OBJECTIVE

Few trials studied the links of food components in different diets with their induced lipidomic changes and related metabolic outcomes. Thus, we investigated specific lipidomic signatures with habitual diets and modified diabetes risk by using a trial and a cohort.

RESEARCH DESIGN AND METHODS

We included 231 Chinese with overweight and prediabetes in a randomized feeding trial with Mediterranean, traditional, or transitional diets (control diet) from February to September 2019. Plasma lipidomic profiles were measured at baseline, third month, and sixth month by high-throughput targeted liquid chromatography–mass spectrometry. Associations of the identified lipids with habitual dietary intakes were examined in another lipidomic database of a Chinese cohort (n = 1,117). The relationships between diet-induced changes of lipidomic species and diabetes risk factors were further investigated through both individual lipids and relevant modules in the trial.

RESULTS

Out of 364 lipidomic species, 26 altered across groups, including 12 triglyceride (TAG) fractions, nine plasmalogens, four phosphatidylcholines (PCs), and one phosphatidylethanolamine. TAG fractions and PCs were associated with habitual fish intake while plasmalogens were associated with red meat intake in the cohort. Of the diet-related lipidomic metabolites, 10 TAG fractions and PC(16:0/22:6) were associated with improved Matsuda index (β = 0.12 to 0.42; PFDR < 0.030). Two plasmalogens were associated with deteriorated fasting glucose (β = 0.29 to 0.31; PFDR < 0.014). Similar results were observed for TAG and plasmalogen related modules.

CONCLUSIONS

These fish- and red meat–related lipidomic signatures sensitively reflected different diets and modified type 2 diabetes risk factors, critical for optimizing dietary patterns.

Type 2 diabetes is a major global public health concern (1). Healthy dietary patterns such as a Mediterranean diet are known to reduce cardiometabolic risks, including type 2 diabetes (2). Similarly, traditional Asian diets enriched with plant-based foods also showed cardiometabolic benefits in some of cohort studies (3,4). In our previous 6-month isocaloric-restricted feeding trial (Dietary Pattern and Metabolic Health, DPMH trial), the traditional Jiangnan diet, a habitual diet originating from Southeast China consisting of plentiful vegetables, soy products, freshwater fish, and whole grains, had benefits comparable to a Mediterranean diet in reducing weight and improving glucose homeostasis, while a transitional diet (control diet) characterized by red meat, mainly pork, and refined rice raised hypoglycemia risk (5). However, it remains a challenge to illustrate specific food(s) or component(s) that affect glucose hemostasis via related metabolic pathway(s) in a given dietary pattern (6). Thus, an innovative approach is essential to capture food-related signatures for optimizing diets in the future.

With advanced omics technology, lipidomics has offered a powerful tool to identify reliable biomarkers of dietary exposures, which is not susceptible to recall bias and measurement error related to questionnaires (7). For instance, we previously found that four dairy-related sphingomyelins were inversely associated with a 6-year change of fasting glucose (8). Indeed, disturbed lipid homeostasis was linked to a higher risk of type 2 diabetes according to a number of longitudinal studies (9). In our earlier study, 14 sphingolipids, 8 glycerophospholipids, and 9 glycerolipids were significantly associated with incident type 2 diabetes (1013). Additionally, lipidomic species with specific molecular features like length and saturation of acyl chains were reported to influence their associations with type 2 diabetes risk in western populations (14). Overall, lipidomics could serve as a promising means to reveal comprehensive insights into lipid metabolism linking between dietary intake and pathogenesis of type 2 diabetes.

To date, only a few intervention trials conducted in western populations explored diet-induced changes of lipid species, including those of phosphatidylcholines (PCs), cholesterol esters, and lysophospholipids (15,16). Notably, genetic diversity and dietary modification could contribute to lipid species with distinctive structures and different health consequences (11). Well-designed feeding trials can definitely provide a unique opportunity to discover reliable diet-related lipidomic signatures in populations with diverse dietary and genetic backgrounds.

Therefore, with repeatedly measured lipidomic profiles in the well-controlled 6-month DPMH feeding trial and another lipidomic database from a cohort (Nutrition and Health of Aging Population in China [NHAPC]), we aimed to investigate: 1) changed lipidomic signatures due to feeding with a Mediterranean diet, traditional Jiangnan diet, and control diet in the DPMH trial; 2) links of the discovered lipid signatures with habitual food intakes in the southern subpopulation of NHAPC cohort; and 3) associations of the food-related lipidomic signatures with modified diabetes risk factors in the DPMH study.

Study Design and Participants

The DPMH is a parallel-arm randomized controlled trial comparing the effects of a Mediterranean diet, traditional Jiangnan diet, and control diet on body weight and glucose homeostasis (ClinicalTrials.gov: NCT03856762), as reported previously (5). In brief, the DPMH included 253 participants age 25–60 years, with BMI ≥24 kg/m2 and fasting glucose ≥5.6 mmol/L, from SAIC Volkswagen Automotive Company in Shanghai, China in 2019. Participants were randomly assigned to a Mediterranean diet (n = 84), traditional Jiangnan diet (n = 85), or control diet (n = 84) for 6 months. A total of 231 participants were included in the present analysis after excluding those who left the trial after less than 3 months.

To further determine the related habitual food groups, the identified lipid species in DPMH were investigated in the baseline lipidomic data among 1,117 Shanghai residents with reasonable energy intake (800–4,000 kcal/day for men or 500–3,500 kcal/day for women) from the NHAPC study, which recruited 3,289 Beijing and Shanghai residents aged 50–70 years in 2005 (17). The protocols of the two studies were approved by the institutional review boards of the Shanghai Institute of Nutritional and Health (Shanghai, China), Chinese Academy of Sciences, and all participants provided written informed consent.

Study Diet and Dietary Assessment

In the DPMH study, a weekly 5-day feeding regimen was followed for 6 months with three isocaloric-restricted diets providing 1,600 kcal/day for men and 1,300 kcal/day for women. In addition to the intake of vegetables, fruits, and whole grains, the Mediterranean diet adopted more Mediterranean ingredients like extra-virgin olive oil, marine fish, and nuts, while the traditional Jiangnan diet was enriched with soy products and freshwater fish according to the traditional cuisines in Southeast China. The control diet was characterized by relatively larger portions of red meat (mainly pork) and refined grains (mainly white rice) to represent the transitional dietary pattern in megacities like Shanghai. Detailed information for dietary intake during the intervention has been reported previously (5).

Dietary intake at baseline in NHAPC was assessed by using a modified and validated 74-item food frequency questionnaire (18), and nutrient intakes were calculated by the Chinese Food Composition Tables (19). Physical activity level in both studies was assessed by using a modified Short-form International Physical Activity Questionnaire (20).

Measurements of Diabetes Risk Factors

In the DPMH study, anthropometric measurements and blood sampling were performed by trained staff members at baseline, third month, and sixth month. Overnight fasting blood samples and samples at 30, 60, and 120 min of the 2-h oral glucose tolerance tests were collected. Plasma concentrations of glucose and insulin were measured by an automatic analyzer (Hitachi 7080) using commercial kits manufactured by Wako Pure Chemical Industries. Insulin sensitivity was evaluated by the Matsuda index (21). β-Cell function was evaluated by insulinogenic index and disposition index (22,23).

Measurements of Lipidomics

A high-throughput targeted liquid chromatography tandem mass spectrometry using electrospray ionization under both the positive and negative modes was performed to measure plasma lipidomics at baseline, third month, and sixth month for the DPMH trial, as well as plasma lipidomics at baseline for NHAPC in the same platform. Detailed information about sample preparation and data acquisition has been reported elsewhere (11). To monitor repeatability of the data, the quality control samples were inserted and analyzed at the beginning and end of the daily sequence, as well as between every 10 samples. Triglyceride (TAG) fractions with the same number of carbon atoms and double bonds but different fatty acyls were identified as separate lipid species. Final analyses included 364 lipids in DPMH and 649 lipids in NHAPC after excluding those with missing values >20% and/or with coefficient of variation >30%.

Statistical Analyses

Descriptive data of participants at baseline and changes during the intervention periods are shown as means ± SD for continuous variables, and number (%) for categorical variables. Missing values for the lipidomics were replaced by half of the minimum detectable values. Baseline individual lipidomic values were naturally log-transformed and scaled to a SD of 1. Changes in lipidomic values were calculated after natural log transformation, and the time differences were also scaled. The statistical assessments of the association between diets, lipid signatures, and diabetes risk factors were conducted in three steps (Supplementary Fig. 1).

