The effect of high-polyphenol Mediterranean diet on visceral adiposity: the DIRECT PLUS randomized controlled trial

DIRECT-PLUS participants (Table 1; Additional file 1: Fig. S1) were 50.8 (10.4) years of age on average, and mostly men (88%), commensurate with the workplace, with abdominal obesity [mean WC = 110 cm (men) and 103 cm (women)], mean BMI = 31.2 (3.9) kg/m², 36.3% participants with pre-diabetes, and 11.3% participants with type 2 diabetes. The mean areas of abdominal adipose tissues and blood biomarkers were similar across the intervention groups (Table 1). Baseline correlations between the three abdominal adipose depots, clinical parameters, and cardiovascular risk score are reported in Additional file 1: Fig. S2. The VAT area at baseline was more strongly correlated with WC (r = 0.54, p < 0.001) than with bodyweight (r = 0.38, p < 0.001). Greater VAT was associated with higher cardiovascular risk score (r = 0.41), systolic (r = 0.36) and diastolic blood pressure (r = 0.28), triglycerides (r = 0.21), glucose (r = 0.30), HOMA-IR (r = 0.47), and IL-6 (r = 0.30) and lower HDL-c levels (r = − 0.18). Deep SAT accumulation was associated with higher HOMA-IR (r = 0.24), weight (r = 0.60), WC (r = 0.63), cardiovascular risk score (r = 0.20), and IL-6 (r = 14). In contrast, greater superficial SAT was associated with lower cardiovascular risk score (r = − 0.36), systolic blood pressure (r = − 0.13), and triglycerides (r = − 0.21).

The DIRECT-PLUS retention rate was 98.3% after 6 months and 89.8% after 18 months; 79.3% had eligible follow-up MRIs. Attribution was due to the lack of motivation and medical reasons unrelated to the study. The 18-month dropout rate did not significantly differ across the intervention groups (p = 0.28). As previously reported [26, 27], the participants assigned to the green-MED diet consumed more green tea and, exclusively, Mankai green shake and reduced their red meat and poultry intake compared to those assigned to the MED diet (p < 0.05 for all comparisons between the MED groups), and all three intervention groups similarly increased their PA levels, measured in MET units. Additional adherence information is presented in S1.

The mean weight loss (HDG: − 0.4% (5.0), MED: − 2.7% (5.6), green-MED: − 3.9% (6.5)) and WC loss (HDG: − 3.6% (5.1), MED: − 4.7% (5.0), green-MED: − 5.7%(5.7)) after 18 months were similar between the two MED diets (p > 0.05 for all) and higher as compared to the HDG (weight: HDG vs. MED: p = 0.02; HDG vs. green+MED: p < 0.001; WC: HDG vs. MED: p = 0.33, HDG vs. green+MED: p = 0.02).

All three abdominal fat depots decreased over 18 months of intervention (p < 0.05 vs. baseline for all). The green-MED group achieved a greater reduction in VAT than the other intervention groups (HDG: − 4.2% (22.5), MED: − 6.0%(31.3), green-MED: − 14.1%(27.7); p < 0.05 green-MED vs. MED or vs. HDG groups). These differences in VAT loss across the groups remained significant after adjusting for age, sex, and 18-month WC change (green-MED vs. MED p = 0.023; green-MED vs. HDG p = 0.002) (Fig. 1). After adjustment for weight change, differences between the MED groups remained significant (p = 0.042), but the difference between the HDG and green-MED groups was attenuated (p = 0.07). Sensitivity analysis among men only (the majority of this cohort) is presented in S2 and Additional file 1: Fig. S3. Changes in superficial SAT and deep SAT were not statistically different between the intervention groups after adjustment for WC or weight change. Illustrative MRI images of three participants with similar baseline characteristics and WC change show greater VAT loss in the green-MED group than in other intervention groups (Additional file 1: Fig. S4). Associations between changes in abdominal adipose depots and cardiometabolic biomarkers adjusted for age, sex, weight changes, and the three intervention groups are reported in Fig. 2. While VAT loss was independently associated with an improved lipid profile, the deep SAT loss was independently and significantly related to beneficial glycemic biomarkers during the intervention (p < 0.05 for all).

Higher dietary consumption of green tea, walnuts, and dietary fiber and reduced red meat consumption were significantly associated with greater %VAT loss (age- and sex-adjusted, p < 0.05 for all) (Fig. 3). Walnut consumption, reduced red meat consumption, and increased dietary fiber consumption remained significantly associated with VAT loss after further adjustment for WC loss (p < 0.05 for all). The association of green tea intake with VAT loss was attenuated after adjustment for WC change. Furthermore, after adjusting for weight loss, increased dietary fiber consumption remained significantly associated with VAT loss (p < 0.001), but all other dietary components were attenuated. Within the green-MED group (the only group with Wolffia globosa (Mankai)), increased intake of Mankai was significantly associated with greater VAT reduction (with a 26% VAT loss at the highest intake level (≥ 3/week); p = 0.04 vs. the lowest level, age-adjusted] (Fig. 3). Higher Mankai consumption was also associated with a reduction in CVD risk and improved lipid profile (total cholesterol: p = 0.023; TG/HDL: p = 0.044; SCORE: p = 0.038; adjusted for age and weight-loss) (Additional file 1: Fig. S5).

