Impact in Plasma Metabolome as Effect of Lifestyle Intervention for Weight-Loss Reveals Metabolic Benefits in Metabolically Healthy Obese Women.

Little is known regarding metabolic benefits of weight loss (WL) on the metabolically healthy obese (MHO) patients. We aimed to examine the impact of a lifestyle weight loss (LWL) treatment on the plasma metabolomic profile in MHO individuals. Plasma samples from 57 MHO women allocated to an intensive LWL treatment group (TG, hypocaloric Mediterranean diet and regular physical activity, n = 30) or to a control group (CG, general recommendations of a healthy diet and physical activity, n = 27) were analyzed using an untargeted 1H NMR metabolomics approach at baseline, after 3 months (intervention), and 12 months (follow-up). The impact of the LWL intervention on plasma metabolome was statistically significant at 3 months but not at follow-up and included higher levels of formate and phosphocreatine and lower levels of LDL/VLDL (signals) and trimethylamine in the TG. These metabolites were also correlated with WL. Higher myo-inositol, methylguanidine, and 3-hydroxybutyrate, and lower proline, were also found in the TG; higher levels of hippurate and asparagine, and lower levels of 2-hydroxybutyrate and creatine, were associated with WL. The current findings suggest that an intensive LWL treatment, and the consequent WL, leads to an improved plasma metabolic profile in MHO women through its impact on energy, amino acid, lipoprotein, and microbial metabolism.


INTRODUCTION 42
Recent studies have shown that even within the obese phenotype, cardiometabolic risk may not 43 necessarily vary primarily in relation to weight or body mass index (BMI) but to other subclinical 44 alterations.1,2 In this sense, there is a subset of obese individuals with lower risk of CVD or all-cause 45 mortality,3,4 which has been referred to as the "obese healthy paradox".5 Although there is currently 46 a lack of consensus on its definition, it has been suggested that metabolically healthy obese (MHO) 47 SampleJet NMR tubes using a SamplePro L liquid handling robot (Bruker BioSpin, Rheinstetten, 128 Germany). 129 1H NMR Spectroscopy. All 1H NMR experiments were performed on an Oxford 800 MHz magnet 130 equipped with a Bruker Avance III HD console and a 3 mm TCI cryoprobe using a water suppression 131 pulse program. Each spectrum was acquired at 298 K applying 128 scans, a spectral width of 20 ppm, 132 a data size of 65 K points, an acquisition time of 2.05 s and a relaxation delay of 3 s. Spectra were 133 processed using TopSpin 3.5pI6 (Bruker GmbH, Rheinstetten, Germany). Processed spectral data 134 were imported into MatLab (Math-Works Inc., Natick, MA) using in-house written scripts. 135 Alignment was achieved using a combination of an in-house peak reference picking function and the 136 "speaq" R-package (version 1.2.1).24 137 Statistical Analyses. All statistical data analyses were performed within the R environment (version 138 3.3.1). Differences in anthropometric and clinical variables at baseline and after 3 and 12 months 139 were assessed by independent or paired Student's t tests according to comparisons between or within 140 groups, respectively. Data are expressed as mean ± SD, unless otherwise stated. To determine 141 discriminant metabolites between control and treatment groups at 3 of intervention and at 12 months 142 of follow-up, we used NMR data of differences in metabolome between baseline and each time point 143 (3 or 12 months) and conducted a supervised analysis based on random forest (RF) modeling within 144 an in-house-developed repeated double crossvalidation framework (rdCV).25,26 In brief, the in-145 house double CV procedure, which has been successfully used in untargeted metabolomics27 and 146 microbiota analysis,28 consists of nested loops (outer "testing" and inner "calibration" loops) to 147 reduce bias from overfitting models to experimental data.25 Feature ranking and selection are 148 performed within the inner loop, to minimize statistical overfitting, by iteratively turning over 149 successively fewer features, removing from each step in the loop the 10% least informative 150 features.27 The rdCV procedure was subjected to 30 repetitions to improve modeling accuracy and 151 with misclassification as the fitness function. The overall validity and degree of overfitting of models 152 were assessed by permutation analysis, following the same rdCV procedure and by reporting the 153 cumulative probability of actual model fitness within a population of fitness measures of randomly permuted classifications (n = 200) based on the assumption of Student's t-distribution. The 155 assumption was confirmed by visual inspection of the histograms of permuted distributions. 156 Secondary analyses of associations of changes in metabolome with changes in weight or other clinical 157 parameters were performed using both the control and treatment groups together, as well as in 158 treatment group alone, by partial leastsquares (PLS) regression within a similar rdCV framework. 159 The quality of each model was evaluated by the R2 (the proportion of the variance of the response 160 variable that is explained by the model) and Q2 (the predictive ability) parameters. Permutation tests 161 (n = 200) were performed similarly to the analysis above, but with Q2 as the fitness measure. 162 Differences in changes of metabolites between groups after the intervention were calculated by fold 163 change (FC), taking the control group as reference, and assessed by independent Student's t tests. 164 The FC here was calculated as follows: FC = ΔTreament/ΔControl, where ΔTreament and Δ 165 Control denote the differences between the NMR intensities of metabolites at either 3 or 12 months 166 and at baseline, for treatment and control groups, repectively. Correlations between significant 167 metabolites selected from multivariate modeling of weight change after the intervention were 168 Of the 115 participants recruited, 58 were excluded due to dropout or failure to show at all visits (n 178 = 43), illness (n = 6), unavailable sample at some time point (at baseline, 3 or 12 months, n = 7), or 179 change of residence (n = 2). Therefore, 57 participants were included in the present data analyses.
Anthropometric measures and clinical parameters at baseline and after 3 and 12 months are presented 181 in Table 1. At baseline, MHO participants had a mean (±SD) age of 45.1 ± 3.45 y and a BMI of 35.8 182 ± 4.93 kg/m2. No differences between the control and treatment groups were observed at baseline 183 regarding menopause, weight, waist circumference, blood pressure, glycaemia, or lipid profile (Table  184 1). 185

