Baseline characteristics
The participants (Table 1) had a mean age of 49.5 ± 10.1 years, with an average BMI of 30.9 ± 3.9 kg/m² and WC of 108.2 ± 9.7 cm (109.1 ± 9.1 cm for males and 101.4 ± 11.0 cm for females). Most participants were men (88.5%), 10.9% of participants had diabetes, and 59.4% had metabolic syndrome. Following the 18-month lifestyle intervention, participants lost − 2.6 ± 5.6 kg body weight and 4.8 ± 5.9 cm of their WC. The changes in abdominal adipose depots were − 17.5 cm2 (-35.4 – -4.0) for superficial SAT (-16.7% (-26.4 – -4.0)), -41.3 cm2 (-76.7 – -9.9) for deep SAT (-19.9% (-31.2 – -4.6)) and − 27.9 cm2 (-52.6 – -6.4) for VAT (-22.5% (-35.9 – -5.7)). The change in VAT proportion absolute units was − 1.3 ± 3.4% (-3.8% (-11.5–3.6)).
Table 1
Sex-specific baseline characteristics of the CENTRAL and DIRECT PLUS clinical trials participants.
Characteristic1 | N | Overall, N = 5722 | Male, N = 5062 | Female, N = 662 | p value3 | q-value4 |
Age | 572 | 49.5 ± 10.1 | 49.3 ± 10.1 | 51.1 ± 10.1 | 0.17 | 0.24 |
Weight, kg | 572 | 92.6 ± 13.9 | 94.3 ± 13.1 | 79.4 ± 13.0 | < 0.001 | < 0.001 |
BMI, kg/m² | 572 | 30.9 ± 3.9 | 30.9 ± 3.7 | 31.0 ± 5.2 | 0.87 | 0.87 |
Waist circumference, cm | 571 | 108.2 ± 9.7 | 109.1 ± 9.1 | 101.4 ± 11.0 | < 0.001 | < 0.001 |
Diabetes | 568 | 62 (10.9) | 58 (11.5) | 4 (6.2) | 0.19 | 0.24 |
Metabolic syndrome | 559 | 332 (59.4) | 305 (61.7) | 27 (41.5) | 0.002 | 0.003 |
SSAT area, cm² | 553 | 119.7 (91.5–162.9) | 113.6 (87.9–151.9) | 197.0 (159.0–256.2) | < 0.001 | < 0.001 |
DSAT area, cm² | 560 | 228.2 (179.9–290.2) | 228.5 (178.5–293.9) | 227.3 (187.4–268.4) | 0.69 | 0.77 |
VAT area, cm² | 564 | 134.8 (103.2–174.3) | 139.3 (108.2–178.1) | 105.4 (77.8–138.4) | < 0.001 | < 0.001 |
VAT proportion, % | 553 | 28.2 ± 9.0 | 29.3 ± 8.7 | 19.8 ± 6.7 | < 0.001 | < 0.001 |
1 BMI, body mass index; SSAT, superficial subcutaneous adipose tissue; DSAT, deep subcutaneous adipose tissue; VAT, visceral adipose tissue. |
2 Values are presented as either the median (p25, p75) or the mean ± standard deviation for continuous variables, depending on their distribution, or as number (%) for categorical variables. |
3 Two Sample t-test; Pearson's Chi-squared test; Wilcoxon rank sum test |
4 False discovery rate correction for multiple testing |
Baseline VAT area and proportion sex-specific deciles showed parallel direct and significant correlation trends with age, blood pressure, and most blood biomarkers (FDR < 0.05) (Table S1 and Figure S1). Specifically, fasting glucose and insulin, homeostatic model assessment of insulin resistance (HOMA-IR), HbA1c, TG, TG/HDLc, and gamma-glutamyl transferase (GGT) were found to increase with higher VAT area and proportion deciles. Conversely, VAT area and proportion exhibited dissimilar associations with anthropometric measurements and specific blood biomarkers. VAT area was positively associated with WC (tau = 0.33, FDR < 0.001), chemerin (tau = 0.18, FDR < 0.001), high-sensitivity C reactive protein (hsCRP) (tau = 0.0.16, FDR < 0.001), alkaline phosphatase (ALKP) and alanine transaminase (ALT) (tau = 0.09, FDR = 0.04 for both). However, the VAT proportion was not associated with these markers. Additionally, while VAT area demonstrated an increasing trend with body weight (tau = 0.21, FDR < 0.001) and leptin (tau = 0.19, FDR < 0.001), VAT proportion presented a decreasing trend with these measurements (tau=-0.13 and − 0.12, FDR < 0.001).