Step 1: Identifying Lipidomic Signatures in the Intervention With Three Diets

An exploratory weighted gene coexpression network analysis was used to determine modules of highly interconnected lipid metabolites that differed between three dietary groups. To identify the lipidomic signatures of the three dietary patterns, changes of individual lipids according to three diet groups were assessed with linear mixed-effect models with unstructured covariance structure adjusted for age, sex, baseline BMI, physical activity, and baseline values of the lipid metabolites. Group, time, and their interactions were treated as fixed effects, while participant was treated as a random effect. Lipid metabolites were analyzed in separate models, and estimated marginal means were the adjusted values from the above models.

Step 2: Specific Food Groups Associated With the Lipidomic Signatures in the Separate Cohort

To investigate food-related lipidomic signatures in different dietary patterns, we analyzed cross-sectional associations between individual lipid species and habitual intakes of eight food groups (i.e., vegetables, fruits, dairy products, fish, nuts and legumes, red meat, processed meat, and refined grains) derived from the commonly used classification in Chinese food choices in the NHAPC study (24). For a given food group, multivariable linear regression was conducted after adjustment for age, sex, baseline BMI, physical activity, and the other seven food groups.

Step 3: Discovering Links Between the Diet-Induced Changes of Lipidomic Signatures and Diabetes Risk Factors

To evaluate the associations of changed lipidomic metabolites with improved diabetes risk factors during the period of the DPMH trial, multivariate linear regression was used for lipid modules, which was adjusted for age, sex, baseline BMI (except when BMI was used as the response variable), physical activity, baseline value of the response variable, and the rest modules. Linear mixed-effect models were used for individual lipids after adjusting for age, sex, baseline BMI (except when BMI was used as the response variable), physical activity, baseline values of the response variable, and the sum of lipid subclass it belongs to. False discovery rate (FDR) corrections were used to define statistical significance. All analyses were performed with R version 3.5.3, and Cytoscape (version 3.7.1) was used to plot the network of lipid modules.

Characteristics of Participants

Baseline characteristics of the study populations from the DPMH trial and NHAPC cohort are shown in Table 1. Compared with the DPMH trial, the NHAPC study included more women (DPMH: 15% vs. NHAPC: 60%) of older age (DPMH: 38 years vs. NHAPC: 59 years) with higher physical activity (percentage of high physical activity, DPMH: 15% vs. NHAPC: 49%) but lower total energy intake (DPMH: 2,346 kcal/day vs. NHAPC: 2,154 kcal/day).

Table 1

Characteristics of participants at the baseline of the DPMH study and Shanghai subjects at the baseline of the NHAPC study

DPMHNHAPC (n = 1,117)
MD (n = 75)TJD (n = 79)CD (n = 77)
Male, n (%) 63 (84) 67 (85) 67 (87) 446 (40) 
Age, years 38 ± 8 38 ± 9 38 ± 9 59 ± 6 
High physical activity, n (%) 7 (10) 14 (18) 13 (17) 544 (49) 
BMI, kg/m2 26.8 ± 3.2 26.8 ± 2.9 26.8 ± 3.0 23.7 ± 3.3 
Cholesterol, mmol/L 5.00 ± 0.85 4.93 ± 0.71 4.95 ± 0.81 4.45 ± 0.90 
HDL cholesterol, mmol/L 1.24 ± 0.37 1.27 ± 0.29 1.21 ± 0.28 1.26 ± 0.33 
LDL cholesterol, mmol/L 3.21 ± 0.78 3.17 ± 0.69 3.24 ± 0.76 3.02 ± 0.87 
TAG, mmol/L 1.90 ± 1.84 1.70 ± 1.18 1.81 ± 0.94 1.30 ± 0.99 
Fasting glucose, mmol/L 6.24 ± 0.76 6.12 ± 0.52 6.22 ± 0.76 5.47 ± 1.31 
Fasting insulin, uIU/mL 14.9 ± 10.2 17.5 ± 12.3 15.2 ± 7.6 17.8 ± 8.7 
HbA1c, % 5.88 ± 0.65 5.85 ± 0.45 5.86 ± 0.66 5.87 ± 0.90 
Dietary intake information     
 Total energy, kcal 2,352 ± 943 2,312 ± 838 2,374 ± 809 2,154 ± 598 
 Carbohydrate, % 55.8 ± 8.9 56.0 ± 9.7 56.9 ± 9.4 62.3 ± 8.5 
 Fat, % 29.1 ± 8.0 28.7 ± 8.3 28.3 ± 8.6 25.8 ± 7.3 
 SFA, % 9.2 ± 2.9 9.1 ± 2.8 9 ± 3 5.8 ± 2.2 
 MUFA, % 11.3 ± 3.4 11.3 ± 3.6 11.1 ± 3.6 9.7 ± 3.0 
 PUFA, % 4.1 ± 1.5 3.9 ± 1 3.7 ± 1.1 8.0 ± 2.6 
 Protein, % 15.1 ± 2.2 15.3 ± 2.3 14.8 ± 2.1 11.8 ± 2.3 
 Cholesterol, mg 616 ± 333 612 ± 295 663 ± 415 284 ± 174 
 Fiber, g/1,000 kcal 5.5 ± 1.8 5.7 ± 2.3 5.1 ± 2.1 4.9 ± 1.9 
 Refined grain, g 264 ± 104 277 ± 117 296 ± 113 362 ± 151 
 Whole grain, g 22 ± 34 25 ± 53 24 ± 39 7 ± 14 
 Vegetables, g 281 ± 195 315 ± 185 281 ± 206 270 ± 160 
 Fruits, g 163 ± 136 154 ± 123 130 ± 99 143 ± 165 
 Red meat, g 133 ± 105 125 ± 95 128 ± 100 30 ± 32 
 Processed meat, g 14 ± 20 12 ± 16 9 ± 12 1 ± 3 
 Poultry, g 37 ± 43 38 ± 41 31 ± 32 9 ± 12 
 Freshwater fish, g 28 ± 27 29 ± 32 33 ± 44 22 ± 24 
 Marine fish, g 20 ± 23 20 ± 32 15 ± 19 14 ± 18 
 Soy products, g 69 ± 55 59 ± 45 55 ± 43 70 ± 48 
 Nuts, g 23 ± 39 19 ± 24 14 ± 19 6 ± 13 
 Dairy products, g 178 ± 183 191 ± 204 185 ± 199 69 ± 113 
DPMHNHAPC (n = 1,117)
MD (n = 75)TJD (n = 79)CD (n = 77)
Male, n (%) 63 (84) 67 (85) 67 (87) 446 (40) 
Age, years 38 ± 8 38 ± 9 38 ± 9 59 ± 6 
High physical activity, n (%) 7 (10) 14 (18) 13 (17) 544 (49) 
BMI, kg/m2 26.8 ± 3.2 26.8 ± 2.9 26.8 ± 3.0 23.7 ± 3.3 
Cholesterol, mmol/L 5.00 ± 0.85 4.93 ± 0.71 4.95 ± 0.81 4.45 ± 0.90 
HDL cholesterol, mmol/L 1.24 ± 0.37 1.27 ± 0.29 1.21 ± 0.28 1.26 ± 0.33 
LDL cholesterol, mmol/L 3.21 ± 0.78 3.17 ± 0.69 3.24 ± 0.76 3.02 ± 0.87 
TAG, mmol/L 1.90 ± 1.84 1.70 ± 1.18 1.81 ± 0.94 1.30 ± 0.99 
Fasting glucose, mmol/L 6.24 ± 0.76 6.12 ± 0.52 6.22 ± 0.76 5.47 ± 1.31 
Fasting insulin, uIU/mL 14.9 ± 10.2 17.5 ± 12.3 15.2 ± 7.6 17.8 ± 8.7 
HbA1c, % 5.88 ± 0.65 5.85 ± 0.45 5.86 ± 0.66 5.87 ± 0.90 
Dietary intake information     
 Total energy, kcal 2,352 ± 943 2,312 ± 838 2,374 ± 809 2,154 ± 598 
 Carbohydrate, % 55.8 ± 8.9 56.0 ± 9.7 56.9 ± 9.4 62.3 ± 8.5 
 Fat, % 29.1 ± 8.0 28.7 ± 8.3 28.3 ± 8.6 25.8 ± 7.3 
 SFA, % 9.2 ± 2.9 9.1 ± 2.8 9 ± 3 5.8 ± 2.2 
 MUFA, % 11.3 ± 3.4 11.3 ± 3.6 11.1 ± 3.6 9.7 ± 3.0 
 PUFA, % 4.1 ± 1.5 3.9 ± 1 3.7 ± 1.1 8.0 ± 2.6 
 Protein, % 15.1 ± 2.2 15.3 ± 2.3 14.8 ± 2.1 11.8 ± 2.3 
 Cholesterol, mg 616 ± 333 612 ± 295 663 ± 415 284 ± 174 
 Fiber, g/1,000 kcal 5.5 ± 1.8 5.7 ± 2.3 5.1 ± 2.1 4.9 ± 1.9 
 Refined grain, g 264 ± 104 277 ± 117 296 ± 113 362 ± 151 
 Whole grain, g 22 ± 34 25 ± 53 24 ± 39 7 ± 14 
 Vegetables, g 281 ± 195 315 ± 185 281 ± 206 270 ± 160 
 Fruits, g 163 ± 136 154 ± 123 130 ± 99 143 ± 165 
 Red meat, g 133 ± 105 125 ± 95 128 ± 100 30 ± 32 
 Processed meat, g 14 ± 20 12 ± 16 9 ± 12 1 ± 3 
 Poultry, g 37 ± 43 38 ± 41 31 ± 32 9 ± 12 
 Freshwater fish, g 28 ± 27 29 ± 32 33 ± 44 22 ± 24 
 Marine fish, g 20 ± 23 20 ± 32 15 ± 19 14 ± 18 
 Soy products, g 69 ± 55 59 ± 45 55 ± 43 70 ± 48 
 Nuts, g 23 ± 39 19 ± 24 14 ± 19 6 ± 13 
 Dairy products, g 178 ± 183 191 ± 204 185 ± 199 69 ± 113 

