A panel of free fatty acid ratios to predict the development of metabolic abnormalities in healthy obese individuals

Increasing evidences support that metabolically healthy obese (MHO) is a transient state. However, little is known about the early markers associated with the development of metabolic abnormalities in MHO individuals. Serum free fatty acids (FFAs) profile is highlighted in its association with obesity-related insulin resistance, type 2 diabetes mellitus (T2DM) and cardiovascular diseases (CVD). To examine the association of endogenous fatty acid metabolism with future development of metabolic abnormalities in MHO individuals, we retrospectively analyzed 24 [product FFA]/[precursor FFA] ratios in fasting sera and clinical data from 481 individuals who participated in three independent studies, including 131 metabolic healthy subjects who completed the 10-year longitudinal Shanghai Diabetes Study (SHDS), 312 subjects cross-sectionally sampled from the Shanghai Obesity Study (SHOS), and 38 subjects who completed an 8-week very low carbohydrate diet (VLCD) intervention study. Results showed that higher baseline level of oleic acid/stearic acid (OA/SA), and lower levels of stearic acid/palmitic acid (SA/PA) and arachidonic acid/dihomo-γ-linolenic acid (AA/DGLA) ratios were associated with higher rate of MHO to MUO conversion in the longitudinal SHDS. Further, the finding was validated in the cross-sectional and interventional studies. This panel of FFA ratios could be used for identification and early intervention of at-risk obese individuals.

CVD 11 , possibly through increased generation of deleterious lipid metabolites 12 , proinflammatory cytokines 13 and inducing oxidative stress and endoplasmic reticulum stress 14,15 . We have recently shown that several circulating unsaturated FFAs were predictive of future metabolic abnormalities in MHO individuals 16 . Compared with concentrations of individual FFAs, the ratios of [product FFA/[precursor FFA] may better reflect the status of endogenous fatty acid metabolism. In this study, we measured 24 [product FFA]/[precursor FFA] ratios, and assessed the associations between these ratios and the metabolic status of obese individuals, in order to uncover novel biomarkers which could be used to identify MHO individuals who are at-risk for metabolic disorders. This study examined samples collected from participants in 3 independent studies, including a 10-year longitudinal cohort of 131 metabolically healthy individuals from the Shanghai Diabetes Study (SHDS) 17 , a cross-sectional cohort of 312 subjects from the Shanghai Obesity Study (SHOS) 18 , and a controlled dietary intervention study of 38 obese subjects who completed 8-week of very low carbohydrate diet (VLCD) 19 .

FFA ratios as predictive markers for the development of metabolic abnormalities in MHO.
The subjects selected from the SHDS consisted of 69 MH-NW and 62 MHO at baseline, with clinical data and sera collected at both baseline and 10-year follow-up. The data were used in the present study to investigate whether MHO had higher risk of developing metabolic abnormalities than MH-NW, and to explore the potential role of FFA ratios as early metabolic markers to predict the development of metabolic abnormalities in MHO. At 10-year follow-up, 33 of 69 baseline MH-NW remained metabolically healthy (stable MH-NW) and other 36 became MU-NW (progressors to MU-NW). By contrast, 12 of 62 baseline MHO remained their metabolic health after 10 years (stable MHO), but 50 baseline MHO developed metabolic abnormalities at the 10-year follow-up (progressors to MUO). The incidence rate of metabolic abnormalities was 1.54-fold higher in MHO than in MH-NW during the 10 years.
The baseline clinical characteristics of MH-NW and MHO groups and four subgroups categorized by follow-up metabolic status were listed in Table 1. The participants in baseline MH-NW group were younger than the MHO. SBP, DBP, 2hPG, fasting insulin (FINS), 2 h INS, TG and HOMA-IR were significantly higher and Matsuda ISI was lower in the MHO group compared with NH-NW. Further, stable MHO had higher levels of 2hPG and TC, and lower level of insulin sensitivity relative to the stable MH-NW. However, all of baseline clinical parameters in Table 1 were comparable between the progressors to MU-NW and progressors to MUO. Only TG was significantly higher in both progressors to MUO and progressors to MU-NW compared with their stable MH subjects, respectively (P < 0.05).