These trends remained following adjustment for age, weight, and intervention trial (CENTRAL and DIRECT-PLUS). In a partial correlation analysis adjusted for these covariates, sex-specific deciles of superficial SAT area, deep SAT area, VAT area, and VAT proportion exhibited distinct associations with anthropometrics and blood biomarkers (Fig. 1A). All fat depot areas were directly and significantly associated with WC, leptin, chemerin, and hsCRP (FDR < 0.01). In contrast to VAT area, which was negatively associated with HDLc (tau=-0.08, FDR < 0.01) and positively associated with TG and TG/HDLc (tau = 0.11, FDR < 0.001), superficial and deep SAT exhibited positive correlations with HDLc (tau = 0.08 and 0.09, FDR < 0.001) and inverse correlations with TG (tau=-0.07 and − 0.06, FDR < 0.05) and TG/HDLc (tau=-0.09 and − 0.08, FDR < 0.01). The VAT proportion association with TG and HDLc (tau = 0.15 and − 0.13, FDR < 0.001) appeared stronger than that of the VAT area (tau = 0.11 and − 0.08, FDR < 0.001). Deep SAT and VAT areas were directly correlated with insulin and HOMA-IR (FDR < 0.05), while superficial SAT was not. The VAT area was positively correlated with systolic and diastolic blood pressure, fasting glucose, HbA1c, ferritin, GGT, and ALKP (FDR < 0.05), while no significant associations were found between superficial and deep SAT and these markers. Superficial SAT was inversely associated with ALT levels (tau=-0.07, FDR < 0.05), whereas VAT was positively associated with ALT levels (tau = 0.09, FDR < 0.01). The VAT area and proportion showed similar trends with most biomarkers, except for leptin and hsCRP. While VAT area had a positive correlation with both hsCRP and leptin (FDR < 0.001), VAT proportion did not correlate with hsCRP and was negatively correlated with leptin (FDR < 0.001). Similarly, in contrast to the direct association of VAT area with WC (FDR < 0.001), VAT proportion was not associated with WC (FDR = 0.99).
Baseline VAT area and proportion in relation to obesity complications
Sex-specific cutoff values of VAT area and proportion were calculated for metabolic syndrome and diabetes status (Table S2, Figure S2). The cutoff values for metabolic syndrome at baseline were 120.72 cm2 VAT area (AUC = 0.69) and 27.84% VAT proportion (AUC = 0.62) for men and 114.8 cm2 VAT area (AUC = 0.82) and 24.39% VAT proportion for women (AUC = 0.74). The cutoff values for diabetes status at baseline were 114.10 cm2 VAT area (AUC = 0.62) and 35.41% VAT proportion (AUC = 0.60) for men and 90 cm2 VAT area (AUC = 0.70) and 25.74% VAT proportion for women (AUC = 0.81). We further compared the VAT area and proportion prediction performances in classifying states of metabolic dysfunction at baseline, in adjustment for trial type, sex, age, and baseline weight (Fig. 2). VAT area and proportion seemed to similarly predict states of metabolic syndrome (AUC = 0.75 for both, p = 0.49), hypertension (AUC = 0.76 for both, p = 0.25), and diabetes (AUC = 0.71, p = 0.48). However, VAT proportion performed better at classifying participants with hypertriglyceridemia (AUC = 0.66) compared to VAT area (AUC = 0.62) (p = 0.01).
VAT area and proportion were correlated with each other (r = 0.68, p < 0.001). Nevertheless, distinct phenotypes of visceral adiposity could be classified for participants whose VAT area was above the median (men = 139 cm2, female = 105 cm2) and whose VAT proportion was below the median (men = 29%, female = 19%), and vice versa. Participants with higher VAT area and decreased VAT proportion had, by definition, higher SAT and increased weight. In multivariable analyses of groups with similar and opposite sex-specific VAT area and proportion medians, participants (13%) characterized by a top-median VAT area and low-median VAT proportion exhibited increased diastolic blood pressure, HbA1c, fasting insulin, HOMA-IR, ALT, AST, leptin, chemerin, and hsCRP, after controlling for weight and deep SAT, compared to the other groups (FDR < 0.05 for all). Alternatively, participants (14.1%) with low-median VAT areas and top-median VAT proportions presented similar adverse lipid profiles to those with higher VAT areas (Supplementary Material Table S3).