Data are mean ± SD or n (%). CD, control diet; MD, Mediterranean diet; MUFA, monounsaturated fatty acid; PUFA, polyunsaturated fatty acid; SFA, saturated fatty acid; TJD, traditional Jiangnan diet.

Level of physical activity was calculated and classified based on the International Physical Activity Questionnaire.

Dietary information was calculated from the food frequency questionnaire.

Lipid Profiling in DPMH and in NHAPC Studies

A total of 364 lipid species spanning 19 lipid subclasses in the DPMH trial were identified and quantitated, and 283 (77.7%) of them were replicated in the NHAPC cohort. Of the 364 lipids, no significant differences were observed between the three dietary groups at baseline of the DPMH study (Supplementary Table 1).

Changes in Lipidomic Signatures According to Different Dietary Patterns During the Intervention

Seven lipid modules were generated by network analysis (Supplementary Fig. 2). Compared with the control group, both Mediterranean diet and traditional Jiangnan diet groups resulted in a lower level of yellow module dominated by alkylphosphatidylethanolamines (PE(O)s) and alkenylphosphatidylethanolamines (PE(P)s) (P = 0.029), and a marginally higher level of the brown module mainly comprising TAG fractions with long-chain highly unsaturated fatty acyls (P = 0.053) (Supplementary Table 2). As shown in Supplementary Table 3, levels of most lipid metabolites decreased in all three diet groups over the 6-month intervention, accompanied with weight loss induced by calorie restriction as reported in other studies (25). The changes in 26 lipids were significantly different between the three dietary patterns during the 6-month intervention (PFDR < 0.05). The direction of changes and significant time effects are presented in Fig. 1A. Compared with the control diet group, 12 TAG fractions with long-chain highly unsaturated fatty acyls or odd chains reduced to a lesser extent in both Mediterranean diet and traditional Jiangnan diet groups, and four of the TAG fractions in the Mediterranean diet group decreased to a lesser extent than those in the traditional Jiangnan diet group, whereas four PE(O)s and five PE(P)s decreased more in the Mediterranean diet and traditional Jiangnan diet groups than in the control diet group. On the other hand, four PCs and one phosphatidylethanolamine (PE) significantly increased in the Mediterranean diet group compared with the other two groups.

Figure 1

Lipids significantly changed between three intervention groups and their associations with food groups. (A) Heatmap of significant changes at third and sixth month of plasma lipids among three intervention groups. Changes in each group were examined by linear mixed models adjusted for age, sex, baseline BMI, and physical activity. Colors represent directions of the third- or sixth-month changes in each group (red: increase; blue: decrease). (B) Heatmap of lipids associated with food groups at NHAPC baseline. Coefficients for associations calculated by multivariate linear regression indicate the changes in lipids per SD increment in daily consumption of specific food groups. Colors represent directions of the association (red: positive; blue: inverse), while the color depth indicates the strength of associations (the darker the stronger). *P < 0.05, **P < 0.01, and ***P < 0.001 after FDR correction.

Figure 1

Lipids significantly changed between three intervention groups and their associations with food groups. (A) Heatmap of significant changes at third and sixth month of plasma lipids among three intervention groups. Changes in each group were examined by linear mixed models adjusted for age, sex, baseline BMI, and physical activity. Colors represent directions of the third- or sixth-month changes in each group (red: increase; blue: decrease). (B) Heatmap of lipids associated with food groups at NHAPC baseline. Coefficients for associations calculated by multivariate linear regression indicate the changes in lipids per SD increment in daily consumption of specific food groups. Colors represent directions of the association (red: positive; blue: inverse), while the color depth indicates the strength of associations (the darker the stronger). *P < 0.05, **P < 0.01, and ***P < 0.001 after FDR correction.

Close modal

Specific Food Groups Associated Lipid Signatures

Of the 26 identified lipids, all 12 less reduced TAG fractions in the Mediterranean diet and traditional Jiangnan diet group were positively associated with fish intake (β = 0.10 to 0.14; PFDR < 0.001) (Fig. 1B). PC(16:0/22:6) increased in the Mediterranean diet group was also associated with fish intake (β = 0.15; PFDR < 0.001). In contrast, six of the nine PE(O)s and PE(P)s, which decreased more in the Mediterranean diet and traditional Jiangnan diet groups, were positively associated with the red meat intake (β = 0.09 to 0.16; PFDR < 0.004), while PE(O-18:0/22:4) was inversely associated with fish intake (β = −0.08; PFDR = 0.024).

Lipid Signatures and Diabetes Risk Factors

The brown module mainly comprising TAG fractions with long-chain highly unsaturated fatty acyls was positively correlated with improved Matsuda index (β = 0.23; PFDR = 0.020) and disposition index (β = 0.25; PFDR = 0.017), whereas the yellow module dominated by PE(O) and PE(P) was positively associated with deteriorated fasting glucose (β = 0.12; PFDR = 0.035) (Table 2). Among the 12 fish-related TAG fractions identified, 3 of them were inversely associated with the changes in BMI (β = −0.06 to −0.07; PFDR = 0.048), while 10 of them were positively associated with changes in the Matsuda index (β = 0.12 to 0.21; PFDR < 0.030); and 5 of them showed a trend to be positively associated with improvements in the disposition index (β = 0.21 to 0.28; PFDR < 0.10) (Table 3). Moreover, elevated levels of PC(16:0/22:6), PC(18:2/22:6), and PE(18:0/14:0) were associated with reduced BMI (β = −0.16 to −0.25; PFDR < 0.001) and improved Matsuda index (β = 0.32 to 0.54; PFDR < 0.001) (Table 3). In contrast, among the red meat–associated lipids, increased levels of PE(P-18:0/20:3) and PE(P-18:1/20:3) were associated with deteriorated fasting glucose (β = 0.29 to 0.31; PFDR = 0.014) and Matsuda index (β = −0.38 to −0.54; PFDR ≤ 0.014) (Table 3). PE(P-18:1/20:3) also showed a marginal association with deteriorated disposition index (β = −0.32; PFDR = 0.080) (Table 3).

