Predictors of Whole-Body Insulin Sensitivity Across Ages and Adiposity in Adult Humans

Abstract

Context:

Numerous factors are purported to influence insulin sensitivity including age, adiposity, mitochondrial function, and physical fitness. Univariate associations cannot address the complexity of insulin resistance or the interrelationship among potential determinants.

Objective:

The objective of the study was to identify significant independent predictors of insulin sensitivity across a range of age and adiposity in humans.

Design, Setting, and Participants:

Peripheral and hepatic insulin sensitivity were measured by two stage hyperinsulinemic-euglycemic clamps in 116 men and women (aged 19–78 y). Insulin-stimulated glucose disposal, the suppression of endogenous glucose production during hyperinsulinemia, and homeostatic model assessment of insulin resistance were tested for associations with 11 potential predictors. Abdominal subcutaneous fat, visceral fat (AF_VISC), intrahepatic lipid, and intramyocellular lipid (IMCL) were quantified by magnetic resonance imaging and spectroscopy. Skeletal muscle mitochondrial respiratory capacity (state 3), coupling efficiency, and reactive oxygen species production were evaluated from muscle biopsies. Aerobic fitness was measured from whole-body maximum oxygen uptake (VO₂ peak), and metabolic flexibility was determined using indirect calorimetry.

Results:

Multiple regression analysis revealed that AF_VISC (P < .0001) and intrahepatic lipid (P = .002) were independent negative predictors of peripheral insulin sensitivity, whereas VO₂ peak (P = .0007) and IMCL (P = .023) were positive predictors. Mitochondrial capacity and efficiency were not independent determinants of peripheral insulin sensitivity. The suppression of endogenous glucose production during hyperinsulinemia model of hepatic insulin sensitivity revealed percentage fat (P < .0001) and AF_VISC (P = .001) as significant negative predictors. Modeling homeostatic model assessment of insulin resistance identified AF_VISC (P < .0001), VO₂ peak (P = .001), and IMCL (P = .01) as independent predictors.

Conclusion:

The reduction in insulin sensitivity observed with aging is driven primarily by age-related changes in the content and distribution of adipose tissue and is independent of muscle mitochondrial function or chronological age.

Multiple factors are intertwined in the decline of insulin sensitivity in humans including age, physical inactivity, and obesity (1–6). There remains uncertainty about the causes and inevitability of age-related changes in insulin sensitivity. Leading hypotheses implicate mitochondrial dysfunction (4), oxidative stress (7), inflammation (8), and lipotoxicity (2) in the pathophysiology of insulin resistance. An accumulating body of evidence links age-related decline in insulin sensitivity to adiposity and reduced physical activity levels, rather than age per se (9–14). Indeed, we recently reported that insulin sensitivity did not differ between lean young and older adults who were similar in levels of adiposity and fitness (15). In a cohort of young and older individuals, body fat rather than age was associated with reduced insulin sensitivity (16). Additionally, regional adipose tissue distribution (ie, sc vs visceral) is purported to strongly influence insulin sensitivity (17, 18), whereas older adults exhibit increased visceral fat commensurate with insulin resistance (5). Ectopic fat accumulation in insulin sensitive tissues such as liver and skeletal muscle is also implicated in the etiology of insulin resistance (19, 20). It is therefore important to consider age-related changes in intrahepatic and intramyocellular lipid when evaluating the influence of aging on insulin sensitivity.

Mitochondria have also emerged as organelles whose function or dysfunction may influence insulin sensitivity (21). Mitochondrial content and function in skeletal muscle decline with aging (3), obesity (22), and type 2 diabetes (23). Furthermore, there are reported associations between insulin sensitivity and muscle mitochondrial function in older adults (3, 4), obesity (22), type 2 diabetes (23), women with polycystic ovary syndrome (24), and offspring of type 2 diabetes patients (25). These observations support the hypothesis that reduced mitochondrial function contributes to the development of insulin resistance (4, 25). This association is important to consider in the context of aging, particularly because we previously demonstrated that reduced maximal mitochondrial ATP production with aging can be largely prevented with exercise (15).

