Cardiorespiratory Fitness and Insulin Sensitivity in Overweight or Obese Subjects May Be Linked Through Intrahepatic Lipid Content

We investigated 138 overweight/obese but otherwise healthy volunteers (107 women and 31 men) in our study. All subjects completed a comprehensive medical evaluation including a dietary record for 7 consecutive days before study participation. Volunteers reported <2 h of physical activity per week and were not taking medications that could affect metabolism or liver function. Subjects consuming >20 g/day of alcohol, with diagnosis of type 2 diabetes, with acute or chronic infections, with any diseases that required treatment, or with known drug abuse were excluded. Our institutional review board approved the study, and written informed consent was obtained before entry. Volunteers were advised to continue their current physical activity level and lifestyle throughout the study.

Volunteers visited the laboratory on 2 separate days for anthropometric, metabolic, and cardiovascular evaluations and exercise testing. After a 10- to 12-h overnight fast, we measured body weight, height, and waist circumference in a standardized fashion. After a resting period of at least 5 min in the seated position, we determined blood pressure and heart rate with an automated blood pressure cuff (Dinamap; Critikon, Tampa, FL). During an oral glucose load (75 g glucose/500 ml), we obtained blood samples at baseline and 15, 30, 45, 60, 90, and 120 min after glucose ingestion to measure glucose and insulin. We assessed lean body and fat mass by bioimpedance analysis (BIA 5 series; Denner, Feldmeilen, Switzerland). Glucose, insulin, and blood lipids were measured by standard laboratory procedures in a certified clinical chemistry laboratory. During the second visit, we quantified visceral and subcutaneous abdominal fat mass and IHL and intramyocellular lipid (IMCL) content by magnetic resonance tomography and spectroscopy. CRF was determined during an exhaustive incremental exercise test on a cycle ergometer. We assessed nutrition including the kind and amount of alcohol consumption using standardized protocols over a 7-day period. On the basis of these records, we estimated the average amount of alcohol consumption in grams per day.

Abdominal fat imaging and fat spectroscopy of the lower leg musculature were performed on a 1.5-tesla magnetic resonance scanner (Magnetom Avanto and Sonata; Siemens Medical Solutions, Erlangen, Germany). An axial T1-weighted and water-suppressed gradient echo technique (TR 80, TE 6.11, FA 80, FOV 500 × 500 mm, 512 × 512 matrix, slice thickness 10 mm, interslice gap 10 mm) was applied for abdominal imaging. During repetitive breath-holds, consecutive datasets were acquired, covering the abdomen from the diaphragm to symphysis. Images were analyzed for the amount of VAT and subcutaneous adipose tissue (SAT) by semiautomated image segmentation software that employs a contour-following algorithm (Vitom, University Duisburg-Essen, Germany). Total abdominal adipose tissue was calculated as the sum of VAT and SAT. IHL was assessed by ¹H magnetic resonance spectroscopy of a single voxel located in segment 7 of the liver (TR 7,000, TE 30, voxel size 30 × 30 × 20 mm, number of averages 24, acquisition during repetitive breath-holds). A fat-to-water ratio was calculated after postprocessing of the spectra including a standard line-fitting procedure and integration of the hepatic triglycerides from 0.5 to 2.8 ppm. The unit of measurement is the ratio of the signal from fat (f) to total signal from fat (f) and water (w) (f/[f + w]%). Furthermore, IMCL content was quantified by ¹H single-voxel spectroscopy of the tibialis anterior muscle. High-resolution T1 weighted spin-echo images of the calf allowed identification of an appropriate voxel position in the tibialis anterior, thereby minimizing spectral contamination from extramyocellular lipids from adipose tissue attached to the muscular fasciae. A spin-echo single-voxel spectroscopy sequence (TR 3,000, TE 30, voxel size 11 × 11 × 20 mm, number of averages 64) with frequency-selective water suppression was applied. After baseline and constant phase correction, a standard line-fitting procedure using prior knowledge about the resonance peaks was performed for quantification. IMCL values were calculated as the area under the curve of the IMCL methylene line normalized to the creatine–CH₃ signal and corrected for differences in T1 and T2, resulting in a dimensionless value.