Table 2

Associations between the changes of lipid modules and the changes of diabetes risk factors

ModuleBMIFasting glucoseMatsuda indexInsulinogenic indexDisposition index
β (95% CI)PFDRβ (95% CI)PFDRβ (95% CI)PFDRβ (95% CI)PFDRβ (95% CI)PFDR
Yellow (mainly PE(P) and PE(O), n = 40) 0.01 (−0.11, 0.13) 0.86 0.12 (0.01, 0.22) 0.035 0.06 (−0.08, 0.20) 0.40 −0.04 (−0.12, 0.04) 0.64 −0.08 (−0.18, 0.01) 0.26 
Blue (mainly SM and Cer, n = 64) −0.10 (−0.22, 0.02) 0.15 −0.02 (−0.11, 0.08) 0.81 0.05 (−0.07, 0.18) 0.43 −0.08 (−0.16, 0.01) 0.31 −0.06 (−0.16, 0.03) 0.39 
Green (PE, PI, and PG, n = 29) −0.08 (−0.21, 0.06) 0.34 −0.08 (−0.18, 0.03) 0.29 0.18 (−0.02, 0.38) 0.099 −0.05 (−0.14, 0.04) 0.58 −0.06 (−0.2, 0.08) 0.51 
Brown (mainly TAG with CN of 54–60 and DB of 5–10, n = 48) −0.06 (−0.24, 0.13) 0.56 0.04 (−0.09, 0.19) 0.56 0.23 (0.01, 0.46) 0.020 0.11 (−0.04, 0.27) 0.31 0.25 (0.08, 0.42) 0.017 
Black (mainly TAG with CN of 51–56 and DB of 3–8, n = 19) 0.10 (−0.09, 0.28) 0.28 0.04 (−0.09, 0.18) 0.57 −0.11 (−0.33, 0.11) 0.24 0.15 (−0.15, 0.45) 0.17 0.08 (−0.1, 0.25) 0.46 
Red (mainly TAG with CN of 46–52 and DB of 0–5, n = 22) 0.05 (−0.24, 0.34) 0.69 0.20 (−0.01, 0.41) 0.070 −0.26 (−0.61, 0.08) 0.099 −0.07 (−0.38, 0.24) 0.62 −0.07 (−0.3, 0.16) 0.63 
Turquoise (other TAGs not included in above modules, n = 66) −0.02 (−0.48, 0.44) 0.92 −0.26 (−0.57, 0.04) 0.12 −0.04 (−0.56, 0.48) 0.83 −0.35 (−0.65, −0.06) 0.093 −0.63 (−0.96, −0.31) 0.002 
ModuleBMIFasting glucoseMatsuda indexInsulinogenic indexDisposition index
β (95% CI)PFDRβ (95% CI)PFDRβ (95% CI)PFDRβ (95% CI)PFDRβ (95% CI)PFDR
Yellow (mainly PE(P) and PE(O), n = 40) 0.01 (−0.11, 0.13) 0.86 0.12 (0.01, 0.22) 0.035 0.06 (−0.08, 0.20) 0.40 −0.04 (−0.12, 0.04) 0.64 −0.08 (−0.18, 0.01) 0.26 
Blue (mainly SM and Cer, n = 64) −0.10 (−0.22, 0.02) 0.15 −0.02 (−0.11, 0.08) 0.81 0.05 (−0.07, 0.18) 0.43 −0.08 (−0.16, 0.01) 0.31 −0.06 (−0.16, 0.03) 0.39 
Green (PE, PI, and PG, n = 29) −0.08 (−0.21, 0.06) 0.34 −0.08 (−0.18, 0.03) 0.29 0.18 (−0.02, 0.38) 0.099 −0.05 (−0.14, 0.04) 0.58 −0.06 (−0.2, 0.08) 0.51 
Brown (mainly TAG with CN of 54–60 and DB of 5–10, n = 48) −0.06 (−0.24, 0.13) 0.56 0.04 (−0.09, 0.19) 0.56 0.23 (0.01, 0.46) 0.020 0.11 (−0.04, 0.27) 0.31 0.25 (0.08, 0.42) 0.017 
Black (mainly TAG with CN of 51–56 and DB of 3–8, n = 19) 0.10 (−0.09, 0.28) 0.28 0.04 (−0.09, 0.18) 0.57 −0.11 (−0.33, 0.11) 0.24 0.15 (−0.15, 0.45) 0.17 0.08 (−0.1, 0.25) 0.46 
Red (mainly TAG with CN of 46–52 and DB of 0–5, n = 22) 0.05 (−0.24, 0.34) 0.69 0.20 (−0.01, 0.41) 0.070 −0.26 (−0.61, 0.08) 0.099 −0.07 (−0.38, 0.24) 0.62 −0.07 (−0.3, 0.16) 0.63 
Turquoise (other TAGs not included in above modules, n = 66) −0.02 (−0.48, 0.44) 0.92 −0.26 (−0.57, 0.04) 0.12 −0.04 (−0.56, 0.48) 0.83 −0.35 (−0.65, −0.06) 0.093 −0.63 (−0.96, −0.31) 0.002 

Associations between the changes of lipid modules and BMI, fasting glucose, Matsuda index, insulinogenic index, and disposition index were calculated using multivariate linear regression, which was adjusted for age, sex, baseline BMI (except when BMI was used as the response variable), physical activity, baseline value of the response variable, and the rest modules. The β-coefficient and 95% CI obtained from linear regression were presented. PFDR were obtained after FDR correction. Cer, ceramide; CN, total number of carbons; DB, number of double bonds; PI, phosphatidylinositol; SM, sphingomyelin.

Table 3

Associations between the changes of lipids related to food groups and the changes of diabetes risk factors

