Study population
This current study was executed within the framework of the PERSonalized glucose Optimization through Nutritional intervention (PERSON) study [12] and includes tissue-specific insulin resistant (MIR or LIR), weight stable (3 months ≤3 kg weight gain/loss) individuals (age 40-75), with a BMI between 25-40 kg/m2. Main exclusion criteria were: pre-diagnosed diabetes type 2, glucose/lipid altering medications, uncontrolled hypertension, alcohol consumption >14 units/week, smoking, and moderate-to-vigorous physical activity (MVPA) >4 hours/week.A table with all exclusion criteria can be found elsewhere in the design paper of the study [12].
In total, 119 participants were included. During the intervention, 7 participants dropped out, resulting in a sample size of 112. Assessments of cardiovascular risk and the OGTT could not be completed for 11 participants due to local COVID-19 lockdowns. Baseline characteristics for the whole group and per intervention arm are shown in Table 2.
Study design
As aforementioned, this research was part of the two-center PERSonalized glucose Optimization through Nutritional intervention (PERSON) study [12]. It involves two centers located in the Netherlands, Maastricht University Medical Center+ and Wageningen University & Research (WUR). The complete design and the CONSORT-diagram, which was approved by the local Medical Ethical Committee (NL63768.068.17), is published elsewhere [12]. The PERSON study was registered at a clinical trial register (ClinicalTrials.gov, NCT03708419) and executed according to the Declaration of Helsinki. Vascular measurements were performed at WUR (n=119) only, thus in a subgroup of the total PERSON study population. Before and during week 12 of the intervention, vascular function, cardiovascular risk factors, IR, and disposition index were assessed (Figure 1), as described in more detail below. The focus of this manuscript is on vascular assessments, which were only performed in this subgroup of the PERSON study. Other results of the PERSON study have been recently published elsewhere [14].
Screening
During screening, glucose and insulin values measured during a 7-point OGTT (time points 0, 15, 30, 45, 60, 90, 120) were used to calculate the muscle insulin sensitivity index (MISI) and hepatic insulin sensitivity index (HIRI). Calculations were based on Abdul-Ghani et al. [7]. The modelling of MISI was optimized by O’Donovan et al. [15]. HIRI and MISI have been validated against the golden standard hyperinsulinemic-euglycemic clamp [7,15]. The first blood sample (t=0) was drawn fasted from an intravenous cannula (antecubital vein). The remaining samples were taken after ingestion of a 200 ml 75 g glucose solution (Novolab). Data from The Maastricht study [16], from which a population with characteristics similar to the PERSON participants was selected, was used for MISI/HIRI tertile reference categories. Participants were classified as having MIR if their MISI was within the lowest tertile, and as LIR, if their HIRI was within the highest tertile [12]. Compared to The Maastricht Study, LIR prevalence was found to be lower in the first 163 participants of the PERSON study, wherefore the median HIRI of the PERSON study was used as cutoff thereafter.
Education level, retirement status, and alcohol consumption habits were assessed during screening with questionnaires. A food frequency questionnaire (FFQ, validated, 163-items) assessed habitual dietary intake [17].
Diet intervention
Participants were randomly allocated to follow either Phenotype diet (PhenoDiet) group A (LFHP for LIR, HMUFA for MIR), or PhenoDiet group B (LFHP for MIR, HMUFA for LIR), using center-specific minimization with randomization factors of 1.0 for the LIR/MIR phenotype, and 0.8 for age and sex, and a base probability of 0.7 by means of biased-coin [12]. During the 12-week intervention, participants had to remain weight stable, in order to assess the effect of the diet rather than weight loss. Participants were instructed to maintain their habitual physical activity levels. A more detailed description of the diet, instructions given to participants and exceptions can be found elsewhere [12].
Due to COVID-19 restrictions, some aspects of the intervention had to be adjusted, as the weekly visits were not possible anymore: on-site visits were substituted by phone/video calls and key products were delivered to participants at home.
Vascular function
CAR was assessed after an overnight fast (>10 h) with ultrasound (Terason uSmart 3300, Burlington, MA, USA) at baseline and during the last week of the intervention. CAR has been associated with coronary artery function, CVD risk, and disease progression in patients with peripheral arterial disease [20,21]. For CAR assessments, additional exclusion criteria applied: angina pectoris, Raynaud disease, chronic pain syndrome affecting the upper extremities, arteriovenous shunt, scleroderma, and heart infarct or heart failure within the last three months. Of the 119 participants, 105 were eligible for vascular function assessment, with 83 participants completing week-12 measurements (6 dropouts, 16 local COVID-19 lockdown). Three ultrasound recordings were excluded due to measurement problems, resulting in a total population for vascular function assessments of 80.