Eighteen-month changes in abdominal adipose tissue depots
Despite the opposite associations at baseline between SAT and VAT regarding lipids profile, the loss of each fat compartment was associated with an improved lipids profile, even after adjustment for age, overall weight loss, and intervention trial (Fig. 1B). Similarly, all fat depot area reductions were associated with reduced WC, fasting insulin, HOMA-IR, and leptin, and none were associated with changes in blood pressure, hsCRP, ALKP, or AST. Both SAT subcompartments’ losses were related to reductions in glucose, but VAT loss was not. The opposite was true for the reductions in HbA1c, ferritin, GGT, and ALT, which were directly associated only with the loss of VAT area. Deep SAT and VAT losses were associated with chemerin reduction, while superficial SAT loss was not. Changes in VAT area and proportion were both directly and significantly correlated with reduced WC, HbA1c, dyslipidemia, chemerin, ferritin, GGT, and ALT (FDR < 0.05). However, some contrasts were noted; while VAT area loss was correlated with reduced insulin (tau = 0.11), HOMA-IR (tau = 0.10), and leptin (tau = 0.12), FDR < 0.01 for all, VAT proportion loss was not (tau = 0.04–0.06, FDR = 0.10–0.30). Alternatively, VAT proportion loss was correlated with reduced AST (tau = 0.09, FDR = 0.01), while VAT area loss was not (tau = 0.06, FDR = 0.12).
Prediction models of VAT baseline and 18-month change
Prediction models were developed for VAT area and proportion baseline and changes, utilizing either anthropometric measurements and demographic data, blood biomarkers, or a combination of both. Each model’s selected variables and performance metrics are presented in Supplementary Material Tables S4-S7.
The best-performing prediction model for baseline VAT area included a combination of anthropometrics, demographics, and blood biomarkers (Supplementary Material Table S4). It was trained on data from 227 DIRECT-PLUS participants, tested on 55 DIRECT-PLUS participants, and validated on 259 CENTRAL participants. Participants assigned to the training data had similar characteristics to those assigned to the testing data (Supplementary Material Table S8). The cross-validation models for choosing the optimal hyperparameters for the final model had RMSE of 0.27 and R2 of 0.44. The final model was applied to the testing and validation datasets, with RMSEs of 0.26 and 0.40 and R2 of 0.53 and 0.50, respectively. This model selected both anthropometric, demographic, and blood biomarkers mesurments, including WC, MAP, age, TG/HDLc, HbA1c, HOMA-IR, glucose, GGT, ALKP, and chemerin (Fig. 3A).
Similarly, the best-performing model for the estimation of baseline VAT proportion also included a combination of anthropometric, demographic and blood biomarkers predictors (Supplementary Material Table S5). The model was trained on n = 218 participants (RMSE = 6.87, R2 = 0.37), tested on n = 53 participants (RMSE = 6.55, R2 = 0.51), and validated on n = 142 participants (RMSE = 6.7, R2 = 0.39). The selected variables were similar to those predictive of VAT area, with additional selected variables: sex, ALT, fetuin-A, ferritin, and leptin (the latter negatively contributing to VAT proportion). Furthermore, female sex was a predictor of a lower VAT proportion (Fig. 3B). In contrast to the VAT area prediction, WC was not selected for the prediction of VAT proportion. However, higher weight was selected as a predictor of a lower VAT proportion in a model developed with only anthropometrics and demographic variables.
As for VAT area´s change, the best-performing predictor on the testing data included only anthropometric measurements (RMSE = 15.33, R2 = 0.59). In contrast, the best performance for the validation data incorporated variables of both anthropometrics and blood biomarkers (RMSE = 52.1, R2 = 0.52). The predictors were trained on 180 participants, tested on 46 participants, and validated on 207–212 participants. Both predictors included changes in weight and WC, with the former including change in MAP and the latter including change in leptin (Fig. 3C, Supplementary Material Table S6).
Lastly, the best-performing predictor of the change in VAT proportion was developed using only anthropometric and demographic data (weight, WC, MAP, age and sex). The model was trained on data from 172 participants (RMSE = 11.32, R2 = 0.15), tested on data from 43 participants (RMSE = 11.80, R2 = 0.24) and validated on data from 212 participants (RMSE = 18.72, R2 = 0.16) (Fig. 3D, Supplementary Material Table S7). Due to insufficient data, sex-specific prediction formulas could not be developed for females (11.5%). However, sex-specific male predictors are available in Supplementary Material Figure S3.