Identification of urine metabolites associated with 5

2 downloads 0 Views 1MB Size Report
Diabetes & Metabolism xxx (2017) xxx–xxx .... Relaxation delay (RD), mixing time (tm) and acquisition time were set ...... Clin Chem Lab Med 2005;43:1069–75.
G Model

DIABET-895; No. of Pages 8 Diabetes & Metabolism xxx (2017) xxx–xxx

Available online at

ScienceDirect www.sciencedirect.com

Original article

Identification of urine metabolites associated with 5-year changes in biomarkers of glucose homoeostasis N. Friedrich a,b,c,*, T. Skaaby a, M. Pietzner b,c, K. Budde b, B.H. Thuesen a, M. Nauck b,c, A. Linneberg a,d,e a

Research centre for prevention and health, the capital region of Denmark, Glostrup, Denmark Institute of clinical chemistry and laboratory medicine, university medicine Greifswald, Greifswald, Germany DZHK (German centre for cardiovascular research), partner site, Greifswald, Germany d Department of clinical experimental research, Rigshospitalet, Denmark e Department of clinical medicine, faculty of health and medical sciences, university of Copenhagen, Copenhagen, Denmark b c

A R T I C L E I N F O

A B S T R A C T

Article history: Received 16 January 2017 Accepted 23 May 2017 Available online xxx

Aim. – Metabolomics provides information on pathogenetic mechanisms and targets for interventions, and may improve risk stratification. During the last decade, metabolomics studies were used to gain deeper insight into the pathogenesis of diabetes mellitus. However, longitudinal metabolomics studies of possible subclinical states of disturbed glucose metabolism are limited. Therefore, the aim of this study was to analyze the associations between baseline urinary metabolites and 5-year changes in continuous markers of glucose homoeostasis, including fasting glucose, HbA1c and homoeostasis model assessment of insulin resistance (HOMA-IR) index values. Methods. – Urine metabolites in 3986 participants at both baseline and 5-year follow-up of the population-based Inter99 study were analyzed by 1H-NMR spectroscopy. Linear regression and analyses of covariance models were used to detect associations between urine metabolites and 5-year changes in markers of glucose homoeostasis. Results. – Higher baseline levels of urinary alanine, betaine, N,N-dimethylglycine (DMG), creatinine and trimethylamine were associated with an increase in HbA1c from baseline to follow-up. In contrast, formic acid and trigonelline levels were associated with a decrease in HbA1c over time. Analyses of 5-year changes in fasting glucose and HOMA-IR index showed similar findings, with high baseline levels of lactic acid, beta-D-glucose, creatinine, alanine and 1-methylnicotinamide associated with increases in both parameters. Conclusion. – Several urine metabolites were found to be associated with detrimental longitudinal changes in biomarkers of glucose homoeostasis. The identified metabolites point to mechanisms involving betaine and coffee metabolism as well as the possible influence of the gut microbiome.

C 2017 Elsevier Masson SAS. All rights reserved.

Keywords: Betaine Coffee consumption Diabetes mellitus Glucose homoeostasis Metabolomics

Introduction For the past several decades, type 2 diabetes mellitus (T2DM) has represented one of the main worldwide health problems set to become more and more important due to the steadily increasing prevalence of overweight and obesity. Over the last 10 years, the profiling of small molecules—called ‘metabolomics’—has been performed to gain deeper insights into the pathogenesis of diabetes and to identify early biomarkers of T2DM [1–3]. There * Corresponding author at: Research centre for prevention and health, centre for health, capital region of Denmark, Rigshospitalet, Glostrup, Ndr. Ringvej 57, building 84/85, DK-2600 Glostrup, Denmark. Fax: +4938633977. E-mail address: [email protected] (N. Friedrich).

are several cross-sectional metabolomics studies with respect to diabetes, but prospective studies are still limited. First, Wang et al. [4] used mass spectrometry (MS)-based metabolomics to reveal that branched-chain and aromatic amino acids (AAs), such as leucine, valine and phenylalanine, are associated with incident T2DM in a prospective study of 2422 individuals. Further prospective studies confirmed these findings [5–8] and also detected further AAs, including alanine and glycine [9–11], associated with T2DM in different settings. However, the predictive ability of these markers over the classic clinical markers is still a subject of discussion [5,10]. Beside AAs, other metabolomics-based detected markers of incident diabetes have included trigonelline [9], 2-aminoadipic acid [12], acylcarnitines [10,11] and lyso-glycerophospholipids

http://dx.doi.org/10.1016/j.diabet.2017.05.007 C 2017 Elsevier Masson SAS. All rights reserved. 1262-3636/

Please cite this article in press as: Friedrich N, et al. Identification of urine metabolites associated with 5-year changes in biomarkers of glucose homoeostasis. Diabetes Metab (2017), http://dx.doi.org/10.1016/j.diabet.2017.05.007

G Model

DIABET-895; No. of Pages 8 2

N. Friedrich et al. / Diabetes & Metabolism xxx (2017) xxx–xxx

[6,10,13,14], which reflect processes contribute to the development or progression of diabetes involving different metabolic pathways. The majority of these studies [4–7,10–14], however, applied MS to measure blood metabolites, whereas only two previous investigations used nuclear magnetic resonance (NMR) spectroscopy and only one study used urine as a biofluid [9]. As previously reviewed [1,3], both methods have their strengths and weaknesses. However, NMR provides highly reproducible results, and is inexpensive and fast and, therefore, highly suitable as a screening tool, especially combined with urine specimens. Nevertheless, a serious concern of such studies is the limited number of incident cases (< 200) [4–7,9,12–14] and lack of assessment of the possible subclinical states of perturbed glucose metabolism. Thus, the aim of the present investigation was to analyze the associations between baseline urine metabolites measured by NMR spectroscopy and 5-year longitudinal changes in continuous markers of glucose homoeostasis, including fasting glucose, glycated haemoglobin (HbA1c) and homoeostasis model assessment of insulin resistance (HOMA-IR).