The CAR test was performed after a minimum of 10 minutes supine rest. CAR measures the diameter change of the right common carotid artery in response to a 3-min cold pressor test (CPT) (sympathetic stimulus). During CPT, the left hand of the participant was immersed in cold water (≤4°C) up to the wrist. The average diameter of a 1-min baseline recording was compared to the maximum diameter response (in 10 second intervals) during the 3-min CPT, using wall-tracking and edge-detection software [22]. Data were filtered manually for major artefacts, caused for example by swallowing, breathing or probe movement. Analysis was done blinded and an independent assessor reviewed the analyses. In response to the CPT, the carotid artery can dilate or constrict. The direction of reactivity was determined by a positive (dilation) or negative (constriction) area under the curve (CARAUC). CAR% was then defined as the maximum dilation or constriction from baseline, divided by the baseline diameter.
Cardiovascular disease risk factors
CVD risk factors measured before the start of the intervention and during week 12 of the intervention include fasting levels of total cholesterol and high-density lipoprotein (Cobas Pentra C400 with ABX Pentra Cholesterol CP reagens or ABX Pentra HDL Direct, respectively). Blood pressure was measured in sitting position after 5 min rest (dominant arm, automated sphygmomanometer, average of two measurements). The Framingham risk score for cardiovascular disease was calculated as described by D’Agostino et al. [23], based on age, total cholesterol, HDL, treated/untreated systolic blood pressure, diabetes and smoking status. Two measurements of height, weight, and waist-/hip circumference were taken and averaged at each point of assessment
Glucose homeostasis
A 7-point OGTT was performed at baseline and repeated during week 12 of the dietary intervention. MISI and HIRI [7,15] were calculated as follows: MISI= (dGlucose/dt)/insulin [mean during OGTT in pmol/L], with dGlucose/dt being the rate of decay of plasma glucose concentration (mmol/L) during the OGTT. HIRI=glucose0-30 [AUC in mmol/L*h] x insulin0-30 [AUC in pmol/L*h].
Matsuda index [24] and Homeostatic Model Assessment of Insulin Resistance (HOMA-IR) [25] were calculated as follows: HOMA-IR: fasting glucose (mmol/L) × (fasting insulin (mU/L)/22.5). Matsuda index: 10,000 ÷ square root of (fasting plasma glucose (mmol/L) x fasting insulin (pmol/L) x (mean glucose T0, T30, T60, T90, T120 (mmol/L) x mean insulin T0, T30, T60, T90, T120 (pmol/L)). In case of one missing timepoint value (N=2), mean glucose/insulin were still calculated with the remaining timepoints.
Disposition index was calculated as: Matsuda index * AUC30 insulin (pmol/L)/AUC30 glucose (mmol/L). AUC30 was calculated as the area under the curve from 0-30 minutes with the trapezoid method.
Physical activity
Physical activity was measured with the activPAL3 micro (PAL Technologies Ltd., Glasgow, UK), starting during the baseline measurements and continuing for ~1 week during the first week of the dietary intervention. At the end of the intervention, physical activity was reassessed starting in week 11, continuing until the end of week 12 (Figure 1). Only ‘free-living’ days (Figure 1, minimum of 1 weekend + 3 week days), where participants did not visit the university or had to fill in extensive questionnaires, were included in physical activity analysis. ActivPAL data were analyzed with an adapted script based on Winkler at al. [26]. Adaptations were made to include sleep/wake diaries filled in by participants.
Statistical analysis
Normally distributed data are presented as mean±SD, non-normal data as median [IQR]. Changes in the outcome variables for the total study population were assessed with a paired t-test or Wilcoxon signed-rank test in case of a non-normal distribution of the delta score (Week 12 - baseline).
Analysis of intervention effects with repeated measures linear mixed models revealed a substantial violation of homoscedasticity for our primary outcome (vascular function). Therefore, differences in delta scores (week 12 - baseline) between interventions (PhenoDiet group A versus PhenoDiet group B; LFHP versus HMUFA) were analyzed with linear regression models, corrected for baseline values. In a second model we corrected additionally for age and sex. As participants lost weight during the intervention, which was not intended, we adjusted for weight change in model 3 (Table S1-S2). In a forth model we corrected for changes in physical activity (Table S1-S2). To this end, physical activity expressed as % of awake time, as it takes into account the interconnectedness between physical activity and sedentary behavior, meaning that a higher percentage of the day spent in physical activity results in a lower percentage spent in sedentary behavior. Analyses where done in R studio, R version 3.6.2 [27].