The etiology of insulin resistance is complex, and the precise mechanisms continue to be explored. Our understanding of the impact of aging on insulin sensitivity is complicated by numerous factors that change concurrently with age and preclude simple univariate association studies. The goal of this study was to determine specific factors that best predict insulin sensitivity when accounting for interdependent and parallel effects. Specifically, we sought to determine whether age is a significant predictor of insulin sensitivity, while evaluating the extent to which whole-body adiposity, regional and ectopic fat distribution, mitochondrial function, and overall aerobic fitness contribute to age-related insulin resistance. An array of gold standard techniques was implemented to assess these variables across a range of ages and adiposities in a relatively large cohort of people (n > 100). Specifically, insulin sensitivity was assessed using pancreatic hyperinsulinemic-euglycemic clamps and was modeled by multiple regression analysis.

Materials and Methods

Participants

Data were collected from three clinical studies (numbers NCT01686568, NCT01497106, and NCT01477164) conducted at the Mayo Clinic, Rochester, between 2010 and 2015. A portion of these data has been published previously (26, 27). A total of 116 nondiabetic participants (19–78 y [54% young, <45 y; 23% middle aged, 45–65 y; and 23% elderly, ≥65 y] and body mass index [BMI; 20–44 kg/m²]) were included in this analysis. All studies were conducted in accordance with the Declaration of Helsinki. Participants provided written informed consent as approved by the Mayo Foundation Institutional Review Board. They were screened by medical history, physical examination, resting electrocardiogram, and blood tests of hematological, hepatic, renal, and metabolic function. Those with diagnosed diabetes, cardiovascular or pulmonary disease, uncontrolled hypothyroidism, chronic inflammatory disease and cancer, or a history of smoking and alcohol or other substance abuse were excluded. Participants were excluded if they used β-blockers. Only sedentary adults participated in the studies and were excluded if they reported exercising more than 2 days per week (>30 min/d). The list of the most representative medications across participants is shown in Supplemental Table 1.

Body composition

Dual-energy X-ray absorptiometry (Lunar DPX-L; Lunar Radiation) was used to measure fat mass, fat-free mass, and percentage body fat (% fat). Subcutaneous abdominal fat (AF_SQ) and visceral abdominal fat (AF_VISC) were measured by magnetic resonance imaging. Serial T1-weighted axial images were acquired through the abdominal region between pelvis and the upper end of the diaphragm during a breath-hold phase. A single trained analyst evaluated the fat distribution within each image using Analyze Software System (Mayo Clinic Biomedical Imaging Resource). Intramyocellular (IMCL) and intrahepatic (IHL) lipid were measured by proton magnetic resonance spectroscopy (¹H-MRS). For skeletal muscle, localized ¹H spectra were acquired from a 10 × 10 × 10 mm³ voxel within the tibialis anterior muscle. The methylene signal at 1.5 ppm was normalized to creatine. Liver ¹H signals were acquired using a short TE STEAM sequence with and without water suppression from a 25 × 25 × 25 mm³ voxel, avoiding large vessels and extrahepatic fat. The proton signal from the liver fat was normalized to the water signal.

Mitochondrial function

Participants were admitted to the Clinical Research Unit of Mayo Clinic Hospital after 3 days of a weight-maintenance diet (20% protein, 50% carbohydrates, and 30% fat). After an overnight fast, skeletal muscle biopsy samples were collected percutaneously from the vastus lateralis muscle under local anesthesia (2% lidocaine) as previously reported (15, 26). Approximately 60 mg of tissue was immediately used for high-resolution respirometry of isolated mitochondria (Oxygraph 2K; Oroboros Instruments), as described previously (28). Muscle tissue was homogenized on ice, and mitochondria were isolated by differential centrifugation and suspended in a respiration buffer (MiR05; Oroboros). Mitochondrial oxidative capacity (state 3) was determined from oxygen consumption rates in the presence of 10 mM glutamate, 2 mM malate, 10 mM succinate, and 2.5 mM ADP. State 3 respiration was normalized to tissue wet weight. Mitochondrial coupling efficiency was measured from the respiratory control ratio (RCR) calculated by state 3/state 4 respiration. Hydrogen peroxide (H₂O₂) production in isolated mitochondria was measured with a spectrofluorometer (HORIBA Jobin Yvon) by continuously monitoring oxidation of Amplex Red under state 4 conditions with oligomycin to inhibit ATP synthase (28). Rates of H₂O₂ reactive oxygen species (ROS) production were normalized to the protein content of the mitochondrial isolate.

