Metformin is the first-line oral medication to increase insulin sensitivity in patients with type 2 diabetes (T2D). Our aim was to investigate the pleiotropic effect of metformin using a nontargeted metabolomics approach. We analyzed 353 metabolites in fasting serum samples of the population-based human KORA (Cooperative Health Research in the Region of Augsburg) follow-up survey 4 cohort. To compare T2D patients treated with metformin (mt-T2D, n = 74) and those without antidiabetes medication (ndt-T2D, n = 115), we used multivariable linear regression models in a cross-sectional study. We applied a generalized estimating equation to confirm the initial findings in longitudinal samples of 683 KORA participants. In a translational approach, we used murine plasma, liver, skeletal muscle, and epididymal adipose tissue samples from metformin-treated db/db mice to further corroborate our findings from the human study. We identified two metabolites significantly (P < 1.42E-04) associated with metformin treatment. Citrulline showed lower relative concentrations and an unknown metabolite X-21365 showed higher relative concentrations in human serum when comparing mt-T2D with ndt-T2D. Citrulline was confirmed to be significantly (P < 2.96E-04) decreased at 7-year follow-up in patients who started metformin treatment. In mice, we validated significantly (P < 4.52E-07) lower citrulline values in plasma, skeletal muscle, and adipose tissue of metformin-treated animals but not in their liver. The lowered values of citrulline we observed by using a nontargeted approach most likely resulted from the pleiotropic effect of metformin on the interlocked urea and nitric oxide cycle. The translational data derived from multiple murine tissues corroborated and complemented the findings from the human cohort.

Metformin became the first-line choice for treatment of type 2 diabetes (T2D) in the course of the UK Prospective Diabetes Study (UKPDS) (1). Additionally, metformin has been reported to have other pleiotropic effects; e.g., it reduces insulin resistance (2), improves the uptake of glucose in muscle (3,4), reduces the risk for cancer (5), and lowers the values of LDL cholesterol (LDL-C) (6). The underlying mechanism of the reduction of LDL-C is, at least in part, due to the activation of the AMPK in the liver (7). Apart from that, AMPK affects several processes such as nitric oxide (NO) production by endothelial NO synthase (eNOS) (8), which is also stimulated by metformin (9). However, the mode of action of metformin is not completely understood (1012).

Our previous study was based on a targeted metabolomics approach to explore the effects of metformin on lipid profiles in the population-based KORA (Cooperative Health Research in the Region of Augsburg) cohort (6,13,14). Irving et al. (15) recently reported decreased levels of arginine and citrulline as an effect of insulin sensitizer therapy in 12 metformin- and pioglitazone-treated individuals and 13 placebo-treated control subjects. Nontargeted metabolomic measurements have been applied to investigate hyperglycemia (16,17) and the effects of metformin treatment in individuals without diabetes (18). However, none of the previous nontargeted metabolomics studies investigated metformin treatment in patients with T2D.

In this study we focused on serum metabolites associated with metformin treatment based on a nontargeted approach in a human population from the KORA cohort. A cross-link from human to mice was corroborated in multiple tissues (plasma, liver, skeletal muscle, and epididymal adipose tissue) from a mouse study. Biologically relevant pathways for the identified metabolites were analyzed using bioinformatical approaches.

Ethics Statement

All participants gave written informed consent. The KORA study was approved by the ethics committee of the Bavarian Medical Association, Munich, Germany.

Approval for Mouse Study

Within this study, all mice were bred and housed in a temperature- and humidity-controlled environment in compliance with Federation of European Laboratory Animal Science Associations protocols. Animal experiments were approved by the District Government of Upper Bavaria (Regierung von Oberbayern, Gz.55.2–1-54–2531–70–07, 55.2–1-2532–153–11).

KORA Cohort

KORA is a population-based cohort study conducted in southern Germany (14). The baseline survey 4 (KORA S4) consists of 4,261 individuals (aged 25–74 years) examined between 1999 and 2001. During the years of 2006 to 2008, 3,080 individuals took part in the follow-up survey 4 (KORA F4). Clinical data for each participant were retrieved from medical records. On the basis of fasting glucose, 2-h postglucose load, and physician-validated and self-reported diagnoses, KORA participants were classified according to the World Health Organization diagnostic criteria. A further grouping of patients with T2D was based on information on medication (19,20) (Table 1). Only participants with metabolite measurements were included in the present analysis (Metabolon, n = 1,768 in KORA F4). We excluded 1) participants with overnight nonfasting serum samples (n = 8), 2) patients suffering from type 1 diabetes and drug-induced (e.g., via steroids) diabetes (n = 6), 3) T2D patients treated with insulin (n = 16) or both insulin and metformin (n = 13), and 4) patients taking glucose-lowering oral medication other than metformin (n = 17). Furthermore, participants with isolated impaired fasting glucose (IFG) (n = 77) were excluded. We have previously shown that IFG and impaired glucose tolerance (IGT) should be considered as two different phenotypes (21).

Table 1

Characteristics of the KORA F4 cross-sectional study samples (n = 1,604)

Clinical parametersNGTIGTndt-T2Dmt-T2D
n
 
1,143
 
272
 
115
 
74
 
Age, years
 
58.9 (8.5)
 
63.8 (8.1)
 
65.1 (7.1)
 
66.2 (7.5)
 
Male, %
 
45
 
50
 
61
 
58
 
BMI, kg/m2
 
27.0 (4.3)
 
29.8 (4.7)
 
31.0 (4.8)
 
32.0 (5.6)
 
Waist, cm
 
92 (12.9)
 
100.2 (14.3)
 
104.9 (12.1)
 
106.7 (13.0)
 
Physical activity, % >1 h per week
 
63
 
54
 
48
 
36
 
High alcohol intake, %*
 
20
 
18
 
25
 
19
 
Smoker, %
 
18
 
8
 
12
 
15
 
Systolic BP, mmHg
 
121.6 (17.7)
 
129.0 (19.0)
 
134.3 (19.2)
 
130.7 (18.2)
 
HDL-C, mg/dL
 
58.9 (14.8)
 
54.4 (14.2)
 
49.6 (11.5)
 
50.4 (9.6)
 
LDL-C, mg/dL
 
140.7 (34.3)
 
144.5 (36.2)
 
136.6 (36.0)
 
124.1 (28.3)
 
Total cholesterol, mg/dL
 
223.4 (37.7)
 
226.4 (41.5)
 
214.1 (37.0)
 
203.8 (37.8)
 
Triglycerides, mg/dL
 
118.5 (84.6)
 
150.7 (89.0)
 
172.5 (128.7)
 
177.5 (140.5)
 
HbA1c, %
 
5.4 (0.3)
 
5.6 (0.3)
 
6.3 (0.9)
 
6.8 (1.1)
 
HbA1c, mmol/mol
 
35.7 (3.2)
 
38.1 (3.9)
 
44.91 (9.9)
 
51.2 (11.6)
 
Fasting glucose, mg/dL
 
93.2 (7.5)
 
101.2 (10.6)
 
126.6 (30.5)
 
142.0 (36.0)
 
2-h postglucose load, mg/dL
 
100.8 (20.6)
 
162.6 (17.5)
 
216.0 (50.7)
 

 
Time since diagnosis, years
 

 

 
1.4 (2.6)
 
7.4 (6.6)
 
Insulin, µIU/mL
 
7.1 (26.1)
 
10.1 (10.9)
 
14.3 (14.3)
 
11.6 (11.0)
 
Leptin, ng/mL
 
17.2 (19.2)
 
