Insulin resistance and circadian rhythm of cardiac autonomic modulation

For this report, we used the data collected for the Air Pollution and Cardiac Risk and its Time Course (APACR) study, which we designed to investigate the mechanisms and the time course of the adverse effects of fine particulate matter (PM_2.5) on cardiac electrophysiology, blood coagulation, and systemic inflammation. Recruitment methods for the APACR study have been published elsewhere [28, 29]. All study participants were recruited from communities in central Pennsylvania, primarily from the Harrisburg metropolitan area. The inclusion criteria for the study included nonsmoking adults, ≥ 45 years old, who had not been diagnosed with severe cardiac problems (defined as diagnosed valvular heart disease, congenital heart disease, acute myocardial infarction or stroke within 6 months, or congestive heart failure). Approximately 75% of the individuals who were contacted and who met our inclusion criteria were enrolled in the APACR study. Our targeted sample size was 100 individuals, and we enrolled and examined 102 individuals in the APACR study.

Study participants were examined in the General Clinical Research Center (GCRC) during the morning hours of Day 1, between 8:00 AM and 10:00 AM. After administering the informed consent, the participants completed a health history questionnaire. A trained research nurse measured seated blood pressure three times, height, weight, and drew 50 ml of blood for blood biomarker assays according to the blood sample preparation protocols. A trained investigator connected the PM_2.5 and Holter electrocardiography (ECG) recorders. Participants were given an hourly activity log to record special events that occurred during 24 hours, including outdoor activities, exposure to traffic on the street, travel in an automobile, and any physical activities. Before participants were released, they received detailed instructions on how to operate both monitors. The entire session lasted approximately 1 hour. The next morning (Day 2), the participants came back to the GCRC to disconnect the PM_2.5 and Holter monitors, return the completed activity log, have another 50 ml of blood drawn, and provide a urine sample. Then, an exercise echocardiography was performed on each participant according to a standardized protocol (Bruce protocol) to measure the participant's ventricular function and structure. The entire Day 2 session lasted approximately 1 hour and 45 minutes. The study protocol was approved by the Penn State University College of Medicine Institutional Review Board (IRB). All participants gave written informed consent prior to their participation in the study. Each participant was reimbursed with $50.00, a breakfast certificate in the hospital cafeteria, and the mileage of the transportation required for participating in the study.

During the Day 1 visit, after an overnight fast, blood was drawn from each participant. Fasting glucose and insulin were measured by the Penn State College of Medicine GCRC central laboratory. HOMA-IR was calculated as [fasting insulin (μU/mL) Ø fasting glucose (mmol/L)/22.5], with higher HOMA-IR values indicative of more IR [5].

The entire 24-hour normal beat-to-beat RR interval data were divided into 30-minute segments of RR data. Thus, each individual provided 48 segments of 30-minute RR data. The RR data for HRV analysis were processed according to current recommendations [9]. Within each segment, any RR interval <400 ms, >2000 ms, or where the ratio from two adjacent RR intervals was <0.80 or >1.20 were excluded from the HRV analysis. The time- and frequency-domain HRV analysis were performed on the remaining normal RR interval data if the total length of such normal RR intervals was greater than 20 minutes (67% of original data), using the HRV Analysis Software v1.1 [30]. When performing frequency-domain HRV analysis, we used Fast Fourier Transformation (FFT). Briefly, the adjacent RR interval data were interpolated using a piecewise cubic spline approach, with a 2 Hz sampling rate. The FFT was performed on the equidistantly interpolated RR time series. We used a second order polynomial model to remove the slow non-stationary trends of the HRV signal. The following HRV indices were calculated as measures of CAM: standard deviation of all RR intervals (SDNN, ms), square root of the mean of the sum of the squares of differences between adjacent RR intervals (RMSSD, ms), power in the high frequency range (0.15-0.40 Hz, HF), power in the low frequency range (0.04-0.15 Hz, LF), the ratio of LF to HF (LF/HF), and in the very low frequency power (0.00-0.04 Hz, VLF). Following current recommendations [9], we performed logarithmic transformations on HF and LF prior to statistical analysis.

