Metabolic profiling of metformin treatment for low-level Pb-induced nephrotoxicity in rat urine

Chronic kidney disease is a worldwide problem, and Pb contamination is a potential risk factor. Since current biomarkers are not sensitive for the diagnosis of Pb-induced nephrotoxicity, novel biomarkers are needed. Metformin has both hypoglycaemic effects and reno-protection ability. However, its mechanism of action is unknown. We aimed to discover the early biomarkers for the diagnosis of low-level Pb-induced nephrotoxicity and understand the mechanism of reno-protection of metformin. Male Wistar rats were randomly divided into control, Pb, Pb + ML, Pb + MH and MH groups. Pb (250 ppm) was given daily via drinking water. Metformin (50 or 100 mg/kg/d) was orally administered. Urine was analysed by nuclear magnetic resonance (NMR)-based metabolomics coupled with multivariate statistical analysis, and potential biomarkers were subsequently quantified. The results showed that Pb-induced nephrotoxicity was closely correlated with the elevation of 5-aminolevulinic acid, d-lactate and guanidinoacetic acid in urine. After co-treatment with metformin, 5-aminolevulinic acid and d-lactate were decreased. This is the first demonstration that urinary 5-aminolevulinic acid, d-lactate and guanidinoacetic acid could be early biomarkers of low-level Pb-induced nephrotoxicity in rats. The reno-protection of metformin might be attributable to the reduction of d-lactate excretion.

Clinical chemistry. The Pb group had marginally increased serum creatinine (p = 0.052) and decreased urinary creatinine (p = 0.05) compared with the control (Fig. 3A). The Pb + ML group had significantly increased serum creatinine (p = 0.008) compared with the control group (Fig. 3A). The Pb + MH group had significantly increased serum creatinine (p = 0.003), and decreased BUN compared with the control (p = 0.001) and the Pb group (p < 0.000), as well as decreased urinary creatinine (p = 0.011) compared with the control group (Fig. 3A). In the MH group, urinary creatinine was significantly decreased compared with the control group (p = 0.002) (Fig. 3A). There were no significant differences in urinary protein, body weight or water consumption between groups ( Fig. 3A-C). The serum AST/ALT ratio was not significantly different between groups ( Supplementary  Fig. S1).

Figure 1.
Study flow chart. First, the animal experiment was carried out, and urine was collected followed by NMR-based metabolomics. Second, NMR spectra were integrated into peak lists followed by multivariate analysis. Third, altered peaks were identified through MetaboHunter or MetaboMiner. Fourth, potential biomarkers were quantified through the HPLC or the LC/MS/MS method. Fifth, the hypothesis was showed according to our results.  Supplementary Table S1). NMR spectra were integrated into peak lists followed by multivariate analysis. PCA analysis showed that there were no outliers. All samples were within the 0.95 Hotelling's T2 ellipse in scores plot (Fig. 5A), and there was no obvious contribution in the loading plot (Fig. 5C). PLS-DA was further applied to analyse peaks according to classification. In PLS-DA analysis, there was good separation between the 5 groups (Fig. 5B), and there was some effective contribution in the loading plot (Fig. 5D). Leave-one-out cross validation (LOOCV) and the permutation test were used to validate the PLS-DA model. Since the difference between R 2 and Q 2 was greater than 0.3 in the LOOCV method (Supplementary Table S2), PLS-DA was successfully validated by the permutation test (p = 0.044). All significantly altered peaks were identified and are detailed in Supplementary Table S3. As a result, sixty-eight metabolites were identified, and fifty-four metabolites among them had a variable importance in projection (VIP) score greater than 1 (Supplementary Table S3).
To understand which pathways were involved in these metabolite changes, all significantly altered metabolites were submitted to pathway analysis. The crucial pathways were arginine and proline metabolism, d-glutamine and d-glutamate metabolism, and alanine, aspartate and glutamate metabolism ( Supplementary  Fig. S2). Based on our knowledge and interests, we chose seven potential biomarkers for further quantification: N-acetyl-β-d-glucosaminidase (NAG), 5-aminolevulinic acid (ALA), d-(d-LA) and l-lactate (l-LA), guanidinoacetic acid (GAA), indoleacetic acid (IAA), and hippuric acid (HA) in rat urine.
Quantification of potential biomarkers. Quantitative data on these metaolites are shown in Table 1.
Value of urinary d-LA and l-LA were well-established in our previous studies 13, [27][28][29] . We validated ALA, GAA, (F) Sum of the tubulointerstitial histological score of Pb-induced nephrotoxicity rats that were co-treated with metformin or not. Scale bar, 50 μm. (★) means degeneration of tubular cell; (→) means inflammatory cell infiltration. **p < 0.01 compared with control group using one-way ANOVA; ## p < 0.01 compared with the Pb group using one-way ANOVA. In the control and MH groups, there was no obvious damage. In Pb-treated rats ( Fig. 1B-D), the renal tissues displayed tubular degeneration and cell infiltration without interstitial fibrosis. Compared to the Pb + MH group, low-dose metformin (Pb + ML) showed more reno-protective ability.