24.2 (21.3)
 
26.8 (21.2)
 
27.8 (25.0)
 
Statin usage, %
 
12
 
15
 
28
 
36
 
β-Blocker usage, %
 
16
 
31
 
43
 
38
 
ACE inhibitor usage, %
 
10
 
21
 
32
 
45
 
ARB usage, %
 
8
 
10
 
17
 
14
 
Insulin therapy, %
 
0
 
0
 
0
 
0
 
Metformin usage, %
 
0
 
0
 
0
 
100
 
Parental T2D, % 28 30 37 49 
Clinical parametersNGTIGTndt-T2Dmt-T2D
n
 
1,143
 
272
 
115
 
74
 
Age, years
 
58.9 (8.5)
 
63.8 (8.1)
 
65.1 (7.1)
 
66.2 (7.5)
 
Male, %
 
45
 
50
 
61
 
58
 
BMI, kg/m2
 
27.0 (4.3)
 
29.8 (4.7)
 
31.0 (4.8)
 
32.0 (5.6)
 
Waist, cm
 
92 (12.9)
 
100.2 (14.3)
 
104.9 (12.1)
 
106.7 (13.0)
 
Physical activity, % >1 h per week
 
63
 
54
 
48
 
36
 
High alcohol intake, %*
 
20
 
18
 
25
 
19
 
Smoker, %
 
18
 
8
 
12
 
15
 
Systolic BP, mmHg
 
121.6 (17.7)
 
129.0 (19.0)
 
134.3 (19.2)
 
130.7 (18.2)
 
HDL-C, mg/dL
 
58.9 (14.8)
 
54.4 (14.2)
 
49.6 (11.5)
 
50.4 (9.6)
 
LDL-C, mg/dL
 
140.7 (34.3)
 
144.5 (36.2)
 
136.6 (36.0)
 
124.1 (28.3)
 
Total cholesterol, mg/dL
 
223.4 (37.7)
 
226.4 (41.5)
 
214.1 (37.0)
 
203.8 (37.8)
 
Triglycerides, mg/dL
 
118.5 (84.6)
 
150.7 (89.0)
 
172.5 (128.7)
 
177.5 (140.5)
 
HbA1c, %
 
5.4 (0.3)
 
5.6 (0.3)
 
6.3 (0.9)
 
6.8 (1.1)
 
HbA1c, mmol/mol
 
35.7 (3.2)
 
38.1 (3.9)
 
44.91 (9.9)
 
51.2 (11.6)
 
Fasting glucose, mg/dL
 
93.2 (7.5)
 
101.2 (10.6)
 
126.6 (30.5)
 
142.0 (36.0)
 
2-h postglucose load, mg/dL
 
100.8 (20.6)
 
162.6 (17.5)
 
216.0 (50.7)
 

 
Time since diagnosis, years
 

 

 
1.4 (2.6)
 
7.4 (6.6)
 
Insulin, µIU/mL
 
7.1 (26.1)
 
10.1 (10.9)
 
14.3 (14.3)
 
11.6 (11.0)
 
Leptin, ng/mL
 
17.2 (19.2)
 
24.2 (21.3)
 
26.8 (21.2)
 
27.8 (25.0)
 
Statin usage, %
 
12
 
15
 
28
 
36
 
β-Blocker usage, %
 
16
 
31
 
43
 
38
 
ACE inhibitor usage, %
 
10
 
21
 
32
 
45
 
ARB usage, %
 
8
 
10
 
17
 
14
 
Insulin therapy, %
 
0
 
0
 
0
 
0
 
Metformin usage, %
 
0
 
0
 
0
 
100
 
Parental T2D, % 28 30 37 49 

KORA F4 study characteristics (including solely subjects with available Metabolon measurements). Percentages of individuals or means (SD) are shown for each variable and each group (NGT, IGT, ndt-T2D, and mt-T2D).

*≥20 g/day for women; ≥40 g/day for men.

n = 81.

‡For newly diagnosed T2D patients (n = 74), years since T2D diagnosis was defined as 0.

In KORA F4, we focused on four groups: 1) participants with normal glucose tolerance (NGT), 2) individuals with prediabetes with IGT, 3) T2D patients without glucose-lowering treatment (non-antidiabetes drug treated, ndt-T2D), and 4) metformin-treated T2D (mt-T2D) patients (Table 1).

The same exclusion and classification criteria were used in the longitudinal analyses. We only considered participants with metabolite measurements in both studies (KORA S4 and F4, n = 818), and we excluded at both time points 1) participants with overnight nonfasting serum samples (n = 88), which included patients suffering from type 1 diabetes or drug-induced diabetes, 2) participants taking oral glucose-lowering medication other than metformin (n = 11), 3) participants undergoing insulin treatment (n = 3), and 4) participants with a missing diabetes status (n = 33). The remaining 683 participants were ndt-T2D individuals with prediabetes and healthy control subjects at KORA S4, 37 of whom started metformin treatment at KORA F4.

The data from KORA S4 and F4, including metabolite concentrations with clinical phenotypes, are available upon request through the platform KORA-PASST (project application self-service tool) (www.helmholtz-muenchen.de/kora-gen).

Blood Sampling

In the KORA cohort study, blood was drawn into Monovette serum tubes (Sarstedt AG & Co., Nümbrecht, Germany) in the morning between 8:00 a.m. and 10:30 a.m. after at least 8 h of fasting. Tubes were gently inverted twice, followed by resting 30 min at room temperature to obtain complete coagulation. For serum collection, blood was centrifuged at 2,750g at 15°C for 10 min. Serum was filled into synthetic straws, which were stored in liquid nitrogen (−196°C) until metabolomic analyses.

Nontargeted Metabolite Profiling

The serum samples from participants of KORA S4 and F4 were measured with the Metabolon analytical system (Metabolon, Inc., Durham, NC). Metabolon applied a nontargeted semiquantitative liquid chromatography–tandem mass spectrometry (LC-MS/MS) and gas chromatography–mass spectrometry (GC-MS) platform for the identification of structurally named and unknown molecules (22,23). We measured 363 (including 109 unknown) metabolites in fasting serum samples from KORA S4. In the 7-year KORA F4, 353 metabolites (including 107 unknown) were determined (24).

In this study, we applied the same criteria for quality control as described by Albrecht et al. (25). In brief, metabolites with more than 20% missing values were excluded, as were samples with more than 10% missing metabolites (25). All normalized relative ion counts were log transformed, and the remaining data were imputed with Multivariate Imputation by Chained Equations (MICE) (26). We used 363 metabolites in KORA S4 and 353 metabolites in KORA F4 (Supplementary Table 1). The number of overlapping metabolites in KORA S4 and F4 was 312. Metabolite names were used according to Shin et al. (27); however, the identity of metabolite ID M32654 and the molecule “3-dehydrocarnitine*” could not be confirmed. We therefore used the name X-21365 (Supplementary Table 1).

Each metabolite was standardized with a mean of zero and an SD of one in each study after the exclusion of nonfasting participants.

