High intensity intermittent exercise improves cardiac structure and function and reduces liver fat in patients with type 2 diabetes: a randomised controlled trial

Patients

Type 2 diabetic patients (stable control with diet and/or metformin for at least 6 months) were randomised to a HIIT (n = 14) or a control group (n = 14). Participant characteristics can be seen in Table 1. Participants were excluded if they had any overt cardiac disease, took part in regular exercise (≥60 min moderate–vigorous activity per week), were being treated with β-blocker medication or had any contraindications to exercise stress testing according to guidelines [18]. The study was approved by the Newcastle and Northeast Tyneside Local Research Ethics Committee and all participants provided written informed consent. Participants were recruited via advertising in local newspapers and diabetes community groups between September 2012 and September 2013. During the study, two participants left for unrelated medical reasons, one participant could not commit the time and two failed to comply with MRI procedures, leaving 12 in the HIIT group and 11 in the control group (Fig. 1).

Table 1

Participant characteristics

Characteristic	Control group	HIIT group	p value
n (men/women)	8/3	10/2
Age (years)	59 ± 9	61 ± 9	0.70
Time since diagnosis (years)	4 ± 2	5 ± 3
BMI (kg/m2)	32 ± 6	31 ± 5	0.71
Height (cm)	169 ± 9	171 ± 8	0.71
Weight (kg)	90 ± 9	90 ± 15	0.95
HbA1c (%)	7 ± 0.5	7 ± 1	0.87
HbA1c (mmol/mol)	55 ± 6	54 ± 11	0.88
Fasting glucose (mmol/l)	7.0 ± 1.0	6.8 ± 1.6	0.69
2 h glucose (mmol/l)	11.7 ± 3.1	12.5 ± 3.1	0.57
Liver fat (%)	7.1 ± 6.8	6.9 ± 6.9	0.94
\( \dot{V}{\mathrm{O}}_{2\mathrm{peak}} \) (ml kg−1 min−1)	20.3 ± 6.1	21.8 ± 5.4	0.54
Medications
Metformin	7	7
Statins	6	7
BP	5	3

\( \dot{V}{\mathrm{O}}_{2\mathrm{peak}} \) is normalised to body mass

Open image in new window

Experimental protocol and randomisation

Following an initial screening visit, cardiac structure and function, liver and visceral fat, body composition, glycaemic control and blood variables were measured at baseline and after 12 weeks of HIIT or continued standard care. Glucose control was measured within 48–72 h of the final exercise session to control for the acute effect of exercise on glucose uptake. Patients were randomised into groups using a simple random allocation sequence (www.randomization.com). Concealed envelopes with consecutive numbers were locked in a drawer and withdrawn in numerical order by the main author (SC).

Screening

To determine underlying cardiac disease, a medical history, physical examination, and resting and exercise 12-lead electrocardiography (Custo med, Ottobrunn, Germany) were performed. During the exercise test, gas exchange was measured to determine peak oxygen consumption (\( \dot{V}{\mathrm{O}}_{2\mathrm{peak}} \)) normalised to body mass. \( \dot{V}{\mathrm{O}}_{2\mathrm{peak}} \) was determined as the point at which participants reached volitional exhaustion, participants could no longer maintain a cycling rate of 60 rev/min, or continuing exercise was contraindicated [18]. The test was performed using an electronically braked semirecumbent cycle ergometer (Corival Lode, Groningen, The Netherlands) and resistance was increased by 1 W every 6 s. Before the exercise test, patients completed a 20 min resting period in which they lay supine while beat-to-beat BP was measured by the vascular unloading technique [19].