Whole-body maximum oxygen uptake (VO₂ peak) was measured by indirect calorimetry on a stationary cycle ergometer with an incremental workload. Electrocardiogram was continuously monitored, blood pressure was measured every 2 minutes, and the subjective level of exhaustion was assessed using the Borg scale. To ensure that subjects achieved maximal oxygen uptake (VO₂ peak), at least two of the following criteria were confirmed: a plateau in oxygen uptake despite increasing exercise intensity, respiratory exchange ratio of 1.1 or greater, and attainment of heart rate within 10 bpm of age-predicted maximum. The VO₂ peak was expressed relative to total body weight.

Insulin sensitivity

A pancreatic hyperinsulinemic-euglycemic clamp was performed at least 7 days after the VO₂ peak test. Participants were admitted to the Clinical Research Unit of Mayo Clinic Hospital after 3 days of a weight-maintenance diet (20% protein, 50% carbohydrates, and 30% fat). Participants consumed no calories after 7:00 pm, except for water, to achieve a 10-hour fast. A retrograde catheter was inserted into a dorsal hand vein and the hand was kept in a heated box (130°F) during blood sample collection. Two more iv catheters were placed in the contralateral arm for infusions. A two-stage euglycemic-hyperinsulinemic clamp was performed for 6 hours. Three hours prior to start of hormone infusion, a primed (6 mg/kg fat free mass [FFM]) infusion of [6,6-²H₂]-D-glucose was administered (at 4 mg/kg FFM · h). For the first 3 hours, regular insulin was infused at 0.62 mU/kg FFM · min and then at a higher rate of 2.3 mU/kg FFM · min for the next 3 hours. For the entire 6 hours, somatostatin (60 ng/kg FFM · min), glucagon (0.65 ng/kg FFM · min) and human GH (3 ng/kg FFM · min) were infused. An infusion of 40% dextrose was administered to maintain euglycemia (90 mg/dL [5.0 mmol/L]) and was adjusted accordingly to blood samples taken at 10-minute intervals with a Beckman glucose analyzer. The dextrose infusion was enriched with 2% [6,6-²H₂]-D-glucose to minimize changes in glucose enrichment and maintain constant plasma enrichment (29).

Mass spectrometry was used to measure [6,6-²H₂]-D-glucose enrichment in plasma samples and in infusates to calculate glucose rates of appearance (Ra) and disappearance. Steele's steady state equation was implemented for use with stable isotope tracers: Ra = glucose rate of disappearance = F_Glu/E_P, where F_Glu is the tracer infusion rate and E_P is the plasma tracer enrichment (30). Peripheral insulin sensitivity was assessd from the glucose infusion rate (GIR) required to maintain euglycemia. Endogenous glucose production (EGP) was calculated from the difference of total Ra and exogenous GIR during the first step of the clamp (29). Because the liver is the main source of endogenous glucose production, with renal glucose production contributing less than 10% in the postabsorptive state, the percentage of EGP suppression during hyperinsulinemia (EGP_SUP) was used as a measure of hepatic insulin sensitivity. For the calculation of EGPSUP, EGP was normalized to insulin concentration levels for both the basal and first insulin stage of the clamp (EGP/insulin), to account for the fact that minor changes in plasma insulin concentration cause appreciable alterations in hepatic insulin action (31). Indirect calorimetry (Parvomedics TrueOne 2400 Canopy system) was used for measuring the energy expenditure and the respiratory quotient (RQ) over 20-minute intervals during the basal postabsorptive and the high insulin states. Metabolic flexibility was then calculated from the change in RQ from baseline to high insulin (ΔRQ). The homeostatic model assessment of insulin resistance index (HOMA-IR) was calculated according to the following formula: fasting glucose (millimoles per liter) × fasting insulin (microinternational units per milliliter)/22.5 (32).

