Pioglitazone-Mediated Changes in Lipoprotein Particle Composition Are Predicted by Changes in Adiponectin Level in Type 2 Diabetes

Abstract

Context:

In animal and observational studies, adiponectin is associated with lipoprotein risk factors for cardiovascular disease.

Objective:

We analyzed data from a randomized clinical trial to evaluate the relationship between changes in adiponectin to changes in lipoprotein risk factors after an intervention that alters adiponectin levels.

Design and Setting:

Adiponectin levels were measured at baseline and follow-up, as were lipoprotein risk factors for cardiovascular disease, at academic medical centers and ambulatory care centers.

Patients and Other Participants:

Participants included 361 men and women with type 2 diabetes.

Intervention:

Intervention included randomization to treatment with glimepiride or pioglitazone for 72 wk.

Main Outcome Measure:

The relationship of treatment-related differences in adiponectin level to treatment-related differences in lipoprotein cardiovascular risk factors at 72 wk was evaluated.

Results:

Pioglitazone led to an increase in adiponectin compared with glimepiride. Compared with baseline, pioglitazone treatment at 72 wk led to an increase in low-density lipoprotein (LDL) and high-density lipoprotein (HDL) particle size and a decrease in very-low-density lipoprotein (VLDL) particle size and LDL particle number. Glimepiride treatment more modestly decreased LDL particle number and increased LDL particle size. At 72 wk, there were significant treatment group differences for HDL, LDL, and VLDL particle size, and triglyceride and HDL cholesterol level. The increase in adiponectin predicted treatment-related improvement for triglyceride and HDL cholesterol level and LDL and HDL particle size.

Conclusion:

Increased adiponectin contributed to treatment-related benefit in lipoprotein cardiovascular disease risk factors in obese diabetic subjects treated with pioglitazone. These results provide support for a model that mechanistically links changes in adiponectin level to changes in lipoprotein composition in humans.

Adiponectin is an abundant adipokine that plays an important role in glucose and lipid metabolism (1). Lower levels of adiponectin are frequently observed in association with obesity (1), metabolic disorders such as diabetes mellitus type 2 (DM2) (2), dyslipidemia, and cardiovascular disease (CVD) (1). Adiponectin also demonstrates antiinflammatory and antiatherogenic effects in some animal models (1, 3). In some cross-sectional and observational studies in humans, higher adiponectin levels predict lower CVD risk, and one longitudinal observational study suggested that most of the beneficial effects of adiponectin may be mediated by its association with a high-density lipoprotein (HDL) cholesterol level (4).

The CHICAGO (Carotid Intima-Media Thickness in Atherosclerosis Using Pioglitazone) study was a prospective, randomized interventional trial conducted in patients with DM2 that evaluated the impact of pioglitazone vs. glimepiride treatment on multiple cardiovascular risk endpoints (5). We have previously reported that treatment with pioglitazone suppressed atherosclerosis progression and resulted in significant improvement in many metabolic and CVD risk factors after 24 wk. As part of the initial design of the CHICAGO trial, we planned to evaluate emerging and novel factors that could contribute to differences in CVD risk between treatment groups. We therefore evaluated the impact of pioglitazone compared with glimepiride therapy on total and high molecular weight (HMW) adiponectin levels in the CHICAGO cohort and further evaluated the relationship of treatment-related changes in total and HMW adiponectin to treatment-related differences in lipid/lipoprotein CVD risk factors.

Subjects and Methods

Subjects and data collection

CHICAGO was a prospective, randomized, double-blind intervention study conducted between October 2003 and May 2006 at 28 clinical sites in the Chicago area (5). Participants included men and women with DM2 between the ages of 45 and 85 yr that were asymptomatic for coronary artery disease at baseline and had a fasting triglyceride level below 500 mg/dl. Randomized treatment consisted of pioglitazone (15–45 mg/d) or glimepiride (1–4 mg/d) for 72 wk. The design and methodological details of the study and primary outcome have been previously reported (5–7). Homeostasis model assessment for insulin resistance was calculated as previously described (8). The study was approved by central and local institutional review board committees, and all participants provided written informed consent.