We observed a significant synergistic interaction effect between decreased red meat consumption and increased serum folate on VAT loss (p, interaction = 0.048, age- and sex-adjusted) (Additional file 1: Fig. S6). VAT loss was significantly higher (− 21.7%) among participants who both reduced red meat consumption and increased serum folate (top tertile) compared to those whose red meat consumption and serum folate remained unchanged (− 5.5%; p = 0.013 between the groups).

As previously reported [28], after 18 months, the total plasma polyphenol levels were higher in both MED groups (0.47 (0.4) mg/L for both) as compared to the HDG group (0.35 (0.4) mg/L; p < 0.05 for both MED vs. HDG).

Higher levels of total plasma polyphenol and serum folate [29] may reflect higher consumption of “green” dietary components, which were significantly associated with greater VAT loss (age- and sex-adjusted, p < 0.05 for all) (Fig. 3). After further adjusting for WC loss, total plasma polyphenol levels remained significantly associated with VAT loss (p = 0.018). Specifically, among plasma polyphenols examined, hippuric acid levels at 18 months were significantly higher in both MED diets than in the HDG group (p < 0.05) and were associated with VAT loss in a multivariate model adjusted for age and sex (Fig. 3) and were still significant after additional adjustment for WC change (p = 0.015, low vs. high tertile).

An exploratory analysis of urine polyphenol compounds showed an increase in urolithin A was strongly correlated with reduced VAT, even after adjusting for multiple comparisons for 139 identified metabolites (r = − 0.241, p < 0.001, q = 0.00036). Urolithin A was significantly associated with lower VAT (Fig. 3; adjusting for age and sex) and remained significant after adjusting for WC change. The increase in urolithin A was significantly correlated with increased consumption of walnuts (r = 0.14, p = 0.035) and Mankai (r = 0.24, p = 0.044).

The DIRECT-PLUS trial (ClinicalTrials.​gov NCT03020186), initiated in May 2017, was conducted in an isolated workplace (Negev Nuclear Research Center, Dimona, Israel), where a monitored lunch was provided. Of the 378 volunteers, 294 met the inclusion criteria: 30+ years of age with abdominal obesity [waist circumference (WC): men > 102 cm, women > 88 cm] or dyslipidemia [triglycerides > 150 mg/dL and high-density lipoprotein cholesterol (HDL-c): men ≤ 40 mg/dL, women: ≤ 50 mg/dL]. The exclusion criteria are fully described in S3. The Soroka University Medical Centre Medical Ethics Board and the Institutional Review Board approved the study protocol for the DIRECT PLUS trial. All participants provided written informed consent and received no financial compensation.

Participants who completed the baseline measurements were randomly assigned to one of three intervention groups (1:1:1 ratio), stratified by sex and working sites: healthy dietary guidelines (HDG), MED diet, or green-MED diet, all included PA recommendations, with a free gym membership and educational sessions promoting moderate-intensity PA [13] with ~ 80% aerobic content (S4). Dietary and PA interventions are fully described in Additional file 1: Table S1. Randomization was conducted in a single phase, as the interventions were conducted simultaneously, and participants were aware of their assigned intervention (open-label protocol). The HDG group received basic health promotion guidelines to achieve a healthy diet. The MED group was instructed to follow a calorie-restricted traditional MED diet, low in simple carbohydrates, similar to the DIRECT [23] and CENTRAL [13] trials. Both MED and green-MED diets were equally calorie-restricted (men: 1500–1800 kcal/day; women: 1200–1400 kcal/day); ~ 40% of total fat was mainly from polyunsaturated fatty acids (PUFA) and monounsaturated fatty acids (MUFA) and consisted of less than 40 g/day carbohydrates in the first 2 months with increased gradual intake up to 80 g/day. In addition, both MED groups included 28 g/day of walnuts (containing ~ 440 mg polyphenols/day; gallic acid equivalents (GAE), including mostly ellagitannins, ellagic acid, and its derivatives, Phenol-Explorer database). Participants from the green-MED diet were instructed to avoid red and processed meat and were guided to consume 3–4 cups/day of green tea and 100 g of frozen Wolffia globosa (Mankai cultivated strain) [26, 27, 47‐49] plant cubes (~ 20 g dry Mankai). A specific strain of Wolffia globosa, an aquatic plant in the duckweed family, is characterized by high protein content (more than 45% of the dry matter) and the presence of 9 essential and 6 conditional amino acids [48]. The Mankai plant is rich in insoluble fibers, vitamins (including vitamin B12 and folic acid), and minerals (including iron and zinc). We guided the participants to prepare a green Mankai shake with additional ingredients, part of the diet regimen (fruits, walnuts, or vegetables) each evening. The green protein shake was partially substituted for dinner, replacing beef/poultry protein sources (S5). In total, both MED diets had the same calorie restriction. Green tea and Mankai provided an additional daily intake of ~ 800 mg polyphenols [47] [GAE, Phenol-Explorer, and Eurofins laboratory analysis] beyond the polyphenol content in the MED diet. Green tea contains mainly epigallocatechin (EGC), epicatechin gallate (ECG), and epigallocatechin gallate (EGCG) [52]. As previously reported [47], Mankai mostly included the following phenolic metabolites: ellagic acid, benzoic acid, naringenin, luteolin, quercetin, p-coumaric acid, and caffeic acid, analyzed by mass spectrometry-based metabolomics methods from three different laboratories. Participants received green tea, walnuts, and Mankai onsite, free of charge. The participants were instructed to follow their lifestyle intervention for 18 months. The lifestyle interventions included 90-min nutritional and PA sessions in the workplace with multidisciplinary guidance (physicians, clinical dietitians, and fitness instructors). These sessions were held every week during the first month, once a month over the following 5 months, and every other month until the 18th month. All lifestyle educational programs were provided at the same intensity to all three groups. Text messages with relevant information for each assigned intervention group were sent at fixed time intervals to keep the participants motivated. In addition, a website listing all nutritional and PA information needed for the participants was accessible to them according to their intervention group. Most clinical and medical measurements, as well as lifestyle intervention sessions, were conducted onsite. We assessed adherence by self-reported dietary intake and lifestyle habit assessment tool, using validated food frequency questionnaires at baseline and after 6 and 18 months [30], including PA, measured in metabolic equivalent (MET) units. The questionnaire includes 127 food items and 3 portion size pictures for 17 selected food items and a physical activity questionnaire. In addition, the participant’s closed workplace enabled monitoring the freely provided lunch and the presence of an onsite clinic. A detailed description of the provided foods is reported in S5.

