1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References


Weight loss helps reduce the symptoms of the metabolic syndrome (MetS) in the obese, but weight regain after active weight loss is common. The changes and predictive role of circulating adipokines and sex hormones for weight regain in men during dietary intervention, and also the effect of basal MetS status on weight regain, were investigated.

Design and Methods

Twenty-four men who continued to lose weight (WL) and 24 men who regained weight (WR) during the 6-month follow-up period after weight loss were selected from the Diogenes Study. Their circulating concentrations of leptin, adiponectin, retinol-binding protein 4 (RBP4), luteinizing hormone, prolactin, progesterone, total and free testosterone, and sex hormone-binding globulin (SHBG) were measured at baseline, after 8-week low-calorie diet-induced active weight loss, and after a subsequent 26-week ad libitum weight maintenance diet, and analyzed together with anthropometrical and physiological parameters.


Overweight and obese men with MetS at baseline had higher risk to regain weight (odds ratio = 2.8, P = 0.015). High baseline RBP4, low total testosterone, and low SHBG are predictors of weight loss regain (different between WR and WL with P = 0.001, 0.038, and 0.044, respectively).


These variables may play roles in the link between MetS and weight loss regain.


  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

One major health impact associated with obesity is the metabolic syndrome (MetS), which is a cluster of interrelated abnormal conditions including impaired control of glycemia, raised blood pressure, higher triglyceride (TG) level, low high-density lipoprotein cholesterol (HDL-C) level, and central obesity, leading to a high risk for cardiovascular disease and diabetes ([1]). Sustained weight loss has proven to be beneficial in improving the metabolic status notably through reversal of the MetS components ([2]). However, successful maintenance of the reduced weight can be difficult and weight regain is common ([3, 4]). In this regard, efforts have been made to understand the physiology of weight loss maintenance or regain and to seek measures to counteract weight regain.

It is already known that the circulating concentrations of adipokines, such as leptin, adiponectin, and retinol-binding protein 4 (RBP4), and sex hormones, such as testosterone and sex hormone-binding globulin (SHBG), are altered by weight loss ([5]). However, only a few studies address the effect of weight loss maintenance. In these limited studies, it has been shown that adipokines and sex hormones have different behavior between weight loss period and follow-up period ([5, 7, 9, 12]). However, they did not address their relation with successful weight loss maintenance.

We recently reported the effects of weight loss and follow-up on sex hormones and adipokines in women of the Diogenes study, and reported predictors of weight loss regain in that cohort, including testosterone and luteinizing hormone (LH) ([13, 14]). Here, we investigated in male participants of the Diogenes study the changes in circulating adipokines and sex hormones, and their roles as predictors of weight regain. In addition, we found an influence of basal MetS status on weight loss regain.


  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Participants and study design

Subjects were participants of the Diogenes study, which is a pan-European, randomized, and controlled dietary intervention study in eight European centers ([15]). This study was conducted according to the guidelines laid down in the Declaration of Helsinki. All procedures involving human subjects were approved by the local ethical committees in the respective countries ([13]), and a written informed consent was obtained from all subjects. The study was registered (Registration Number: NCT00390637) at

In brief, the intervention was composed of two phases: an active weight-loss phase of 8 weeks with low-calorie diet (LCD) about 3.3-4.2 MJ/d and a follow-up phase of 26 weeks (equivalent to 6 months) with ad libitum one of these maintenance diets: control diet and four moderate-fat intervention diets: low protein (LP) and low glycemic index (GI) (LGI), LP and high GI (HGI), high protein (HP) and LGI, and HP and HGI ([15]). Dietary consultation was given every 2-4 weeks and participants were advised to maintain their reduced weight.

The proportion of weight loss regained was used to score the outcome of the follow-up.

  • display math

In total, 181 adult male subjects completed the intervention ([14]). From them, we focused on nondiabetic and nondyslipidemic at baseline Caucasians ([14]). We selected 48 subjects evenly from the four intervention dietary groups and the control diet group was excluded. In each dietary group, based on the weight loss regain score (s), the extreme subjects beyond the 10-90 percentiles were excluded. Then six men with the most high positive s value were taken as weight regain subjects (WR) and six men with the most low negative s value were taken as continued weight loss subjects (WL).