Statistical methods

Descriptive data are presented using means and sd if data were normally distributed and using medians and quartiles if data were not normally distributed for both baseline values and change in values at 72 wk. Spearman correlation coefficients were calculated comparing changes in adiponectin with changes in HDL cholesterol for the entire cohort. Paired t tests were used to compare 72-wk observed values with baseline values for lipoprotein particle number and size. We tested for treatment group differences for each metabolic and cardiovascular risk factor at 72 wk using analysis of covariance (ANCOVA). Log transformation of the data was performed when necessary to achieve normality of variable distribution. These analyses were adjusted for site, treatment group, and baseline values of the metabolic or cardiovascular risk factor. There were no treatment group differences at baseline as previously reported (5). To account for the testing of multiple models, a Bonferroni adjusted cutoff of 0.005 was used for significance in these comparisons. In a previous publication, we performed similar analyses to test for treatment group differences at 24 wk (6).

The primary aim of the current study was focused on evaluating treatment-related changes in adiponectin as a predictor and potential mediator of treatment-related changes in lipoprotein parameters at 72 wk. To accomplish this, base ANCOVA models were employed that included treatment group, site, and baseline value of the lipoprotein parameter. We then added 72-wk change in total or HMW adiponectin to this base model. The percent change in treatment effect estimate produced by the addition of adiponectin to the base model was calculated, as was the significance value for the treatment effect in the new model. For comparison, similar analyses were performed evaluating the impact of the addition of other potential mediators. Consistent with our previously described analysis (6), predictors that resulted in the treatment effect becoming nonsignificant were identified as predictors, and therefore potential mediators, of the treatment effect. Analyses were performed using the 18.0 PC package of SPSS statistical software (SPSS, Inc., Chicago, IL).

Results

The analysis included 361 men and women with DM2 who were randomized to glimepiride (n = 186) or pioglitazone (n = 175) therapy for 72 wk. As reported previously, the two groups did not differ in any baseline characteristics or baseline laboratory measurements including lipids and glycemic or inflammatory measures (5, 6). The baseline low-density lipoprotein (LDL) cholesterol approximated 112 mg/dl (5). Fifty-two percent of subjects were on statin therapy and another 6% used an alternate lipid-lowering therapy, and these were not different between treatment groups.

Both total (before 4.1 ± 1.9; after, 9.7 ± 5.1) and HMW (before, 1.8 ± 1.1; after, 5.8 ± 4.2) adiponectin levels rose significantly with pioglitazone treatment (P < 0.0001 for both), whereas they did not change with glimepiride treatment. Treatment with pioglitazone for 72 wk led to significant decreases in LDL particle number and very-low-density lipoprotein (VLDL) particle size and significant increases in LDL and HDL particle size (Table 1, P < 0.0001 for all changes). Treatment with glimepiride for 72 wk led to a significant decrease in LDL particle number (Table 1, P < 0.0001) and increase in LDL particle size (Table 1, P < 0.001), but these changes were more modest in magnitude compared with changes with pioglitazone. Different from pioglitazone, glimepiride therapy had no effect on VLDL or HDL particle size (Table 1).

Table 1.

Descriptive statistics within treatment group for baseline values and 72-wk change in lipoprotein particle number and size

Treatment	Pioglitazone		Glimepiride
	Baseline	72-wk change	Baseline	72-wk change
Particle number (nmol/liter)
VLDL	72.4 ± 46.0	1.4 ± 45.6	72.1 ± 52.4	−5.2 ± 37.9
LDL	1472.8 ± 404.3	−193.0 ± 470.1a	1412.6 ± 393.9	−111.1 ± 349.2a
HDL	31.9 ± 7.1	−0.3 ± 5.8	31.5 ± 5.9	−0.9 ± 5.3
Particle size (nm)
VLDL	52.6 ± 9.4	−5.8 ± 9.9a	53.0 ± 10.7	−0.8 ± 10.0
LDL	20.5 ± 0.8	0.6 ± 0.7a	20.5 ± 0.7	0.2 ± 0.6b
HDL	8.6 ± 0.4	0.2 ± 0.3a	8.6 ± 0.3	−0.02 ± 0.29

Values are shown as mean ± sd. Tests for change at 72 wk were based on log₁₀ values for each variable.