Discussion
To our knowledge, this is the first investigation of the effect of metformin on low-level Pb-induced renal injury in rats by urinary metabolomics. The routes of environmental exposure to Pb are inhalation, ingestion and dermal contact. After absorption, Pb particularly accumulates in the kidney and liver 30 . To inspect the lesion of Pb toxicity, both kidney and liver tissues were collected and evaluated by histopathology and clinical chemistry. Serum AST and ALT are not specific for the diagnosis of liver injury, and the AST/ALT ratio is more clinically useful 31 . Our results showed that 4-week oral administration of 250 ppm Pb did not influence liver tissues, as there was no elevation of the AST/ALT ratio and no liver damage. Although there was no obvious change in serum creatinine or BUN, renal tissues were certainly damaged by Pb (Fig. 2). Therefore, the following discussion focuses on Pb-induced nephrotoxicity.
Our study showed that Pb-induced renal injuries were related to nuclear inclusion bodies, proximal tubular apoptosis and inflammatory cell infiltration without glomerular damage. Our findings are consisted with other reports, which have shown that Pb binds to proteins and forms insoluble intranuclear inclusion bodies in proximal tubules 32 , and particularly damages proximal tubules 33 . In our clinical chemistry analysis, Pb did not alter serum creatinine or BUN compared with control. Even if serum creatinine and BUN are regarded as renal injury markers, these markers fail to increase in Pb-induced renal injury 11,12 . When these tests are found abnormal, the nephropathy has already reached the irreversible phase that may lead to renal insufficiency 14,34 . On the other Pb-induced nephrotoxicity rats. *p < 0.05, **p < 0.01 compared with control group using one-way ANOVA; # p < 0.05, ## p < 0.01 compared with Pb group using one-way ANOVA. In Pb-exposed rats, serum creatinine was marginally increased compared with control rats. After co-treatment with metformin, serum creatinine was increased in both Pb + ML and Pb + MH rats compared with control rats. BUN was reduced in Pb + MH rats compared with control and Pb rats. Pb-exposed rats had marginally decreased urinary creatinine compared with control rats. After co-treatment with metformin, urinary creatinine was significantly decreased in Pb + MH and MH rats compared with control rats. There were no significant differences in urinary protein, body weight or water consumption.
hand, CKD increases in prevalence year by year across the world. Pb contamination is one of the causes of CKD, and it is not easy to diagnose 3 . Thus, the discovery of sensitive biomarkers for the early diagnosis of Pb-induced nephrotoxicity is urgent. After co-treatment with metformin, the alteration of creatinine in both serum and urine might be attributed to renal acidification 13 . Moreover, the reduction of BUN in the Pb + MH group might have (D) 5-7 ppm (32× magnification) and (E) 7-9.5 ppm (16× magnification). Arrows point to peaks that were significantly altered. There were 21 NMR peaks that were significantly altered between the 5 groups in 1D-NMR spectra according to one-way ANOVA.
SCientifiC RepoRts | (2018) 8:14587 | DOI:10.1038/s41598-018-32501-3 been secondary to suppression of urea biosynthesis owing to inhibition of gluconeogenesis by high-dose metformin 35 . These data are consistent with our previous findings that reno-protection of metformin is more predominant in a low dose than a high dose 18,19 .
NMR-based metabolomics is a powerful tool for systematic investigation of potential biomarkers in biofluids 24 . Urine, a waste and filtrate from kidney, is non-invasive and directly reflects renal status 21,36 . Therefore, urinary metabolomics coupled with multivariate analysis provide a powerful approach to investigate metformin's effect on Pb-induced renal injury. MetaboAnalyst, a comprehensive tool for metabolomics analysis and interpretation, provides an effective approach to discover the potential biomarkers of disease statuses 37 . In PCA analysis, there were no observations lying outside the 0.95 Hotelling's T2 ellipse (Fig. 5A). The loading plot of PCA showed no significant contribution (Fig. 5C). Therefore, PLS-DA was used to discriminate according to classification. Since PLS-DA analysis can have a 'data overfitting' problem, LOOCV and the permutation test was used to validate our PLS-DA model 38 . In LOOCV, the difference between R 2 and Q 2 values was used as the indicator to determine if PLS-DA was overfitted. This difference should be less than 0.3 39 . In our study, the difference between R 2 and Q 2 was greater than 0.3 (Supplementary Table S2). Therefore, the permuation test was used to validate the PLS-DA model. The permutation test was performed in a 2,000-iteration re-analysis, and only 88 of those results were significantly better than the original dataset (p = 88/2000 = 0.044 < 0.05). Taking these data together, PLS-DA was validated and suitable for estimating potential biomarkers. According to the score plot of PLS-DA, the discrimination between the 5 groups was more effective in metformin-treatment than in Pb (A-E) represents control, Pb, Pb + ML, Pb + MH and MH groups, respectively. In the PCA score plot, there was no obvious discrimination, and no obvious features were seen in the PCA loading plot. Therefore, PLS-DA was used to separate metabolites according to classification. The PLS-DA score plot shows good separation, and some features were obviously discriminated in the loading plot.