Cardiac MRI

All examinations were performed using the 3.0 T Philips Achieva MRI scanner with a six channel cardiac array (Philips Medical Systems, Best, The Netherlands). During breath holding, a stack of balanced steady-state free precession images were obtained in the short axis view, covering the entire left ventricle (field of view [FOV] = 350 mm, repetition time [TR] / echo time [TE] = 3.7/1.9 ms, turbo factor 17, flip angle 40°, slice thickness 8 mm, 0 mm gap, 14 slices, 25 phases, resolution 1.84 × 1.37 mm with zero filling to 1.37 × 1.37 mm and temporal duration approximately 40 ms per phase, dependent on heart rate). Using a Viewforum workstation (Philips Medical Systems), the short axis slices at end-diastole and end-systole were used (Fig. 2a) to manually trace endocardial and epicardial borders, with papillary muscles excluded from volume calculations but included in calculations of left ventricular mass. The apical slice was defined as the last slice showing inter-cavity blood pools, and the basal slice as the last slice in which at least 50% of the blood volume was surrounded by myocardium. The analysis was performed by a single observer blinded to group allocation. Details of the algorithm for contour selection and calculation of left ventricular mass, systolic and diastolic variables have been previously published [20]. The eccentricity ratio is a measure of concentric remodelling, and was calculated by dividing left ventricular mass by end-diastolic blood volume. Longitudinal shortening was determined from cine MRI in the four-chamber view by determining the perpendicular distance from the plane of the mitral valve to the apex in systole and diastole.

Open image in new window

Cardiac tagging

Tagged short axis images were acquired. Cardiac tagging works by applying radiofrequency pulses to cancel the MR signal from the myocardium in diastole in a rectangular grid pattern and tracking the deformation of these tags through the rest of the cardiac cycle (Fig. 2a). A turbo-field echo sequence with acceleration factor 9 was used (TR = 4.9, TE = 3.1, flip angle = 10°, number of excitations = 1, SENSE factor 2, FOV 350 × 350 mm, voxel size 1.37 × 1.37 mm with no zero filling and an orthogonal complementary spatial modulation of magnetistion grid with tag spacing of 7 mm). Short axis slices of 10 mm thickness were prescribed. The Cardiac Image Modelling package (University of Auckland, New Zealand) was used to analyse the tagging data by aligning a mesh on the tags between the endo- and epicardial contours. Details of the calculation of strain and torsion variables have been previously published [21].

Cardiac spectroscopy

Cardiac high-energy phosphate metabolism was assessed using ³¹P-magnetic resonance spectroscopy (see Fig. 2b for an example cardiac spectrum). Cardiac acquisition was vectorcardiogram-gated with the participant in a prone position during free breathing, with a 10 cm diameter ³¹P surface coil used for transmission/reception (PulseTeq, Chobham, UK) and the scanner body coil used for localising images. A cardiac gated one-dimensional chemical shift imaging sequence was used, and to eliminate contamination from liver, a 7 cm slice was selected in the foot–head direction using a selective pulse (of the ‘spredrex’ type, as outlined in [22]). The flip angle subtended at the myocardium was fixed to achieve 50° excitation at the target depth and measured using a variable flip angle sequence on a gadolinium-doped 20 mmol/l phenyl phosphonic acid phantom at the centre of the coil. Sixteen coronal phase-encoding steps yielded spectra from 10 mm slices (TR = heart rate, 192 averages, acquisition time approximately 20 min), using a trigger delay of 400 ms. A cosine apodisation filter was applied. The first spectrum containing signal beyond the chest wall and solely from cardiac tissue was selected. The spectrum was analysed using an AMARES time domain fit algorithm in jMRUI [23] and the ATP peak area was corrected for blood contamination [24].

Liver and visceral fat MRI

Liver fat was assessed by ¹H-magnetic resonance spectroscopy (TR/TR = 3,000 ms/35, 50, 75, 100, 125 or 150 ms, 3 × 3 × 3 × 3 cm voxel, SENSE torso array, one signal average and an expiration breath hold of 17 s). The water and CH₂ resonances were analysed using the AMARES algorithm in jMRUI [23]. The T2 relaxation times of the water and CH₂ resonances for each participant were calculated from the data by monoexponential fitting of signal intensity vs the six echo times, the signals at 35 ms were then corrected for T2 decay and the fat fraction was calculated as a percentage of the total signal from that volume [25].

Visceral fat was estimated at the L4–L5 junction [26] using a three-point Dixon sequence (TR = 50 ms, TE = 3.45, 4.60 or 5.75 ms, number of averages = 1, flip angle = 5°, voxel size 2.5 × 2.5 mm, slice thickness 10 mm, zero filled to 1.4 × 1.4 mm, and median FOV 440 mm [range 400–480 mm to suit participant size], with 70% phase FOV) [27]. Binary gating and a watershed algorithm was used to divide the binary image into distinct areas (Fig. 3); this allowed easy separation and quantification of the subcutaneous and visceral fat areas using ImageJ [28]. Liver and visceral fat analyses were performed by a single observer blinded to group allocation.

