Assessment of clinical characteristics and plasma lipidomic features
The clinical information, including the physiological and anthropometric indicators of the individuals included in this cohort, is summarized in Table 1. The participants were divided into three groups according to their BMI values. The level of SBP, waist-hip ratio, fat mass, body fat percentage and visceral fat area were significantly higher in both overweight and obese individuals than in the control group, with obese participants exhibiting higher values compared with overweight individuals (Kruskal-Wallis test, P<0.001).
We evaluated both coverage and reproducibility of the non-targeted lipidomic data on our sample. Using Progenesis QI 2.0 and metaX, the non-targeted metabolomics analysis yielded 51135 positive ion modes (Additional file 2) and 8988 negative ion modes (Additional file 3).
Overweight and obesity-related features
Because of the observed effects in obese adolescents on the lipid profiles, we performed a blocked Kruskal-Wallis test, using the obese group as the blocking factor, followed by Dunn’s hoc test for paired comparisons. As shown in Additional file 4 and 5, 876 positive and 544 negative features were gradually up-regulated among the 3 groups. Also, there were 1081 positive and 353 negative features down-regulated in Additional file 6 and 7. Of these, there are lipids or lipid-like compounds, also including organ-oxygen compounds, amino acids, peptides, and analogs, benzyl alcohols, glycerophospholipids and triacylglycerol. As shown in Fig. 1, paired comparisons revealed that 460 features (290 positive features in Additional file 8 and negative features in Additional file 11) exhibited significant differences between the control and obese group, whereas 231 and 244 features (Additional file 9 and 12, Additional file 10 and 13 in both positive and negative, respectively) showed obvious differences between the overweight versus control and obese group, respectively (P<0.05). Of these significantly changed metabolites, we screened out eight (six positive and two negative) metabolites with significant differences in expression among the three groups. The number of variables distinguishing overweight and obesity suggested that changes in a large fraction of the lipid profiles in overweight and obesity were shared, implying that compared with the control group, the overweight and obese group share similar metabolites.
To quantify the differential features among the 3 groups, all detected features were assessed using criteria: 1) variable importance of the projection (VIP) >1.0 estimated by partial least squares discriminant analysis (PLS-DA); 2) fold change in mass intensity ≥1.2 or ≤0.83; 3) P<0.05.
Table 1
Basic characteristics of the three groups in the study.
Variables
|
Control (n=30)
|
Overweight (n=26)
|
Obese (n=34)
|
P value b
|
Obese vs Overweight c
|
Obese vs Control c
|
Overweight vs Control c
|
Sex (female %), no. (%) a
|
18 (60.00)
|
16 (53.33)
|
14 (46.67)
|
0.594
|
——
|
——
|
——
|
Age, year
|
12.50±0.51
|
12.73±0.45
|
12.77±0.47
|
0.058
|
0.958
|
0.072
|
0.132
|
BMI, Kg/m2
|
17.49±1.41
|
23.76±1.00
|
29.89±3.17
|
<0.0001
|
<0.0001
|
<0.0001
|
<0.0001
|
SBP, mmHg
|
111.93±9.77
|
120.27±7.18
|
123.13±6.23
|
<0.0001
|
<0.0001
|
<0.0001
|
0.339
|
DBP, mmHg
|
68.13±6.77
|
68.73±4.68
|
70.07±7.10
|
0.477
|
0.69
|
0.46
|
0.927
|
TG, mmol/L
|
0.88±0.36
|
1.08±0.67
|
1.12±0.61
|
0.207
|
0.964
|
0.225
|
0.342
|
CHO, mmol/L
|
4.16±0.76
|
4.03±0.80
|
4.34±0.77
|
0.297
|
0.268
|
0.631
|
0.796
|
HDL, mmol/L
|
1.39±0.29
|
1.24±0.22
|
1.22±0.20
|
0.015
|
0.885
|
0.018
|
0.061
|
LDL, mmol/L
|
2.32±0.53
|
2.35±0.61
|
2.71±0.66
|
0.031
|
0.069
|
0.048
|
0.987
|
Waist-hip ratio
|
0.79±0.03
|
0.85±0.04
|
0.91±0.05
|
<0.0001
|
<0.0001
|
<0.0001
|
<0.0001
|
FBG, mmol/L
|
5.52±0.37
|
5.56±0.42
|
5.60±0.42
|
0.740
|
0.907
|
0.718
|
0.933
|
Fat mass, Kg
|
9.00±3.33
|
19.70±4.05
|
28.82±6.96
|
<0.0001
|
<0.0001
|
<0.0001
|
<0.0001
|
Body fat percent, %
|
19.86±6.06
|
31.63±5.44
|
38.00±6.60
|
<0.0001
|
<0.0001
|
<0.0001
|
<0.0001
|
Visceral fat area, cm2
|
38.78±14.26
|
88.76±24.86
|
138.81±39.78
|
<0.0001
|
<0.0001
|
<0.0001
|
<0.0001
|
Values are given as mean ± SD or number of individuals (%). BMI: body mass index; SBP: systolic pressure; DBP: diastolic pressure; TG: triglyceride; CHO: cholesterol; HDL: high density lipoprotein; LDL: Low density lipoprotein; FBG: fast blood glucose.
a P value of chi-square test.
b P value of Kruskal–Wallis test.
c P-value of Dunn’s post hoc test.