NAG is a lysosomal enzyme highly expressed in kidney. Though elevation of urinary NAG is positively correlated with renal injury 28 , it might fail to change in Pb-exposed workers, which is consistent with our study 40 . Altogether, urinary NAG might not be suitable for diagnosis of early low-level Pb-induced renal injury.
ALA, a precursor of haem synthesis, accumulates due to inhibition of ALA dehydratase (ALA-D) by Pb and is subsequently eliminated into urine 41 . The elevation of urinary ALA supports other studies that showed that it is highly increased after exposure to Pb 42 . However, some still hold the opposite opinion that urinary ALA is not a suitable indicator of exposure to Pb in human 43 . There is no direct correlation between renal morphology and urinary ALA level. To our knowledge, this is the first study that simultaneously determined urinary ALA and renal morphology in Pb-exposed rats. Our study showed that low-level Pb resulted in a 40-fold increase in urinary ALA compared with control rats. After co-treatment with a high dose of metformin, urinary ALA was profoundly decreased compared with Pb rats (Table 1). Even though metformin had no influence on ALAD activity in vitro 44 , the effect of metformin on ALAD activity remains unclear in vivo. On the other hand, metformin enhances ALA-related photodynamic therapy by activating AMPK autophagy signalling in vitro 45 ; hence, the relationship between metformin and ALA still needs to be elucidated. According to our results, we regard urinary ALA as an early biomarker for diagnosis of low-level Pb-induced renal injury.
Lactate (LA), containing a chiral centre, has two enantiomers, d-and l-LA. d-LA, a downstream metabolite of methylglyoxal, is eliminated from kidney 46 . Due to the high reactivity and deleterious characteristics of methylglyoxal, accumulation of methylglyoxal and its metabolite, d-LA, is associated with renal injury 13,18,19,47 . l-LA is formed from pyruvate through l-lactate dehydrogenase 48 . In this study, urinary d-LA, not l-LA, was increased in Pb rats compared with control rats (Table 1). Similarly, our previous studies demonstrated that urinary d-LA is a powerful biomarker of renal injury in rodents 13,27,28 and human 29 . After co-treatment with a low dose of metformin, urinary d-LA was reduced compared with Pb rats. In this regard, our previous study demonstrated that metformin could reduce renal methylglyoxal and urinary d-LA in Pb-induced renal injury rats 13 . Again, urinary d-LA was shown to be a promising biomarker for assessing renal injury. However, high-dose metformin did not attenuate Pb-induced renal injury, and it produced higher urinary d-and l-LA compared with control rats. High urinary l-LA might be attributed to inhibition of gluconeogenesis by high-dose metformin 49 . On the other hand, d-LA can be produced by intestinal Lactobacillus species 50 . Administration of metformin could change gut microbiota that enriches Lactobacillus in high-fat diet-induced obesity in rats 51 . We speculate that high urinary d-LA in high-dose metformin might be secondary to the alteration of gut microbiota. Taking these findings together, we suggest that low-dose metformin is more suitable for recovering renal injury in Pb-induced nephrotoxicity than high-dose.