BMIFasting glucoseMatsuda indexInsulinogenic indexDisposition index
β (95% CI)PFDRβ (95% CI)PFDRβ (95% CI)PFDRβ (95% CI)PFDRβ (95% CI)PFDR
TAG(54:7)-FA(22:6) −0.02 (−0.07, 0.03) 0.40 0 (−0.12, 0.11) 0.97 0.11 (0, 0.23) 0.079 −0.1 (−0.3, 0.11) 0.78 0.13 (−0.09, 0.35) 0.39 
TAG(55:2)-FA(18:2) −0.03 (−0.08, 0.02) 0.27 −0.04 (−0.16, 0.08) 0.58 0.18 (0.07, 0.29) 0.005 0.18 (0.03, 0.32) 0.70 0.28 (0.13, 0.44) 0.063 
TAG(54:8)-FA(22:6) 0(.01 (−0.04, 0.05) 0.69 0.06 (−0.06, 0.17) 0.40 0.07 (−0.04, 0.18) 0.28 0.14 (−0.02, 0.3) 0.76 0.26 (0.06, 0.45) 0.080 
TAG(56:9)-FA(22:6) −0.02 (−0.06, 0.02) 0.36 −0.03 (−0.14, 0.08) 0.59 0.12 (0.02, 0.23) 0.030 0.1 (−0.04, 0.24) 0.78 0.26 (0.1, 0.42) 0.063 
TAG(55:1)-FA(16:0) −0.04 (−0.08, 0.01) 0.19 −0.04 (−0.15, 0.08) 0.60 0.20 (0.08, 0.32) 0.002 0.06 (−0.11, 0.23) 0.78 0.2 (0.03, 0.38) 0.13 
TAG(57:2)-FA(18:1) −0.05 (−0.09, 0) 0.08 −0.08 (−0.18, 0.02) 0.31 0.17 (0.07, 0.26) 0.002 0.06 (−0.06, 0.17) 0.78 0.18 (0.06, 0.3) 0.11 
TAG(56:8)-FA(22:6) −0.06 (−0.11, −0.02) 0.048 −0.06 (−0.18, 0.05) 0.40 0.19 (0.08, 0.30) 0.002 0.06 (−0.09, 0.2) 0.78 0.21 (0.05, 0.38) 0.080 
TAG(58:9)-FA(22:6) −0.05 (−0.09, 0) 0.080 −0.06 (−0.17, 0.04) 0.38 0.18 (0.08, 0.28) 0.002 0.12 (−0.06, 0.29) 0.74 0.23 (0.11, 0.35) 0.063 
TAG(56:6)-FA(22:6) −0.07 (−0.12, −0.02) 0.048 −0.06 (−0.19, 0.07) 0.43 0.19 (0.07, 0.32) 0.006 −0.11 (−0.23, 0.01) 0.78 0.06 (−0.13, 0.24) 0.71 
TAG(56:7)-FA(22:6) −0.06 (−0.11, −0.01) 0.048 −0.07 (−0.20, 0.05) 0.38 0.21 (0.09, 0.33) 0.002 −0.06 (−0.15, 0.03) 0.78 0.11 (−0.04, 0.27) 0.38 
TAG(58:7)-FA(22:6) −0.03 (−0.08, 0.02) 0.27 −0.06 (−0.18, 0.06) 0.40 0.16 (0.05, 0.28) 0.012 −0.04 (−0.12, 0.04) 0.78 0.08 (−0.06, 0.22) 0.55 
TAG(58:8)-FA(22:6) −0.05 (−0.09, 0) 0.083 −0.07 (−0.18, 0.04) 0.38 0.17 (0.07, 0.28) 0.004 0.04 (−0.06, 0.14) 0.78 0.16 (0.03, 0.28) 0.20 
PE(O-16:0/22:5) 0.07 (−0.02, 0.15) 0.19 0.23 (0.01, 0.44) 0.17 −0.04 (−0.24, 0.17) 0.81 0.06 (−0.12, 0.23) 0.78 0.01 (−0.26, 0.28) 0.95 
PE(P-18:0/16:0) 0.06 (−0.05, 0.17) 0.35 0.04 (−0.23, 0.31) 0.31 −0.02 (−0.28, 0.24) 0.87 −0.02 (−0.16, 0.13) 0.90 −0.12 (−0.32, 0.08) 0.48 
PE(P-16:0/22:5) 0.10 (0, 0.21) 0.11 0.22 (−0.04, 0.48) 0.10 0.02 (−0.22, 0.26) 0.87 −0.06 (−0.2, 0.09) 0.80 0.03 (−0.22, 0.28) 0.93 
PE(O-16:0/20:4) −0.09 (−0.18, 0.01) 0.12 −0.14 (−0.38, 0.09) 0.38 0.32 (0.09, 0.54) 0.011 −0.13 (−0.27, 0) 0.80 −0.08 (−0.26, 0.1) 0.71 
PE(P-18:1/20:4) −0.11 (−0.23, 0) 0.11 0.02 (−0.27, 0.30) 0.27 0.20 (−0.08, 0.47) 0.24 −0.18 (−0.36, −0.01) 0.76 −0.23 (−0.44, −0.03) 0.23 
PE(O-18:0/22:4) −0.05 (−0.11, 0.02) 0.22 0.09 (−0.08, 0.26) 0.40 0.08 (−0.08, 0.23) 0.40 −0.03 (−0.12, 0.06) 0.80 0.01 (−0.15, 0.18) 0.95 
PE(O-18:0/20:3) 0.05 (−0.02, 0.11) 0.22 0.14 (−0.02, 0.3) 0.27 −0.10 (−0.26, 0.05) 0.26 −0.25 (−0.49, 0) 0.60 −0.23 (−0.43, −0.03) 0.15 
PE(P-18:0/20:3) 0.09 (−0.03, 0.21) 0.18 0.31 (0.01, 0.60) 0.014 −0.38 (−0.67, −0.1) 0.014 −0.28 (−0.65, 0.08) 0.60 −0.22 (−0.49, 0.04) 0.20 
PE(P-18:1/20:3) 0.15 (0.03, 0.27) 0.048 0.29 (0.01, 0.58) 0.014 −0.54 (−0.82, −0.26) 0.001 −0.2 (−0.43, 0.03) 0.70 −0.32 (−0.55, −0.08) 0.080 
PC(16:0/22:6) −0.20 (−0.27, −0.13) <0.001 −0.12 (−0.30, 0.05) 0.33 0.42 (0.25, 0.58) < 0.001 −0.02 (−0.12, 0.08) 0.81 0.1 (−0.02, 0.22) 0.37 
PE(18:0/14:0) −0.25 (−0.32, −0.18) <0.001 −0.34 (−0.52, −0.17) 0.003 0.54 (0.37, 0.72) < 0.001 0.05 (−0.15, 0.26) 0.80 0.11 (−0.03, 0.25) 0.37 
PC(18:2/22:6) −0.16 (−0.22, −0.09) <0.001 −0.19 (−0.35, −0.04) 0.076 0.32 (0.17, 0.47) < 0.001 0.08 (−0.12, 0.28) 0.80 0.12 (0.02, 0.22) 0.37 
PC(16:0/18:1) −0.12 (−0.22, −0.02) 0.056 −0.18 (−0.42, 0.06) 0.31 0.16 (−0.07, 0.4) 0.24 −0.17 (−0.45, 0.11) 0.70 0.01 (−0.18, 0.2) 0.95 
PC(18:1/18:1) −0.07 (−0.13, 0) 0.090 −0.06 (−0.22, 0.09) 0.51 0.02 (−0.14, 0.18) 0.84 0.09 (−0.01, 0.2) 0.80 0.1 (−0.04, 0.23) 0.38 
BMIFasting glucoseMatsuda indexInsulinogenic indexDisposition index
β (95% CI)PFDRβ (95% CI)PFDRβ (95% CI)PFDRβ (95% CI)PFDRβ (95% CI)PFDR
TAG(54:7)-FA(22:6) −0.02 (−0.07, 0.03) 0.40 0 (−0.12, 0.11) 0.97 0.11 (0, 0.23) 0.079 −0.1 (−0.3, 0.11) 0.78 0.13 (−0.09, 0.35) 0.39 
TAG(55:2)-FA(18:2) −0.03 (−0.08, 0.02) 0.27 −0.04 (−0.16, 0.08) 0.58 0.18 (0.07, 0.29) 0.005 0.18 (0.03, 0.32) 0.70 0.28 (0.13, 0.44) 0.063 
TAG(54:8)-FA(22:6) 0(.01 (−0.04, 0.05) 0.69 0.06 (−0.06, 0.17) 0.40 0.07 (−0.04, 0.18) 0.28 0.14 (−0.02, 0.3) 0.76 0.26 (0.06, 0.45) 0.080 
TAG(56:9)-FA(22:6) −0.02 (−0.06, 0.02) 0.36 −0.03 (−0.14, 0.08) 0.59 0.12 (0.02, 0.23) 0.030 0.1 (−0.04, 0.24) 0.78 0.26 (0.1, 0.42) 0.063 
TAG(55:1)-FA(16:0) −0.04 (−0.08, 0.01) 0.19 −0.04 (−0.15, 0.08) 0.60 0.20 (0.08, 0.32) 0.002 0.06 (−0.11, 0.23) 0.78 0.2 (0.03, 0.38) 0.13 
TAG(57:2)-FA(18:1) −0.05 (−0.09, 0) 0.08 −0.08 (−0.18, 0.02) 0.31 0.17 (0.07, 0.26) 0.002 0.06 (−0.06, 0.17) 0.78 0.18 (0.06, 0.3) 0.11 
TAG(56:8)-FA(22:6) −0.06 (−0.11, −0.02) 0.048 −0.06 (−0.18, 0.05) 0.40 0.19 (0.08, 0.30) 0.002 0.06 (−0.09, 0.2) 0.78 0.21 (0.05, 0.38) 0.080 
TAG(58:9)-FA(22:6) −0.05 (−0.09, 0) 0.080 −0.06 (−0.17, 0.04) 0.38 0.18 (0.08, 0.28) 0.002 0.12 (−0.06, 0.29) 0.74 0.23 (0.11, 0.35) 0.063 
TAG(56:6)-FA(22:6) −0.07 (−0.12, −0.02) 0.048 −0.06 (−0.19, 0.07) 0.43 0.19 (0.07, 0.32) 0.006 −0.11 (−0.23, 0.01) 0.78 0.06 (−0.13, 0.24) 0.71 
TAG(56:7)-FA(22:6) −0.06 (−0.11, −0.01) 0.048 −0.07 (−0.20, 0.05) 0.38 0.21 (0.09, 0.33) 0.002 −0.06 (−0.15, 0.03) 0.78 0.11 (−0.04, 0.27) 0.38 
TAG(58:7)-FA(22:6) −0.03 (−0.08, 0.02) 0.27 −0.06 (−0.18, 0.06) 0.40 0.16 (0.05, 0.28) 0.012 −0.04 (−0.12, 0.04) 0.78 0.08 (−0.06, 0.22) 0.55 
TAG(58:8)-FA(22:6) −0.05 (−0.09, 0) 0.083 −0.07 (−0.18, 0.04) 0.38 0.17 (0.07, 0.28) 0.004 0.04 (−0.06, 0.14) 0.78 0.16 (0.03, 0.28) 0.20 
PE(O-16:0/22:5) 0.07 (−0.02, 0.15) 0.19 0.23 (0.01, 0.44) 0.17 −0.04 (−0.24, 0.17) 0.81 0.06 (−0.12, 0.23) 0.78 0.01 (−0.26, 0.28) 0.95 
PE(P-18:0/16:0) 0.06 (−0.05, 0.17) 0.35 0.04 (−0.23, 0.31) 0.31 −0.02 (−0.28, 0.24) 0.87 −0.02 (−0.16, 0.13) 0.90 −0.12 (−0.32, 0.08) 0.48 
PE(P-16:0/22:5) 0.10 (0, 0.21) 0.11 0.22 (−0.04, 0.48) 0.10 0.02 (−0.22, 0.26) 0.87 −0.06 (−0.2, 0.09) 0.80 0.03 (−0.22, 0.28) 0.93 
PE(O-16:0/20:4) −0.09 (−0.18, 0.01) 0.12 −0.14 (−0.38, 0.09) 0.38 0.32 (0.09, 0.54) 0.011 −0.13 (−0.27, 0) 0.80 −0.08 (−0.26, 0.1) 0.71 
PE(P-18:1/20:4) −0.11 (−0.23, 0) 0.11 0.02 (−0.27, 0.30) 0.27 0.20 (−0.08, 0.47) 0.24 −0.18 (−0.36, −0.01) 0.76 −0.23 (−0.44, −0.03) 0.23 
PE(O-18:0/22:4) −0.05 (−0.11, 0.02) 0.22 0.09 (−0.08, 0.26) 0.40 0.08 (−0.08, 0.23) 0.40 −0.03 (−0.12, 0.06) 0.80 0.01 (−0.15, 0.18) 0.95 
PE(O-18:0/20:3) 0.05 (−0.02, 0.11) 0.22 0.14 (−0.02, 0.3) 0.27 −0.10 (−0.26, 0.05) 0.26 −0.25 (−0.49, 0) 0.60 −0.23 (−0.43, −0.03) 0.15 
PE(P-18:0/20:3) 0.09 (−0.03, 0.21) 0.18 0.31 (0.01, 0.60) 0.014 −0.38 (−0.67, −0.1) 0.014 −0.28 (−0.65, 0.08) 0.60 −0.22 (−0.49, 0.04) 0.20 
PE(P-18:1/20:3) 0.15 (0.03, 0.27) 0.048 0.29 (0.01, 0.58) 0.014 −0.54 (−0.82, −0.26) 0.001 −0.2 (−0.43, 0.03) 0.70 −0.32 (−0.55, −0.08) 0.080 
PC(16:0/22:6) −0.20 (−0.27, −0.13) <0.001 −0.12 (−0.30, 0.05) 0.33 0.42 (0.25, 0.58) < 0.001 −0.02 (−0.12, 0.08) 0.81 0.1 (−0.02, 0.22) 0.37 
PE(18:0/14:0) −0.25 (−0.32, −0.18) <0.001 −0.34 (−0.52, −0.17) 0.003 0.54 (0.37, 0.72) < 0.001 0.05 (−0.15, 0.26) 0.80 0.11 (−0.03, 0.25) 0.37 
PC(18:2/22:6) −0.16 (−0.22, −0.09) <0.001 −0.19 (−0.35, −0.04) 0.076 0.32 (0.17, 0.47) < 0.001 0.08 (−0.12, 0.28) 0.80 0.12 (0.02, 0.22) 0.37 
PC(16:0/18:1) −0.12 (−0.22, −0.02) 0.056 −0.18 (−0.42, 0.06) 0.31 0.16 (−0.07, 0.4) 0.24 −0.17 (−0.45, 0.11) 0.70 0.01 (−0.18, 0.2) 0.95 
PC(18:1/18:1) −0.07 (−0.13, 0) 0.090 −0.06 (−0.22, 0.09) 0.51 0.02 (−0.14, 0.18) 0.84 0.09 (−0.01, 0.2) 0.80 0.1 (−0.04, 0.23) 0.38 