Open image in new window

Glycaemic control, blood variables and body composition

After an 8 h minimum overnight fast, a 75 g OGTT was performed in which samples were drawn every 15 min and analysed for whole blood glucose (YSI 2300 Stat Plus-D, Yellow Springs Instruments, Yellow Springs, OH, USA) and plasma insulin (Mercodia Iso-Insulin ELISA, no. 10-1128-01, Mercodia, Uppsala, Sweden). Insulin resistance and beta cell function were predicted using the HOMA2 [29], and the area under the glucose curve (AUGC) was calculated using the trapezoidal rule [30]. Fasting plasma samples were analysed in an accredited clinical pathology laboratory (Department of Clinical Biochemistry, Newcastle upon Tyne Hospital NHS Foundation Trust) for alanine phosphatase (ALP), alanine transaminase (ALT), aspartate aminotransferase (AST), total cholesterol, triacylglycerols and HbA_1c. Total cholesterol, triacylglycerols, ALP, AST and ALT were measured using a Roche P800 Modular Analyzer (Basel, Switzerland) and HbA_1c was measured using a TOSOH HLC-723G8 (Minato, Tokyo, Japan). Body composition was measured using air displacement plethysmography (BodPod, Life Measurement, CA, USA).

Intervention

The HIIT group performed 36-cycle ergometry sessions over 12 weeks (3 sessions per week on non-consecutive days) at a local gym. Patients were required to perform at least 32 sessions (89% of total sessions) for inclusion in the analysis. Intensity was based on the 6–20 point Borg Rating of Perceived Exertion (RPE) [31]. Each session included a 5 min warm up in which participants would progress from RPE 9 to 13 (‘very light’ to ‘somewhat hard’), followed by five intervals, each with a pedal cadence of >80 rev/min, reaching a RPE 16–17 (‘very hard’). The final interval was then followed by a 3 min recovery cycle. Intervals lasted 2 min in week 1 and progressed by 10 s increments each week such that week 12 consisted of 3 min 50 s intervals. Three min recovery periods interspersed each interval, which consisted of 90 s passive recovery, 60 s of band-resisted upper body exercise and 30 s to prepare for the subsequent interval. The arm resistance bands (Bodymax Fitness, Clydebank, UK) were used as light recovery and involved one exercise per recovery period in the following order: face pull, horizontal push, horizontal pull and 30° push. The initial session was supervised; thereafter, participants were guided through each session by voice-recorded instructions using an iPod shuffle (Apple, CA, USA). An exercise diary was completed to monitor exercise adherence.

Apart from HIIT sessions, all study participants were instructed to continue their normal routine and care for 12 weeks and not to change medication, habitual physical activity, diet or body weight. Weekly phone calls were made to assess adherence, and habitual physical activity was assessed over 7 days pre- and post-intervention using a validated multisensory armband (Sensewear; Bodymedia, Pittsburgh, PA, USA) [32].

Statistical analysis

Statistical power was based on the change in HbA_1c. We selected a sample size of 12 to provide a statistical power of 80% to detect a difference of 0.6% in HbA_1c [33]. A sample size of 14 was used to allow for two dropouts per group. A per-protocol analysis was adopted because the intention of this study was to assess efficacy and mechanisms of change, not effectiveness. All analyses were performed using IBM SPSS Statistics software (version 19, NY, USA) and data are presented as means ± SD unless otherwise stated. Continuous data were tested for normality using the Shapiro–Wilk test. Comparisons of key baseline variables were made using independent sample t tests. Between-group comparisons were made using ANCOVA with the baseline value as the covariate. Within-group changes were assessed by paired-sample t tests or the non-parametric alternative (Wilcoxon signed-rank test) for non-normally distributed data. Adjustment for multiple comparisons was not made because of co-linearity between variables, hypothesis-driven comparisons and the increased risk of type II error following adjustment [34]. Pearson’s correlation or the non-parametric alternative (Spearman’s rank correlation) was used to calculate r values among body composition, metabolic and cardiac variables. P values <0.05 were considered statistically significant.