Metformin Mouse Intervention Study

Pharmacological studies were conducted in 20 male 8-week-old diabetic BKS.Cg-Dock7m+/+Leprdb/J (db/db) mice that were bred and housed in a temperature- and humidity-controlled environment in compliance with Federation of European Laboratory Animal Science Associations protocols. To exclude estrous cycle–related influences on metabolic parameters, only male mice were included in this study. From age 3 weeks, all mice were fed a high-fat diet (S0372-E010; ssniff Spezialdiäten, Soest, Germany) containing (gm%) palm fat (13.5), sunflower oil (13.5), starch (30), saccharose (10), casein (20), lignocellulose (5), mineral+vitamin mix (5+2), safflower oil (0.5), and linseed oil (0.5) to manifest a uniform diabetic phenotype. Animals received either vehicle (5% solutol/95% hydroxyethylcellulose) without (n = 10) or with metformin (300 mg/kg; Sigma Aldrich, Taufkirchen, Germany; n = 10) via gavage once daily between 5:00 and 6:00 p.m. before dark-phase onset (6:00 p.m.) for 14 days. At 18 ± 2 h after the last treatment, 4-h fasted mice were sacrificed with an isoflurane overdose, and organs and blood were immediately collected (4). Murine plasma was prepared from whole blood by centrifugation at 4°C, and tissues were freeze-clamped; both were stored at −80°C until further analyses. All samples were measured with the Metabolon analytical system. Metabolites with more than 20% missing values were excluded, as were samples with more than 10% missing metabolites (25). All normalized relative ion counts were log transformed, and the remaining data were imputed with MICE (26). Linear regression was done on metabolite values for metformin-treated mice as the cases as well as for the nonmetformin-treated, vehicle-gavaged mice as the controls. A metabolomics examination was done for plasma, liver, skeletal muscle, and epididymal adipose tissue (Table 5 and Fig. 1B).

Figure 1

Differences in relative metabolite concentrations in a human study, in a mouse study, and in organ-specific pathways. A: Mean relative residuals of the concentrations (with SEM) of two metabolites for the NGT, IGT, ndt-T2D, and mt-T2D groups derived in cross-sectional analysis of KORA F4. Residuals were calculated from a linear regression model with full adjustments. B: Mean relative concentrations (with SEM) of two metabolites in four different mouse tissues (plasma, liver, skeletal muscle, and epididymal adipose tissue). C: The connections indicated by liver, muscle, and blood (plasma and serum) show organ specificity between metabolites, pathway-related proteins, metformin targets, and metformin. The metabolites (ellipses) were connected to metformin treatment (straight-sided hexagons), proteins (hexagons), and metformin targets (rectangles). The activation/stimulation is indicated with arrows. For further information, see Tables 35 and Supplementary Tables 2 and 3.

Figure 1

Differences in relative metabolite concentrations in a human study, in a mouse study, and in organ-specific pathways. A: Mean relative residuals of the concentrations (with SEM) of two metabolites for the NGT, IGT, ndt-T2D, and mt-T2D groups derived in cross-sectional analysis of KORA F4. Residuals were calculated from a linear regression model with full adjustments. B: Mean relative concentrations (with SEM) of two metabolites in four different mouse tissues (plasma, liver, skeletal muscle, and epididymal adipose tissue). C: The connections indicated by liver, muscle, and blood (plasma and serum) show organ specificity between metabolites, pathway-related proteins, metformin targets, and metformin. The metabolites (ellipses) were connected to metformin treatment (straight-sided hexagons), proteins (hexagons), and metformin targets (rectangles). The activation/stimulation is indicated with arrows. For further information, see Tables 35 and Supplementary Tables 2 and 3.

Statistical Analysis

To evaluate the effect of metformin treatment on certain metabolites, multivariable linear regression models were conducted with the relative metabolite concentration values as outcome and the grouping variable as predictor. Each metabolite was assessed individually. To consider potential risk factors and confounding parameters with known effect on metabolite profiles (6,13,2832), two models were used: 1) adjusted for age and sex as the crude model and 2) adjusted for age, sex, BMI, physical activity, high alcohol intake, smoking status, systolic blood pressure (BP), HbA1c, fasting glucose, HDL cholesterol (HDL-C), and triglycerides as well as the use of statins, β-blockers, ACE inhibitors, and angiotensin receptor blockers (ARB) as the full model. The association of conventional risk factors of T2D as well as other population characteristics with metformin treatment was calculated via χ2 test for categorical variables. Shapiro–Wilk test was applied to test continuous variables for normal distribution (P ≤ 0.05 for nonnormally distributed variables, P > 0.05 for normally distributed variables), followed by Student's t test for normally distributed continuous variables and Wilcoxon test for nonnormally distributed continuous variables.

To account for multiple testing for the linear models, Bonferroni correction was applied, and only metabolites with P < 0.05/353 = 1.42E-04 were considered to be statistically significantly different in KORA F4. In addition, we calculated the adjusted P value with the false discovery rate (FDR) using the Benjamini-Hochberg method, which is not as stringent as the Bonferroni correction. For the full linear models, participants were excluded because of missing information of considered confounders. This led to 1,138 NGT (after exclusion of five individuals because of missing confounding information), 272 IGT, 114 ndt-T2D (after exclusion of one individual because of missing confounding information), and 70 mt-T2D (after exclusion of four individuals because of missing confounding information) participants.

In the KORA S4 to F4 longitudinal study (S4 → F4), generalized estimating equations (GEE) were used to validate the significant metabolites in both crude and full models.

All statistical analyses were performed in R (version 3.2.2) (33).

Pathway Analysis

Pathways were explored using databases, considering tissue and organ specificity. The link from observed significant metabolites to the interacting enzymes was drawn using the Human Metabolome Database (34). Protein–protein interactions were analyzed with the Search Tool for the Retrieval of Interacting Genes (35) and the Kyoto Encyclopedia of Genes and Genomes (36). To consider drug-related effects of metformin on certain targets, we used DrugBank (37). The link between metformin targets and the protein network was analyzed using the Kyoto Encyclopedia of Genes and Genomes (36).

Population Characteristics of Human and Mouse Studies

On the basis of the available nontargeted metabolomic profiles, our human discovery study, KORA F4, includes 1,143 NGT, 272 IGT, 115 ndt-T2D, and 74 mt-T2D participants (Table 1). Among the four groups, mt-T2D patients were the oldest, were more frequently men, and had the highest values of HbA1c, fasting glucose, triglycerides, BMI, and waist circumference (Table 1).

The longitudinal KORA study includes samples of 683 participants without metformin treatment at baseline, 37 of whom were treated with metformin in the 7-year follow-up (Table 2).

Table 2

Characteristics of the KORA S4 → F4 prospective study samples (n = 683)

Clinical parametersKORA S4 w/o metformin§ → KORA F4 w/o metformin§
KORA S4 w/o metformin§ → KORA F4 w/ metformin
S4F4P valueS4F4P value
n
 
646
 
646
 

 
37
 
37
 

 
Age, years
 
61.4 (4.2)
 
68.5 (4.2)
 

 
63.4 (3.8)
 
70.4 (3.9)
 

 
Male, %
 
51
 
51
 

 
54
 
54
 

 
BMI, kg/m2
 
27.9 (3.9)
 
28.2 (4.2)
 
0.2
 
32.8 (4.3)
 
32 (4.5)
 
0.49
 
Waist, cm
 
93.9 (11.0)
 
96.8 (11.9)
 
1.72E-05
 
106.3 (11.3)
 
106.7 (12.3)
 
0.87
 
Physical activity, % >1 h per week
 
47
 
57
 
5.47E-04
 
32
 
43
 
0.47
 
High alcohol intake, %*
 
20
 
19
 
0.65
 
27
 
16
 
0.4
 
Smoker, %
 
13
 
8
 
9.00E-03
 
14
 
8
 
0.71
 
Systolic BP, mmHg
 
132.1 (18.7)
 
128.2 (19.6)
 
3.12E-04
 
144.9 (18.1)
 
131.6 (18.3)
 