Subjects were classified as having the MetS ([1]), if a subject was carrying 3 of 5 risk factors: waist circumference ≥94 cm, TG ≥ 1.7 mmol/L, HDL-C < 1.03 mmol/L, blood pressure systolic (SBP)≥ 130 and/or diastolic (DBP) ≥85 mm Hg, and fasting glucose ≥5.6 mmol/L.

After overnight fasting, the anthropometrical and physiological parameters were measured, and blood was taken using the same standardized protocol at each center at baseline (week -8), after weight loss (week 0), and after the follow-up (week 26). Serum glucose, TG, and cholesterols were measured ([15]). The post-intervention dietary intake levels of protein and GI were obtained from 36 (75%) subjects, for whom dietary records were available at week 26.

Measurement for RBP4, SHBG, and sex hormones

Adipokines' concentrations were all measured in EDTA-plasma. Leptin and RBP4 were measured with the Quantikine human leptin and RBP4 ELISA kits, respectively (R&D systems, Minneapolis, Minnesota, USA). Adiponectin was quantified using the Humane Adiponectin ELISA kit (BioVendor GmbH, Heidelberg, Germany).

Sex hormones and SHBG were measured in serum by immunoassays in the Clinical Chemistry laboratory of Maastricht University Hospital. Testosterone (total) was measured using the total testosterone Count-A-Count® technique from Siemens (Los Angeles, California, USA). Progesterone (total), LH, and prolactin were quantified with the AutoDelfia (Thermofisher, PerkinElmer, Turku, Finland). SHBG was measured on the Immulite 2000 (Siemens). The free testosterone concentration was calculated based on serum testosterone and SHBG using the formula proposed by Vermeulen et al ([16]).

Data analysis

The modern multivariate classification analysis to identify the important predictors was done with the Random Forests (RF) method using “randomForest” package version 4.5-34 with R version 2.10.1 ([17]). RF is able to give unbiased results without need for cross-validation. In this method, “Mean Decreased Gini” index is provided to rank the importance of variables in the classification. Prior to RF analysis, the missing values were imputed using the Probabilistic PCA (PPCA) method, and all values were normalized by being centered to the mean and divided by the SD of each variable with MetaboAnalyst 2.0 ([18]).

Conventional statistical analyses were carried out with SPSS version 15.0 for Windows (SPSS Inc, Chicago, USA). The values of the parameters were expressed as mean ± SD. For leptin, prolactin, and SHBG, whose values were significantly not normally distributed, geometric means were used instead of arithmetic means.

The difference between the weight change groups or dietary groups at each time point was analyzed using multi-way ANOVA as indicated in the notes of Table 1 and Figure 2. For hormones, age was included as covariate in the model. The dynamic change and group difference on the dynamic changes during weight loss and/or follow-up periods were analyzed using mixed-model ANOVA with repeated measures on time.


Figure 1. The baseline metabolic syndrome status related to the weight regain during 6 month follow-up period. Subjects who regained weight (gray) or continued to lose weight (white) during the follow-up were analyzed for their baseline metabolic syndrome status by χ2-test (P = 0.015).

Download figure to PowerPoint


Figure 2. The changes of hormones in men during LCD-mediated weight loss and the follow-up period by the weight change outcome of the follow-up, dietary intervention, and baseline metabolic state. Week -8 (baseline) to week 0 was the LCD-mediated weight loss period, and week 0 to week 26 was the follow-up period with intervention diets having variation on dietary protein and glycemic index. GI, glycemic index; LCD, low caloric diet; LH, luteinizing hormone; MetS, metabolic syndrome; RBP4, retinol-binding protein 4; SHBG, sex hormone-binding globulin. Data expressed as mean ± SD at the fasting state, except for leptin, prolactin, and SHBG, which were geometric mean ± SD. Possible differences on the concentrations between groups were assessed using 4-way ANOVA controlled for age with the weight change outcome of the follow-up (regain or continue lose weight), dietary protein level (low or high), dietary GI level (low or high), and the baseline metabolic state (with or without MetS) as main effects. Possible differences on the changes between groups were assessed by mixed-model of ANOVA with repeated measure of the hormone along time and on weight change outcome of the follow-up, dietary protein level, dietary GI level, and the baseline metabolic status as main effects, together with the interaction of time × main effect. A two-sided P-value <0.05 was taken as significant. * above the data points: significant difference between the groups on the static concentration. * above the connecting line: significant difference between the groups on the changes.