Subjects underwent a stepwise incremental exercise test on a bicycle ergometer (VIAsprint 150P; Ergoline, Bitz, Germany) until volitional exhaustion. Exercise was performed in a temperature-controlled room (21–22°C) ∼2 h after subjects had ingested a standardized breakfast (containing ∼520 kcal: 24% fat, 68% carbohydrate, and 8% protein). Alcohol and caffeine were not permitted 48 h before the exercise test. After 3 min in the seated position, resting measurements were recorded. Exercise was then started at a workload of 25 W. Workload was increased every 2 min by 25 W until the subjects could not maintain the requested 60 rpm pedal frequency. We monitored gas exchange continuously during the test to assess oxygen uptake and power output. Using an open spirometric system (Vmax Spectra model 229D analyzer; SensorMedics, Yorba Linda, CA), the time course of oxygen uptake and carbon dioxide production was recorded breath-by-breath and averaged in 10-s intervals. Heart rate was recorded by an electrocardiogram (GE Medical Systems, Waukesha, WI) throughout the exercise test. We assumed that subjects had reached maximal oxygen uptake (Vo_2max) when at least two of the following criteria were met: 1) respiratory exchange ratio > 1.10, 2) Vo₂ leveling off despite increase in power output, and 3) heart rate within 10 beats × min⁻¹ of the predicted maximum heart rate. To consider the individual differences in body weight, oxygen uptake was expressed as kilograms of body weight (Vo₂: ml⁻¹ × min⁻¹ × kg⁻¹).

Glucose (millimoles per liter), insulin (microunits per milliliter), and alanine aminotransferase (units per liter) were determined by standard methods in a certified clinical chemistry laboratory. Insulin resistance was estimated by homeostasis model assessment index (HOMA) derived from fasting glucose and insulin concentrations. HOMA was calculated from fasting insulin and glucose by (insulin [microunits per milliliter] × glucose [millimoles per liter])/22.5) (22). Areas under the curve for insulin (AUC_INS) and glucose (AUC_GLU) were assessed using the trapezoid method. Insulin sensitivity was calculated by the composite insulin-sensitivity index (C-ISI) (23): C-ISI = 10,000/√[(FPG × FPI) × (G × I)], where FPG and FPI are fasting plasma glucose (milligrams per deciliter) and fasting plasma insulin (microunits per milliliter), respectively, and G (milligrams per deciliter) and I microunits per milliliter) are the mean glucose and mean insulin concentration during the 2-h oral glucose tolerance test, respectively. Serum high sensitivity C-reactive protein (hs-CRP) (in micrograms per milliliter), total adiponectin as well as low-, middle-, and high-molecular-weight adiponectin multimers (all in micrograms per milliliter), fetuin-A (in nanograms per milliliter), and retinol-binding protein 4 (RBP4) (in micrograms per milliliter) were measured by sandwich enzyme-linked immunosorbent assay with the following characteristics: hs-CRP (RH961CRP01HR; BioVendor, Heidelberg, Germany), intra-assay coefficient of variation 3.8% and interassay coefficient of variation 5.2%; adiponectin (47-ADPHU-E01; ALPCO Immunoassays, Salem, NH), intra-assay coefficient of variation between 5.1 and 9.8% for the different multimers and interassay coefficient of variation between 4.8 and 6.5%; fetuin-A (RD191037100; BioVendor), intra-assay coefficient of variation 4.9% and interassay coefficient of variation 5.7%; and RBP4 (AG-45A-0011EK-KI01; AdipoGen, Seoul, Korea#), intraassay coefficient of variation 3.9% and interassay coefficient of variation 8.1%.

Data were first tested for normal distribution and variance homogeneity with Kolmogorov-Smirnov test and the Levene test, respectively. Pearson correlation coefficients were used to determine the relationship between CRF and insulin sensitivity. Partial correlations were used to control for total body fat mass, VAT, SAT, and intrahepatic lipid content. To specify the effect of CRF on insulin sensitivity, anthropometric and metabolic risk marker subjects were categorized into CRF tertiles (Vo_2max). Then a one-way ANOVA was performed to examine differences in subject characteristics across fitness levels. When the ANOVA result was significant, a Tukey post hoc comparison test was used to identify specific between-group differences. To further analyze influence of fat redistribution, we subgrouped subjects among CRF tertiles on the basis of sex-specific VAT levels (<2.9 or ≥2.9 kg for men and <1.5 or ≥1.5 kg for women), IHL content (<7.4 or ≥7.4% for men and <4.6% or ≥4.6% for women), and total body fat mass (<29.8 or ≥29.8 kg for men and <33.6 or ≥33.6 kg for women). Finally, a stepwise multivariate regression analysis was performed to identify predictors of insulin sensitivity. All statistical analyses were performed with SPSS 16 (SPSS, Chicago, IL). Significance was accepted at P < 0.05. Values are given as means ± SD.

