.K. wrote and edited the manuscript. S.L.H., F.L.H., L.F.W., M.H., K.M.L., and C.H. organized and carried out the training intervention. A.-M.L., F.L.H., K.A.S., J.R.H., A.K.S., C.H.O., C.S.C., L.F.W., M.H., K.M.L., C.H., and T.E.J. contributed to the experiments. All authors contributed to the manuscript and approved the final version of the manuscript. B.K. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.