Download figure to PowerPoint

Table 1. Characteristics of subjects by diet and the weight change outcome in the follow-up at baseline (week −8), after weight loss (week 0) and after the follow-up (week 26)
  LP/LGILP/HGIHP/LGIHP/HGIGroup difference 
          P-value, staticP-value, dynamic 
ParametersTime (week)Continued Weight Loser N = 6Weight regainer N = 6Continued Weight Loser N = 6Weight regainer N = 6Continued Weight Loser N = 6Weight regainer N = 6Continued Weight loser N = 6Weight regainer N = 6Weight changeDietPeriod, weekTime* weight change groupTime* diet groupPooled P-value Time
  1. Values are mean ± SD at the fasting state. Bold P-values are significant (P < 0.05).

  2. BMI, body mass index; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; HOMA, homeostasis model assessment; MetS, metabolic syndrome.

  3. The “P-value, static” results from analyzing the difference between groups at each time point by 2-way ANOVA with diet and the weight change outcome of the follow-up as effects.

  4. The “P-value, dynamic” results from analyzing the differences of the changes between groups during the weight loss period (week -8 to week 0), the follow-up period (week 0 to week 26), and the whole intervention (week -8 to week 26), respectively, by mixed-model ANOVA with repeated measures on time, effects of diet, weight change outcome of the follow-up, and the interactions.