Methods Study population The Inter99, a population-based non-pharmacological lifestyle intervention study, was initiated in March 1999 and carried out by the Capital Region of Denmark Research Centre for Prevention and Health. The study design and intervention have been described in detail elsewhere [15], and can also be found on the website www. inter99.dk. Subjects were drawn randomly from the Civil Registration System. The study population (n = 61,301) comprised all individuals in specific age groups (30, 35, 40, 45, 50, 55 and 60 years) from a defined area of Copenhagen. From this study population, 13016 were randomly drawn for the intervention, of whom 6784 (52.5%) were examined at baseline. Of these, 4511 subjects participated in a 5-year follow-up examination. The present project was approved by the Ethics Committee of the Capital Region of Denmark (H-15004167) and Danish Data Protection Agency. Baseline NMR measurements were available for 4117 subjects who participated in both the baseline and 5-year follow-up examinations. Of these, subjects with missing values for baseline/ follow-up HbA1c or baseline confounders were excluded, resulting in a total of 3986 subjects included in the analyses. Measurements Information on lifestyle, such as smoking habits and coffee consumption, were assessed by questionnaire. Blood and urine spot samples were collected after an overnight fast. Height and weight were measured without shoes and with light clothing. Body mass index (BMI) was calculated as weight (kg) divided by height squared (m2). All participants had their blood pressure measured twice with a mercury sphygmomanometer (Mercuro 300; Speidel & Keller GmbH & Co, Jungingen, Germany) and appropriate cuff size after 5 min of rest, in supine position, and the average of the two recorded measurements was used. HbA1c was measured in all participants using ion-exchange high-performance liquid chromatography (VARIANTTM Hemoglobin A1c, Bio-Rad Laboratories, Hercules, CA, USA). Fasting glucose concentrations were analyzed by hexokinase/ glucose-6-phosphate dehydrogenase assay (Roche Diagnostics Corporation, Indianapolis, IN, USA). Insulin levels were measured by a fluoroimmunoassay technique (Dako Diagnostics Ltd., Ely, Cambridgeshire, UK), and the HOMA-IR index was calculated as: [fasting glucose (mmol/L)  fasting insulin (mU/L)]/22.5.

Low-density lipoprotein (LDL) cholesterol was measured using enzymatic colorimetric methods (Roche Diagnostics). 1

H-NMR spectroscopy analysis of urine specimens

After thawing, urine specimens were centrifuged for 5 min at 3000 g, and the supernatant was used for spectroscopy analysis: 450 mL of urine was mixed with 50 mL of phosphate buffer to stabilize urine pH at 7.0 ( 0.35). The buffer was prepared with D2O and contained sodium 3-trimethylsilyl-(2,2,3,3-D4)-1-propionate (TSP) as reference. Spectra were recorded at the University Medicine Greifswald, Germany, on a Bruker DRX-400 NMR spectrometer (Bruker BioSpin GmbH, Rheinstetten, Germany), operating at a 1H frequency of 400.13 MHz, and equipped with a 4-mm selective inverse flow probe (FISEI; 120 mL of active volume) with z gradients. Specimens were automatically delivered to the spectrometer via flow injection. Acquisition temperature was calibrated to 300  0.1 K, and a standard one-dimensional 1H-NMR pulse sequence with suppression of water peak (NOESYGPPR1D) was used: RD—P(908)—4 ms— P(908)—tm—P(908)—acquisition of free induction decay (FID). For each sample, the non-selective 908 hard pulse [P(908)] was individually calibrated in full automation, using the Bruker automation program PULSECAL. Relaxation delay (RD), mixing time (tm) and acquisition time were set to 4 s, 10 ms and 3.96 s, respectively, resulting in a total recycle time of  8.0 s. Low-power continuouswave irradiation on water resonance at a radiofrequency field strength of 25 Hz was applied during the RD and tm for presaturation, and 1-ms z gradients were applied between RD and P(908) and between tm and P(908) to further reduce the residual solvent signal. After application of four dummy scans, NS 32 were collected into 65536 (64 K) complex data points, using a spectral width of 20.689 parts per million (ppm) and a receiver gain (RG) setting of 128. FIDs were multiplied with an exponential function corresponding to a line broadening of 0.3 Hz before Fourier transformation. TopSpin Version TS2.1pl6 was generally used for automated data acquisition and data processing. Spectra were automatically phasecorrected and referenced to the internal standard (TSP–0.0 ppm), using Bruker’s processing programme APK0.NOE. Bucketing of 1H-NMR spectra Processed spectra were segmented into 500 consecutive integrated spectral regions (buckets) of fixed bucket width (0.018 ppm), covering a range of 0.5 ppm to 9.5 ppm (R version 3.0.1, R Foundation for Statistical Computing, Vienna, Austria). The 4.5– 5.0 ppm chemical shift region (28 buckets) was left out of the analysis to remove the effects of variations in suppression of water resonance and variations in the urea signal, caused by partial crosssolvent saturation through solvent-exchanging protons. To account for urine dilution based on the remaining 472 buckets, a dilution factor for each sample was obtained by probabilistic quotient normalization (PQN) [16]. This procedure involves calculation of a median reference spectrum within the population (overall Inter99 observations) and, subsequently, estimation of the median quotient of each sample to this reference. Thereafter, the derived dilution factor was used for normalization of the buckets. In addition, the Fourier-transformed and baseline corrected spectra were processed, using the published FOCUS workflow, including a proposed alignment and peak-picking algorithm called RUNAS [17]. Again, for this purpose, spectral regions containing the water peak (4.5– 5.0 ppm) were excluded and the algorithm was run with the default parameters, whereas the threshold for such peak occurrences was adapted to the present signal intensities. As a result, 153 distinct peak entities that markedly reduced the undesired shift in ppm signals due to slight differences in pH or molecular interactions were obtained. The FOCUS-aligned peaks were also PQN normalized.