2.43E-03
 
HDL-C, mg/dL
 
59.1 (16.3)
 
56.7 (14.2)
 
0.02
 
53.4 (11.8)
 
52.4 (7.9)
 
0.69
 
LDL-C, mg/dL
 
154.5 (40.8)
 
142.5 (36.9)
 
1.86E-08
 
143.5 (37.6)
 
123.5 (23.6)
 
0.02
 
Total cholesterol, mg/dL
 
245.5 (42.0)
 
225.2 (40.7)
 
2.2E-16
 
234.1 (41.0)
 
202.1 (33.6)
 
5.20E-04
 
Triglycerides, mg/dL
 
129.9 (76.4)
 
132.7 (83.4)
 
0.56
 
170.6 (169)
 
154.5 (159.2)
 
0.79
 
HbA1c, %
 
5.6 (0.3)
 
5.6 (0.5)
 
0.53
 
6.4 (0.9)
 
6.6 (0.7)
 
0.14
 
HbA1c, mmol/mol
 
37.6 (3.7)
 
38.0 (5.7)
 
0.53
 
46.8 (10.3)
 
48.3 (7.8)
 
0.14
 
Fasting glucose, mg/dL
 
99.7 (10.9)
 
100.7 (17.6)
 
0.53
 
130.7 (30.1)
 
129.6 (28.3)
 
0.9
 
2-h postglucose load, mg/dL
 
115.1 (37.1)
 
126.9 (40.8)
 
8.34E-09
 
205.5 (76.8)
 

 

 
Statin usage, %
 
10
 
22
 
8.05E-10
 
8
 
29
 
0.04
 
β-Blocker usage, %
 
17
 
31
 
5.16E-09
 
16
 
29
 
0.27
 
ACE inhibitor usage, %
 
8
 
21
 
2.72E-11
 
16
 
51
 
3.18E-03
 
ARB usage, %
 
3
 
12
 
3.67E-10
 
3
 
14
 
0.2
 
Insulin therapy, %
 
0
 
0
 

 
0
 
0
 

 
Metformin usage, %
 
0
 
0
 

 
0
 
100
 

 
Parental T2D, % 25 25  47 47  
Clinical parametersKORA S4 w/o metformin§ → KORA F4 w/o metformin§
KORA S4 w/o metformin§ → KORA F4 w/ metformin
S4F4P valueS4F4P value
n
 
646
 
646
 

 
37
 
37
 

 
Age, years
 
61.4 (4.2)
 
68.5 (4.2)
 

 
63.4 (3.8)
 
70.4 (3.9)
 

 
Male, %
 
51
 
51
 

 
54
 
54
 

 
BMI, kg/m2
 
27.9 (3.9)
 
28.2 (4.2)
 
0.2
 
32.8 (4.3)
 
32 (4.5)
 
0.49
 
Waist, cm
 
93.9 (11.0)
 
96.8 (11.9)
 
1.72E-05
 
106.3 (11.3)
 
106.7 (12.3)
 
0.87
 
Physical activity, % >1 h per week
 
47
 
57
 
5.47E-04
 
32
 
43
 
0.47
 
High alcohol intake, %*
 
20
 
19
 
0.65
 
27
 
16
 
0.4
 
Smoker, %
 
13
 
8
 
9.00E-03
 
14
 
8
 
0.71
 
Systolic BP, mmHg
 
132.1 (18.7)
 
128.2 (19.6)
 
3.12E-04
 
144.9 (18.1)
 
131.6 (18.3)
 
2.43E-03
 
HDL-C, mg/dL
 
59.1 (16.3)
 
56.7 (14.2)
 
0.02
 
53.4 (11.8)
 
52.4 (7.9)
 
0.69
 
LDL-C, mg/dL
 
154.5 (40.8)
 
142.5 (36.9)
 
1.86E-08
 
143.5 (37.6)
 
123.5 (23.6)
 
0.02
 
Total cholesterol, mg/dL
 
245.5 (42.0)
 
225.2 (40.7)
 
2.2E-16
 
234.1 (41.0)
 
202.1 (33.6)
 
5.20E-04
 
Triglycerides, mg/dL
 
129.9 (76.4)
 
132.7 (83.4)
 
0.56
 
170.6 (169)
 
154.5 (159.2)
 
0.79
 
HbA1c, %
 
5.6 (0.3)
 
5.6 (0.5)
 
0.53
 
6.4 (0.9)
 
6.6 (0.7)
 
0.14
 
HbA1c, mmol/mol
 
37.6 (3.7)
 
38.0 (5.7)
 
0.53
 
46.8 (10.3)
 
48.3 (7.8)
 
0.14
 
Fasting glucose, mg/dL
 
99.7 (10.9)
 
100.7 (17.6)
 
0.53
 
130.7 (30.1)
 
129.6 (28.3)
 
0.9
 
2-h postglucose load, mg/dL
 
115.1 (37.1)
 
126.9 (40.8)
 
8.34E-09
 
205.5 (76.8)
 

 

 
Statin usage, %
 
10
 
22
 
8.05E-10
 
8
 
29
 
0.04
 
β-Blocker usage, %
 
17
 
31
 
5.16E-09
 
16
 
29
 
0.27
 
ACE inhibitor usage, %
 
8
 
21
 
2.72E-11
 
16
 
51
 
3.18E-03
 
ARB usage, %
 
3
 
12
 
3.67E-10
 
3
 
14
 
0.2
 
Insulin therapy, %
 
0
 
0
 

 
0
 
0
 

 
Metformin usage, %
 
0
 
0
 

 
0
 
100
 

 
Parental T2D, % 25 25  47 47  

Percentages of individuals or means (SD) of participants (with available Metabolon measurements for KORA S4 and F4) are shown for each variable and each group. w/o, without; w/, with.

*>40 g/day in men; >20 g/day in women.

§Includes participants with NGT, isolated IFG, IGT, and ntd-T2D.

‖Normally distributed (every other distribution is not normally distributed).

From the metformin-treated mice, we obtained 10 samples for plasma, liver, and epididymal adipose tissue and 9 samples for skeletal muscle. In the same amount of vehicle-gavaged control mice, we obtained 10 samples for plasma, liver, epididymal adipose tissue, and skeletal muscle.

Two Metabolites Are Associated With Metformin Treatment in a Human Cross-sectional Study

Two out of the 353 used metabolites (citrulline and X-21365) were found to be significantly (P < 1.42E-04) associated with metformin treatment when comparing mt-T2D with ndt-T2D patients in the cross-sectional KORA F4 study (Table 3 and Fig. 1A). Using multivariable linear regression models, we detected negative β-estimates for both the crude (β = –0.75, P = 2.31E-05) and full adjustment (β = –0.79, P = 2.54E-05) for citrulline. Hence, the relative concentration of citrulline is significantly lower in mt-T2D compared with ndt-T2D patients. By contrast, the relative concentration of X-21365 was significantly higher in mt-T2D patients than in ndt-T2D patients (Table 3 and Fig. 1A). When applying the FDR, no additional associations were found to be significant in both crude and full models (Supplementary Table 3). When applying a significance cutoff of P < 0.05 to the comparison of mt-T2D with ndt-T2D for the models with crude and full adjustment, 44 additional metabolites were found, including ornithine, arginine, and urea (Supplementary Table 3).