  5. The P-value of time was obtained by the aforementioned mixed-model of ANOVA.

Age, y−845.7 ± 3.942.3 ± 5.640.5 ± 2.846.2 ± 6.246.3 ± 6.143.0 ± 5.441.5 ± 6.842.8 ± 3.10.9570.702
Weight loss, kg−811.7 ± 1.914.1 ± 2.913.5 ± 3.612.9 ± 3.614.0 ± 4.312.3 ± 2.412.4 ± 2.611.0 ± 2.10.7020.609   
Weight loss, %−810.7 ± 1.613.3 ± 3.212.7 ± 2.811.4 ± 2.113.9 ± 3.811.6 ± 2.011.4 ± 1.511.4 ± 2.10.7510.613
Weight regain, kg0–26−3.0 ± 1.15.2 ± 1.1−1.9 ± 1.16.9 ± 1.7−6.1 ± 2.25.9 ± 1.4−2.8 ± 0.94.5 ± 1.6<0.0010.005
Weight regain, %0–26−26.8 ± 12.236.9 ± 6.7−13.9 ± 7.754.9 ± 13.1−46.2 ± 19.747.8 ± 2.2−22.5 ± 6.741.0 ± 10.7<0.0010.003
BMI, kg/m−2−833.5 ± 5.432.5 ± 2.935.5 ± 6.735.2 ± 5.933.4 ± 4.232.5 ± 3.036.6 ± 5.131.1 ± 3.10.1690.573−8–00.3010.664<0.001
029.9 ± 5.028.2 ± 2.830.9 ± 5.831.2 ± 4.928.7 ± 3.728.7 ± 2.832.4 ± 4.527.6 ± 3.40.2020.5340–26<0.0010.004<0.001
2629.0 ± 4.929.8 ± 2.730.3 ± 5.633.3 ± 5.026.7 ± 4.030.5 ± 3.031.5 ± 4.629.0 ± 3.40.3030.312−8–26<0.0010.321<0.001
Waist, cm−8111 ± 13107 ± 9113 ± 13118 ± 14113 ± 14108 ± 5115 ± 11109 ± 60.4700.556−8–00.7630.934<0.001
0101 ± 1096 ± 8101 ± 12107 ± 1199 ± 1497 ± 4104 ± 1098 ± 100.5150.4790–26<0.0010.8910.963
2695 ± 7102 ± 895 ± 10114 ± 1192 ± 14104 ± 5103 ± 11102 ± 90.0020.309−8–26<0.0010.817<0.001
Systolic blood pressure, mm Hg−8131 ± 12132 ± 10140 ± 13134 ± 12120 ± 13137 ± 20133 ± 8128 ± 150.5690.477−8–00.6460.214<0.001
0123 ± 10119 ± 10132 ± 12122 ± 15110 ± 8128 ± 16129 ± 11128 ± 90.8180.2310–260.0010.0310.002
26127 ± 10136 ± 12126 ± 9138 ± 19117 ± 10141 ± 13128 ± 13128 ± 100.0050.824−8–260.0020.4140.118
Diastolic blood pressure, mm Hg−883 ± 1183 ± 887 ± 1086 ± 976 ± 1483 ± 2086 ± 781 ± 90.9460.523−8–00.7150.395<0.001
078 ± 972 ± 879 ± 1176 ± 869 ± 979 ± 1080 ± 1079 ± 80.8310.4990–260.0010.1740.001
2676 ± 779 ± 978 ± 1186 ± 873 ± 1488 ± 980 ± 1280 ± 60.0270.725−8–260.0140.2970.025
Cholesterol, mmol/L−85.3 ± 1.14.8 ± 1.05.3 ± 0.64.5 ± 1.15.2 ± 0.74.3 ± 0.74.6 ± 1.34.8 ± 1.40.1050.814−8–00.5200.662<0.001
04.1 ± 0.83.8 ± 0.54.5 ± 0.73.8 ± 1.34.0 ± 0.93.5 ± 1.03.7 ± 1.03.8 ± 1.20.1830.7180–260.0770.982<0.001
264.8 ± 0.44.9 ± 0.45.0 ± 0.65.1 ± 0.64.9 ± 0.74.6 ± 0.94.5 ± 1.54.9 ± 1.10.8650.754−8–260.0200.7340.877
Triglycerides, mmol/L−81.3 ± 0.71.6 ± 0.61.5 ± 0.42.0 ± 0.81.1 ± 0.31.3 ± 0.51.2 ± 0.61.3 ± 0.30.0680.075−8–00.0280.478<0.001
01.4 ± 1.01.1 ± 0.21.2 ± 0.91.2 ± 0.70.8 ± 0.10.9 ± 0.50.8 ± 0.30.9 ± 0.40.8180.1260–260.0140.9480.019
261.1 ± 0.51.7 ± 0.71.2 ± 0.61.7 ± 1.00.8 ± 0.21.3 ± 0.51.1 ± 0.71.2 ± 0.50.0190.324−8–260.3500.7110.024
HDL-C, mmol/L−81.3 ± 0.61.2 ± 0.21.1 ± 0.30.9 ± 0.21.3 ± 0.31.0 ± 0.21.1 ± 0.21.1 ± 0.20.1100.486−8–00.0990.3640.344
01.0 ± 0.31.1 ± 0.31.1 ± 0.21.0 ± 0.21.2 ± 0.31.0 ± 0.11.0 ± 0.21.2 ± 0.30.6830.6800–260.1390.949<0.001
261.2 ± 0.31.3 ± 0.41.4 ± 0.31.2 ± 0.41.6 ± 0.41.2 ± 0.21.3 ± 0.41.3 ± 0.20.1770.911−8–260.9820.447<0.001
LDL-C, mmol/L−83.4 ± 0.82.9 ± 1.03.5 ± 0.62.7 ± 1.03.4 ± 0.82.7 ± 0.52.9 ± 1.33.1 ± 1.20.0770.980−8–00.3510.608<0.001
02.5 ± 0.52.1 ± 0.42.8 ± 0.52.3 ± 1.12.4 ± 0.92.1 ± 0.72.3 ± 1.02.2 ± 0.90.1620.8030–260.1080.979<0.001
263.1 ± 0.42.7 ± 0.53.1 ± 0.63.1 ± 0.53.0 ± 0.82.9 ± 0.82.7 ± 1.33.1 ± 0.90.9410.905−8–260.0290.8650.193
Glucose, mmol/L−85.5 ± 0.95.0 ± 0.84.7 ± 0.35.3 ± 0.55.4 ± 0.55.0 ± 0.35.1 ± 0.35.6 ± 0.60.8060.730−8–00.2590.100<0.001
05.4 ± 0.65.0 ± 0.44.9 ± 0.55.0 ± 0.64.8 ± 0.34.6 ± 0.44.7 ± 0.25.0 ± 0.40.7260.0980–260.0760.1560.053
265.2 ± 0.45.0 ± 0.45.3 ± 0.35.2 ± 0.54.6 ± 0.54.9 ± 0.34.8 ± 0.45.3 ± 0.70.3020.144−8–260.5380.1080.101
Insulin, μIU/mL−811.5 ± 3.08.2 ± 4.69.2 ± 3.113.9 ± 4.210.2 ± 3.612.2 ± 7.211.9 ± 3.012.6 ± 6.80.4230.630−8–00.8280.144<0.001
05.7 ± 2.57.0 ± 2.26.8 ± 2.99.6 ± 2.54.6 ± 5.15.9 ± 2.46.5 ± 3.59.3 ± 5.00.0430.1280–260.1650.9710.005
269.8 ± 7.28.0 ± 4.28.1 ± 3.812.0 ± 2.74.4 ± 2.310.0 ± 3.68.3 ± 2.710.6 ± 9.70.1750.798−8–260.5980.4420.008
HOMA−83.0 ± 1.21.9 ± 1.31.9 ± 0.63.3 ± 1.12.4 ± 0.82.7 ± 1.52.7 ± 0.73.6 ± 1.80.2880.520−8–00.2600.209<0.001
01.3 ± 0.51.5 ± 0.61.5 ± 0.72.2 ± 0.81.0 ± 1.11.1 ± 0.51.3 ± 0.82.1 ± 1.10.0720.1370–260.1300.9710.005
262.3 ± 1.91.8 ± 1.11.9 ± 1.02.7 ± 0.60.9 ± 0.52.2 ± 0.81.8 ± 0.62.6 ± 2.60.2000.736−8–260.8360.4940.009