Statistical analysis

All analyses were performed using JMP statistical software (SAS Institute). Spearman's rank correlation coefficients were used to assess univariate correlations between predictors of whole-body insulin sensitivity (GIR), hepatic insulin sensitivity (EGP_SUP), and HOMA-IR. Predictors and samples with greater than 20% missing data were dropped from the analysis. A total of 116 samples and 11 predictors remained (age, BMI, % fat, AF_SQ, AF_VISC, IHL, IMCL, VO₂ peak, state 3, RCR, and ROS). Univariate correlation results showed significant correlations (P < .05) between several of the remaining predictors. Hence, we chose a two-step approach for determining the GIR regression model. Prior to modeling, predictors with extreme skew were log transformed (including IHL and IMCL), and all predictors were centered and scaled. In the first step of modeling, a nonlinear iterative partial least squares method was used to identify predictors of importance. Nonlinear iterative partial least squares modeling was configured to use a restricted maximum likelihood method for imputing missing values. Predictors with variability importance in the projection scores of greater than 0.85 were considered as candidate predictors of GIR response. In the next step, a stepwise regression method was configured to generate all possible GIR linear regression models that contain up to eight of the candidate predictors. We used a minimalistic model that maximized the R² between predicted and measured GIR while minimizing the degree of multicollinearity between the model's constituent variables, which was assessed using variance inflation factor (VIF; >5 indicated multicollinearity). The same procedure was followed to generate the EGP_SUP and HOMA-IR model. ΔRQ was excluded from the multivariate analyses because there are not enough mechanistic data to support the role of metabolic flexibility as a determinant rather than an outcome of insulin sensitivity. For the EGP_SUP model of hepatic insulin sensitivity, skeletal muscle parameters (IMCL, state 3, RCR, and ROS) were excluded from the analyses.

Results

The descriptive statistics for all potential predictors (age, BMI, % fat, AF_SQ, AF_VISC, IHL, IMCL, VO₂ peak, state 3, RCR, ROS, and ΔRQ) are displayed in Table 1. Simple univariate associations among the individual variables are displayed in Table 2. For peripheral insulin sensitivity, there were significant univariate negative associations between GIR and measures of adiposity: BMI, % fat, AF_SQ, AF_VISC, and IHL but not IMCL. GIR was positively associated with VO₂ peak and ΔRQ, but no significant associations were detected with ROS production or mitochondrial coupling efficiency (RCR). There was a nonsignificant trend for a positive association between GIR and state 3 and a negative association with age, which also did not reach significance.

Table 1.

Descriptive Statistics

	n	Mean ± SEM	IQR
Physical characteristics
Age, y	116	43.5 ± 4.0	27.0–59.2
Height, cm	116	171.2 ± 0.98	164.4–178.8
Weight, kg	116	89.0 ± 2.1	72.8–102.4
Blood pressure systolic, mm Hg	116	123.7 ± 1.18	114–131
Blood pressure diastolic, mm Hg	116	73.8 ± 0.8	68–79
Total cholesterol, mg/dL	116	188 ± 3.4	159–214
LDL, mg/dL	116	107.4 ± 2.8	87–122.5
Triglycerides, mg/dL	116	123.4 ± 5.9	74–161
HDL, mg/dL	116	55.9 ± 1.6	45.5–62.3
Glucose, mg/dL	116	100 ± 0.81	94.5–105
Insulin, μIU/mL	116	7.3 ± 4.5	4.4–9.0
Insulin low stage, μIU/mL	102	18.9 ± 0.4	15.7–21.8
Body composition
BMI, kg/m2	116	30.2 ± 2.8	25.4–34.2
FFM, kg	116	50.7 ± 1.1	41.1–57.6
Body fat, %	116	39.5 ± 3.7	33.4–46.7
AFSQ, cm2	105	33 769 ± 3295	21 713–46 209
AFVisc, cm2	105	9252 ± 902.9	5044–11 932
IMCL, IMCL/Cr	104	3.6 ± 0.36	2.12–4.38
IHL (lipid/water)	106	0.06 ± 0.006	0.01–0.07
Mitochondrial function
State 3, pmol/sec · mg tissue	110	491 ± 46.8	391–566
RCR	110	7.1 ± 0.67	5.8–8.0
ROS production, nM H2O2 per sec per mg tissue	80	16.2 ± 1.8	11.9–19.5
VO2 peak, ml/kg BW per min	83	25.6 ± 2.8	19.9–31.5
ΔRQ	97	0.11 ± 0.01	0.05–0.13
Insulin sensitivity
GIR, μmol/kg FFM · min	114	11.8 ± 1.1	9.0–15.2
HOMA-IR	102	1.8 ± 0.18	1.1–2.2
EGP basal, μmol/kg FFM · min	102	15.3 ± 0.34	13.6–17.5
EGP low insulin, μmol/kg FFM · min	102	7.8 ± 0.45	4.4–11.2
EGPSUP normalized to insulin, %	102	78.6 ± 7.8	70.8–90.9