Among 24 [product FFA]/[precursor FFA] ratios, 3 were significantly altered (P < 0.01) in progressors to MUO compared with stable MHO, and these 3 ratios were SA/PA (C18:0/C16:0), OA/SA (C18:1 n9/C18:0), and AA/DGLA (C20:4 n6/C20:3 n6) ( Fig. 1 and Supplementary Table S1). The predictive powers of the 3 ratios for the risk of developing metabolic abnormalities among the MHO were evaluated using multivariate logistic regression analysis ( predictive value of 3 FFA ratios were better than it, with an area under the curve (AUC) of 0.755~0.767 compared to the AUC of 0.684 for TG. When the SA/PA, OA/SA, AA/DGLA ratios were combined with TG, the potential to distinguish the progressors to MUO and stable MHO was 0.837 as measured by ROC AUC (Fig. 2). By contrast, none of FFA ratios had significant difference between stable MH-NW and progressors to MU-NW.
Associations of SA/PA, OA/SA and AA/DGLA with metabolic health and BMI. To further confirm the associations of SA/PA, OA/SA, AA/DGLA ratios with metabolic and BMI, we compared the 3 FFA ratios among the MH-NW, MHO and MUO groups in the 312 subjects selected from the SHOS. Table 3 shows the clinical characteristics of the three study groups. The MHO and MUO subjects had similar BMIs which were significantly higher than that of the MH-NW. Subjects in the MUO group were ~10 years older than those in the other two groups. As expected, FPG, 2hPG and HbA1c concentrations were all significantly higher in the MUO subjects compared with the other two groups. FINS, TG, LDL-C and HOMA-IR were significantly higher and HDL-C was lower in the MHO and MUO subjects compared with NH-NW, and these parameters were significantly different between the MHO and the MUO. The 24 [product FFA]/[precursor FFA] ratios are summarized in Supplementary Table S2. Strikingly, the SA/PA, OA/SA and AA/DGLA ratios were all significantly altered in the MUO (P < 0.01) compared with both MH-NW and MHO groups (Table 3). Correlation analysis showed OA/SA was positively correlated with FPG, 2 h PG, HbA1c, TG, LDL-c, HOMA-IR, and negatively correlated with Matsuda ISI, while SA/PA and AA/DGLA were both positively correlated with Matsuda ISI, and negatively correlated with FPG, 2 h PG, HbA1c, TG, LDL-c, HOMA-IR (| r| = 0.15 ~ 0.59, P < 0.05) (Supplementary Table S3).
Effects of diet-induced weight loss on SA/PA, OA/SA, and AA/DGLA ratios. SA/PA, OA/SA, and AA/DGLA ratios were analyzed for the 38 obese subjects from the dietary intervention study to examine the effect of weight loss on these FFA ratios. The clinical characteristics of the participants are summarized in Table 4. Eight weeks of VLCD resulted in a significant decrease of BMI) (3.1 kg/m 2 ), significant reduction of subcutaneous and visceral fat areas, and improvement of insulin sensitivity. Compared with the baseline levels, OA/SA was significantly decreased while SA/PA and AA/DGLA increased after 8 weeks of VLCD intervention (P < 0.01 or 0.05) ( Table 4). Correlation analysis showed the changes of OA/SA and AA/DGLA were negatively (r = − 0.35, p = 0.03) and positively (r = 0.52, p < 0.01) correlated with the change of Matsuda ISI, respectively.