To calculate whether baseline MetS state is associated with the weight change, either WR or WL, we performed a χ2-test with WL as reference.

The correlations were analyzed using Spearman's rho (ρ) test on the values or concentrations, or on the changes expressed as fold change. Generally, a two-sided P-value <0.05 was considered as significant. To correct for multiple testing in multiple correlation analyses, we set the significant P-value as <0.01.

Taking WR as the outcome event and WL as reference, the possible predictors were further analyzed using multinomial logistic regression with age and diet always as forced entry variables, entering other predictors in the backward stepwise manner.


  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Subject characteristics and metabolic improvement by weight loss and 6-month follow-up

Selected 48 overweight/obese (BMI 33.8 ± 4.1 kg/m2, range 27.5-44.8 kg/m2) but otherwise healthy adult men were 43.5 ± 5.2 (range 28-57) years old. They had lost 12 ± 3 % of their initial body weight, corresponding to 12.7 ± 3.0 kg, during the active weight loss period. Twenty-two of them (46%) had MetS at baseline. The anthropometrical and physiological parameters of all subjects (Table 1) were not different at baseline with respect to the outcome of the follow-up and diet groups. Except for insulin and changes of TG and MetS status, the parameters were also not different after active weight loss, supporting that groups were reasonably matched (Table 1). The baseline state of MetS tended to be different between WR and WL using ANOVA. By more proper χ2-test, it showed that in the WR group more subjects were with MetS at baseline compared to that in the WL group (odds ratio = 2.8, P = 0.015, Figure 1).

After 6-month follow-up, WL lost another 3.4 ± 2.1 kg, and WR gained 5.6 ± 1.6 kg. However, both groups had lower body weight compared with the baseline (all P < 0.001).

There was a modest but significant difference in the protein intake between the low and high dietary protein groups (18.2 ± 3.7% vs. 21.6 ± 4.9% of energy, P = 0.029) and also in the GI of the diet between low and high GI groups (57.6 ± 5.8 vs. 61.6 ± 5.0, P = 0.033). These levels were similar to the results on the whole Diogenes cohort ([15]). The maintenance diet had a significant effect on weight loss regain (Table 1). Subjects taking the HP/LGI diet showed less weight loss regain compared with the subjects on LP/HGI diet, which is in line with the outcome of the whole Diogenes study ([15]).

The whole intervention significantly reduced the waist circumference, DBP, TG, insulin, and homeostasis model assessment (HOMA) index, and elevated HDL-C. Compared to WR, WL generally had a more significant improvement of metabolic markers, particularly on lower waist, blood pressure, cholesterol, and LDL-C (Table 1).

The changes of adipokines and sex hormones during weight loss and 6-month follow-up

As shown in Figure 2, both leptin and RBP4 decreased during active weight loss (P < 0.001 and =0.001, respectively) and increased during the follow-up (both P < 0.001). The final concentration of leptin was lower than baseline (P < 0.001), whereas the final level of RBP4 was close to baseline (P = 0.073). The same pattern was observed for LH and free testosterone, whose final levels were not different from baseline (P = 0.416 and 0.913, respectively). Prolactin first decreased (P < 0.001), then was stable during the follow-up period (P = 0.127), resulting in a final level lower than baseline (P = 0.010).