Abbreviation: BW, body weight; IQR, interquartile range. Values are means ± SD.

Table 2.

Univariate Correlations Among Variables

	Age	BMI	% Fat	AFSQ	AFVISC	IHL	IMCL	VO2 Peak	State 3	RCR	ROS	ΔRQ	HOMA-IR	EGPSUP	GIR
Age		0.076a	0.009a	0.491a	<0.0001a	<0.0001a	0.256a	<0.0001a	0.0008a	0.877a	0.703a	0.550a	0.549a	0.088a	0.071a
BMI	0.18		<0.0001a	<0.0001a	<0.0001a	<0.0001a	0.888a	<0.0001a	0.501a	0.850a	0.688a	<0.0001a	<0.0001a	<0.0001a	<0.0001a
% Fat	0.26	0.69		<0.0001a	<0.0001a	0.002a	0.365a	<0.0001a	0.031a	0.512a	0.900a	0.022a	<0.0001a	<0.0001a	0.001a
AFSQ	0.07	0.81	0.86		<0.0001a	0.012a	0.746a	<0.0001a	0.368a	0.199a	0.611a	0.003a	<0.0001a	<0.0001a	<0.0001a
AFVISC	0.59	0.66	0.44	0.42		<0.0001a	0.051a	<0.0001a	0.076a	0.669a	0.180a	<0.0001a	<0.0001a	0.004a	<0.0001a
IHL	0.40	0.54	0.31	0.26	0.70		0.371a	0.0004a	0.932a	0.017a	0.161a	<0.0001a	<0.0001a	0.032a	<0.0001a
IMCL	0.12	0.02	0.10	0.04	0.21	0.10		0.100a	0.452a	0.192a	0.636a	0.863a	0.977a	0.704a	0.883a
VO2 peak	−0.76	−0.52	−0.77	−0.53	−0.65	−0.40	−0.20		<0.0001a	0.282a	0.178a	0.058a	0.0019a	0.126a	0.005a
State 3	−0.33	−0.07	−0.22	−0.09	−0.18	−0.01	0.08	0.54		0.0009a	0.069a	0.059a	0.343a	0.558a	0.056a
RCR	−0.02	−0.02	−0.07	−0.13	−0.05	0.24	0.14	0.16	0.33		0.654a	0.605a	0.854a	0.555a	0.736a
ROS	−0.04	−0.05	−0.01	−0.06	−0.19	−0.16	0.06	0.16	0.21	−0.05		0.076a	0.348a	0.922a	0.877a
ΔRQ	−0.06	−0.43	−0.23	−0.31	−0.47	−0.44	0.02	0.22	0.19	0.05	0.20		<0.0001a	0.0004a	<0.0001a
HOMA-IR	0.06	0.69	0.40	0.52	0.62	0.53	−0.003	−0.35	−0.10	0.02	−0.11	−0.57		<0.0001a	<0.0001a
EGPSUP	0.17	−0.24	−0.47	−0.52	−0.29	−0.22	0.04	0.18	0.06	0.06	0.01	0.35	−0.70		<0.0001a
GIR	−0.18	−0.51	−0.32	−0.40	−0.63	−0.58	0.02	0.32	0.19	0.19	−0.02	0.60	−0.71	0.12

Unlettered values provide the corresponding Spearman's ρ. Bold values indicate a significant association.

^aValues contain the P values for the association between intersecting variables.