Of 213 screened subjects, 172 subjects met the inclusion criteria of our study. Fourteen women and 2 men smoked regularly between 5 and 20 cigarettes per day. Nineteen women and 2 men did not meet the criteria for valid Vo_2max estimation. MR studies were unsuccessful in 11 subjects due to claustrophobia (n = 7), equipment failure (n = 1), waist circumference exceeding the magnetic resonance scanner limits table (n = 1), and poor magnetic resonance image quality (n = 2). Two subjects dropped out for personal reasons. Anthropometric and metabolic characteristics of the remaining 138 overweight and obese women and men, classified into CRF tertiles, are given in Table 1. Vo_2max was 25.4 ± 4.1 ml/min/kg in men and 21.1 ± 3.5 ml/min/kg in women (P < 0.05). Furthermore, VAT mass and IHL content were greater in men (VAT 3.49 ± 1.06 vs. 1.63 ± 0.79 kg, P < 0.01; IHL 11.63 ± 4.7 vs. 7.9 ± 3.6%, P < 0.05), whereas insulin sensitivity was lower than in women (C-ISI 4.8 ± 2.7 vs. 6.1 ± 2.4, P < 0.05; HOMA 2.1 ± 1.1 vs. 1.6 ± 1.2, P = 0.69). Alanine aminotransferase was higher in men (21.7 ± 8.1 vs. 36.1 ± 11.2 units/l, P > 0.05). Average alcohol consumption was 6.6 ± 4.8 g/day in women and 9.5 ± 5.2 g/day in men.

Table 2 shows Pearson correlation coefficients between CRF and selected cardiometabolic risk markers for women and for men before and after adjustment for either VAT or IHL. CRF correlated significantly negatively with BMI, percent body fat, systolic blood pressure, subcutaneous fat mass, and total abdominal fat mass. Diastolic blood pressure correlated significantly negatively to CRF in women only. In men, VAT tended to inversely correlate with CRF. In women, we observed a trend for an inverse correlation between IHL content and CRF. The inverse correlation between IHL and CRF in men was mainly driven by the particularly low IHL content in the highest fitness tertile (Table 1).

For all subjects, alanine aminotransferase was related to BMI (r = 0.27, P < 0.01), IHL (r = 0.47, P < 0.001), VAT (r = 0.36, P < 0.01), fetuin-A (r = 0.20, P < 0.05), and C-ISI (r = −0.32, P < 0.001) and tended to do so for CRF (r = −0.17, P = 0.076). Critical serum parameters (fetuin-A, adiponectin, hs-CRP, and RBP4) and their relation to adiposity and fat distribution are given in Table 4. Insulin sensitivity was positively related to total (r = 0.36, P < 0.01) and high-molecular-weight adiponectin (r = 0.37, P < 0.01) and negatively related to fetuin-A (r = −0.19, P < 0.05) and hs-CRP (r = −0.20, P < 0.05) but not to RBP4 (r = 0.04). Vo_2max showed a correlation to hs-CRP (r = −0.23, P < 0.05) but not to fetuin-A, adiponectin multimers, or RBP4. Alcohol consumption (inclusion criteria <20 g/day) showed no significant relation to IHL content (r = −0.05) or measures of liver function (alanine aminotransferase r = 0.12).

In partial correlation analysis for both sexes, we observed that the direct, sex-independent relationship between CRF and insulin sensitivity (r = 0.32, P < 0.05) was abolished after controlling for IHL (r = 0.14, NS), shown in Fig. 1, whereas for the adjustment of percent body fat (r = 0.28, P < 0.01), VAT (r = 0.25, P < 0.05), and SAT (r = 0.29, P < 0.01), the correlation between CRF and insulin sensitivity was attenuated but remained significant. Adjustment for alanine aminotransferase (r = 0.19, NS) and hs-CRP (r = 0.17, NS) also abolished the direct correlation between CRF and insulin sensitivity.