In contrast, SHBG increased during active weight loss (P < 0.001) and decreased during the follow-up (P < 0.001). Total testosterone first increased (P = 0.007) and then stayed at this higher level during the follow-up (P = 0.560). Both final levels were higher than baseline (P < 0.001 and =0.037, respectively).

Adiponectin and progesterone did not change significantly during the active weight loss. During the follow-up adiponectin increased and resulted to a final increase (P < 0.001), whereas progesterone decreased (P < 0.001), but the final level was not different from baseline (P = 0.630).

The changes during the active weight loss period and the changes during the follow-up period were negatively correlated (all P < 0.05), in line with the return-back/reversal effect of sex hormones and other proteins that we observed in women ([14]).

Except for leptin and SHBG, the changes of measured analytes during the active weight loss and during the subsequent 6 months were not different with respect to the outcome of the follow-up, dietary protein, and GI levels, nor to baseline MetS status (Figure 2). The decrease of leptin by LCD was more in subjects without MetS (= 0.046). Its increase during the follow-up was larger in WR (P = 0.001). The decrease of SHBG during the follow-up period was more pronounced in WR than that in WL (P = 0.008) (Figure 2).

There were significant associations between adipokines, SHBG, and some sex hormones with the parameters of obesity and MetS on both fasting values (Table 2) and dynamic changes (Table 3). SHBG, adiponectin, and testosterone showed mainly negative correlations with parameters of obesity and of MetS. In contrast, leptin, RBP4, LH, and prolactin showed a positive relation with parameters of obesity and MetS.

Table 2. The correlation between hormone concentrations and the values of parameters of obesity and the metabolic syndrome over the intervention study
  1. Spearman's rho Correlation Coefficient based on pooled data of concentrations on three time points (week -8, week 0, and week 26).

  2. Bold values are significant (two-sided P-value <0.01 to correct multiple testing).

  3. BMI, body mass index; SBP, systolic blood pressure; DBP, diastolic blood pressure; TG, triglycerides; HDL-C, high density lipoprotein cholesterol; HOMA, homeostasis model assessment; RBP4, retinol-binding protein 4; LH, luteinizing hormone; SHBG, sex hormone-binding globulin.

Fat mass %0.430.26−−0.11−0.010.24
Adiponectin 0.04−0.14−0.02−0.06−0.01−0.050.13
RBP4  0.11−0.160.07−
Prolactin    0.17−0.100.00−0.15
Progesterone     0.050.15−0.12
Testosterone      0.770.57
Free testosterone       0.00
Table 3. The dynamic correlation between the changes of hormone concentration and the changes of the parameters of obesity and the metabolic syndrome over the intervention study
 LeptinAdiponectinRBP4LHProlactinProgesteroneTestosteroneFree TestosteroneSHBG
  1. Spearman's rho Correlation Coefficient based on pooled data of changes during three periods (week -8 to week 0, week 0 to week 26, and week -8 to week 26).

  2. Bold values are significant (two-sided P-value < 0.01 to correct multiple testing).

  3. See notes for abbreviations in Table 2.

Fat mass %0.610.010.320.040.35−0.06−
Adiponectin 0.17−0.07−0.04−0.08−0.12−0.04−0.15
RBP4  0.150.28−0.16−
LH   0.33−
Testosterone      0.810.23
Free Testosterone       0.31

Prediction of weight regain by RBP4, SHBG, and testosterone

We used RF analysis on all 19 measured baseline variables (as listed in Table 2), together with age, to quickly filter out predictors that were associated with weight regain at 6 months after active weight. As aforementioned, MetS at baseline was associated with weight loss regain. Whereas this suggests that parameters of obesity and MetS would predispose to weight regain, RF analysis showed that steroids and plasma proteins, but not anthropometric or physiological parameters, were the top important predictors of weight regain (Figure 3).


Figure 3. The important variables at baseline to predict the weight regain after weight loss. (A) the importance of top 10 parameters at baseline based on its Mean Decreased Gini in Random Forests analysis for the classification of weight regainers and continued weight losers. The values of the parameter are indicated as low (white) or high (black) level in these two groups. BMI, body mass index; SBP, systolic blood pressure; DBP, diastolic blood pressure; TG, triglycerides; HOMA, homeostasis model assessment; RBP4, retinol-binding protein 4; LH, luteinizing hormone; SHBG, sex hormone-binding globulin; free T, free testosterone. (B) Multidimensional scaling plot of proximity matrix from Random Forests on the separation of continued weight losers (•) and weight regainers (▵) by the profile of the top 10 parameters at baseline.