EGP_SUP strongly and inversely correlated with BMI, % fat, and AF_SQ, and had a moderate association with AF_VISC and IHL. No associations were observed among EGP_SUP and skeletal muscle-dependent variables, including IMCL, mitochondrial function (state 3, RCR, ROS) and whole-body VO₂ peak. The association of EGP_SUP with age was positive but nonsignificant (Table 2).

With increasing age, there was increased adiposity (% fat, AF_VISC, and IHL), decreased aerobic fitness (VO₂ peak), and decreased mitochondrial oxidative capacity (state 3). IHL showed positive associations with other measures of adiposity (BMI, % fat, AF_SQ, and AF_VISC), whereas IMCL did not correlate significantly with any of the examined variables. ΔRQ was not associated with age, but there were significant negative associations with adiposity (BMI, % fat, AF_SQ, AF_VISC, and IHL) and a trend for a positive association with state 3 and VO₂ peak (Table 2).

Three different multiple regression models were generated to identify independent predictors of insulin sensitivity. The first model included peripheral insulin sensitivity measured from GIR (Table 3). The model in Table 3 explained 57% of the variance in GIR and identified AF_VISC and IHL as independent negative predictors, whereas VO₂ peak, IMCL, and age were independent positive predictors. Although age was a significant positive predictor, it explained only 1% of the variance in GIR. Notably, no indices of mitochondrial function were found to be important independent predictors of GIR. The second model included hepatic insulin sensitivity for the dependent variable EGP_SUP (Table 4). The model in Table 4 explained 37% of the variability in EGP_SUP and identified % fat and AF_VISC as important negative predictors, and age as an important positive predictor of hepatic insulin sensitivity, albeit accounting for only 3% of the variance in hepatic insulin sensitivity. The third model included HOMA-IR (Table 5). The model in Table 5 explained 61% of the variance in HOMA-IR and identified AF_VISC as an independent positive predictor, whereas VO₂ peak, IMCL, and age were independent negative predictors. As with the previous two models, age accounted for a very small percentage (less than 1%) of the variance in HOMA-IR.

Table 3.

Multiple Linear Regression for the Dependent Variable GIR

Variable	Model R2 (Partial R2)	Estimate	SEE	VIF	P Value
Intercept	0.57	−0.021	0.062		.737
Age	(0.01)	0.410	0.087	1.91	<.0001
AFVISC	(0.46)	−0.480	0.092	2.16	<.0001
VO2peak	(0.11)	0.321	0.091	2.12	.0007
IHL	(0.11)	−0.277	0.086	1.87	.002
IMCL	(0.04)	0.147	0.064	1.03	.023

Abbreviation: SEE, SE of the estimate.

Table 4.

Multiple Linear Regression for the Dependent Variable EGP_SUP

Variable	Model R2 (Partial R2)	Estimate	SEE	VIF	P Value
Intercept	0.37	0.115	0.066		.084
Age	(0.03)	0.353	0.076	1.33	<.0001
% Fat	(0.29)	−0.358	0.071	1.17	<.0001
AFVISC	(0.11)	−0.261	0.079	1.43	.001

Abbreviation: SEE, SE of the estimate.

Table 5.

Multiple Linear Regression for the Dependent Variable HOMA-IR

Variable	Model R2 (Partial R2)	Estimate	SEE	VIF	P Value
Intercept	0.60	−0.085	0.045		.063
Age	(0.0006)	−0.418	0.067	0.27	<.0001
AFVISC	(0.55)	0.610	0.061	1.51	<.0001
VO2MAX	(0.11)	−0.231	0.070	2.42	.001
IMCL	(0.06)	−0.119	0.045	1.03	.011

Abbreviation: SEE, SE of the estimate.

Discussion

The key findings from the current study are that peripheral insulin sensitivity was negatively associated with visceral fat and intrahepatic lipid accumulation, whereas it was positively associated with aerobic fitness, and intramyocellular lipid content when all of these variables were simultaneously accounted for in the multiple regression model. Skeletal muscle mitochondrial function did not emerge as an independent predictor of peripheral insulin sensitivity when accounting for age, adiposity, and cardiorespiratory fitness. Hepatic insulin sensitivity was negatively associated with body fat and visceral fat. Intrahepatic lipid content was not an independent determinant of hepatic insulin sensitivity.