Changes in Anthropometric and Clinical Measurements 186
At both 3 and 12 months, the treatment group showed greater WL and more pronounced reductions 187 in BMI and WC than the control group (Table 1). Compared to baseline, both groups showed a 188 decrease in total cholesterol and changes in HDL at 3 and 12 months. In particular, at 3 months, the 189 levels of HDL were decreased in the treatment group and increased in the control group. Moreover, 190 at 3 months, only participants in the treatment group showed decreases in LDL cholesterol and at 12 191 months decreases in SBP, glucose, and triglycerides, whereas at 12 months, only the control group 192 showed decreases in LDL (Table 1)

Modulatory Effect of Intervention on Plasmatic Metabolites 205
Changes in metabolome after 3 months of intervention included higher levels in the treatment group 206 of 3-hydroxybutyrate (3-HB), formate, methylguanidine, myoinositol, and phosphocreatine, as well 207 as lower levels of LDL/VLDL signals, proline, trimethylamine (TMA), and three unassigned 208 compounds (U3.32, U4.35, and U6.40) ( Table 2). Absolute FC in 3-HB, methylguanidine, 209 phosphocreatine, myo-inositol, proline, U4.35, and U6.40 were ≥2 (two-times or more) higher in the 210 treatment group than in the control group. Because of the poor multivariate classification between 211 groups at 12 months, discriminant metabolites at 3 months were further investigated by t test at 12 212 months of follow-up (Table 2). From this analysis, differences between the treatment and control 213 groups at 12 months were only observed in U3.32 (p < 0.05) and phosphocreatine (p < 0.05). 214 However, compared to at 3 months, fold changes in these metabolites at 12 months indicated a more 215 accentuated change in U3.32 and a change from upregulation to downregulation in phosphocreatine. 216