Please cite this article in press as: Friedrich N, et al. Identification of urine metabolites associated with 5-year changes in biomarkers of glucose homoeostasis. Diabetes Metab (2017), http://dx.doi.org/10.1016/j.diabet.2017.05.007

G Model

DIABET-895; No. of Pages 8 N. Friedrich et al. / Diabetes & Metabolism xxx (2017) xxx–xxx

Metabolite quantification A total of 18 metabolites were automatically annotated and quantified using the Bruker Remote Data Analysis Server. In brief, quantification is performed via signal-fitting using a simplex algorithm under consideration of constraints. Constraints are metabolite- and signal-specific parameters and their ranges, including:  metabolite parameters (molecular mass);  parameters/strategies for identification of each considered signal of each metabolite (number of protons, relaxation time, multiplicity, coupling constant and/or peak patterns, search range for signal detection);  and definition of additional signal-fit parameters, their start, and allowed minimum and maximum values (fit range, chemical shift, line width, coupling constant, Gauss–Lorentz ratio, baseline offset and slope). In all, 17 metabolites were used in the present study. Correlation between metabolites at baseline or follow-up as well as metabolite-specific correlation between baseline and follow-up values are presented in Fig. S1 (see supplementary material associated with this article online). To account for urine dilution, metabolite concentrations were normalized by the bucket-based derived dilution factor, as mentioned above. This was of particular importance to avoid biased results arising from the usually applied creatinine normalization, as urine creatinine levels are significantly associated with the outcome of interest. Statistical analysis Continuous data are expressed as medians (quartile 1; quartile 3) and nominal data as percentages. For bivariate statistics (Table 1), the Wilcoxon rank-sum test (continuous data) or x2-test (nominal data) was used. For longitudinal association analyses, weighting for the inverse probability of attrition (IPA) was applied to account for any potential attrition bias [18]. To derive the IPA weights, a logistic-regression model for participation at follow-up was applied, using baseline variables such as age, sex, BMI,

3

smoking habits, physical activity, education and HbA1c as independent variables. Based on this regression model, the individual’s probability of follow-up participation and IPA weight as the reciprocal of individual probability were calculated. The main aim of the present study was to analyze the association between baseline urine metabolite levels and 5-year changes in HbA1c. Initial analyses revealed a negative correlation between baseline HbA1c and changes, indicating regression to the mean (Fig. S2; see supplementary material associated with this article online). This means that subjects with low HbA1c levels at baseline generally deteriorate more than subjects with high baseline HbA1c levels. Therefore, two analyses were used to account for this. In the first-step linear regression analyses of the 5year follow-up HbA1c levels as outcome and baseline urine metabolite levels as exposure adjusted for baseline HbA1c levels were performed. All models were further adjusted for age, sex, BMI, LDL cholesterol and systolic blood pressure (SBP). For each metabolite-specific regression analysis, metabolites were logtransformed and subjects with extreme metabolite levels were excluded (metabolites > mean + 3 * standard deviation (SD) or metabolite < mean–3 * SD). In the second step, analyses of covariance (ANCOVAs) were performed with baseline metabolites categorized into four groups by quartiles of their distribution. For each metabolite-specific analysis, only the subset that matched the initial HbA1c ( 0.2%) levels across all four groups was used. Models were adjusted for age, sex, BMI, LDL cholesterol and SBP. Besides the metabolite-based analyses, all linear regression models were further performed with buckets as well as FOCUS-aligned peaks as exposure variables. Again, extreme buckets or peak values were excluded (mean  3 * SD approach). Both linear regression and ANCOVA were performed on the whole population as well as after exclusion of 2639 subjects with self-reported diabetes or taking diabetes medication, HbA1c  5.7% or an estimated glomerular flirtation rate (eGFR)  60 mL/min. In addition, subgroup analyses were performed by repeating the linear regression analyses in lean (BMI < 25 kg/m2) and overweight/obese (BMI  25 kg/m2) subjects. To account for multiple testing, P-values were adjusted by controlling for a false discovery rate (FDR) at 5%, using the Benjamini–Hochberg procedure. All statistical analyses were performed using SAS statistical software, version 9.4 (SAS Institute Inc., Cary, NC, USA).

Table 1 General baseline characteristics of the study population. Whole population

Age, years Men, % Smoking, % Body mass index, kg/m2 SBP, mmHg DBP, mmHg Total cholesterol, mmol/L LDL cholesterol, mmol/L HDL cholesterol, mmol/L eGFR, mL/min Fasting glucose, mmol/L Insulin, mU/L HOMA-IR index HbA1c, % Diabetesa, %

All (n = 3986)

Men (n = 1970)

Women (n = 2016)

P

45 (40; 50) 49 28.0 25.5 (23.0; 28.2) 129 (120; 140) 80 (75; 90) 5.4 (4.7; 6.1) 3.4 (2.8; 4.0) 1.41 (1.17; 1.69) 94 (83; 105) 5.4 (5.1; 5.8) 3.3 (2.3; 4.9) 0.79 (0.54; 1.22) 5.8 (5.5; 6.1) 5.6

45 (40; 55) – 29.8 26.2 (24.1; 28.6) 130 (121; 140) 84 (79; 90) 5.5 (4.8; 6.2) 3.6 (3.0; 4.2) 1.26 (1.07; 1.51) 97 (87; 108) 5.6 (5.3; 6.0) 3.5 (2.3; 5.2) 0.87 (0.58; 1.34) 5.9 (5.6; 6.1) 7.6

45 (40; 50) – 26.2 24.4 (22.2; 27.7) 123 (115; 135) 80 (71; 86) 5.2 (4.6; 6.0) 3.2 (2.6; 3.8) 1.54 (1.30; 1.82) 90 (80; 101) 5.3 (5.0; 5.6) 3.1 (2.2; 4.6) 0.74 (0.51; 1.12) 5.7 (5.5; 6.0) 4.0

0.01 – 0.01 < 0.01 < 0.01 < 0.01 < 0.01 < 0.01 < 0.01 < 0.01 < 0.01 < 0.01 < 0.01 < 0.01 < 0.01

No diabetes, HbA1c < 5.7% or eGFR > 60 mL/min (n = 1347)

With diabetes, HbA1c  5.7% or eGFR  60 mL/min (n = 2639)

P

44 (40; 50) 41 17.9 24.7 (22.6; 27.4) 125 (115; 137) 80 (73; 88) 5.1 (4.5; 5.7) 3.1 (2.5; 3.7) 1.45 (1.20; 1.74) 94 (85; 105) 5.3 (5.0; 5.6) 3.1 (2.2; 4.6) 0.74 (0.50; 1.10) 5.4 (5.3; 5.6) –

50 (45; 55) 54 33.1 25.8 (23.3; 28.6) 130 (120; 140) 80 (75; 90) 5.6 (4.9; 6.3) 3.5 (2.9; 4.2) 1.38 (1.15; 1.67) 93 (83; 104) 5.5 (5.2; 5.9) 3.4 (2.3; 5.1) 0.82 (0.55; 0.51) 6.0 (5.8; 6.2) –

< 0.01 < 0.01 < 0.01 < 0.01 < 0.01 < 0.01 < 0.01 < 0.01 < 0.01 0.01 < 0.01 < 0.01 < 0.01 < 0.01 –

Continuous data are expressed as medians (quartile 1; quartile 3), nominal data as percentages; Wilcoxon rank-sum test for continuous data, x2-test for nominal data. eGFR: estimated glomerular filtration rate; SBP/DBP: systolic/diastolic blood pressure; LDL/HDL: low-density/high-density lipoprotein; HOMA-IR: homoeostasis model assessment of insulin resistance. a Self-reported or use of diabetes medication, HbA1c  6.5%.