Associations between the changes of lipids and BMI, fasting glucose, Matsuda index, insulinogenic index, and disposition index were calculated using linear mixed models adjusted for age, sex, baseline BMI (except when BMI was used as the response variable), physical activity, baseline value of the response variable, and the sum of lipid subclass the lipid belongs to. The β-coefficient and 95% CI obtained from linear mixed models were presented. PFDR were obtained after FDR correction.

To our best knowledge, this is the first time a feeding trial has shown that changed lipidomic signatures could sensitively reflect dietary modification with Mediterranean diet and traditional Chinese diet and the transitional diet in Chinese with overweight and prediabetes. The links of those identified lipidomic signatures in the trial with habitual diet/foods were further examined, and fish-related TAG fractions and red meat–related plasmalogens, including 12 lipid species not reported earlier, were discovered in an independent cohort. The fish-related 12 TAG fractions and PC(16:0/22:6) had favorable, while red meat related 6 PE(O)s and PE(P)s had unfavorable, impacts on BMI, fasting glucose, insulin sensitivity, and β-cell function, highlighting that altering specific food components in the context of healthy diet(s) might modify overall type 2 diabetes risk via correcting disturbed lipid metabolisms.

Our results highlighted that the 12 fish-related TAG fractions maintained at relatively higher levels over a 6-month intervention with both a Mediterranean diet and a traditional Jiangnan diet than with a control diet. Noteworthily, nine of them contained long-chain highly unsaturated fatty acyl 22:6, which might reflect enhanced TAG synthesis with abundant substrates of docosahexaenoic acid (DHA; 22:6n-3). Indeed, DHA per se, but not the TAGs with 22:6, was reported as a fish-related biomarker in various observational studies (26). Similar to our findings, a 4-week crossover feeding trial also showed the associations between TAGs with C22:6 or C20:5 and fatty fish consumption among 19 participants with obesity (27). However, unlike the odd-chain fatty acids that are well-known biomarkers of dairy products (28), three odd-chain TAG fractions in our NHAPC study showed significant associations with fish consumption, in line with the results from a cross-sectional study with over 3,000 participants (29). The discrepancy between the studies could be due to the fact that low dairy consumption (69 g/day) in our study population might unmask the link between odd-chain fatty acids and fish intake. In addition, we also documented relatively higher levels of PC(16:0/22:6) and PC(18:2/22:6) in the Mediterranean diet than in the other two diets, though only PC(16:0/22:6) was significantly associated with fish intake after multiple testing corrections in the NHAPC study. Nonetheless, the associations between the PCs with C22:6 and fish intake were also supported by previous interventions with the Mediterranean diet or Nordic diet (15,30). Taken together, we expanded the species of TAG fractions (6 of 12 not reported earlier) to be promising biomarkers to capture signs of fish intake in the diets, with previously reported PCs.

In contrast, concentrations of nine PE(P)s and PE(O)s in both the Mediterranean diet and traditional Jiangnan diet groups were relatively lower than in the CD group, and six of these discovered lipidomic biomarkers showed significant associations with red meat consumption. In fact, an earlier food composition analysis revealed that red meats were not only a rich source of arachidonic acid (20:4n-6) but also contained substantial amounts of plasmalogens (31), which may provide abundant substrate to promote synthesis of specific PE(P)s and PE(O)s. However, since pork was the major red meat consumed by our study populations, the relationships between these plasmalogens and other types of red meats need to be validated in other populations. Interestingly, PC(16:0/18:1) and PC(18:1/18:1) showed larger increases in the Mediterranean diet group than in the other two groups, while both PCs had null associations with any food group in the NHAPC study. The inconsistent findings of both PCs in the DPMH trial and in the NHAPC cohort might be explained by the fact that the Mediterranean diet was cooked exclusively with olive oil, which is rich in oleic acid (18:1n-9) (32), while the traditional Jiangnan diet was prepared exclusively with sunflower oil during the 6-month intervention. Moreover, our finding was also supported by a trial showing elevated PCs containing C18:1 after 3-week olive oil consumption among 33 hypercholesterolemic subjects (33). On the other hand, lack of significant association in NHAPC could be ascribed to the fact that none of the participants consumed olive oil at baseline. Again, the changed plasmalogens and PCs as consequences of altered red meat and/or types of cooking oil in the Mediterranean diet and/or traditional Jiangnan diet implicated associations between the given dietary components and involvement of specific lipid metabolic pathway(s). Certainly, it needs to be clarified whether these mainly pork-related plasmalogens in our study populations could also reflect intake of other kinds of red meats.

In addition to the response to different dietary patterns, the identified 10 fish-related TAG fractions with long-chain highly unsaturated fatty acyls like 22:6 or odd chains, individually or collectively, were associated with improvements in type 2 diabetes risk factors like BMI and/or Matsuda index during the 6-month intervention, which was in agreement with the results from some observational studies (14,34,35). For instance, in two nested case-control studies, TAGs with higher carbon numbers (>54) and double bonds (>4) were reported to be associated with lower diabetes risk in Chinese (34), while odd-chain TAGs were suggested to be inversely associated with 3.8-year risk of type 2 diabetes according to the Prevención con Dieta Mediterránea (PREDIMED) study (35). Moreover, our study also documented favorable associations between the fish-related PC(16:0/22:6) and improved BMI and Matsuda index, consistent with the findings of a Swedish cohort study, which showed that 22:6-containing PCs partially accounted for the beneficial effects of fish intake on the risk factors of type 2 diabetes (36). With respect to excessive intake of red meat, we also evidenced that red meat–related plasmalogens with 20:3, individually or collectively, were associated with exacerbated fasting glucose, somewhat similar to the positive associations between the 20:3-containing plasmalogens and estimated cardiovascular risk score in a 4-week intervention trial among 44 participants with overweight (37). Therefore, our feeding trial brought more evidence that changed TAG fractions, PCs, and plasmalogens indicated the links between high fish/low red meat intake and reduced diabetes risk in response to healthy diets like the Mediterranean diet and traditional Jiangnan diet.