Comparison between control and overweight, overweight and obese, and control and obese using random forest classifier and ROC curves
As the qualitative and quantitative analyses revealed significant differences in the metabolites levels among the three groups and indicated a gradual change from control to obese via overweight, we investigated if the metabolites could predict the risk of further obesity development. To assess this possibility, we used a random forest classifier.
As illustrated in Fig. 1, 8 metabolites were generated. The relationships among the three groups were analyzed by the random forest classifier and receiver operating characteristic (ROC) curves. Fig 2A-C shows that the area under the ROC curve (AUC) is 61.90% (95% confidence interval (CI) = 42.00–85.60%), 62.80% (95% CI = 21.50–86.50%), and 74.30% (95% CI = 56.00–91.00%) between control and overweight, overweight and obese, and control and obese in down-regulated both positive and negative ion mode. For up-regulated, the AUC is 59.70% (95% CI = 19.50–82.50%), 65.40% (95% CI = 34.10–75.50%), and 72.10% (95% CI = 49.00–93.50%) in Fig 2D-F. Together, these results indicate that the lipidomic profiles are regulated in a complex manner during the development of overweight and obesity.
The level of selected metabolites in the control, overweight and obese groups
As illustrated in Fig. 1, eight metabolites were selected from both positive and negative ion mode lipidomic profiling. The expression of the selected metabolites is shown in Fig 3. Fig. 3A and Fig. 3B indicate that 6.10_861.5490m/z and 1.82_480.3095m/z in negative ion mode were gradually decreased in the control, overweight and obese groups. Fig. 3D and Fig. 3H exhibit 1.11_396.2412m/z and 10.13_949.7263m/z in selected positive ion mode were gradually increased in the control, overweight and obese groups. However, 4.86_902.5761m/z was gradually decreased in Fig. 3E, 4.84_530.4012n and 4.96_546.3962n peaked in the overweight group (Fig. 3F-G). In summary, the development of obesity may go through the process of overweight in most cases, but it may directly develop into obesity through the alterations of some lipid metabolites.
Correlations between the selected metabolites and clinical parameters
In the body of overweight and obese people, metabolism is inevitably changed. Hence, the metabolites are also changed. To investigate the relationship between the selected metabolites and clinical parameters, we performed a correlation analysis. As shown in Fig.4A, 6.10_861.5490m/z was negatively correlated with BMI, visceral fat area, body fat percent, and waist/hip ratio. 1.82_480.3095m/z was negatively correlated with BMI, visceral fat area, body fat percent, and waist/hip ratio, but positively correlated with triglyceride in Fig. 4B. 4.84_530.4012n was negatively correlated with total cholesterol (CHO) in Fig. 4C. 1.11_396.2412m/z was positively correlated with BMI, visceral fat area, and waist/hip ratio in Fig. 4D. 4.86_902.5761m/z was negatively correlated with BMI, but positively with triglyceride (Fig. 4E). 10.13_949.7263m/z was positively correlated with BMI, visceral fat area, waist/hip ratio, triglyceride, and body fat percent (Fig. 4F).
Phospholipids phosphatidylcholine (PC) and phosphatidylethanolamine (PE) are the two most abundant phospholipid species in eukaryotic cells [17]. Lysophosphatidylcholine (LPC), an important signaling molecule and fatty acid carrier, constitutes 5–20% of total plasma phospholipids [18]. Phosphatidylinositol (PI) plays an important role in cell morphology, metabolic regulation, signal transduction and various physiological functions. 1.82_480.3095m/z was annotated as PC (15:0/0:0), PE (18:0/0:0), LPC (15:0), and LPE (0:0/18:0). 6.10_861.5490m/z was annotated as PI (14:0/22:2(13Z, 16Z))- PI (22:2(13Z,16Z)/14:0) (Additional file 3). 1.11_396.2412m/z was annotated as 18-hydroxycortisol, isohumulinone A, and 11-dihydro-12-norneoquassin; 4.86_902.5761m/z was annotated as PI (18:0/20:5 (5Z,8Z,11Z,14Z,17Z)); and 10.13_949.7263m/z was annotated as TG (20:4 (5Z,8Z,11Z,14Z) /20:3(5Z,8Z,11Z) /18:3 (9Z,12Z,15Z)). The levels of TG, 18-hydroxycortisol, isohumulinone A, and 11-dihydro-12-norneoquassin were up-regulated in the obese group, while PC, PE, LPC, LPE, and PI were significantly down-regulated in the obese group than in control and overweight individuals (Additional file 2).