Please cite this article in press as: Friedrich N, et al. Identification of urine metabolites associated with 5-year changes in biomarkers of glucose homoeostasis. Diabetes Metab (2017), http://dx.doi.org/10.1016/j.diabet.2017.05.007

G Model

DIABET-895; No. of Pages 8 N. Friedrich et al. / Diabetes & Metabolism xxx (2017) xxx–xxx

4

Results General characteristics General baseline characteristics of the whole study population by sex and study subpopulation are displayed in Table 1. Men, in general, showed a more adverse profile in relation to BMI, lipid levels, blood pressure and markers of glucose homoeostasis than did women. The same finding was observed for the healthy subpopulation compared with the rest of the participants (Table 1). For the whole study population (n = 3986), small decreases were observed in glucose (mean: 0.060 mmol/L [95% confidence interval (CI): 0.077; 0.042], signed-rank test: P < 0.01), HbA1c [ 0.022% ( 0.032; 0.013); P < 0.01] and HOMA-IR index [ 0.040 ( 0.064; 0.016); P < 0.01] over the 5-year follow-up period. In the healthy subpopulation (n = 1347), an increase in HbA1c [0.123% (95% CI: 0.110; 0.135); P < 0.01] was observed, whereas fasting glucose [ 0.100 mmol/L ( 0.120; 0.072); P < 0.01] and HOMAIR [ 0.082 ( 0.119; 0.046); P < 0.01] decreased over time. Association between urine metabolites and follow-up glycaemic parameters Baseline urinary levels of alanine, betaine, DMG, creatinine, trigonelline and trimethylamine (TMA) were the most noticeable metabolites associated with 5-year changes in HbA1c in both the whole study population and apparently healthy subpopulation (Fig. 1). Higher baseline levels of alanine, betaine, DMG, creatinine and TMA were associated with an increase in HbA1c from baseline to follow-up. In contrast, formic acid and trigonelline levels were associated with a decrease in HbA1c from baseline to follow-up, although formic acid was only significant after correction in the whole population. In addition, in the whole population, but not the healthy subpopulation, higher baseline urinary levels of beta-D-glucose and dimethylamine were associated with an increase in HbA1c. With respect to buckets and FOCUS-aligned peaks, several ppm ranges were observed that corresponded to the quantified

metabolites strongly associated with follow-up HbA1c (Figs. 2 and 3 and Fig. S3 [see supplementary material associated with this article online]). However, several unknown ppm regions were also found as shown for five selected regions in figure 3. Despite addressing queries to comprehensive databases [www.hmdb.ca (23161693) and Chenomx NMR Suite 8.1], no unambiguous or even a tentative assignment for these regions could be made. Analyses for 5-year changes in fasting glucose and HOMA-IR showed similar findings, with high baseline levels of lactic acid, beta-D-glucose, creatinine, alanine (only HOMA-IR) and 1-methylnicotinamide (MNA) associated with increases in both parameters (Fig. 1). Concerning spectral analyses, selected regions with respect to fasting glucose are shown in Fig. 3. BMI-related changes in associations between urine metabolites and HbA1c changes To assess the influence of overweight/obesity on these associations, linear regression analyses were repeated separately for lean (BMI < 25 kg/m2) and overweight/obese (BMI  25 kg/m2) subjects (Fig. 4; Fig. S3 [see supplementary materials associated with this article online]). The above-mentioned strong associations between baseline levels of alanine, creatinine, formic acid and trigonelline and the increases in HbA1c during follow-up were confirmed in both lean and overweight/obese subjects. However, with respect to DMG, the positive association remained significant only in lean subjects. Also, higher baseline glycine and acetic acid levels were related to an increase in HbA1c in lean, but not overweight/obese subjects, whereas hippuric acid was negatively associated only in overweight/obese individuals. With respect to changes in fasting glucose and HOMA-IR, the positive relations of MNA and beta-D-glucose with glucose and of MNA, lactic acid and creatinine with HOMA-IR were stable in both lean and overweight/obese subjects. In addition, differences in terms of ppm regions were found (Fig. 4; Fig. S4 [see supplementary materials associated with this article online]), particularly for HbA1c levels. Regions between 7.5 and 8.0 ppm were more strongly

Fig. 1. Associations of baseline metabolite levels with longitudinal changes in continuous markers of glucose homoeostasis during the 5-year follow-up in the whole study population (black squares) and healthy subpopulation (red squares; *no diabetes or use of diabetes medication, HbA1c < 5.7% and estimated glomerular filtration rate [eGFR] > 60 mL/min). The b coefficient and 95% CI for linear regression analyses adjusted for age, sex, body mass index (BMI), low-density lipoprotein (LDL) cholesterol and systolic blood pressure (SBP); changes from baseline glycated haemoglobin (HbA1c) are displayed. +: significant associations based on adjusted analyses of covariance; FDR: false discovery rate (at 5% for P, Benjamini–Hochberg procedure).

Please cite this article in press as: Friedrich N, et al. Identification of urine metabolites associated with 5-year changes in biomarkers of glucose homoeostasis. Diabetes Metab (2017), http://dx.doi.org/10.1016/j.diabet.2017.05.007

G Model

DIABET-895; No. of Pages 8 N. Friedrich et al. / Diabetes & Metabolism xxx (2017) xxx–xxx

5

Fig. 2. Corrected P-values for associations of baseline metabolite levels (right) or ppm (buckets, left) with longitudinal changes in continuous markers of glucose homoeostasis during the 5-year follow-up. The subpopulation had no diabetes or no diabetes medication use, HbA1c < 5.7% and eGFR > 60 mL/min. Linear regression models were adjusted for age, sex, BMI, LDL cholesterol, SBP and baseline HbA1c levels. FDR: false discovery rate; HOMA-IR: homoeostasis model assessment of insulin resistance.

related to HbA1c in lean than in obese subjects, whereas regions of around 2.0 ppm were more pronounced in obese subjects.

increases in fasting glucose and HOMA-IR values during the 5-year follow-up.