A major point of emphasis of the current study is that age did not emerge as a major independent predictor of insulin sensitivity in any of the three models when accounting for adiposity and fat distribution. Although there was a nonsignificant negative association of GIR with increasing age in the univariate analysis, this relationship was reversed in the multivariate model and accounted for a very small amount of the variance in GIR. The observation suggests that increasing age is not a major determinant of insulin sensitivity when accounting for the age-related changes in adiposity and aerobic fitness. This is in contrast with the common belief that insulin resistance is an obligatory consequence of aging (1, 4–6). An emerging body of literature supports the notion that adiposity increases with age and is a primary determinant of age-related insulin resistance (9–14). Indeed, we previously reported that when lean older adults were compared with young adults, peripheral and hepatic insulin sensitivity did not differ (15, 16). In agreement, other studies have also identified visceral fat as an independent explanatory variable of insulin sensitivity (33, 34). The current study included a larger cohort of participants across ages and adiposity to demonstrate that adiposity, specifically the distribution of fat in abdominal regions and tissues, are the primary determinants of insulin sensitivity rather than age per se. Nevertheless, it is important to recognize that, although a wide age range was included in the study, the mean age of participants was 43.5 years with an interquartile range of 27–59, indicating that individuals at the older end of the age spectrum may not be fully represented.

The deposition of fat in nonadipose tissues (ie, ectopic lipid) has also been linked with the development of insulin resistance. Insulin resistance is associated with IMCL content measured from muscle biopsies (35) and noninvasively by ¹H-MRS (36, 37). Krssak et al (37) found that IMCL content of the plantarflexor muscles was inversely correlated with insulin sensitivity (r = −0.579, P = .0037) as measured by hyperinsulinemic-euglycemic clamp and predicted insulin sensitivity in multiple regression analysis independently of other possible predictors, such as BMI or age. In the present study, we measured IMCL in the tibialis anterior muscle and found a positive predictive role of IMCL for peripheral insulin sensitivity (Table 3). We cannot exclude the possibility that the contrasting results may be related to the muscle group under investigation. Also, our interpretation may be limited by the fact that ¹H-MRS cannot identify the subcellular localization of IMCL, which is indicative of the fate of IMCL toward storage or β-oxidation in the mitochondria. There has been, however, accumulating evidence that lipid droplets in muscle appear to be benign to insulin sensitivity. Instead, the content of lipid intermediates (eg, diacylglycerol, ceramides, long chain acyl-CoAs, acyl carnitine intermediates) is what appears to disrupt insulin signaling in skeletal muscle (38, 39).

Another interesting finding was that mitochondrial function was not identified as an independent predictor of insulin sensitivity. Reduced mitochondrial abundance and oxidative capacity in skeletal muscle have been observed concurrently with decreased insulin sensitivity (3, 4, 22–24), prompting a hypothesis that mitochondria play an etiological role in the development of insulin resistance (25). Whereas some evidence supports this hypothesis, there are other examples of mitochondrial function being dissociated from insulin sensitivity (40, 41), casting some uncertainty on their precise relationship. Here we evaluated muscle mitochondrial oxidative capacity (state 3), mitochondrial coupling efficiency, and ROS production from skeletal muscle biopsies collected during a basal state (fasted and resting). Although muscle mitochondrial oxidative capacity was inversely associated with age and tended to be positively associated with GIR when examined in univariate relationships, none of the measurements of mitochondrial function emerged as independent predictors of peripheral insulin sensitivity. On the other hand, whole-body VO₂ peak was a strong predictor of peripheral insulin sensitivity, in perfect consonance with present literature (42, 43). Despite its strong negative interaction with age and adiposity indices, VO₂ peak contributed significantly to the total variance of the GIR model.