Discussion
The present study highlighted the predictive value of SA/PA, OA/SA and AA/DGLA ratios for the development of metabolic abnormalities in MHO, i.e., MHO with higher OA/SA ratio had greater chance of being metabolically unhealthy within a 10-year period, while MHO with higher SA/PA and AA/DGLA ratios had lower chance of developing metabolic abnormalities. The predictive value of these 3 ratios is independent of age, sex, BMI, TG    21 . Furthermore, a systematic review evaluating the associations of metabolic health and BMI with CVD events reported that MHO individuals are at increased risk for adverse long-term CVD outcomes compared with MH-NW (RR = 1.24; 95% CI, 1.02 to 1.55), even in the absence of metabolic abnormalities 22 . In our longitudinal SHDS at 10-year follow-up, we found a higher incidence rate of metabolic abnormalities in MHO (80.6%, n = 62) compared with their MH-NW counterpart (52.2%, n = 69), which further proves that MHO is at higher risk of developing metabolic abnormalities. Factors that benefit metabolic health in obese people are yet to be fully understood. Previous cross-sectional analyses suggested MHO subject were more likely to be younger and female, with higher physical activity and fitness 23,24 . In addition, greater insulin sensitivity, smaller visceral and ectopic fat accumulation, higher adiponectin level and decreased circulating concentration of high-sensitive C-reactive protein (hsCRP), a surrogate marker of systemic inflammation, were suggested to be the protective factors in MHO phenotype 21,25,26 . In the analysis of samples collected from the SHDS longitudinal study, we found that, baseline TG was higher in the stable MHO group than the progressors to MUO (p = 0.049), but other baseline markers such as sex, age, plasma glucose and insulin, LDL, HDL and insulin sensitivity were comparable between the two groups. Notably different studies used different criteria to identify MHO. Here we defined MHO as the absence of components of metabolic syndrome by CDS, which was in line with some analyses 22 , but different from most studies in which allowed MHO to have one 4,27 , or even two 5,21 metabolic syndrome components. Chronic exposure to elevated concentration of FFAs promotes insulin resistance 28 and impairs compensatory β -cell response 29 , which increased the subsequent risks of MetS, T2DM, and CVD. Previous studies confirmed that MHO had lower visceral, liver, and muscle fat contents than insulin-resistant obese people did, suggesting that MHO have a better abilities to trap FFAs in the adipose tissue 21,30 . Consistently, a cross-sectional study by Succurro et al. 31 also observed that MUO subjects have increased levels of plasma total FFA compared with the MHO.
The profile of circulating FFAs is considered as a surrogate marker of fatty acid composition in the adipose tissue 32 . However, to our best knowledge, few studies have used individual FFA or the ratios of [product FFA]/[precursor FFA] to predict the incidence of metabolic abnormalities in MHO subjects. Challenge may be mainly from the development of easy, fast and reliable method to quantitate individual FFAs in the blood 11 which only contain a fairly small fraction of FFAs among the total fatty acids in the circulation (0.2-0.7 mM FFAs vs. 10 mM total fatty acids) 33 . Our previous work developed a targeted metabolomics approach and quantitatively determined 40 serum FFAs to assess their associations with metabolic health in obese individuals, and we found that several unsaturated FFAs were associated with the risk of MetS and T2DM 16 . Because circulating FFA composition to some extent reflects the function of elongase and desaturase during endogenous fatty acid metabolism, we hypothesized that some key [product FFA]/[precursor FFA] ratios might be better metabolic markers for adverse outcomes of obesity. We found that SA/PA, OA/SA, and AA/DGLA ratios, which reflect stearoyl-CoA desaturase 1 (SCD-1), delta 5 desaturase (D5D) and ELOVL 6 activities, respectively, independently associated with the incidence of metabolic abnormalities in baseline MHO at 10-year follow-up. This novel finding was corroborated by a cross-sectional sampling from the SHOS, and a dietary intervention study. Our findings highlighted the importance of fatty acid metabolism in the maintenance of metabolic health in overweight/obese subject, and warrant future studies to investigate the roles of SCD-1, D5D, and ELOVL6 in adipocyte, which may affect FFA metabolism, transportation, storage and utilization. The effects of some confounding factors such as gender and age on the changes of three key FFAs ratios were examined in this study. We performed gender-based subgroups analyses on MH-NW, MHO and MUO participants in the cross-sectional study, respectively. Results showed that at similar BMI and metabolic status, the levels of the OA/SA, SA/PA and AA/DGLA ratios in male and female participants had not statistically significant difference (p > 0.05). Similarly, none of them had the significant correlation with age in both SHDS and SHOS studies. Considering that the participants in baseline MH-NW group were younger (age range 21~39) than the MHO in the SHDS, we further investigate whether MHO had a higher risk of developing metabolic abnormalities than MH-NW under age-matched condition. Among 29 subjects less than 40 years old selected from 62 baseline MHO subjects, 23 of them (79.3%) developed metabolic abnormalities at 10-year follow-up, similar to the present results ignoring the age in that 50 of 62 baseline MHO (80.6%) progressed to MUO. These results confirmed that the confounding factors of gender and age did not affect our present findings.