Download figure to PowerPoint

Conventional analysis was applied to quantitate the prediction power. WR had 28% more RBP4 (P = 0.001), 26% less LH (P = 0.007), 20% less testosterone (P = 0.038), and 24% less SHBG (P = 0.044) than WL. After active weight loss and subsequent 6-month follow-up, LH was not different anymore (P = 0.138), whereas RBP4, testosterone, and SHBG remained different (P = 0.001, 0.007, and 0.005, respectively) between WR and WL (Figure 2). The differences were calculated after controlling for age and baseline MetS status. This implies that their predictive power for weight regain is independent of MetS status.

Although insulin and HOMA ranked high in prediction by RF, their difference between WR and WL was not significant according to ANOVA analysis (Table 1). A further analysis indicated that insulin had interaction with testosterone in the prediction, which caused insulin to rank high among the variables in RF. When the cohort was divided into low or high baseline insulin level groups from the median insulin value (11.0 μIU/mL), testosterone was a significant predictor (P = 0.004) only in subjects with high insulin level, but not significant in subjects with low insulin level (P = 0.280).

By continuous variable analysis, the weight loss regain score was correlated with baseline RBP4 (ρ = 0.455, P = 0.001), testosterone (ρ = −0.376, P = 0.009), and SHBG (ρ = −0.294, P = 0.045). Among anthropometric and physiological parameters at baseline, only TG (ρ = 0.336, P = 0.020) and HDL-C (ρ = −0.289, P = 0.048) were related to weight loss regain score.

We further analyzed the relation among these candidates in the prediction for weight regain by logistic regression. The analysis showed that the contribution of baseline MetS status, glucose, insulin, leptin, BMI, TG, HDL-C, LH, and SHBG were not independent from that of RBP4 and testosterone. And there was no interaction between the diet or dietary component levels with RBP4, SHBG, and testosterone with regard to the prediction.


  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

It has been reported that persons with type 2 diabetes are more difficult to lose weight ([19]). Here, we reported that in men the weight regain after weight loss is affected by the baseline MetS status, which, to our knowledge, has not been addressed before. Interestingly, the physiological components of MetS are not significantly different per se between WR and WL. More likely, the hormones: RBP4, testosterone and SHBG, which are associated with MetS and shown to be predictors and biomarkers for weight regain, may be the molecular basis behind the high weight regain potency of subjects with MetS. Although 6-month follow-up is relatively short, the 1-year follow-up on part of the Diogenes study showed a high correlation between 6-month and 1-year body weight regain (data not shown). Therefore, these predictors might also be valuable for the long-term weight maintenance. This study is based on well-characterized subjects selected from a large intervention cohort. Nevertheless, the sample size is limited. Thus, our conclusions require validation from more and larger studies.

In this study, we did not observe correlations on the plasma concentration of RBP4 with other adipokines and obesity parameters (leptin, adiponectin, body fat mass percentage, BMI, and waist circumference). Besides a limited sample size, a plausible explanation for this is that our cohort was only composed of obese and overweight men without lean subjects. This made it difficult to find a true relation with obesity. A reanalysis of published data ([20]) showed that the linear relationship between BMI and serum RBP4 was mainly determined by low levels in lean subjects, whereas in the overweight and obese subjects there was no linear relation anymore. This suggests that after progressing from lean to obese, the circulating RBP4 level is not particularly determined anymore by the contribution from adipocytes, but rather, the ectopic fat storage in the liver might disturb the hepatic production of RBP4 and change the circulating RBP4 level. In a cohort with a similar age range as the present study cohort, it was shown that the fat content of the liver was the only factor positively associated with the plasma RBP4 level, but not fat in other parts of the body ([21]). Supporting this observation, we found that plasma RBP4 was mostly associated with fasting TGs, which are mainly attributed by the liver.

RBP4-carrying retinol is a precursor of ligands of the retinoid X receptor (RXR). This receptor works together with peroxisome proliferator-activated receptors (PPARs) regulating the transcription of genes involved in fat metabolism and adipogenesis ([22]). It has been shown that weight regain after weight loss is associated with adipogenesis ([23]). In this regard, a high RBP4 level reflects a high tendency for adipogenesis, which can lead to weight regain.