Figure 2 presents C-ISI values in women and in men classified into sex-specific CRF tertiles and then further subdivided into groups with high versus low VAT mass, IHL content, and body fat mass based on the sex-specific median for each measurement. For VAT levels, there was a significant difference in the lowest and moderate CRF tertile, with lower insulin sensitivity among subjects with high levels of VAT. However, in the high CRF tertile, there was no difference between the high versus low VAT subgroup. We observed a similar pattern between subgroups classified on the basis of total body fat mass. Whereas in the low and moderate CRF tertiles insulin sensitivity was lower among men and women in the subgroup with high body fat, there was no influence of body fat on the relationship between CRF and insulin sensitivity in the highest fitness tertile. The influence of VAT and body fat on insulin sensitivity across fitness levels was unchanged when we excluded men from the analysis. However, in subjects with high IHL content, insulin sensitivity was lower compared with that in subjects with low IHL content, irrespective of fitness level. The relationship remained significant when men were excluded from the analysis (P < 0.05).

The correlation of insulin sensitivity with percent body fat, VAT and SAT mass, and IHL content were similar between men and women (% body fat: women r = −0.19, NS, men r = −0.16, NS; VAT: women r = −0.46, P < 0.05, men r = −0.51, P < 0.05; SAT: women r = −0.21, NS, men r = −0.25, NS; IHL: women r = −0.44, P < 0.01, men r = −0.58, P < 0.01), whereas a correlation between C-ISI and IMCL was observed in women only (r = −0.29, P < 0.05).

To assess determinants of insulin sensitivity in more detail, we conducted a multivariate regression analysis with insulin sensitivity as the dependent variable. In women, the significant direct correlation for insulin sensitivity with visceral fat mass and IHL persisted even after inclusion of age, BMI, percent body fat, subcutaneous fat mass, triglycerides, and Vo_2max as covariables (Table 3). Also, Vo_2max (CRF) became a weak but significant predictive variable. The model that included age, BMI, percent body fat, visceral and subcutaneous fat mass, triglycerides, IHL, and Vo_2max as independent variables explained 41% of the variation in insulin sensitivity. In men, multivariate regression with the same variables included (Table 3) revealed IHL, visceral fat mass, and Vo_2max as significant predictors of insulin sensitivity and explained 45% of the variation in insulin sensitivity. When men and women were analyzed together, IHL (r² = −0.43, P < 0.01), VAT (r² = −0.34, P < 0.01), and CRF (r² = 0.29, P < 0.05) were observed as independent predictors of insulin sensitivity. IHL, VAT, and CRF explained 33 and 39% of the variation in insulin sensitivity in women and men, respectively (Table 3, model 1). When we excluded IHL from the multivariate regression model 5 (Table 3), the predictive power of CRF on insulin sensitivity was stronger, both in women (from r = 0.21, P < 0.05 to r = 0.34, P < 0.01) and in men (from r = 0.37, P < 0.01 to r = 0.43, P < 0.001).

The novel finding of our study is that the positive relationship between CRF and insulin sensitivity in overweight and obese subjects was no longer present after controlling for intrahepatic fat content. In contrast, the relationship between CRF and insulin sensitivity was largely unaffected after adjustment for total body, abdominal subcutaneous, abdominal visceral, and intramyocellular fat. Thus, our study suggests that the positive effect of increased CRF in overweight on insulin sensitivity to obese subjects may be mediated indirectly through IHL reduction. Moreover, high CRF is “protective” in patients with excessive visceral fat but does not negate the metabolic effect of increased IHL.

We obtained multislice whole-abdomen images providing an accurate estimate of visceral and subcutaneous fat. In previous studies, abdominal fat area was measured from a single-slice computed tomography scan at the level of the fourth or fifth vertebra (8,10,11,24,25) or the umbilicus (26). Moreover, we measured both IHL and IMCL in the same subjects. Thus, we were able to assess the individual contribution of fat distribution and organ fat on glucose metabolism in obese subjects.

Increased intrahepatic and intramyocellular lipids are associated with hyperinsulinemia, impaired glucose tolerance, and hepatic insulin resistance in diabetic and nondiabetic subjects (9,13,14). Recent studies reported that intrahepatic fat has a stronger impact on insulin sensitivity than VAT (27,–29). Furthermore, the detrimental influence of intrahepatic lipid accumulation on metabolic function appears to be independent of VAT mass (27). In stepwise regression analysis, IHL content was a stronger predictor of insulin sensitivity than visceral fat mass or IMCL content, independent of sex, age, or BMI, underscoring the central role of the liver in the pathogenesis of obesity-associated metabolic disease. Furthermore, the finding suggests IHL as a prime candidate for explaining the relationship between CRF and glucose metabolism.