It was interesting to note that all identified TAG fractions with 22:6 were included in the module positively correlated with improved Matsuda index and disposition index, opposite to the fact that all identified plasmalogens were included in the module positively associated with deteriorated fasting glucose levels. Previous studies also suggested that improved insulin sensitivity and β-cell function indicated by the Matsuda index and disposition index were associated with lower diabetes risk (22,38). For instance, the findings from the follow-up of the US Diabetes Prevention Study suggested that reversion from prediabetes status to normal glucose via dietary and lifestyle intervention could lower diabetes risk by 56% in high-risk subjects (38). Meanwhile, subjects with better insulin sensitivity and β-cell function after having intervention were related to a lower risk of diabetes (hazard ratio = 0.80–0.83) (38). Although the underlying mechanism(s) was not fully understood, studies on animal models suggested that fish-source DHA could act as an anti-inflammatory agent that interacts with major inflammatory signaling pathways, such as eicosanoid production, cytokines production, and Toll-like receptors signaling (39). Conversely, plasmalogens rich in C20:3 and C20:4, originated due to red meat intake, could serve as the substrates for biosynthesizing pathways of eicosanoid and platelet activating factor, which consequently aggravated oxidative stress and inflammation (32). Considering the strong connections of enhanced inflammation with adipose accumulation and insulin resistance, our results emphasized that applying a lipidomic approach in dietary intervention could lead to a thorough understanding of how to modify diets with particular food components that could trigger alterations of certain metabolic pathway(s) and pathogenetic mechanisms of type 2 diabetes, which might be important in clinical settings to optimize dietary patterns for diabetes prevention.

The lipidomic signatures identified in the DPMH were sensitive to dietary patterns beyond the lipid-lowering effect of moderate calorie restriction. Previous trials demonstrated that calorie restriction and induced weight loss could alter the levels of global lipidomic profiles (25). For instance, calorie restriction was reported to lower concentrations of most TAG fractions and PCs, but not for TAG fractions with long-chain fatty acyls according to a 33-week intervention trial conducted in Swedish with abnormal glucose (25). Thus, it is plausible that calorie restriction may diminish the chance to identify some of the lipidomic signatures such as TAG fractions with short-chain highly saturated fatty acyls, but have little impact on other lipidomic varieties. Of note, all three dietary groups with isocaloric restriction in the DPMH achieved similar weight loss (5), and those identified lipids might reflect certain food components in given dietary patterns beyond calorie restriction per se. Indeed, the relationships between the identified lipidomic signatures and habitual dietary intake were evidenced in a separate cohort without caloric restriction. Certainly, future studies are merited to systematically determine whether and to what extent lipidomic signatures could be influenced by diets with or without caloric restriction.

The strengths of our current study included that the lipidomics were repeatedly measured in the well-controlled feeding trial with the 5-day/week feeding regimen. We also included another lipidomic database from a separate cohort to evaluate the associations between the identified lipidomic signatures and habitual dietary intakes among the free-living population in the same geographic location. Moreover, our study covered a wider range of lipidomic profiles than the majority of earlier intervention trials, which allowed us to discover more novel biomarkers related to the intakes of fish and red meat, as well as linked metabolic benefits. However, the limitations of our study included that the 6-month intervention duration could not evaluate long-term associations of the diet-related lipidomic biomarkers with weight maintenance and incident type 2 diabetes. Though we observed the cochanges of lipidomic signatures and diabetes risk factors after modifying dietary patterns, we still could not fully confirm the causal relationship between them. Our findings were of a biologically exploratory nature and could not fully exclude the possibility that the identified lipidomic signatures were impacted by subclinical disease progression (e.g., reduced body weight) other than diet alone. Certain lipidomic signatures within different dietary assignments might be masked by caloric restriction and/or induced weight loss in DPMH trial. We could not fully rule out possible nonlinear associations when applying linear mixed models between diet and each outcome or metabolite. Our findings may not be generalizable to other populations, since we only included subjects with overweight and prediabetes in the DPMH study, as well as middle-aged and elderly Chinese in the NHAPC study. Potential misclassification of trial arms may exist in the DPMH study.

In summary, combining broad coverage of lipidomics from the 6-month feeding trial and a separate free-living cohort, we identified a number of novel signatures and confirmed some of the reported lipidomic signatures, which could sensitively reflect the modifications of fish and red meat as components in the Mediterranean diet and/or traditional Jiangnan diet and be linked with improved type 2 diabetes risk factors in Chinese with high risks. Our findings merit being confirmed in different study populations with different dietary habits.

Clinical trial reg. no. NCT03856762, clinicaltrials.gov

Y.L., L.S., and Q.W. contributed equally to this study as co-first authors.

L.S., R.Z., and X.L. contributed equally to this study as senior authors.

This article contains supplementary material online at https://doi.org/10.2337/figshare.23605377.

Acknowledgments. The authors thank all participants in the DPMH study and NHAPC study. We are also grateful to Shaofeng Huo, Quan Xiong, Huan Yun, Shuangshuang Chen, Zhenhua Niu, Di Wang, and Qianlu Jin from Shanghai Institute of Nutrition and Health, University of Chinese Academy of Sciences, Chinese Academy of Sciences, Shanghai, China.

Funding. The study was supported by the Strategic Priority CAS Project (XDB38010300 and XDB38020000), Shanghai Municipal Science and Technology Major Project (2017SHZDZX01), Chinese Academy of Sciences (ZDBS-SSW-DQC-02 and KJZD-EW-G20-02), and the National Natural Science Foundation of China (81970684 and 81700700).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. Y.L., L.S., Q.W., B.S., Y.W., X.Y., P.Z., Z.N., H.Z., H.L., W.G., J.W., G.N., R.Z., and X.L. were involved in the conception, design, and conduct of the study. Y.L. analyzed the data with support from L.S., and Q.W. and Y.L. wrote the initial paper. L.S. and X.L. edited the paper. All authors reviewed the manuscript and approved the final version. R.Z. and X.L. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