A standardized questionnaire administered on Day 1 to the participants was used to collect the following individual-level information: (1) demographic variables, including age, race, sex, and highest education level; (2) medication uses, including anti-hypertensive medication and glucose lowering medications; and (3) physician diagnosed chronic disease history, including coronary heart disease (e.g., revascularization procedures and myocardial infarction), hypertension, and diabetes. Body weight and standing height were obtained in the morning of Day 1 and were used to calculate body mass index (BMI) as weight (kg)/height (m²). Seated blood pressure was obtained three times after 5 minutes of resting. The averages of the second and third measures of seated systolic and diastolic blood pressures were used to represent a participant's blood pressure levels. Hypertension was defined as systolic blood pressure ≥ 140 mmHg, diastolic blood pressure ≥ 90 mmHg, or physician diagnosed hypertension and currently using anti-hypertensive medication. History of CVD was defined as having a physician diagnosed history of myocardial infarction or coronary arterial revascularization procedures. Type 2 diabetes was defined as fasting glucose > 126 mg/dL, physician diagnosed type 2 diabetes, or currently using glucose lowering medication.

From the 102 individuals, we excluded 8 participants with type 2 diabetes. As a result, this report used the data from the remaining 94 individuals. Each individual contributed up to 48 segments of 30-minute RR interval data. After excluding segments where the total length of RR interval data was less than 20 minutes, we analyzed 4404 segments of 30-minute HRV data from 94 individuals.

A two-stage analysis was performed to assess the relationship between IR measures and the circadian pattern of HRV. At the first stage, for each individual we fit the HRV data based on all available 30-minute segments to a cosine periodic regression model [31] using nonlinear least squares: HRV_i(t) = M_i + A_i•cos [2π •(t-θ_i)/T] + ε_i, i = 1, ..., 94, in which M_i is the daily average of HRV of the i^th subject, A_i is the amplitude of HRV of the i^th subject around M_i, t is the time-specific segment order number, T is the total number of 30-minute segments in 24 hours, θ_i is the acrophase (the lag from the reference time point (9:00 AM) to the time of the zenith of the cosine curve fit to the data of the i^th subject), and ε_i is the error term of the i^th subject. One unit of t corresponds to 30 minutes, with 1 indicating 9:00 AM to 9:30 AM, 2 indicating 9:30 AM to 10:00 AM, etc. Thus, from the above described cosine model, we obtained the estimated individual-level cosine periodic regression parameters, namely the M, Â, and θ to quantify the periodicity of the HRV variables. At the second stage, we used random-effects meta-analyses to obtain overall estimates of M, Â and θ, and their 95% confident intervals (CIs) to assess the associations between IR and the three components of the circadian pattern of HRV [32]. Because HOMA-IR and fasting insulin were not normally distributed, we used a logarithmic transformation for these two IR variables before performing the analysis. To visualize the impact of elevated IR on the three circadian parameters, we plotted the population level circadian pattern of HF, and superimposed the circadian pattern predicted HF from one standard deviation increase in HOMA-IR. To confirm the associations between the IR and the means from the random-effects meta-analysis models, we also applied linear mixed-effects models, specifying a first-order autoregressive covariance structure, to assess the associations between HRV variables and HOMA-IR using 24-hour data, analyzing the daytime (9 AM to 9 PM) and nighttime (9 PM to 9 AM next day) data separately. This model will also enable us to examine whether IR affect daytime and nighttime autonomic modulation differently. In this approach we ignored the assumption of the specific cosine form for the data and treated the HRV variables from each 30-minute segment as repeated measures. All analyses were performed using SAS version 9.1 software (SAS Institute Inc., Cary, NC, USA).