Discussion

Betaine and N,N-dimethylglycine (DMG)

The present study aimed to investigate the relationship between the urinary metabolic profile and longitudinal (5-year) changes in biomarkers of glucose homoeostasis. The most pronounced associations were found for HbA1c with high baseline levels of alanine, betaine, DMG, and TMA as well as low levels of formic acid and trigonelline being related to an increase in HbA1c levels over time. Importantly, no substantial differences in these associations were found between lean vs overweight/obese subjects. Lactic acid, alanine and MNA were associated with

Betaine acts as a methyl donor for homocysteine methylation catalyzed by betaine–homocysteine methyltransferase, with methionine and DMG being the products. Therefore, betaine represents a determinant of plasma homocysteine levels [19]. In the present study, higher urinary baseline levels of betaine as well as its metabolite DMG were linked to an increase in HbA1c over time. Our findings are in concordance with a study reporting positive associations with incident diabetes for both urine metabolites

Fig. 3. Unknown ppm regions associated with follow-up HbA1c or fasting glucose identified by linear regression analyses (Fig. 2): (upper) mean nuclear magnetic resonance spectra for selected ppm regions by quartiles of baseline-adjusted follow-up HbA1c (left) and fasting glucose (right); and assigned high-resolution 1H-NMR spectrum of a human urine sample (lower).

Please cite this article in press as: Friedrich N, et al. Identification of urine metabolites associated with 5-year changes in biomarkers of glucose homoeostasis. Diabetes Metab (2017), http://dx.doi.org/10.1016/j.diabet.2017.05.007

G Model

DIABET-895; No. of Pages 8 N. Friedrich et al. / Diabetes & Metabolism xxx (2017) xxx–xxx

6

Fig. 4. Corrected P-values for associations of baseline metabolite levels (right) or ppm (buckets; left) with longitudinal changes in continuous markers of glucose homoeostasis during the 5-year follow-up in lean (BMI < 25 kg/m2) and in overweight/obese (BMI  25 kg/m2) subjects. Linear regression models were adjusted for age, sex, BMI, LDL cholesterol, SBP and baseline HbA1c levels. FDR: false discovery rate; HOMA-IR: homoeostasis model assessment of insulin resistance.

[9]. Another study, however, found a negative relationship between plasma betaine levels and incident diabetes [20]. A possible explanation for this discrepancy might be the rather low correlation between plasma and urine levels of betaine, as shown in a study investigating betaine supplementation [21]. With respect to our present findings, previous investigations had already revealed altered betaine metabolism in diabetes patients, with significantly higher urinary betaine excretion in such patients [22,23] linked to hyperglycaemia and proximal tubular dysfunction [22] as well as plasma glucose or HbA1c levels [22]. The latter authors further showed that plasma glucose infusion did not lead to increased betaine excretion in female sheep and suggested that increased betaine excretion may not be directly caused by elevated plasma glucose levels or glycosuria in diabetes [24]. A more recent study [25] based on 2400 subjects confirmed the increased betaine excretion in diabetes patients, and also found a non-linear association between betaine excretion and HbA1c with a strong positive association for HbA1c levels > 6%. Interestingly, among diabetes patients, betaine excretion was inversely linked to plasma betaine levels [25], thereby providing a further explanation for the above-mentioned discrepant findings for plasma and urine betaine levels in cases of incident diabetes [9,20]. In concordance with our present results, betaine excretion was a significant and strong predictor of incident diabetes making a strong argument that urinary betaine might be a useful diagnostic tool for diabetes [25]. Possible mechanisms behind the positive association include tubular impairment, reflecting the role of betaine as a renal osmolyte [26]. An association of urinary betaine with excretion of retinol-binding protein, a known marker of proximal tubular dysfunction, is also strengthened this suggestion [22]. As a consequence, urinary betaine may indeed be an indicator of the adverse effects of altered glucose metabolism on renal function. Trigonelline This alkaloid is present in many plants, including fenugreek seeds, oats and potatoes, and represents a bioactive compound in coffee with a concentration of about 1% [27]. In the human body, trigonelline is not substantially metabolized and is therefore

mainly excreted in urine [27,28]. It has also been proposed as a urinary surrogate marker of coffee consumption [29]. The present study has shown that higher baseline urinary trigonelline levels are related to a decrease in HbA1c over time. This finding is in accordance with a previous experimental study showing that trigonelline administration leads to improved insulin resistance, increased glucokinase activity in rodents [30] and lower glucose levels in tolerance tests [31]. A beneficial effect of trigonelline in humans has also been found: oral trigonelline intakes resulted in acutely reduced early glucose and insulin responses during oral glucose tolerance tests (OGTTs) in overweight men, although no long-term effects were found [32]. Coffee is a complex mix of more than 1000 partly bioactive compounds, including caffeine and its metabolites as well as a large number of additional molecules, such as trigonelline, phenolic acids, chlorogenic acid, kahweol and cafestol [27]. Meta-analyses [33–35] have supported the beneficial effects of coffee consumption on diabetes risk. As recently reviewed [36,37], several plausible pathophysiological mechanisms might explain this association. A positive effect of coffee on glucose metabolism could be mediated via increased glucose absorption or storage, whereas a beneficial effect on lipid metabolism may be attributed to an increased rate of fatty acid oxidation as well as depressed lipogenesis. Trimethylamine (TMA) Dietary precursors of TMA include choline, betaine, phosphatidylcholine and L-carnitine, all of which are degraded to TMA by microbial enzymes, thereby explaining the powerful impact of intestinal microbiota composition on TMA levels [38]. TMA is further rapidly absorbed through the gut wall and oxidized to trimethylamine-N-oxide (TMAO) by hepatic flavin-containing monooxygenase 3 (FMO3) [39], demonstrating a close host–gut microbiome interaction [40]. Over the past decade, TMA or, more specifically, TMAO levels have been linked to several diseases, including cancer, renal or (cardio)metabolic diseases [39]. Our present study found a positive association between urinary baseline TMA levels and HbA1c changes over time. This finding is consistent with the reported higher TMAO levels in diabetic mice [41] and patients [42,43], as well as the positive relationship