In our previous investigation in women of the same study, the association between RBP4 and weight regain was not pronounced ([13]). Different methods for measuring RBP4 concentration in men (normal ELISA) and women (multiplex assay) may contribute to this difference. Also, discrepancies in clinical studies on RBP4 are common ([24]). However, we cannot exclude a genuine sex difference. In another study on premenopausal women, no difference in basal RBP4 levels or diet-induced variations of RBP4 was found in lean women and two groups of obese women with high and low insulin sensitivity ([5]). Although women have a higher fat percentage in the body than men, their RBP4 levels are lower ([25]). This again implies that circulating RBP4 is not purely determined by adipocytes. Moreover, estrogen can regulate RBP4 expression in vitro ([26]). Thus, in women, the RBP4 expression may be masked by female hormones and may lose its association with weight control and insulin resistance.

Testosterone, another important predictor of weight regain after weight loss, does have sex-associated relations with obesity and MetS. A low testosterone level is associated with obesity, high BMI, and high visceral fat deposit and MetS in men ([27, 28]), but in women, particularly in post-menopausal women, it is associated with a low BMI and small fat mass and has a negative relation with MetS ([28, 29]). Our data also confirm these conclusions in men.

The increase of testosterone by weight loss might be the effect of releasing the testis from a leptin-induced inhibition ([30]), or could be driven by the demand for an active lipolysis during calorie restriction ([31]). In both men and women, we observed lower testosterone level associated with weight regain. However, in women, not baseline level but only the level after active weight loss was the predictor ([13]).

The present finding that high testosterone level predicts continued weight loss is in line with the previous report that high plasma dihydrotestosterone level predicts better weight loss maintenance in women after 2.5 years follow-up ([32]). Dihydrotestosterone is derived from testosterone and cannot be converted to estradiol by aromatase anymore. This suggests that the predictive role of testosterone is via its own androgen receptor, but not caused by converting to estradiol and subsequent binding to estrogen receptors. Nevertheless, measurement of circulating estrogens would be of some value for future application. A testosterone/estradiol ratio may acquire more clinical relevance than testosterone alone.

Other studies have reported that the free testosterone level increased with weight loss ([7, 9]), but we observed a decrease together with LH, which is the master hormone of testosterone in the hypothalamic-pituitary-gonadal axis ([33]). The discrepancy could be because free testosterone is not directly measured and can be derived with different formulas from total testosterone and SHBG, leading to much less precision of its determination compared to total testosterone or SHBG.

Among all hormones and related proteins investigated here, SHBG showed the strongest correlations with obesity and MetS. Additionally, it is the most sensitive factor to the weight change, in line with what others reported for weight losing women ([34]). SHBG is produced in the liver, and its production is inhibited by the monosaccharide-induced lipogenesis ([35]). It is likely that during LCD-mediated weight loss the negative energy balance reduces hepatic lipogenesis, thus enhancing the production of SHBG. The increased SHBG then drives the testosterone production to maintain the required free testosterone level in the body. The increased total testosterone level might interact with its receptors on adipocytes, leading to increased fat lipolysis and mobilization in response to the increased energy demand from fat while less glucose is available. Higher baseline SHBG and total testosterone levels might reflect a less lipogenesis in the liver and more active fat mobilization in the adipocytes, which make it not so easy to regain weight.

Interestingly, both weight regain predictors, SHBG and RBP4, are produced by the liver. Recent work has shown that the liver-specific decrease of lipogenesis protected mice from diet-/age-induced obesity ([36]). The association between liver-specific proteins with weight regain suggests that the liver is not only important for obesity development, but may also have an important role in weight maintenance or regain after weight loss. It would be interesting to investigate this role of liver fat and function. However, this is out of the scope of the present study.


  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

This work is part of the Diogenes project, supported by European Community (Contract No. FP6-2005-513946). PW, EM, AA, NV, DL, and WS designed the study; AA, MB, TL, SJ, AK, AP, JAM, THD, and PA led the intervention; PW, PM, and MA analyzed samples; PW analyzed data; PW and EM wrote the paper. We thank Sanne van Otterdijk for excellent technical assistance in RBP4 analysis.


  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References