Changes in Metabolome Associated with Weight Loss 217
A total of 11 metabolites were moderately associated with a change in weight from baseline in both 218 groups at 3 months (Table 3). Keeping in mind that an association of metabolites with WL was 219 established as an inverse association with weight change (i.e., a positive association with weight 220 change means an inverse association with WL), WL was inversely associated with 2-hydroxybutyrate 221 (2-HB), creatine, LDL/VLDL signals, TMA, and three unknown compounds (U.sugar, U2.96, and 222 U3.32) and directly associated with asparagine, formate, hippurate, and phosphocreatine. 223 Interestingly, from this model, the changes in formate, phosphocreatine, LDL/VLDL signals, TMA, 224 and U3.32 were found to be in the same direction as those observed in the previous model with 225 treatment (Figure 3). 226

DISCUSSION 227
Using untargeted 1H NMR-based metabolomics and multivariate modeling, we were able to 228 determine changes in the plasma metabolome associated with a LWL treatment based on a 229 hypocaloric diet and physical activity in MHO women. Within this context, we further investigated 230 the association of WL with changes in the metabolome. As expected, compared to the control group, 231 individuals in the treatment group underwent greater WL. It is important to highlight that participants 232 of the current metabolomics study were a subpopulation of other larger study aimed to assess the 233 effect of WL on cardiometabolic risk markers.21 Findings in the present study regarding changes in 234 clinical parameters were similar to that larger study. Consequently, the discussion in the current work 235 focuses on the impact of LWL intervention on the plasma metabolome. Differences in the plasma 236 metabolome between the treatment and control groups were more pronounced at 3 than at 12 months 237 ( Figure 1; Supporting Information, Figure S-1). One reason could be the similar WL achieved during 238 the period between the third and 12th months after beginning the intervention (Table 1)  association between long-term successful WL and lower plasma proline levels.39 However, this was 257 not supported in the present study since proline was not directly associated with WL. The positive 258 association between asparagine and WL found in our study is consistent with previous reports, which 259 have shown an inverse association between this amino acid and obesity.40,41 Circulating levels of 260 asparagine can be obtained from dietary sources or synthesized from endogenous oxaloacetate via 261 aspartate. Studies conducted in animal models have demonstrated that supplementation with aspartate 262 and asparagine increased the glucose uptake and glycogen content in skeletal muscle, possibly 263 through the incorporation of glucose transporters type 4 or vesicles into the glycogen complex. 42 We 264 therefore speculate that along with WL, an increase of asparagine may be associated, in part, with an 265 improved glucose homeostasis. However, future studies are warranted to better determine the 266 functional role of asparagine in WL. Taken together, the observed associations of 3-HB, 2-HB, and 267 asparagine with the current LWL intervention and WL strongly suggest a positive impact on glucose 268 homeostasis in the MHO phenotype, which could also be interpreted as a decreased risk of T2D. Also 269 related to amino acid metabolism, we found that phosphocreatine increased with both treatment and 270 WL, whereas creatine decreased with WL. Creatine is mainly produced in the liver and skeletal 271 muscle from glycine and arginine and can further be phosphorylated to form phosphocreatine by the 272 enzymatic action of creatine kinase (CK). 43 We therefore speculate that the contrasting association 273 of creatine and phosphocreatine with WL may be related to a modulatory effect of WL on CK activity. which has been shown to be both proatherogenic and associated with cardiovascular disease 288 risk.51,52 We hypothesize that the lower levels of TMA associated with treatment and WL are related 289 to either a lower intake of its dietary precursors (i.e., eggs and meat)52,53 or modulation of choline 290 and carnitine metabolism, and consequently point to a lower risk of CVD due to a likely reduced 291 synthesis of TMAO. The observed treatment-related changes in these microbial metabolites may be 292 related with dietary intake. In the current study, however, data on food intake at 3 months that would 293 have allowed us to better establish this relationship were unfortunately lacking. Other unidentified 294 metabolites, including unassigned signals corresponding to sugars, were found to be related to 295 treatment and WL (Tables 1 and 2). Of particular interest is the unknown U3.32, which was not only 296 found to be related to treatment and WL at 3 months, but also remained significant in the treatment 297 group at 12-month follow-up. Further research to identify this compound to understand its role in WL 298 in the short and long-term is needed. The