There are several limitations in this study. Firstly, the subjects in the longitudinal SHDS were randomly recruited from two urban communities of Shanghai, China, but only ~20% (1250 of 5994) completed 10-year follow-up visit, which may result in selection bias in the cohort used in the present study. Secondly, the size of the cohort selected from the SHDS is moderate, and therefore the predictive value of three ratios should be validated larger cohorts, particularly in population-based prospective studies. Thirdly, the FFAs ratios at 10-year follow up were not measured. Fourthly, information on lifestyle and diet were not collected, which may result in covariate imbalance. Fifthly, a recent study reported that insulin sensitivity improved by about 26% in MUO but decreased by about 13% in MHO after a 6-month energy-restricted diet 34 , and it would be interesting to assess the differences of these key FFA ratios between MHO and MUO individuals after achieving similar diet-induced weight loss. Lastly, due to the potential effect of dietary habit on the fatty acids composition in adipose tissue, whether the findings in this study could translate across into other populations needs further validation. The main strength of this study is that as a retrospective study, we utilized clinical information and serum samples collected in 3 independent studies. Results from the longitudinal SHDS established potential cause-and-effect relationship between the FFA ratios and the incidence of metabolic abnormalities in the MHO, the cross-sectional sampling from the SHOS corroborated the association between the FFA ratios and metabolic abnormalities, and the dietary intervention study demonstrated that diet and BMI were influential factors on the FFA ratios.
In conclusion, we showed that MHO individuals had a higher risk of developing metabolic abnormalities over time compared with MH-NW in a communities-based Chinese population. The SA/PA, OA/SA and AA/DGLA ratios were associated with the incidence of metabolic abnormalities in MHO subjects, and had relatively high predictive value to identify overweight/obese individuals who are at-risk for metabolic abnormalities. Further study is needed to elucidate whether desaturase and elongase, such as SCD-1, D5D, and ELOVL6, play a causal role in metabolic dysregulation and could be novel therapeutic targets for prevention and treatment of metabolic deterioration in obese population.

Methods
Study design and participants. This retrospective study utilized clinical data and samples collected from participants enrolled in the following studies: The protocol for this metabolomics study using samples from the three aforementioned studies was approved by the ethics committee at the Shanghai Jiao Tong University Affiliated Sixth People's Hospital. Written informed consent was obtained from all participants.