Table 3

Two human serum metabolites significantly associated with metformin treatment in a cross-sectional analyses (KORA F4)

Crude linear model
mt-T2D (n = 74) vs. ndt-T2D (n = 115)Full linear model
mt-T2D (n = 70) vs. ndt-T2D (n = 114)
Metaboliteβ (95% CI) per SD
P valueFDRβ (95% CI) per SDP valueFDR
Citrulline
 
−0.75 (−1.09, −0.41)
 
2.31E-05
 
2.83E-04
 
−0.79 (−1.15, −0.43)
 
2.54E-05
 
2.83E-04
 
X-21365 0.67 (0.38, 0.96) 7.54E-06 1.42E-04 0.65 (0.34, 0.97) 5.20E-05 1.42E-04 
Crude linear model
mt-T2D (n = 74) vs. ndt-T2D (n = 115)Full linear model
mt-T2D (n = 70) vs. ndt-T2D (n = 114)
Metaboliteβ (95% CI) per SD
P valueFDRβ (95% CI) per SDP valueFDR
Citrulline
 
−0.75 (−1.09, −0.41)
 
2.31E-05
 
2.83E-04
 
−0.79 (−1.15, −0.43)
 
2.54E-05
 
2.83E-04
 
X-21365 0.67 (0.38, 0.96) 7.54E-06 1.42E-04 0.65 (0.34, 0.97) 5.20E-05 1.42E-04 

Estimates (β) and P values for the comparison of 189 participants (74 mt-T2D and 115 ndt-T2D) were calculated using linear regression analysis with the crude and full adjustments. Because of missing confounding information, the models with full adjustment were based on fewer participants. Significant metabolites are highlighted in boldface type with respect to Bonferroni correction (P < 0.05/353 = 1.42E-04) or the FDR.

¶After exclusion of individuals because of missing confounding information.

Five additional pairwise comparisons between the four groups (NGT, IGT, ndt-T2D, mt-T2D) confirmed that these two metabolites are specific for metformin treatment and not due to the progression of the disease. The relative concentration of citrulline was significantly lower in the mt-T2D than in the NGT and IGT groups, whereas the concentration of X-21365 was significantly higher (Fig. 1A and Supplementary Table 2). Consistently, neither of the two metabolites showed a significantly different relative concentration in the pairwise comparisons among the three groups without metformin treatment, i.e., NGT, IGT, and ndt-T2D (Fig. 1A and Supplementary Table 2).

The Spearman correlation coefficient between the two metabolites was low (r = 0.06). We observed similar associations between the two metabolites with a number of risk factors of T2D (−0.19 < r < 0.19) when considering 189 individuals with ndt-T2D and mt-T2D in KORA F4 (Supplementary Fig. 1).

Metformin Treatment Is Associated With Decreased Blood Citrulline Values in a Human Longitudinal Cohort

The two metformin-associated metabolites were further investigated in the prospective KORA study. In 37 patients who started metformin treatment during the 7-year follow-up, citrulline was found to be significantly (Bonferroni cutoff for two identified metabolites P < 0.05/2 = 0.025) decreased in longitudinal in both the crude (β = –0.67, P = 2.03E-05) and the full model (β = –0.61, 2.96E-04, Table 4). In the same group, X-21365 was significantly increased in the crude (β = 0.41, P = 5.62E-03) but not in the full model (β = 0.16, P = 0.374, Table 4).

Table 4

Citrulline remains significantly associated with metformin treatment in human serum in a longitudinal analysis (KORA S4 → F4)

Crude GEE model
mt-T2D (n = 37) vs. nonmetformin-treated (n = 646) participants§Full GEE model
mt-T2D (n = 33) vs. nonmetformin-treated (n = 629) participants§
Metaboliteβ (95% CI) per SD
P valueFDRβ (95% CI) per SDP valueFDR
Citrulline
 
−0.67 (−0.98, −0.36)
 
2.03E-05
 
1.76E-03
 
−0.61 (−0.94, −0.28)
 
2.96E-04
 
3.21E-04
 
X-21365 0.41 (0.12, 0.69) 5.62E-03 0.011 0.14 (−0.17, 0.45) 0.374 0.024 
Crude GEE model
mt-T2D (n = 37) vs. nonmetformin-treated (n = 646) participants§Full GEE model
mt-T2D (n = 33) vs. nonmetformin-treated (n = 629) participants§
Metaboliteβ (95% CI) per SD
P valueFDRβ (95% CI) per SDP valueFDR
Citrulline
 
−0.67 (−0.98, −0.36)
 
2.03E-05
 
1.76E-03
 
−0.61 (−0.94, −0.28)
 
2.96E-04
 
3.21E-04
 
X-21365 0.41 (0.12, 0.69) 5.62E-03 0.011 0.14 (−0.17, 0.45) 0.374 0.024 

GEE model with crude and full adjustment was used to assess the associations between metformin treatment and metabolite serum values in the longitudinal study of 683 participants with no antidiabetes medical treatment at KORA S4. Of these participants, 37 started metformin treatment after KORA S4. Because of missing confounding information, the models with full adjustment were based on fewer participants. Significant metabolites are highlighted in boldface type with respect to Bonferroni correction (P < 0.05/2 = 0.025) and the FDR.

§Includes participants with NGT, isolated IFG, IGT, and ntd-T2D.

¶After exclusion of individuals because of missing confounding information.

Lower Citrulline Relative Concentrations in Plasma, Skeletal Muscle, and Epididymal Adipose Tissue Confirmed in Metformin-Treated Mice

We observed significantly lower plasma citrulline relative concentrations in db/db mice following daily, subchronic metformin treatment compared with the vehicle-gavaged control mice (β = –0.39, P = 2.56E-07, Table 5 and Fig. 1B), which is consistent with the results observed in humans. In addition, we found significantly lower values of citrulline in both skeletal muscle (β = –0.35, P = 1.79E-09) and epididymal adipose tissue (β = –0.26, P = 4.52E-07). However, citrulline values in the liver did not differ between the metformin-treated and vehicle-gavaged non–metformin-treated db/db animals (β = –0.02, P = 0.258, Table 5 and Fig. 1B). Significantly different relative concentrations of X-21365 were not found in plasma, skeletal muscle, epididymal adipose tissue, or liver of metformin-treated mice when compared with the controls (Table 5 and Fig. 1B).

Table 5

Metabolites significantly associated with metformin treatment in mouse models

MetaboliteIn plasma
In liver
In skeletal muscle
In adipose tissue
β (95% CI) per SDP valueβ (95% CI) per SDP valueβ (95% CI) per SDP valueβ (95% CI) per SDP value
Citrulline
 
−0.39 (−0.49, −0.28)
 
2.56E-07
 
−0.02 (−0.06, 0.02)
 
0.258
 
−0.35 (−0.41, −0.28)
 
1.79E-09
 
−0.26 (−0.33, −0.19)
 
4.52E-07
 
X-21365 −0.08 (−0.24, 0.08) 0.311 −0.12 (−0.34, 0.11) 0.295 −0.12 (−0.27, 0.04) 0.126 −0.23 (−0.46, 0.01) 0.063 
MetaboliteIn plasma
In liver
In skeletal muscle
In adipose tissue
β (95% CI) per SDP valueβ (95% CI) per SDP valueβ (95% CI) per SDP valueβ (95% CI) per SDP value
Citrulline
 
−0.39 (−0.49, −0.28)
 
2.56E-07
 
−0.02 (−0.06, 0.02)
 
0.258
 
−0.35 (−0.41, −0.28)
 
1.79E-09
 
−0.26 (−0.33, −0.19)
 
4.52E-07
 
X-21365 −0.08 (−0.24, 0.08) 0.311 −0.12 (−0.34, 0.11) 0.295 −0.12 (−0.27, 0.04) 0.126 −0.23 (−0.46, 0.01) 0.063 

Estimates (β) and P values for the comparison between metformin-treated (n = 10, in skeletal muscle [n = 9]) and nontreated mice (n = 10) sacrificed at 4 h after the last treatment. Significant metabolites are highlighted in boldface type (P < 0.05).