Metabolic flexibility refers to the ability to adjust fuel selection to prevailing nutritional conditions (44) and appears to be impaired in insulin resistant individuals in response to infusion of insulin and glucose (44, 45). The precise cellular determinants of metabolic flexibility remain a topic of investigation, but defects in mitochondrial metabolism and fuel oxidation have been suggested to mediate metabolic inflexibility. Here we found that the change in RQ from postabsorptive basal conditions to hyperinsulinemia (ΔRQ) was not significantly associated with mitochondrial oxidative capacity (Table 2). A strength of the study is that RQ was measured after a 3-day period during which all subjects were in energy balance and were fed standardized meals with similar macronutrient compositions. It is therefore unlikely that ΔRQ was influenced by variability in the food quotient. Metabolic flexibility was not included as a potential predictor in the multiple regression models because of the likelihood that it is a reflection of insulin sensitivity rather than a factor that influences insulin sensitivity.

Previous studies examined the effects of aging on whole body insulin sensitivity, but few have specifically evaluated the influence of aging on EGP representing largely hepatic insulin sensitivity (15, 46). A unique aspect of the current study is the pancreatic clamp to suppress endogenous insulin production and maintain basal concentrations of glucagon during the two-stage insulin clamp. Infusion of isotopically labeled glucose and control of circulating insulin levels allowed us to evaluate the extent to which EGP was suppressed in response to insulin infusion. Simple univariate analyses indicated that EGP_SUP was significantly and negatively associated with indices of adiposity, including BMI, % fat, AF_SQ, AF_VISC, and IHL but not with age (Table 2). Our findings with regard to the effects of aging are consistent with precedent literature, indicating that hepatic insulin sensitivity is not impaired with advancing age (15, 46). Of note, when these variables were combined in a multiple regression model, we found that age was a significant positive predictor of EGP_SUP when adjusting for adiposity but explained a very small fraction of the variance in hepatic insulin sensitivity (Table 4). This reinforces the concept that age-related changes in body composition are the most important determinants of age-related insulin resistance.

It is important to note that age, % fat, and AF_VISC explained only 37% of the variance in EGP_SUP. It is therefore clear that there are additional key determinants of hepatic insulin sensitivity that were not measured in the present study. This underscores the gap in current knowledge of understanding hepatic insulin sensitivity. Other variables including changes in circulating free fatty acids (47) and liver mitochondria metabolism (48, 49) would be of interest to study with regard to hepatic insulin sensitivity. Interestingly, IHL did not contribute to the final model of EGP_SUP, which challenges recent reports stating a causal relationship between liver fat and hepatic insulin action (50). Based on the univariate relationship and the multivariate model, we conclude that there is minimal if any correlation of IHL with hepatic insulin sensitivity. This counterintuitive finding may be a consequence of the strict inclusion and exclusion criteria. We excluded people with diabetes and other comorbidities, which likely excluded many individuals with hepatic steatosis that may have otherwise bolstered the relationship between IHL and EGP suppression. Another limitation of the study is that magnetic resonance spectroscopy measures IHL in a localized area and therefore does not take into account the heterogeneous distribution of intrahepatic lipid. The lack of specificity of ¹H-MRS could partially explain the discrepancy on this matter. However, similarly to IMCL, increased lipid metabolites, particularly diacylglycerol content, have been associated with hepatic insulin resistance rather than triglycerides (51). It is also of interest that IHL emerged as a significant predictor in the model of peripheral insulin sensitivity. Because there is no established direct relationship between IHL and muscle glucose disposal, it is possible that IHL may influence GIR through potential effects on hepatic glucose disposal during hyperinsulinemia.

A predictive model of HOMA-IR was included in the current study because it is a commonly used index of insulin sensitivity for large cohort studies in which gold standard measurements are impractical. The univariate associations revealed strong correlation of HOMA-IR with all variables of adiposity, metabolic flexibility, and aerobic fitness. Nevertheless, there was no association with skeletal muscle descriptive variables, including mitochondrial function and IMCL (Table 2). In the multivariate model, AF_VISC was a significant positive predictor, whereas age, VO₂ peak, and IMCL were negative predictors and explained 61% of the variance in HOMA-IR (Table 5) in this cohort of nondiabetic individuals.

Collectively this work highlights the importance of adiposity, especially visceral fat, and regional distribution of fat on whole-body insulin sensitivity and suggests that neither chronological age nor skeletal muscle mitochondrial function appears to be important independent determinants of insulin sensitivity.