Based on the changes of metabolic parameters between baseline and follow-up, participants in the SHDS longitudinal study were separated into four groups: 1) stable MH-NW: BMI < 24 kg/m 2 and no metabolic abnormalities at baseline and follow-up; 2) progressors to MU-NW: BMI < 24 kg/m 2 no metabolic abnormalities at baseline, remained normal weight but became metabolically unhealthy at follow-up; 3) stable MHO, BMI ≥ 24 kg/m 2 and no metabolic abnormalities at baseline and follow-up; 4) progressors to MUO: BMI ≥ 24 kg/m 2 and no metabolic abnormalities at baseline, remained being overweight or obese and became metabolically unhealthy at follow-up. Participants selected from the SHOS were divided into three group: 1) MH-NW: BMI < 24 kg/m 2 and no metabolic abnormalities; 2) MHO: BMI ≥ 24 kg/m 2 and no metabolic abnormalities; 3) MUO: BMI ≥ 24 kg/m 2 and met at least two CD metabolic syndrome diagnostic criteria.

Anthropometric indices and laboratory measurements. Each participant had a physical examination
including measurement of height, weight, waist circumference and blood pressure. BMI was calculated as weight (kg) divided by squared height (m 2 ). Waist circumference (WC) was measured at the horizontal plane between the inferior costal margin and the iliac crest on the mid-axillary line. Blood pressure was the average of three measurements made with a sphygmomanometer at two minute intervals.
After a fasting venous blood sample was collected, each participant received a 75 g oral glucose tolerance test. Plasma glucose levels were measured by the glucose oxidase method. Serum insulin was assayed using bio-antibody technique (Linco, St Louis, MO, USA). Serum lipid profiles including total cholesterol (TC), serum total triglyceride (TG), high-density lipoprotein cholesterol (HDL-C) and low-density lipoprotein cholesterol (LDL-C), were determined by standard commercial methods on a parallel, multichannel analyser (Hitachi 7600-020, Tokyo, Japan). An experienced technician measured glycated haemoglobin A1c (HbA1c) of de-identified samples using high performance liquid chromatography (HLC-73G7, Tosoh, Japan). Insulin resistance and insulin sensitivity were evaluated using the following formulas: (1) homeostasis model assessment of insulin resistance (HOMA-IR) = fasting insulin× fasting glucose/22.5; (2) Matsuda insulin sensitivity index (Matsuda ISI) = 10,000/√ (fasting insulin× fasting glucose× mean insulin during OGTT× mean glucose during OGTT) 37 .
Subcutaneous and visceral fat areas were assessed by Philips Achieva 3.0T MRI system (Philips Medical Systems, Eindhoven, The Netherlands) using standard array coils with the subject in a supine position as previously reported 19 .
Sample preparation and FFAs analysis. All of serum samples were stored at − 80 °C until use. Sample preparation were performed according to a modified protocol based on Püttmann et al. 33 . Forty FFAs were analyzed using an ACQUITY ultra-performance liquid chromatography (UPLC) system (Waters Corporation, Milford, USA) equipped with a binary solvent delivery system and an auto-sampler (Waters Corporation, Milford, USA), coupled to a tandem quadrupole-time-of-flight (Q-TOF) mass spectrometry (Waters Corporation, Milford, USA) 16 . For details, see the Supplementary information. A mixture of all the reference standards at an appropriate concentration was prepared and run after every ten serum samples for quality control. Twenty-four FFAs ratios were calculated by the absolute concentrations of product-to-precursor. Statistical analysis. The raw data produced by UPLC-QTOFMS were initially processed using TargetLynx applications manager version 4.1 (Waters Corp., Milford, MA) to detect peak signals, obtain calibration equations, and calculate the absolute concentration of each FFA 16 . Manual examination and correction were needed to ensure data quality.
Data are presented as mean ± SD or median ± SE. Group values were compared using the Mann-Whitney test for continuous variables or χ 2 for nominal variables. Simple associations were tested by spearman correlation coefficients (r) between calculated FFAs ratios and clinical measures. Logistic regression was used to predict metabolic abnormalities risk; odds ratios and 95% CIs were presented. Performance of different logistic models was assessed as the receiver operating characteristic area under the curve (ROC AUC). Statistical significance was defined as two-tailed P < 0.05. All statistical computing and graphics were carried out using SPSS 16.0 (SPSS, Chicago, IL, USA).