The abdominal fat depots were assessed at two time points, baseline and 18 months thereafter, using 3-T MRI (Philips Ingenia 3.0T) scans [13]. We quantified abdominal fat using MATLAB-based semiautomatic software [13] and blinded to the intervention group and calculated mean VAT, deep SAT, and superficial SAT along with two axial slices: L5-S1 and L4-L5. Interclass [13] and intraclass reliability were r > 0.96 (p < 0.001). The full protocol is reported in S6. Anthropometric parameters (i.e., weight and WC) and blood and urine biomarkers were taken at three time points, baseline and after 6 months and 18 months of intervention(S7). Polyphenol compounds were measured in both plasma and urine samples(S7).

The co-primary outcomes of the DIRECT PLUS study were 18-month changes in abdominal fat, the previously published intrahepatic fat (IHF) [28], and obesity. A flow chart of the study is presented in Additional file 1: Fig. S1. For this report, we examined the effect of the green-MED diet on changes in VAT and both deep and superficial SAT over 18 months. We further evaluated the associations between the changes in abdominal adipose tissues with changes in blood biomarkers [13] and consumption of specific foods. Changes in abdominal fat tissues were computed as changes relative to baseline [(time 18 − time0)/time 0 × 100]. Continuous variables are presented as the means (standard deviations). Nominal variables are expressed as numbers and percentages. The Kolmogorov-Smirnov test was used to determine whether variables were normally distributed, and natural log transformations were applied when necessary to achieve normal distributions. Differences in values over time were tested using a paired sample T-test or the Wilcoxon test for 18-month changes or three time points using ANOVA for repeated measures. The differences across the groups were tested using ANOVA, the Kruskal-Wallis test, or the chi-square statistic. Correlations were tested using Spearman’s or Pearson’s correlation analysis. The Kendall tau correlation was used to examine the trend of p. Multiple comparisons were adjusted using the Tukey post hoc test (for ANOVA) and Bonferroni correction (for Kruskal-Wallis). We used general and generalized linear regression models for adjustments and interaction models (with the specific adjustments detailed in the results). We calculated the cardiovascular risk score using the Systematic Coronary Risk Evaluation (SCORE) [53]. Of the 294 participants, almost all MRI scans at baseline (n = 286; 97%) were eligible for abdominal adipose tissue analyses; losses were due to technical reasons. The 18-month primary analyses of abdominal adipose tissues included all 286 participants with intention-to-treat (ITT) analysis [13], imputing for missing observations of 59 participants at follow-up using multiple imputation techniques [54], wherein the following predictors were used in the imputation model: age, sex, baseline weight, and WC at 18 months [13, 28]. We used the most recent values [33] for missing weight and WC data, as was used in the IHF paper as part of the DIRECT PLUS [28]. A sensitivity analysis revealed similar results from the per-protocol analysis, using data from completers only and the ITT analysis (S2). For urine polyphenolic compounds in untargeted analyses, zero values were imputed to the lowest detected, value once log2 transformed. Benjamini-Hochberg correction [55] with a 5% false discovery rate (FDR) was applied to control for multiple comparisons when comparing correlations between change in polyphenolic compounds and VAT (presented in Fig. 3 as urolithin A, the only polyphenolic compound that was statistically significant). Sample size and power calculations are reported in S8. Statistical significance was set at a two-sided α = 0.05; analyses were performed using SPSS (version 25.0).