Lipidomic profiling reveals distinct differences in plasma lipid composition in healthy, prediabetic, and type 2 diabetic individuals

Abstract The relationship between dyslipidemia and type 2 diabetes mellitus (T2D) has been extensively reported, but the global lipid profiles, especially in the East Asia population, associated with the development of T2D remain to be characterized. Liquid chromatography coupled to tandem mass spectrometry was applied to detect the global lipidome in the fasting plasma of 293 Chinese individuals, including 114 T2D patients, 81 prediabetic subjects, and 98 individuals with normal glucose tolerance (NGT). Both qualitative and quantitative analyses revealed a gradual change in plasma lipid features with T2D patients exhibiting characteristics close to those of prediabetic individuals, whereas they differed significantly from individuals with NGT. We constructed and validated a random forest classifier with 28 lipidomic features that effectively discriminated T2D from NGT or prediabetes. Most of the selected features significantly correlated with diabetic clinical indices. Hydroxybutyrylcarnitine was positively correlated with fasting plasma glucose, 2-hour postprandial glucose, glycated hemoglobin, and insulin resistance index (HOMA-IR). Lysophosphatidylcholines such as lysophosphatidylcholine (18:0), lysophosphatidylcholine (18:1), and lysophosphatidylcholine (18:2) were all negatively correlated with HOMA-IR. The altered plasma lipidome in Chinese T2D and prediabetic subjects suggests that lipid features may play a role in the pathogenesis of T2D and that such features may provide a basis for evaluating risk and monitoring disease development.


126
As depicted in Fig. 2, pairwise comparisons revealed that 1269 features displayed significant differences 127 between NGT individuals and T2D patients, whereas 785 and 578 features displayed significant differences 128 between prediabetic individuals vs. NGT individuals and T2D patients, respectively (Additional file 6, p<0.05). 129 The low number of variables distinguishing prediabetes and T2D, suggested that changes in a large fraction of 130 the lipid features in prediabetes and T2D were shared, implying that compared with NGT, lipid profiles 131 characterizing prediabetes and T2D are similar. Further, 117 features maintained significances after controlling 132 the FDR by Benjamini-Hochberg multiple testing correction (Additional file 6, FDR <0.05).

133
To quantify the differential features among the three groups, all detected features were assessed using criteria of higher levels in T2D than NGT overlapped with metabolites detected at higher levels in prediabetes than NGT.

141
Hence, quantitative comparison supports the notion that changes in the lipid profile may reflect a continuous 142 change from NGT to T2D via prediabetes. The finding that only 18.31% (39/213) of the features detected at 143 lower levels in T2D and prediabetes than NGT were overlapping, may suggest that down regulated changes are 144 not continuous.

294
In summary, by using LC-MS/MS based untargeted lipidomics analysis, our study is the first large-scale study 295 to explore the alterations in the plasma lipid patterns in individuals with NGT, prediabetes and T2D from East 296 China. We describe a large number of plasma lipids providing a broad coverage of major lipid categories. We 297 identify thousands of plasma lipids exhibiting remarkable difference in abundance between the three diagnostic 298 groups, with a large proportion displaying similar trends in prediabetes and type 2 diabetes. Additionally, we 299 describe stratification of predicted diabetes risk between subgroups of prediabetes based on 28 selected plasma 300 lipids. Several of the diabetes related candidates have not previously been reported. Together, this study 301 provides a better biological understanding of the insidious progression to diabetes from a lipid perspective.

Figure 1 Flowchart for participant recruitment and data processing
The recruitment of participants was based on the 2011 WHO criteria for diabetes and prediabetes diagnoses.
Blood and clinical data were acquired from 293 qualifying subjects, and untargeted lipidomics LC-MS/MS analysis was performed. The raw data were preprocessed with Progenesis QI 2.0 to extract metabolic features.
Unqualified variables and samples were detected and discarded using the BGI in-house program metaX [15].