high number of subjects misclassified at 12 months suggests 299 a larger similarity in the changes in metabolome between groups, presumably due to either loss of 300 compliance or adaptation to changes after the first 3 months in the treatment group. Several factors, 301 including physiological, behavioral, and environmental ones, are key to both compliance and dropout 302 in long-term programs for WL.54 It is well documented that although lifestyle interventions can be 303 effective for long-term WL and improvements on cardiometabolic markers, maximum WL is 304 normally achieved between 1 and 6 months, followed by variable weight maintenance or weight 305 regain.55 However, in the present study, no differences in WL could be observed within the groups 306 from 3 to 12 months, thereby suggesting a maintenance phase. Therefore, based on the poor 307 multivariate predictions at 12 months combined with the maintained WL between 3 and 12 months, 308 we hypothesize that a metabolic adaptation occurs during this maintenance stage. To the best of our 309 knowledge, however, there are no reports on this type of metabolic/metabolome adaptation as a result 310 of longer-term WL interventions. Furthermore, although participants were defined a priori as 311 belonging to MHO, the combined results at 3 months, in terms of changes in clinical parameters and 312 metabolome, indicate that the current weight loss intervention caused shifts toward a healthier 313 phenotype with reduced risk of CVD. However, the positive metabolic regulations appeared to be 314 attenuated in the longer term, even though weight loss was maintained. The reasons for this 315 attenuation remain unclear. We recognize that our study has a number of limitations and strengths. 316 For instance, the sample size is relatively small and the study participants were exclusively 317 Caucasian, women, and middle-aged. Thus, we cannot extrapolate our conclusions to the general 318 population. In this sense, it would be interesting, for example, to determine the effect of a LWL in 319 MUO individuals as well as in men. Another limitation of our study was that compliance of physical 320 activity practice during all study and of MedDiet at 3 months, in both groups, was not measured, thus 321 leading to a lack of information regarding adherence to parts of the applied intervention. We, 322 however, hypothesize that due to the larger WL at both 3 and 12 months, the practice of physical 323 activity and intake of hypocaloric MedDiet were significantly higher in treatment than in control 324 group, as expected. On the other hand, because we were not able to assign the identity of unknown 325 compounds, potentially important information about metabolic perturbations in relation to 326 intervention and WL was unavailable. Future research aimed at identifying these unknown 327 compounds is warranted. As was also pointed out above, our study also had several strengths. The 328 current findings demonstrate that even with a relative healthy condition, the adoption of LWL is 329 always a recommended strategy to reduce the cardiovascular risk and complications in obese 330 individuals. This would be supported, for instance, with the observed inverse association between 331 WL and an early biomarker of impaired glucose regulation (2-HB), suggesting a modulatory effect 332 of WL on the diabetes risk. Furthermore, the untargeted workflow employed peak picking instead of 333 binning, thereby expanding and improving the available information content in the original data. The 334 multivariate modeling procedure and validation framework employed a data-driven, robust approach 335 to maximize information density while minimizing the likelihood of false-positive findings, thereby 336 focusing automatically on the most relevant metabolic perturbations in relation to the WL 337 intervention.27 Finally, our findings reinforce the utility of metabolomics in the identification of 338 biomarkers (beyond clinical parameters) of LWL interventions in individuals with moderate risk of 339 CVD. These biomarkers could be used in future research as additional targets of LWL interventions. 340

CONCLUSIONS
In conclusion, using untargeted 1H NMR metabolomics and multivariate modeling, we determined 342 that the impact on plasma metabolome of MHO women after a lifestyle intervention for weight loss, 343 based on hypocaloric Mediterranean diet and regular physical activity, was driven by changes in 344 amino acid, lipoprotein and microbial metabolism. Furthermore, we found that changes in the 345 metabolome were associated with weight loss within the frame of the same intervention. Taken 346 together, the lifestyle intervention and weight loss regulated plasma metabolome of MHO toward a 347 healthier phenotype. Such regulations were only observed at 3 months. Although weight loss was 348 maintained at 12 months, the metabolic changes driven by intervention were substantially attenuated 349 at 12 months, suggesting metabolic adaptation. The inverse association between WL and 2-HB, in