Please cite this article in press as: Friedrich N, et al. Identification of urine metabolites associated with 5-year changes in biomarkers of glucose homoeostasis. Diabetes Metab (2017), http://dx.doi.org/10.1016/j.diabet.2017.05.007

G Model

DIABET-895; No. of Pages 8 N. Friedrich et al. / Diabetes & Metabolism xxx (2017) xxx–xxx

between TMAO and incident diabetes in women [9]. In an experimental study of mice fed a high-fat diet, dietary TMAO intake led to obstruction of the hepatic insulin-signalling pathway by affecting the expression of genes related to insulin signalling, gluconeogenesis, glycogen synthesis and glucose transport [44]. Consequently, increased fasting insulin levels and impaired glucose tolerance were also observed. Furthermore, FMO3 knockdown in mice resulted in decreased levels of TMAO, glucose and insulin as well as a reduced expression of genes involved in gluconeogenesis [45]. As a possible molecular mechanism, the authors suggested inhibition of peroxisome proliferator-activated receptor (PPAR)-a expression, a key transcription factor in hepatic energy metabolism. Given that TMA/TMAO levels are determined by dietary intakes of TMA precursors and the composition of intestinal microbiota, the observed association with diabetes might also reflect unhealthy nutritional behaviour. In men, increased excretion of TMA or TMAO was observed following oral intakes of choline and carnitine, with 60% and 30% of the dose, respectively, excreted as TMAO [46]. Large amounts of carnitine are found in red meat and in dairy products. Recently, the consumption of red meat has been extensively discussed as a risk factor for diabetes [47], with two large studies, including more than 7500 [48] and 12400 [49] incident cases, providing strong evidence that increased consumption of red meat is linked to a higher risk of the subsequent development of diabetes. Nevertheless, although several possible mechanisms behind the relationship between TMA and diabetes were proposed, the exact processes are still unknown and require further investigation. 1-methylnicotinamide (MNA) MNA is derived from nicotinamide (vitamin B3, niacin) by the transfer of a methyl group from S-adenosylmethionine. This enzymatic reaction is exclusively catabolized by nicotinamide Nmethyltransferase (NNMT). In the present study, higher baseline urinary MNA levels were related to increases in HbA1c, fasting glucose and HOMA-IR values during the 5-year follow-up. Higher NNMT expression was previously described in adipose tissue in diabetic mice [50] and in patients with either diabetes or insulin resistance [51]. Moreover, NNMT knockdown in mice protected against diet-induced obesity due to enhanced cellular energy expenditures [50]. Plasma MNA levels correlate with NNMT expression in adipose tissue in diabetes patients, but not in healthy subjects without diabetes [51]. Conclusively, higher urinary MNA levels were found in patients with diabetes [52], and a correlation between plasma MNA and degree of insulin resistance was seen in insulin-resistant subjects [51]. Moreover, elevated serum MNA levels were associated with an increased risk of overweight/ obesity as well as diabetes in a large population of 1160 Chinese subjects [53]. Taken together with our present results, these previous studies argue in favour of a protective effect of low MNA levels, most likely reflecting low NNMT activity in adipose tissue. Study strengths and limitations

7

observed in the considered markers of glucose homoeostasis over the 5-year follow-up period. Reasons for this finding include:  selection bias due to loss to follow-up;  positive effects of the intervention carried out in the Inter99 study;  or changes towards an overall healthier lifestyle in the general Danish population. With respect to selection bias due to dropouts, IPA weighting was applied to account for any possible attrition bias, and the results were only slightly altered compared with findings based on analyses without IPA weighting. This indicates that, in general, there is no evidence that such effects would bias the association between urine metabolites and changes in metabolic biomarkers. A further possible factor influencing our reported findings might be the applied repeated lifestyle intervention in the Inter99 study. However, previous investigations of the intervention effect on the development of ischaemic heart disease, diabetes and mortality within a 10-year follow-up period showed no effects [54,55]. Thus, the performed lifestyle intervention is not likely to have had a major impact on the present findings of an association between urinary metabolic profile and longitudinal changes in biomarkers of glucose homoeostasis. Conclusion and perspectives Several urine metabolites, including alanine, betaine, MNA, TMA and trigonelline, were associated with detrimental longitudinal changes in biomarkers of glucose homoeostasis. The identified metabolites point to mechanisms within betaine and coffee metabolism as well as the possible influence of the gut microbiome, and were independent of obesity. Such knowledge may provide clues to pathogenetic mechanisms and targets for interventions, and might even improve risk stratification based on a readily obtained biofluid during routine clinical examinations. Disclosure of interest The authors declare that they have no competing interest. Acknowledgements We would like to thank the participants in the Inter99 study, and all members of the Inter99 staff at the Research Centre for Prevention and Health and the Steering Committee. The Inter99 study was supported by the Danish Medical Research Council, Danish Centre for Evaluation and Health Technology Assessment, Novo Nordisk, Copenhagen County, Danish Heart Foundation, Danish Pharmaceutical Association, Augustinus Foundation, Ib Henriksen Foundation and Beckett Foundation. This project has also received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska–Curie Grant Agreement number 657595. Appendix A. Supplementary data

The present study represents one of the largest studies using NMR-based metabolomics to investigate glucose homoeostasis in a population-based longitudinal setting. However, although NMR spectroscopy provides highly reproducible results, it has lower sensitivity than MS. Several unknown compounds showed strong associations with the glucose parameters under investigation, but no unambiguous assignment was possible, thus representing a limitation of our present approach. However, further improvements in assignment algorithms might elucidate these signals in future. With respect to the investigated study population, a decrease was

Supplementary materials (Figs. S1–S4) associated with this article can be found at http://www.scincedirect.com at http://dx. doi.org/10.1016/j.diabet.2017.05.007. References [1] Friedrich N. Metabolomics in diabetes research. J Endocrinol 2012;215:29–42. [2] Gonzalez-Franquesa A, Burkart AM, Isganaitis E, Patti ME. What have metabolomics approaches taught us about type 2 diabetes? Curr Diab Rep 2016;16:74.