The main characteristics of the study population are shown in Table 1. From the first 102 APACR study enrollees, we excluded 8 individuals who had a history of type 2 diabetes. The sample size for this analysis was 94 individuals. The mean (SD) age of the entire study population was 56.5 (7.8) years old, with 37% male and 77% white. The prevalence of CVD was 7.4%. In addition, the mean (SD) for BMI (kg/m²), insulin (μU/mL), glucose (mg/dL), and HOMA-IR of the entire cohort were 27.1 (4.9), 6.6 (6.7), 84.4 (11.3), and 1.45 (1.73) respectively. The means (SD) for the HRV indices: log HF, log LF, LF/HF ratio, log VLF, SDNN, RMSSD, and HR were 4.4 (0.9), 5.4 (0.8), 3.8 (2.2), 6.9 (0.6), 61.8 (18.4), 29.6 (14.8), and 76.5 (9.7), respectively.

The estimated means of the three cosine periodic regression parameters and their 95% CIs for each of the HRV indices are presented in Table 2. The acrophases of log HF and RMSSD (both are reflective of the vagal modulation) were very similar, 4:00 AM and 4:15 AM, respectively. The acrophases of log LF and SDNN (both are reflective of sympathetic and vagal modulation) were also very similar - the point estimates are around 6:00 AM ± 45 minutes. The acrophase of log VLF was 8:00 AM, LF/HF ratio was 6:00 PM, and that of heart rate was 2:00 PM. The tests of zero amplitude were highly significant (P value < 0.0001) for every HRV variable, suggesting a significant circadian variation.

The associations between the IR measurements and cosine parameters of the circadian pattern (M, Â, and θ) estimated from the entire sample using random-effects meta-analysis models are presented in Table 3. Two different random-effects meta-analysis regression models were reported: Model 1, unadjusted; Model 2, adjusted for age sex, race, hypertension, and history of CVD. In general, higher levels of IR, especially IR measured by HOMA-IR, was significantly associated with lower M of all HRV indices and higher HR in both unadjusted and multivariable adjusted models, except that LF/HF ratio was not significantly associated with any measure of IR. In this healthy non-diabetes sample, IR was not significantly associated with the  or θ of any HRV indices. Log VLF was also included in Table 3 as a HRV variable that may be reflective of nocturnal sympathetic hyperactivity. The association between IR measures and VLF is very similar to that of HF and LF - elevated insulin resistance is only associated with lower mean levels of VLF, but not with the amplitude or the acrophase.

The circadian variation of Log HF calculated from the entire population is graphically presented in Figure 1. The predicted circadian pattern of Log HF due to a one SD increase in HOMA-IR is superimposed in Figure 1 to graphically illustrate the impact of elevated IR on the circadian pattern of log HF. It clearly show, similar to the numeric numbers in Table 3, that one SD increase in IR is associated with lower overall mean [β (M)] of Log HF, but not with the other two circadian parameters (Â and θ).

We also performed additional analyses examining the associations between HOMA-IR and HRV cosine parameters of circadian pattern in 8 individuals with physician-diagnosed type 2 diabetes. These analyses were preformed in an unadjusted model because small sample size. The results are presented in Table 4. Compared to that regression coefficients in Table 3 Model 1, the magnitude of effect due to higher Log HOMA-IR on the mean (M), amplitude (Â), and acrophase time (θ), are much larger among persons with type 2 diabetes, a group that clinically have more severe form of insulin resistance. Because of the small sample size, most of the statistical tests of the regression coefficients in Table 4 did not reach traditional significant level (P value < 0.05).

Ignoring the specific cosine form for the data and treating HRV variables from each 30-minute segment as repeated measures, the regression coefficients, SE, and P value relating HOMA-IR and HRV indices according to daytime (9 AM to 9 PM) and nighttime (9 PM to 9 AM next day) are presented in Table 5. In general, these results confirmed that higher IR is associated with lower HRV indices, and the patterns of association are similar daytime and night time.