We found significantly lower values of citrulline and significantly higher values of X-21365 in the serum of T2D patients who underwent metformin treatment compared with the nontreated patients. Additionally, using longitudinal settings, we observed that the values of citrulline significantly decreased in patients after they started metformin treatment during the follow-up. A mouse intervention study using metformin confirmed the lower values of citrulline in plasma, as well as in skeletal muscle and epididymal adipose tissue, but not in liver. Citrulline is a nonproteinogenic amino acid, the product of anabolic and the substrate of catabolic processes (38,39). It is synthesized from arginine by releasing NO, which is involved in the regulation of numerous processes in the nervous system, the immune system, and the cardiovascular system (8). Additionally, citrulline is produced from ornithine in the urea cycle (38). We observed ornithine, urea, and arginine to be lowered in human serum (Fig. 1C). Consistently, in our previous study, which was based on a targeted metabolomics approach, ornithine was found to be significantly lower in the metformin-treated T2D patients of the KORA F4 study. Citrulline was not measured in the targeted panel we used (6).

Metformin activates AMPK in the liver and muscle (7,40). AMPK in turn may stimulate eNOS by its phosphorylation (8,41), which suggests a consequent increase of the NO production in the NO cycle (Fig. 1C). It is known that elevated production of NO is reflected by increased values of citrulline in urine (42), as citrulline can be used as a surrogate marker for NO (43). The decreased values of citrulline and its precursors in blood, skeletal muscle, and epididymal adipose tissue, as were observed in our study, are most likely due to an accountable, increased excretion of this metabolite. However, urine samples were not available in this study. To confirm this assumption, further studies are necessary.

Furthermore, the lower values of citrulline and arginine we observed are likely to be a consequence of the activation of eNOS. In the NO cycle, eNOS catalyzes the reaction from arginine to citrulline, thereby releasing NO (9,38). NO in turn has beneficial cardiovascular effects. The reason is that NO influences smooth muscles and activates their relaxation (44). This underlies the clinical practice guidelines, which have recommended the use of metformin as first-line therapy in T2D patients with cardiovascular disease, mainly in patients with observed reduced NO levels (45). Additional intake of citrulline to compensate for the lower values of citrulline and arginine might even increase the beneficial effects of metformin on cardiovascular disease (46).

Additionally, citrulline is synthesized in the urea cycle, which is strongly interlocked with the NO cycle (Fig. 1C). In mammals, both cycles primarily take place in the liver, but they also take place in the kidney (47). The same accounts for the NO cycle, in which arginine also plays an important role. In fact, similar effects of metformin on the urea and NO cycle were mentioned by Irving et al. (15). Their study design focused on plasma samples of 25 male overweight or obese participants. Furthermore, all 12 metformin-treated participants were additionally treated with pioglitazone (15). Our findings in multiple tissues of mice that were exclusively treated with metformin and in serum of 189 T2D patients enable a deeper understanding of the underlying mode of action for metformin.

The observation that the citrulline values are not affected in the liver of metformin-treated mice is presumably a consequence of the hepatic localization of the consecutive production of citrulline in both the NO and the urea cycle (38), which conserves a state of equilibrium. This is in line with observations in a recent study (18). Furthermore, significantly decreased ornithine values were found in plasma of individuals without diabetes (18).

Apart from the NO and urea cycles, there are additional physiological processes that produce citrulline. The metabolite is also synthesized from other amino acids. Examples of such precursors are glutamine, which is converted in the enterocytes, proline, and glutamate (38). However, we did not observe any significant concentration difference for these metabolites in our human cohort.

X-21365 was not found to be significantly higher in the fully adjusted longitudinal analyses of the KORA S4 → F4 cohort, although it was significant in both cross-sectional analyses and in the longitudinal analyses with crude adjustment. In mice, we did not observe significant differences of X-21365 in any of the examined tissues. Recent advances in the identification of metabolites spectra suggest that this unknown metabolite (X-21365) might be 5-trimethylaminovalerate and therefore closely related to the gut microbiome, which is in line with a recent study (48). Additional studies using both blood and stool samples have to be conducted to confirm this.

The values of metabolites in humans of the KORA study are influenced by multiple factors such as age, sex, BMI, lifestyle, clinical measurements, and medication (6,13,2832). We therefore considered these factors in the models underlying our cross-sectional discovery and longitudinal investigations in a human cohort. Considering the mouse study, there was no need for a comparable adjustment, as the animals were kept under strict laboratory conditions.

Because of the physiological similarity, we used data from a mouse study not only to corroborate our findings in humans but to extend our investigations on other tissues. However, our findings are limited by the comparison of metabolic analytes in two different blood matrices and species: human serum and mouse plasma. In theory, the analytical method could be affected by the difference in matrix, and delicate analytes could deteriorate during the prolonged preparation time of serum compared with that of plasma. Therefore, a direct comparison between the matrices serum and plasma has limitations (49). With respect to this, we compared the serum metabolites only within humans, the plasma metabolites only within mice, and each mice tissue separately (50). The observational nature of cohort studies and the applied methods are of purely statistical character, yet still they offer the opportunity to identify unknown coherences and to design study settings to confirm underlying mechanisms. Because of the fact that NO is below the mass cutoff imposed on the instruments, our investigations did not contain measurements of this chemical compound. Nevertheless, our observations suggest further investigations with a specific design to address the involvement of the NO and urea cycle in metformin treatment.

In summary, we observed that serum values of citrulline were reduced under metformin treatment in human patients with T2D and, in a translational approach, also in plasma, skeletal muscle, and epididymal adipose tissue of diabetic mice. The underlying mechanism is most likely the metformin-induced activation of AMPK and its consequent increase of eNOS activity, which is linked to citrulline by the NO cycle.

See accompanying article, p. 3537.

Acknowledgments. The authors express their appreciation to all KORA study participants for donating their blood and time. The authors thank the field staff in Augsburg, Germany, for conducting the KORA studies. The authors are grateful to the staff (Julia Scarpa, Katharina Faschinger, Franziska Scharl, Nadine Lindemann, Humberto Chavez, and Arsin Sabunchi) from the Institute of Epidemiology at the Helmholtz Zentrum München and the Genome Analysis Center Metabolomics Platform who helped with the sample logistics, data and straw collection, and metabolomic measurements. Additionally, the authors thank the staff (especially Andrea Schneider, Anja Ludolph, Sladjana Jelic, and Birgit Langer) from the Institute of Genetic Epidemiology at the Helmholtz Zentrum München and the platform KORA project application self-service tool (KORA.PASST) for their help with KORA data logistics. The authors also thank the people of the Institute of Diabetes and Regeneration Research (Anett Seelig and Jürgen Schultheiß) and the Institute of Experimental Genetics (Moya Wu), as well as the animal caretaker staff of the German Mouse Clinic, for excellent technical assistance.