Please cite this article in press as: Friedrich N, et al. Identification of urine metabolites associated with 5-year changes in biomarkers of glucose homoeostasis. Diabetes Metab (2017), http://dx.doi.org/10.1016/j.diabet.2017.05.007

G Model

DIABET-895; No. of Pages 8 8

N. Friedrich et al. / Diabetes & Metabolism xxx (2017) xxx–xxx

[3] Suhre K. Metabolic profiling in diabetes. J Endocrinol 2014;221:R75–85. [4] Wang TJ, Larson MG, Vasan RS, Cheng S, Rhee EP, McCabe E, et al. Metabolite profiles and the risk of developing diabetes. Nat Med 2011;17:448–53. [5] Chen T, Ni Y, Ma X, Bao Y, Liu J, Huang F, et al. Branched-chain and aromatic amino acid profiles and diabetes risk in Chinese populations. Sci Rep 2016;6:20594. [6] Lu Y, Wang Y, Ong CN, Subramaniam T, Choi HW, Yuan JM, et al. Metabolic signatures and risk of type 2 diabetes in a Chinese population: an untargeted metabolomics study using both LC-MS and GC-MS. Diabetologia 2016;59:2349–59. [7] Yamakado M, Nagao K, Imaizumi A, Tani M, Toda A, Tanaka T, et al. Plasma free amino acid profiles predict four-year risk of developing diabetes, metabolic syndrome, dyslipidemia, and hypertension in Japanese population. Sci Rep 2015;5:11918. [8] Tillin T, Hughes AD, Wang Q, Wurtz P, Ala-Korpela M, Sattar N, et al. Diabetes risk and amino acid profiles: cross-sectional and prospective analyses of ethnicity, amino acids and diabetes in a South Asian and European cohort from the SABRE (Southall And Brent REvisited) Study. Diabetologia 2015;58:968–79. [9] Friedrich N, Budde K, Suhre K, Vo¨lker U, John U, Felix SB, et al. Sex differences in urine metabolites related with risk of diabetes using NMR spectroscopy: results of the study of health in pomerania. Metabolomics 2015;11:1405–15. [10] Floegel A, Stefan N, Yu Z, Muhlenbruch K, Drogan D, Joost HG, et al. Identification of serum metabolites associated with risk of type 2 diabetes using a targeted metabolomic approach. Diabetes 2013;62:639–48. [11] Wang-Sattler R, Yu Z, Herder C, Messias AC, Floegel A, He Y, et al. Novel biomarkers for pre-diabetes identified by metabolomics. Mol Syst Biol 2012;8:615. [12] Wang TJ, Ngo D, Psychogios N, Dejam A, Larson MG, Vasan RS, et al. 2Aminoadipic acid is a biomarker for diabetes risk. J Clin Invest 2013;123:4309–17. [13] Fall T, Salihovic S, Brandmaier S, Nowak C, Ganna A, Gustafsson S, et al. Nontargeted metabolomics combined with genetic analyses identifies bile acid synthesis and phospholipid metabolism as being associated with incident type 2 diabetes. Diabetologia 2016;59:2114–24. [14] Zhao J, Zhu Y, Hyun N, Zeng D, Uppal K, Tran VT, et al. Novel metabolic markers for the risk of diabetes development in American Indians. Diabetes Care 2015;38:220–7. [15] Jorgensen T, Borch-Johnsen K, Thomsen TF, Ibsen H, Glumer C, Pisinger C. A randomized non-pharmacological intervention study for prevention of ischaemic heart disease: baseline results Inter99. Eur J Cardiovasc Prev Rehabil 2003;10:377–86. [16] Dieterle F, Ross A, Schlotterbeck G, Senn H. Probabilistic quotient normalization as robust method to account for dilution of complex biological mixtures. Application in 1H NMR metabonomics. Anal Chem 2006;78:4281–90. [17] Alonso A, Rodriguez MA, Vinaixa M, Tortosa R, Correig X, Julia A, et al. Focus: a robust workflow for one-dimensional NMR spectral analysis. Anal Chem 2014;86:1160–9. [18] Hogan JW, Roy J, Korkontzelou C. Handling drop-out in longitudinal studies. Stat Med 2004;23:1455–97. [19] Ueland PM, Holm PI, Hustad S. Betaine: a key modulator of one-carbon metabolism and homocysteine status. Clin Chem Lab Med 2005;43:1069–75. [20] Walford GA, Ma Y, Clish C, Florez JC, Wang TJ, Gerszten RE, et al. Metabolite profiles of diabetes incidence and intervention response in the diabetes prevention program. Diabetes 2016;65:1424–33. [21] Atkinson W, Elmslie J, Lever M, Chambers ST, George PM. Dietary and supplementary betaine: acute effects on plasma betaine and homocysteine concentrations under standard and postmethionine load conditions in healthy male subjects. Am J Clin Nutr 2008;87:577–85. [22] Dellow WJ, Chambers ST, Lever M, Lunt H, Robson RA. Elevated glycine betaine excretion in diabetes mellitus patients is associated with proximal tubular dysfunction and hyperglycemia. Diabetes Res Clin Pract 1999;43:91–9. [23] Lever M, Sizeland PC, Bason LM, Hayman CM, Robson RA, Chambers ST. Abnormal glycine betaine content of the blood and urine of diabetic and renal patients. Clin Chim Acta 1994;230:69–79. [24] Dellow WJ, Chambers ST, Barrell GK, Lever M, Robson RA. Glycine betaine excretion is not directly linked to plasma glucose concentrations in hyperglycaemia. Diabetes Res Clin Pract 2001;52:165–9. [25] Schartum-Hansen H, Ueland PM, Pedersen ER, Meyer K, Ebbing M, Bleie O, et al. Assessment of urinary betaine as a marker of diabetes mellitus in cardiovascular patients. PLoS One 2013;8:e69454. [26] Kempson SA, Montrose MH. Osmotic regulation of renal betaine transport: transcription and beyond. Pflugers Arch 2004;449:227–34. [27] Ludwig IA, Clifford MN, Lean ME, Ashihara H, Crozier A. Coffee: biochemistry and potential impact on health. Food Funct 2014;5:1695–717. [28] Lang R, Wahl A, Skurk T, Yagar EF, Schmiech L, Eggers R, et al. Development of a hydrophilic liquid interaction chromatography-high-performance liquid chromatography-tandem mass spectrometry based stable isotope dilution analysis and pharmacokinetic studies on bioactive pyridines in human plasma and urine after coffee consumption. Anal Chem 2010;82:1486–97.