^aP < 0.0001 for change within treatment group at 72 wk.

^bP < 0.001 for change within treatment group at 72 wk.

Comparing 72-wk differences between the treatment groups, the following variables were significantly higher (P < 0.005) in the pioglitazone group: total and HMW adiponectin, HDL cholesterol, LDL size, HDL size, and body mass index (BMI). The following were significantly lower (P < 0.005) in the pioglitazone group: triglyceride, insulin, VLDL size, and glycated hemoglobin (HbA1c). We next focused on evaluating the treatment-related differences in adiponectin as a predictor of significant treatment-related differences in lipid and lipoprotein parameters. For comparison, we also evaluated treatment-related differences in BMI, HbA1c, and insulin level as predictors of treatment-related differences in lipid and lipoprotein parameters, because each of these has a well-developed physiological basis for influencing lipid and lipoprotein metabolism. This evaluation was performed, following our previously published approach (6), by adding 72-wk change in each potential predictor/mediator to an ANCOVA model that already included treatment group, site, and baseline values of the lipid/lipoprotein outcome parameter and then measuring how this addition impacted the treatment effect on the lipid/lipoprotein outcome parameter (6). The addition of factors that led to loss of the treatment effect on the lipid/lipoprotein outcome parameters are identified as predictors, and potential mediators, of the treatment-related changes in these outcome parameters (6) As shown in Table 2, addition of changes in adiponectin or HMW adiponectin to the baseline ANCOVA model importantly reduced the treatment coefficient and resulted in a loss of significance for the treatment effect on HDL cholesterol, triglyceride, and LDL and HDL particle size. On the other hand, the addition of changes in BMI, insulin level, or HbA1c to the baseline ANCOVA model had no impact on the significance of the treatment effect on any lipid/lipoprotein parameter except for triglyceride level. Similar to fasting insulin, inclusion of homeostasis model assessment for insulin resistance had no impact on the significance of the treatment effect (not shown). In addition, the relationship between treatment-related changes in adiponectin level and treatment-related changes in lipoprotein parameters were identical in statin users and nonusers (not shown). These results are consistent with the conclusion that the changes in adiponectin level associated with pioglitazone treatment importantly contribute to the beneficial effect of this treatment on multiple lipid/lipoprotein parameters. Consistent with this analysis, the change in adiponectin or HMW adiponectin in the entire cohort over the course of the study was significantly correlated with change in HDL cholesterol level (r = 0.425 and P < 0.001 for total adiponectin, and r = 0.403 and P < 0.001 for HMW adiponectin).

Table 2.

Change in treatment coefficient and significance of treatment effect on lipid and lipoprotein parameters after addition of individual predictors