Funding. The KORA study was initiated and financed by the Helmholtz Zentrum München German Research Center for Environmental Health, which is funded by the German Federal Ministry of Education and Research (Bundesministerium für Bildung und Forschung) and by the State of Bavaria. Furthermore, KORA research was supported by the Munich Center of Health Sciences (MC-Health), Ludwig-Maximilians-Universität, as part of LMUinnovativ. Part of this project was supported by European Union Seventh Framework Programme grants HEALTH-2009-2.2.1-3/242114 (Project OPTiMiSE) and HEALTH-2013-2.4.2-1/602936 (Project CarTarDis). The German Diabetes Center is funded by the German Federal Ministry of Health (Bundesministerium für Gesundheit, Berlin, Germany) and the Ministry of Innovation, Science and Research of the State of North Rhine-Westphalia (Düsseldorf, Germany). The diabetes part of the KORA F4 study was funded by a grant from the Deutsche Forschungsgemeinschaft (DFG; RA 459/3-1). This study was supported in part by a grant from the Bundesministerium für Bildung und Forschung to the German Center for Diabetes Research (DZD) and by the research consortium “Systems Biology of Metabotypes” (SysMBo grant 0315494A). K.Su. is supported by Biomedical Research Program funds at Weill Cornell Medical College in Qatar, a program funded by the Qatar Foundation.

Duality of Interest. M.F.S. was employed at Helmholtz Zentrum München GmbH during the execution of this study. M.F.S. is currently an employee of the Diabetes Medical Department of AstraZeneca GmbH (Wedel, Germany); however, the company was not involved in work related to data and manuscript generation. R.P.M. is an employee of Metabolon, Inc. S.N. was employed by the Helmholtz Zentrum München GmbH during the execution of this study. S.N. is currently an employee of Sanofi Deutschland GmbH; however, the company was not involved in work related to data and manuscript generation. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. J.Adam, S.B., and R.W.-S. wrote the manuscript. J.L., M.F.S., R.P.M., M.Rot., M.T., S.C., C.H., Y.L., D.A., T.M., M.Rod., and S.N. assisted in manuscript generation. J.Adam, J.L., T.X., and J.B analyzed the data and interpreted the results. M.F.S., R.P.M., M.H., C.H., W.R., G.G., J.Adams., T.I., K.St., C.G., A.P., K.Su., M.H.d.A., S.N., and G.K. performed the experiments, including metabolic profiling. G.K. and R.W.-S. conceived and designed the study. R.W.-S. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