1.
American Diabetes Association Professional Practice Committee
.
2. Classification and diagnosis of diabetes: Standards of Medical Care in Diabetes-2022
.
Diabetes Care
2022
;
45
(
Suppl. 1
):
S17
S38
2.
Evert
AB
,
Dennison
M
,
Gardner
CD
, et al
.
Nutrition therapy for adults with diabetes or prediabetes: a consensus report
.
Diabetes Care
2019
;
42
:
731
754
3.
Li
Y
,
Wang
DD
,
Ley
SH
, et al
.
Time trends of dietary and lifestyle factors and their potential impact on diabetes burden in China
.
Diabetes Care
2017
;
40
:
1685
1694
4.
Qian
F
,
Liu
G
,
Hu
FB
,
Bhupathiraju
SN
,
Sun
Q
.
Association between plant-based dietary patterns and risk of type 2 diabetes: a systematic review and meta-analysis
.
JAMA Intern Med
2019
;
179
:
1335
1344
5.
Luo
Y
,
Wang
J
,
Sun
L
, et al
.
Isocaloric-restricted Mediterranean diet and Chinese diets high or low in plants in adults with prediabetes
.
J Clin Endocrinol Metab
2022
;
107
:
2216
2227
6.
Rodgers
GP
,
Collins
FS
.
Precision nutrition—the answer to “what to eat to stay healthy.”
JAMA
2020
;
324
:
735
736
7.
Kaaks
RJ
.
Biochemical markers as additional measurements in studies of the accuracy of dietary questionnaire measurements: conceptual issues
.
Am J Clin Nutr
1997
;
65
(
Suppl.
):
1232S
1239S
8.
Yun
H
,
Sun
L
,
Wu
Q
, et al
.
Lipidomic signatures of dairy consumption and associated changes in blood pressure and other cardiovascular risk factors among Chinese adults
.
Hypertension
2022
;
79
:
1617
1628
9.
Ibáñez
C
,
Mouhid
L
,
Reglero
G
,
Ramírez de Molina
A
.
Lipidomics insights in health and nutritional intervention studies
.
J Agric Food Chem
2017
;
65
:
7827
7842
10.
Niu
Z
,
Wu
Q
,
Sun
L
, et al
.
Circulating glycerolipids, fatty liver index, and incidence of type 2 diabetes: a prospective study among Chinese
.
J Clin Endocrinol Metab
2021
;
106
:
2010
2020
11.
Yun
H
,
Sun
L
,
Wu
Q
, et al
.
Associations among circulating sphingolipids, β-cell function, and risk of developing type 2 diabetes: a population-based cohort study in China
.
PLoS Med
2020
;
17
:
e1003451
12.
Chen
S
,
Zong
G
,
Wu
Q
, et al
.
Associations of plasma glycerophospholipid profile with modifiable lifestyles and incident diabetes in middle-aged and older Chinese
.
Diabetologia
2022
;
65
:
315
328
13.
Niu
Z
,
Wu
Q
,
Luo
Y
, et al
.
Plasma lipidomic subclasses and risk of hypertension in middle-aged and elderly Chinese
.
Phenomics
2022
;
2
:
283
294
14.
Rhee
EP
,
Cheng
S
,
Larson
MG
, et al
.
Lipid profiling identifies a triacylglycerol signature of insulin resistance and improves diabetes prediction in humans
.
J Clin Invest
2011
;
121
:
1402
1411
15.
Bondia-Pons
I
,
Martinez
JA
,
de la Iglesia
R
, et al
.
Effects of short- and long-term Mediterranean-based dietary treatment on plasma LC-QTOF/MS metabolic profiling of subjects with metabolic syndrome features: The Metabolic Syndrome Reduction in Navarra (RESMENA) randomized controlled trial
.
Mol Nutr Food Res
2015
;
59
:
711
728
16.
Toledo
E
,
Wang
DD
,
Ruiz-Canela
M
, et al
.
Plasma lipidomic profiles and cardiovascular events in a randomized intervention trial with the Mediterranean diet
.
Am J Clin Nutr
2017
;
106
:
973
983
17.
Ye
X
,
Yu
Z
,
Li
H
,
Franco
OH
,
Liu
Y
,
Lin
X
.
Distributions of C-reactive protein and its association with metabolic syndrome in middle-aged and older Chinese people
.
J Am Coll Cardiol
2007
;
49
:
1798
1805
18.
Zhao
W
,
Hasegawa
K
,
Chen
J
.
The use of food-frequency questionnaires for various purposes in China
.
Public Health Nutr
2002
;
5
(
6A
):
829
833
19.
Zong
G
,
Sun
Q
,
Yu
D
, et al
.
Dairy consumption, type 2 diabetes, and changes in cardiometabolic traits: a prospective cohort study of middle-aged and older Chinese in Beijing and Shanghai
.
Diabetes Care
2014
;
37
:
56
63
20.
Craig
CL
,
Marshall
AL
,
Sjöström
M
, et al
.
International physical activity questionnaire: 12-country reliability and validity
.
Med Sci Sports Exerc
2003
;
35
:
1381
1395
21.
Matsuda
M
,
DeFronzo
RA
.
Insulin sensitivity indices obtained from oral glucose tolerance testing: comparison with the euglycemic insulin clamp
.
Diabetes Care
1999
;
22
:
1462
1470
22.
Ohn
JH
,
Kwak
SH
,
Cho
YM
, et al
.
10-year trajectory of β-cell function and insulin sensitivity in the development of type 2 diabetes: a community-based prospective cohort study
.
Lancet Diabetes Endocrinol
2016
;
4
:
27
34
23.
Phillips
DI
,
Clark
PM
,
Hales
CN
,
Osmond
C
.
Understanding oral glucose tolerance: comparison of glucose or insulin measurements during the oral glucose tolerance test with specific measurements of insulin resistance and insulin secretion
.
Diabet Med
1994
;
11
:
286
292
24.
Yu
D
,
Zheng
W
,
Cai
H
, et al
.
Long-term diet quality and risk of type 2 diabetes among urban Chinese adults
.
Diabetes Care
2018
;
41
:
723
730
25.
Schwab
U
,
Seppänen-Laakso
T
,
Yetukuri
L
, et al.;
GENOBIN Study Group
.
Triacylglycerol fatty acid composition in diet-induced weight loss in subjects with abnormal glucose metabolism - the GENOBIN study
.
PLoS One
2008
;
3
:
e2630
26.
Wallin
A
,
Di Giuseppe
D
,
Orsini
N
,
Patel
PS
,
Forouhi
NG
,
Wolk
A
.
Fish consumption, dietary long-chain n-3 fatty acids, and risk of type 2 diabetes: systematic review and meta-analysis of prospective studies
.
Diabetes Care
2012
;
35
:
918
929
27.
Žáček
P
,
Bukowski
M
,
Johnson
L
,
Raatz
SK
,
Picklo
M
.
Selective enrichment of n-3 fatty acids in human plasma lipid motifs following intake of marine fish
.
J Nutr Biochem
2018
;
54
:
57
65
28.
Pranger
IG
,
Joustra
ML
,
Corpeleijn
E
, et al
.
Fatty acids as biomarkers of total dairy and dairy fat intakes: a systematic review and meta-analysis
.
Nutr Rev
2019
;
77
:
46
63
29.
Saadatian-Elahi
M
,
Slimani
N
,
Chajès
V
, et al
.
Plasma phospholipid fatty acid profiles and their association with food intakes: results from a cross-sectional study within the European Prospective Investigation into Cancer and Nutrition
.
Am J Clin Nutr
2009
;
89
:
331
346
30.
Lankinen
M
,
Schwab
U
,
Kolehmainen
M
, et al
.
A healthy Nordic diet alters the plasma lipidomic profile in adults with features of metabolic syndrome in a multicenter randomized dietary intervention
.
J Nutr
2015
;
146
:
662
672
31.
Blank
ML
,
Cress
EA
,
Smith
ZL
,
Snyder
F
.
Meats and fish consumed in the American diet contain substantial amounts of ether-linked phospholipids
.
J Nutr
1992
;
122
:
1656
1661
32.
Serra-Majem
L
,
Román-Viñas
B
,
Sanchez-Villegas
A
,
Guasch-Ferré
M
,
Corella
D
,
La Vecchia
C
.
Benefits of the Mediterranean diet: epidemiological and molecular aspects
.
Mol Aspects Med
2019
;
67
:
1
55
33.
Fernández-Castillejo
S
,
Pedret
A
,
Catalán
Ú
, et al
.
Virgin olive oil phenolic compounds modulate the HDL lipidome in hypercholesterolaemic subjects: a lipidomic analysis of the VOHF study
.
Mol Nutr Food Res
2021
;
65
:
e2001192
34.
Lu
J
,
Lam
SM
,
Wan
Q
, et al
.
High-coverage targeted lipidomics reveals novel serum lipid predictors and lipid pathway dysregulation antecedent to type 2 diabetes onset in normoglycemic Chinese adults
.
Diabetes Care
2019
;
42
:
2117
2126
35.
Razquin
C
,
Toledo
E
,
Clish
CB
, et al
.
Plasma lipidomic profiling and risk of type 2 diabetes in the PREDIMED trial
.
Diabetes Care
2018
;
41
:
2617
2624
36.
Shi
L
,
Brunius
C
,
Bergdahl
IA
, et al
.
Joint analysis of metabolite markers of fish intake and persistent organic pollutants in relation to type 2 diabetes risk in Swedish adults
.
J Nutr
2019
;
149
:
1413
1423
37.
Tovar
J
,
de Mello
VD
,
Nilsson
A
, et al
.
Reduction in cardiometabolic risk factors by a multifunctional diet is mediated via several branches of metabolism as evidenced by nontargeted metabolite profiling approach
.
Mol Nutr Food Res
2017
;
61
:
1600552
38.
Perreault
L
,
Pan
Q
,
Mather
KJ
,
Watson
KE
,
Hamman
RF
;
Diabetes Prevention Program Research Group
.
Effect of regression from prediabetes to normal glucose regulation on long-term reduction in diabetes risk: results from the Diabetes Prevention Program Outcomes Study
.
Lancet
2012
;
379
:
2243
2251
39.
Wall
R
,
Ross
RP
,
Fitzgerald
GF
,
Stanton
C
.
Fatty acids from fish: the anti-inflammatory potential of long-chain omega-3 fatty acids
.
Nutr Rev
2010
;
68
:
280
289
Readers may use this article as long as the work is properly cited, the use is educational and not for profit, and the work is not altered. More information is available at https://www.diabetesjournals.org/journals/pages/license.