Outcome variable with predictors added	Treatment effect coefficient (se)	P for treatment effect
HDL cholesterol
Baseline model	0.049 (0.008)	<0.0001
Δ Adiponectin	0.021 (0.012)	0.09
Δ HMW adiponectin	0.018 (0.013)	0.2
Δ BMI	0.047 (0.008)	<0.0001
Δ HbA1c	0.048 (0.008)	<0.0001
Δ Insulin	0.044 (0.009)	<0.0001
Triglyceride
Baseline model	−0.06 (0.02)	0.004
Δ Adiponectin	0.004 (0.03)	0.9
Δ HMW adiponectin	0.022 (0.04)	0.5
Δ BMI	−0.07 (0.02)	0.001
Δ HbA1c	−0.05 (0.02)	0.02
Δ Insulin	−0.05 (0.02)	0.04
VLDL size
Baseline model	−0.043 (0.008)	<0.0001
Δ Adiponectin	−0.036 (0.011)	0.002
Δ HMW adiponectin	−0.032 (0.013)	0.01
Δ BMI	−0.045 (0.008)	<0.0001
Δ HbA1c	−0.042 (0.008)	<0.0001
Δ Insulin	−0.036 (0.008)	<0.0001
LDL size
Baseline model	0.008 (0.001)	<0.0001
Δ Adiponectin	0.0004 (0.002)	0.8
Δ HMW adiponectin	0.0003 (0.003)	0.9
Δ BMI	0.009 (0.002)	<0.0001
Δ HbA1c	0.008 (0.002)	<0.0001
Δ Insulin	0.008 (0.002)	<0.0001
HDL size
Baseline model	0.008 (0.002)	<0.0001
Δ Adiponectin	0.001 (0.002)	0.6
Δ HMW adiponectin	0.0003 (0.003)	0.9
Δ BMI	0.008 (0.002)	<0.0001
Δ HbA1c	0.008 (0.002)	<0.0001
Δ Insulin	0.006 (0.002)	<0.0001

The Δ log values (log₁₀ 72-wk minus log₁₀ baseline values) were used for insulin and total and HMW adiponectin. The base model included baseline value of the variable, site, and treatment group. P values in bold indicate where addition of the individual predictor to the model leads to loss of the treatment effect.

Discussion

Treatment with pioglitazone for 72 wk significantly increased total and HMW adiponectin levels in obese subjects with DM2, whereas treatment with glimepiride over the same time period did not. Within treatment groups, pioglitazone therapy significantly reduced LDL particle number and VLDL particle size and increased LDL and HDL particle size. Glimepiride therapy had more modest but significant effects for reducing LDL particle number and increasing LDL particle size but did not influence VLDL or HDL particle size (Table 1). Significant differences between treatment groups at 72 wk included changes in total and HMW adiponectin, HDL cholesterol, triglyceride, VLDL size, LDL size, HDL size, BMI, and HbA1c.

Based on human cross-sectional and observational studies, and on studies in animals, it has been suggested that adiponectin or insulin resistance can influence lipid and lipoprotein metabolism and thereby influence lipid and lipoprotein level and particle composition (1, 3, 4, 9, 10) The most important aspect of the current analysis is that we address this relationship in a randomized human intervention trial by evaluating whether treatment-related differences in total and HMW adiponectin level predict treatment-related differences in lipid and lipoprotein parameters. As shown in Table 2, evaluation of models that included treatment-related changes in BMI, HbA1c, or insulin as predictors did not diminish the significance of the treatment effect on most lipid/lipoprotein outcomes. Adjusting the baseline treatment model for treatment-related change in either total adiponectin or HMW adiponectin substantially reduced the treatment coefficient and eliminated the significance of the treatment effect on all of the lipid/lipoprotein parameters with the exception of VLDL particle size. This result provides support for a mechanistic model in which pioglitazone-related changes in total or HMW adiponectin level play a major role in mediating its effect on HDL cholesterol, triglyceride, and LDL and HDL particle size. Of all of these lipid and lipoprotein effects, increased HDL cholesterol is best established as a potential atheroprotective change in both humans and animal models. Previous cross-sectional and observational studies have demonstrated significant positive associations between adiponectin and HDL levels (4, 10). Mechanistic studies in animals have confirmed a number of pathways by which adiponectin could influence HDL metabolism in a way predicted to increase HDL cholesterol levels (11–13). Our study provides data from a human intervention trial that supports an important link between pioglitazone-related increases in adiponectin levels and increased HDL cholesterol.

Our analysis cannot exclude other unmeasured factors that may have changed in concert with adiponectin levels as a result of treatment and contributed to changes in lipid and lipoprotein parameters. Therefore, a direct cause-and-effect relationship between treatment-related changes in adiponectin and treatment-related changes in lipid and lipoprotein parameters cannot be unequivocally established by our analysis. However, our results, in combination with those already in the literature (11–13), provide support for a model in which changes in adiponectin are mechanistically linked to changes in lipoprotein particle composition.