1.
Turner
R
;
UK Prospective Diabetes Study (UKPDS) Group
.
Intensive blood-glucose control with sulphonylureas or insulin compared with conventional treatment and risk of complications in patients with type 2 diabetes (UKPDS 33)
.
Lancet
1998
;
352
:
837
853
[PubMed]
2.
Dunaif
A
.
Insulin resistance and the polycystic ovary syndrome: mechanism and implications for pathogenesis
.
Endocr Rev
1997
;
18
:
774
800
[PubMed]
3.
Giugliano
D
,
De Rosa
N
,
Di Maro
G
, et al
.
Metformin improves glucose, lipid metabolism, and reduces blood pressure in hypertensive, obese women
.
Diabetes Care
1993
;
16
:
1387
1390
[PubMed]
4.
Neschen
S
,
Scheerer
M
,
Seelig
A
, et al
.
Metformin supports the antidiabetic effect of a sodium glucose cotransporter 2 inhibitor by suppressing endogenous glucose production in diabetic mice
.
Diabetes
2015
;
64
:
284
290
[PubMed]
5.
Decensi
A
,
Puntoni
M
,
Goodwin
P
, et al
.
Metformin and cancer risk in diabetic patients: a systematic review and meta-analysis
.
Cancer Prev Res (Phila)
2010
;
3
:
1451
1461
[PubMed]
6.
Xu
T
,
Brandmaier
S
,
Messias
AC
, et al
.
Effects of metformin on metabolite profiles and LDL cholesterol in patients with type 2 diabetes
.
Diabetes Care
2015
;
38
:
1858
1867
[PubMed]
7.
Zhou
G
,
Myers
R
,
Li
Y
, et al
.
Role of AMP-activated protein kinase in mechanism of metformin action
.
J Clin Invest
2001
;
108
:
1167
1174
[PubMed]
8.
Chen
ZP
,
Mitchelhill
KI
,
Michell
BJ
, et al
.
AMP-activated protein kinase phosphorylation of endothelial NO synthase
.
FEBS Lett
1999
;
443
:
285
289
[PubMed]
9.
Davis
BJ
,
Xie
Z
,
Viollet
B
,
Zou
MH
.
Activation of the AMP-activated kinase by antidiabetes drug metformin stimulates nitric oxide synthesis in vivo by promoting the association of heat shock protein 90 and endothelial nitric oxide synthase
.
Diabetes
2006
;
55
:
496
505
[PubMed]
10.
Konopka
AR
,
Esponda
RR
,
Robinson
MM
, et al
.
Hyperglucagonemia mitigates the effect of metformin on glucose production in prediabetes
.
Cell Reports
2016
;
15
:
1394
1400
[PubMed]
11.
Huo
T
,
Cai
S
,
Lu
X
,
Sha
Y
,
Yu
M
,
Li
F
.
Metabonomic study of biochemical changes in the serum of type 2 diabetes mellitus patients after the treatment of metformin hydrochloride
.
J Pharm Biomed Anal
2009
;
49
:
976
982
[PubMed]
12.
Walford
GA
,
Davis
J
,
Warner
AS
, et al
.
Branched chain and aromatic amino acids change acutely following two medical therapies for type 2 diabetes mellitus
.
Metabolism
2013
;
62
:
1772
1778
[PubMed]
13.
Brandmaier
S
,
Xu
T
,
Illig
T
,
Suhre
K
,
Adamski
J
,
Wang-Sattler
R
.
Response to comment on Xu et al. Effects of metformin on metabolite profiles and LDL cholesterol in patients with type 2 diabetes. Diabetes Care 2015;38:1858–1867
.
Diabetes Care
2015
;
38
:
e216
e217
[PubMed]
14.
Holle R, Happich M, Löwel H, Wichmann HE.; MONICA/KORA Study Group. KORA--a research platform for population based health research. Gesundheitswesen 2005;67(Suppl. 1):S19–S25
15.
Irving
BA
,
Carter
RE
,
Soop
M
, et al
.
Effect of insulin sensitizer therapy on amino acids and their metabolites
.
Metabolism
2015
;
64
:
720
728
[PubMed]
16.
Menni
C
,
Fauman
E
,
Erte
I
, et al
.
Biomarkers for type 2 diabetes and impaired fasting glucose using a nontargeted metabolomics approach
.
Diabetes
2013
;
62
:
4270
4276
[PubMed]
17.
Suhre
K
,
Meisinger
C
,
Döring
A
, et al
.
Metabolic footprint of diabetes: a multiplatform metabolomics study in an epidemiological setting
.
PLoS One
2010
;
5
:
e13953
[PubMed]
18.
Rotroff
DM
,
Oki
NO
,
Liang
X
, et al
.
Pharmacometabolomic assessment of metformin in non-diabetic, African Americans
.
Front Pharmacol
2016
;
7
:
135
[PubMed]
19.
Rathmann
W
,
Kowall
B
,
Heier
M
, et al
.
Prediction models for incident type 2 diabetes mellitus in the older population: KORA S4/F4 cohort study
.
Diabet Med
2010
;
27
:
1116
1123
[PubMed]
20.
Meisinger
C
,
Strassburger
K
,
Heier
M
, et al
.
Prevalence of undiagnosed diabetes and impaired glucose regulation in 35-59-year-old individuals in Southern Germany: the KORA F4 Study
.
Diabet Med
2010
;
27
:
360
362
[PubMed]
21.
Wang-Sattler
R
,
Yu
Z
,
Herder
C
, et al
.
Novel biomarkers for pre-diabetes identified by metabolomics
.
Mol Syst Biol
2012
;
8
:
615
[PubMed]
22.
Evans
AM
,
DeHaven
CD
,
Barrett
T
,
Mitchell
M
,
Milgram
E
.
Integrated, nontargeted ultrahigh performance liquid chromatography/electrospray ionization tandem mass spectrometry platform for the identification and relative quantification of the small-molecule complement of biological systems
.
Anal Chem
2009
;
81
:
6656
6667
[PubMed]
23.
Ohta
T
,
Masutomi
N
,
Tsutsui
N
, et al
.
Untargeted metabolomic profiling as an evaluative tool of fenofibrate-induced toxicology in Fischer 344 male rats
.
Toxicol Pathol
2009
;
37
:
521
535
[PubMed]
24.
Suhre
K
,
Shin
S-Y
,
Petersen
A-K
, et al.;
CARDIoGRAM
.
Human metabolic individuality in biomedical and pharmaceutical research
.
Nature
2011
;
477
:
54
60
[PubMed]
25.
Albrecht
E
,
Waldenberger
M
,
Krumsiek
J
, et al
.
Metabolite profiling reveals new insights into the regulation of serum urate in humans
.
Metabolomics
2014
;
10
:
141
151
[PubMed]
26.
van Buuren
S
,
Groothuis-Oudshoorn
K
.
Mice: multivariate imputation by chained equations in R
.
J Stat Softw
2011
;
45
:
1
67
27.
Shin
S-Y
,
Fauman
EB
,
Petersen
A-K
, et al.;
Multiple Tissue Human Expression Resource (MuTHER) Consortium
.
An atlas of genetic influences on human blood metabolites
.
Nat Genet
2014
;
46
:
543
550
[PubMed]
28.
Yu
Z
,
Zhai
G
,
Singmann
P
, et al
.
Human serum metabolic profiles are age dependent
.
Aging Cell
2012
;
11
:
960
967
[PubMed]
29.
Xu
T
,
Holzapfel
C
,
Dong
X
, et al
.
Effects of smoking and smoking cessation on human serum metabolite profile: results from the KORA cohort study
.
BMC Med
2013
;
11
:
60
[PubMed]
30.
Jaremek M, Yu Z, Mangino M, et al. Alcohol-induced metabolomic differences in humans. Transl Psychiatry 2013;3:e276
31.
Altmaier
E
,
Fobo
G
,
Heier
M
, et al
.
Metabolomics approach reveals effects of antihypertensives and lipid-lowering drugs on the human metabolism
.
Eur J Epidemiol
2014
;
29
:
325
336
[PubMed]
32.
Jourdan
C
,
Petersen
A-K
,
Gieger
C
, et al
.
Body fat free mass is associated with the serum metabolite profile in a population-based study
.
PLoS One
2012
;
7
:
e40009
[PubMed]
33.
R Development Core Team. R: A Language and Environment for Statistical Computing. Vienna, Austria. R Foundation for Statistical Computing, 2009
34.
Wishart
DS
,
Tzur
D
,
Knox
C
, et al
.
HMDB: the Human Metabolome Database
.
Nucleic Acids Res
2007
;
35
:
D521
D526
[PubMed]
35.
Szklarczyk D, Franceschini A, Kuhn M, et al. The STRING database in 2011: functional interaction networks of proteins, globally integrated and scored. Nucleic Acids Res 2011;39:D561–D568
36.
Kanehisa
M
,
Goto
S
,
Sato
Y
,
Kawashima
M
,
Furumichi
M
,
Tanabe
M
.
Data, information, knowledge and principle: back to metabolism in KEGG
.
Nucleic Acids Res
2014
;
42
:
D199
D205
[PubMed]
37.
Law
V
,
Knox
C
,
Djoumbou
Y
, et al
.
DrugBank 4.0: shedding new light on drug metabolism
.
Nucleic Acids Res
2014
;
42
:
D1091
D1097
[PubMed]
38.
Curis
E
,
Nicolis
I
,
Moinard
C
, et al
.
Almost all about citrulline in mammals
.
Amino Acids
2005
;
29
:
177
205
[PubMed]
39.
Baur
H
,
Stalon
V
,
Falmagne
P
,
Luethi
E
,
Haas
D
.
Primary and quaternary structure of the catabolic ornithine carbamoyltransferase from Pseudomonas aeruginosa. Extensive sequence homology with the anabolic ornithine carbamoyltransferases of Escherichia coli
.
Eur J Biochem
1987
;
166
:
111
117
[PubMed]
40.
Musi
N
,
Hirshman
MF
,
Nygren
J
, et al
.
Metformin increases AMP-activated protein kinase activity in skeletal muscle of subjects with type 2 diabetes
.
Diabetes
2002
;
51
:
2074
2081
[PubMed]
41.
Zhang
Y
,
Lee
TS
,
Kolb
EM
, et al
.
AMP-activated protein kinase is involved in endothelial NO synthase activation in response to shear stress
.
Arterioscler Thromb Vasc Biol
2006
;
26
:
1281
1287
[PubMed]
42.
Wanchu
A
,
Khullar
M
,
Sud
A
,
Deodhar
SD
,
Bambery
P
.
Elevated urinary nitrite and citrulline levels in patients with rheumatoid arthritis
.
Inflammopharmacology
1999
;
7
:
155
161
[PubMed]
43.
Sud
A
,
Khullar
M
,
Wanchu
A
,
Bambery
P
.
Increased nitric oxide production in patients with systemic sclerosis
.
Nitric Oxide
2000
;
4
:
615
619
[PubMed]
44.
Conti
V
,
Russomanno
G
,
Corbi
G
,
Izzo
V
,
Vecchione
C
,
Filippelli
A
.
Adrenoreceptors and nitric oxide in the cardiovascular system
.
Front Physiol
2013
;
4
:
321
[PubMed]
45.
Tessari
P
,
Cecchet
D
,
Cosma
A
, et al
.
Nitric oxide synthesis is reduced in subjects with type 2 diabetes and nephropathy
.
Diabetes
2010
;
59
:
2152
2159
[PubMed]
46.
Romero
MJ
,
Platt
DH
,
Caldwell
RB
,
Caldwell
RW
.
Therapeutic use of citrulline in cardiovascular disease
.
Cardiovasc Drug Rev
2006
;
24
:
275
290
[PubMed]
47.
Husson
A
,
Brasse-Lagnel
C
,
Fairand
A
,
Renouf
S
,
Lavoinne
A
.
Argininosuccinate synthetase from the urea cycle to the citrulline-NO cycle
.
Eur J Biochem
2003
;
270
:
1887
1899
[PubMed]
48.
Zhernakova
A
,
Kurilshikov
A
,
Bonder
MJ
, et al.;
LifeLines cohort study
.
Population-based metagenomics analysis reveals markers for gut microbiome composition and diversity
.
Science
2016
;
352
:
565
569
[PubMed]
49.
Becker
S
,
Kortz
L
,
Helmschrodt
C
,
Thiery
J
,
Ceglarek
U
.
LC-MS-based metabolomics in the clinical laboratory
.
J Chromatogr B Analyt Technol Biomed Life Sci
2012
;
883-884
:
68
75
50.
Yu
Z
,
Kastenmüller
G
,
He
Y
, et al
.
Differences between human plasma and serum metabolite profiles
.
PLoS One
2011
;
6
:
e21230
[PubMed]
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 http://www.diabetesjournals.org/content/license.

Supplementary data