[29] Lang R, Wahl A, Stark T, Hofmann T. Urinary N-methylpyridinium and trigonelline as candidate dietary biomarkers of coffee consumption. Mol Nutr Food Res 2011;55:1613–23. [30] Yoshinari O, Sato H, Igarashi K. Anti-diabetic effects of pumpkin and its components, trigonelline and nicotinic acid, on Goto-Kakizaki rats. Biosci Biotechnol Biochem 2009;73:1033–41. [31] Yoshinari O, Igarashi K. Anti-diabetic effect of trigonelline and nicotinic acid, on KK-A(y) mice. Curr Med Chem 2010;17:2196–202. [32] van Dijk AE, Olthof MR, Meeuse JC, Seebus E, Heine RJ, van Dam RM. Acute effects of decaffeinated coffee and the major coffee components chlorogenic acid and trigonelline on glucose tolerance. Diabetes Care 2009;32:1023–5. [33] Jiang X, Zhang D, Jiang W. Coffee and caffeine intake and incidence of type 2 diabetes mellitus: a meta-analysis of prospective studies. Eur J Nutr 2014;53:25–38. [34] Ding M, Bhupathiraju SN, Chen M, van Dam RM, Hu FB. Caffeinated and decaffeinated coffee consumption and risk of type 2 diabetes: a systematic review and a dose-response meta-analysis. Diabetes Care 2014;37:569–86. [35] Merlotti C, Morabito A, Pontiroli AE. Prevention of type 2 diabetes; a systematic review and meta-analysis of different intervention strategies. Diabetes Obes Metab 2014;16:719–27. [36] Natella F, Scaccini C. Role of coffee in modulation of diabetes risk. Nutr Rev 2012;70:207–17. [37] Santos RM, Lima DR. Coffee consumption, obesity and type 2 diabetes: a minireview. Eur J Nutr 2016;55:1345–58. [38] Chhibber-Goel J, Gaur A, Singhal V, Parakh N, Bhargava B, Sharma A. The complex metabolism of trimethylamine in humans: endogenous and exogenous sources. Expert Rev Mol Med 2016;18:e8. [39] Fennema D, Phillips IR, Shephard EA. Trimethylamine and trimethylamine Noxide, a flavin-containing monooxygenase 3 (FMO3)-mediated host-microbiome metabolic axis implicated in health and disease. Drug Metab Dispos 2016;44:1839–50. [40] Nicholson JK, Holmes E, Kinross J, Burcelin R, Gibson G, Jia W, et al. Host-gut microbiota metabolic interactions. Science 2012;336:1262–7. [41] Dambrova M, Latkovskis G, Kuka J, Strele I, Konrade I, Grinberga S, et al. Diabetes is associated with higher trimethylamine N-oxide plasma levels. Exp Clin Endocrinol Diabetes 2016;124:251–6. [42] Messana I, Forni F, Ferrari F, Rossi C, Giardina B, Zuppi C. Proton nuclear magnetic resonance spectral profiles of urine in type II diabetic patients. Clin Chem 1998;44:1529–34. [43] Tang WH, Wang Z, Wu Y, Fan Y, Koeth RA, Hazen SL. Gut flora metabolite trimethylamine N-oxide predicts incident cardiovascular risks in both stable non-diabetics and diabetic subjects. J Am Coll Cardiol 2013;61:E1398. [44] Gao X, Liu X, Xu J, Xue C, Xue Y, Wang Y. Dietary trimethylamine N-oxide exacerbates impaired glucose tolerance in mice fed a high fat diet. J Biosci Bioeng 2014;118:476–81. [45] Shih DM, Wang Z, Lee R, Meng Y, Che N, Charugundla S, et al. Flavin containing monooxygenase 3 exerts broad effects on glucose and lipid metabolism and atherosclerosis. J Lipid Res 2015;56:22–37. [46] Zhang AQ, Mitchell SC, Smith RL. Dietary precursors of trimethylamine in man: a pilot study. Food Chem Toxicol 1999;37:515–20. [47] Battaglia Richi E, Baumer B, Conrad B, Darioli R, Schmid A, Keller U. Health risks associated with meat consumption: a review of epidemiological studies. Int J Vitam Nutr Res 2015;85:70–8. [48] Pan A, Sun Q, Bernstein AM, Manson JE, Willett WC, Hu FB. Changes in red meat consumption and subsequent risk of type 2 diabetes mellitus: three cohorts of US men and women. JAMA Intern Med 2013;173:1328–35. [49] InterAct C, Bendinelli B, Palli D, Masala G, Sharp SJ, Schulze MB, et al. Association between dietary meat consumption and incident type 2 diabetes: the EPIC-InterAct study. Diabetologia 2013;56:47–59. [50] Kraus D, Yang Q, Kong D, Banks AS, Zhang L, Rodgers JT, et al. Nicotinamide Nmethyltransferase knockdown protects against diet-induced obesity. Nature 2014;508:258–62. [51] Kannt A, Pfenninger A, Teichert L, Tonjes A, Dietrich A, Schon MR, et al. Association of nicotinamide-N-methyltransferase mRNA expression in human adipose tissue and the plasma concentration of its product, 1-methylnicotinamide, with insulin resistance. Diabetologia 2015;58:799–808. [52] Salek RM, Maguire ML, Bentley E, Rubtsov DV, Hough T, Cheeseman M, et al. A metabolomic comparison of urinary changes in type 2 diabetes in mouse, rat, and human. Physiol Genomics 2007;29:99–108. [53] Liu M, Li L, Chu J, Zhu B, Zhang Q, Yin X, et al. Serum N(1)-methylnicotinamide is associated with obesity and diabetes in Chinese. J Clin Endocrinol Metab 2015;100:3112–7. [54] Jorgensen T, Jacobsen RK, Toft U, Aadahl M, Glumer C, Pisinger C. Effect of screening and lifestyle counselling on incidence of ischaemic heart disease in general population: Inter99 randomised trial. BMJ 2014;348:g3617. [55] Lau CJ, Pisinger C, Husemoen LL, Jacobsen RK, Linneberg A, Jorgensen T, et al. Effect of general health screening and lifestyle counselling on incidence of diabetes in general population: Inter99 randomised trial. Prev Med 2016;91:172–9.

Please cite this article in press as: Friedrich N, et al. Identification of urine metabolites associated with 5-year changes in biomarkers of glucose homoeostasis. Diabetes Metab (2017), http://dx.doi.org/10.1016/j.diabet.2017.05.007