In our study, individuals with high CRF were more insulin sensitive than less fit individuals. Similarly, previous studies identified low CRF as a strong and independent predictor of incident metabolic syndrome. Moderate to high CRF lowered the risk of all-cause and cardiovascular mortality independent of BMI (1,2,30). These studies focused on VAT accumulation when evaluating the relationship between CRF levels and metabolic risk markers. Some investigators observed an influence of VAT on the association of CRF with metabolic risk (10,25). We and others failed to show such a relationship (8). Differences in adiposity and sex distribution among studies could contribute to the discrepancy. Men in the latter study were older, and the prevalence of obesity was higher than that for men in the study by Arsenault et al. (10). Only one study quantified VAT in overweight and obese women to assess mechanistic links between CRF and metabolism (25). Yet, obese women are prone to develop metabolic disease (20,31). Given the discrepancy among studies, we suggest that VAT may not be the crucial factor linking CRF and metabolism.

The positive relationship between CRF and insulin sensitivity in our study was no longer significant after controlling for intrahepatic fat content or alanine aminotransferase, as an indicator of liver function (32). This finding suggests that the relationship between insulin sensitivity and CRF in overweight to obese subjects is mediated by IHL content rather than total body, abdominal, or intramyocellular fat accumulation. Moreover, in subjects in the highest CRF tertile, insulin sensitivity was unaffected by VAT mass or total body fat. In contrast, insulin sensitivity was consistently impaired in subjects with increased IHL regardless of physical fitness. Therefore, the beneficial effect of high CRF on insulin sensitivity appears to be limited to individuals with low intrahepatic fat content.

Potential mechanisms linking CRF and IHL include factors regulating hepatic lipid oxidation (33,,–36). Substrate oxidation is tightly coupled to mitochondrial oxidative capacity (37,38). Mitochondria occupy ∼18% of the liver cell volume (39). Mitochondrial function, a strong determinant of fitness, (40) could conceivably affect hepatic lipid oxidation. In fact, variation in the genes encoding peroxisome proliferated–activated receptor (PPAR) δ, PPAR coactivator 1α, and PPARγ affects mitochondrial function, responsiveness to physical training (33), and liver fat content (41). Very recent experiments in rats showed that low aerobic fitness causes reduced hepatic mitochondrial oxidative capacity, which increased susceptibility to hepatic steatosis and liver injury (42). Plasma biomarker analysis revealed that hs-CRP could also contribute, at least partly, to the observed association between CRF, insulin sensitivity, and IHL accumulation. hs-CRP is elevated in liver disease and predicts the incidence of type 2 diabetes in humans (43,44). Our findings underscore the importance of IHL in obesity-associated insulin resistance and type 2 diabetes (12,45).

The strength of the relationship between fat distribution, CRF, and insulin sensitivity differs between men and women. The sex difference may be explained in part by relatively low CRF (46) and IHL content in our women.

In conclusion, our study suggests mechanisms through which CRF improves cardiovascular and metabolic risk factors independently of body weight. The interaction between CRF and insulin sensitivity seems to be mediated by hepatic lipid content rather than the amount of total, visceral, subcutaneous, or intramyocellular fat. Regular physical activity improves whole-body and abdominal fat mass as well as IHL accumulation and insulin sensitivity (47,–49). Physical fitness noticeably improves metabolic risk and contributes independently to metabolic health even without body fat reduction (8,47,50). Optimization of nutritional, exercise, and pharmacological interventions such that lipid mobilization from the liver is maximized may be particularly beneficial in terms of metabolic risk reduction. Our study provides a strong impetus to test this hypothesis in prospective studies.

This study was part of a joint project between metanomics (Berlin, Germany) and Charité University Medical School that was supported by the Federal Ministry of Education and Research (BMBF-0313868) and in part by the Commission of the European Communities (Collaborative Project ADAPT, contract HEALTH-F2-2008-201100) and the German Obesity Network of Competence (Collaborative Project ADIPOSETARGET, 01 Gl0830 to S.E. and J.J.).

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

We thank Gritt Stoffels, Anke Strauss, and Elke Nickel-Sczcech for expert technical help with patient recruitment and study procedures. Furthermore, we thank Andreas Busjahn for statistical advice.