GAA is an essential precursor of creatine and is synthesized by transamidation from glycine and arginine in the proximal tubules 52 . In nondialysed patients with chronic renal sufficiency, the reduction of creatinine clearance is concomitant with a decrease of GAA excretion 53 . The suppression of GAA excretion might be attributable to an inhibition of amidinotransferase or an impairment of renal production 54,55 . However, there is evidence that urinary GAA is initially increased and subsequently decreased in acute renal failure in rats 55 . Kuwagaki et al. demonstrated that the elevation of urinary GAA at day 1 is due to proximal tubular damage, which would cause less effective reabsorption of GAA. Then, inhibition of renal amidinotransferase activity leads to decreased urinary GAA 55 . Furthermore, urinary GAA is profoundly increased in traditional herbal medicine-induced nephrotoxicity 56 . Elevation of GAA excretion might have been secondary to disturbance of renal reabsorption in our study. This is the first study demonstrating that urinary GAA could be an early biomarker for low-level Pb-induced renal injury.   Table 1. Metformin's effects on potential biomarkers of Pb-induced nephrotoxicity in rat urine. N-acetyl-βd-glucosaminidase (NAG); 5-aminolevulinic acid (ALA); d-lactate (d-LA); l-lactate (l-LA); guanidinoacetic acid (GAA); indole-3-acetic acid (IAA); hippuric acid (HA) *p < 0.05, **p < 0.01 compared with control group using one-way ANOVA; # p < 0.05 compared with Pb group using one-way ANOVA; $ p < 0.05, $$ p < 0.01 compared with control group using two-tailed Student's t-test. @ p < 0.05 compared with Pb group using twotailed Student's t-test. According to quantitative analysis, urinary NAG, IAA and HA were not significantly different between the 5 groups, except for urinary IAA in the MH group. In Pb-exposed rats, urinary ALA, d-LA and GAA were profoundly increased compared with control rats. After co-treatment with metformin, urinary ALA and d-LA were significantly decreased in Pb + MH and Pb + ML rats compared with Pb-exposed rats, respectively. However, urinary GAA was not altered after co-treatment with metformin. In high-dose metformin (MH group), urinary d-and l-LA showed elevation compared with control rats. Recently, gut microbiota have received attention concerning renal injury 57 . Subchronic exposure to Pb interferes with intestinal microbiota such as Ruminococcaceae, Akkermansia and Lachnoclostridium 58,59 . In addition, metformin could significantly increase the abundances of gut microbiota, Akkermansia muciniphila and Clostridium cocleatum, in high-fat-diet induced obesity mice 60 . Hippuric acid or indoleacetic acid is the product of benzoate or tryptophan metabolism by a range of gut microbiota, for example Clostridium sp., Bacteroides, Clostridia, and Escherichia coli [61][62][63][64] . No reports revealed that urinary HA and IAA were related to Pb-induced nephrotoxicity with metformin treatment. However, both urinary HA and IAA were not significantly altered in low-level Pb exposure in this study (Table 1). Glomerular filtration contributes to HA excretion 65 . Therefore, the lack of alteration of urinary HA and IAA might be secondary to mild renal injury and might not influence the microbiota as a result of low-level Pb 58 . In the MH group, the elevation of IAA might attribute to the effect of metformin on gut microbiota mentioned above.
In human population, Lalau et al. investigated that patients were administered metformin 850 or 1,700 mg daily according to CKD stage 2 to 3 or 1 to 2, respectively. Authors concluded that metformin can be efficiently cleared in mild-to-moderate CKD patients 66 .
In a population open cohort study showed that the risk of severe complications of diabetes was significantly decreased in metformin user compared to non-user 67 . A retrospective cohort study investigated the effect of initiators (metformin or sulfonylurea) on veterans with diabetes. The result showed that metformin treatment had a lower risk of decline in kidney function or death 68

. Inzucchi et al.
proposed that a maximal daily dose of metformin in patients with CKD stage 1 to 3b are ranged from 2250 to 1000 mg daily 69 . In our study, 50 and 100 mg/kg of metformin in rats was equal to 606 and 1211 mg/day in human dosage, according to practice guide for dose conversion 70 . In present study, the low-dose metformin (50 mg/kg) provide the better renoprotective effect than the high-dose one. Based on our results, we speculated that administration of 600 mg metformin might possibly attenuate low-level Pb-induced renal injury in human population, especially in the early stage of renal injury without hyperglycemia.
Altogether, the mechanism of low-level Pb-induced nephrotoxicity and reno-protection ability of metformin is described in Fig. 6. To our knowledge, this is the first study demonstrating that low-level Pb-induced renal injury is simultaneously related to excretion of d-LA, ALA and GAA. Although no new mechanisms of metformin were found, its reno-protection seemed to occur through reduced d-LA excretion through the methylglyoxal pathway. However, the possible mechanisms that metformin protects the kidney in rats might involve in activating AMP-activated protein kinase and antioxidant activities and suppressing interstitial fibrosis and inflammation responses [71][72][73] . But the assumption of renoprotective effects of metformin on Pb-induced nephrotoxicity still need to be confirmed by further studies. Additionally, high-dose metformin was more influential in metabolic profiling than low-dose, for example, lactate and IAA. These perturbances of metabolites in rats treated with high-dose metformin suggest that adequate doses of metformin have clinical utility.

Conclusion
Combining metabolomics with quantitative analysis has promise as non-invasive tool to discover potential biomarkers. Our results suggest that the elevation of urinary d-LA, ALA and GAA can be biomarkers for early diagnosis of low-level Pb-induced nephrotoxicity. The reno-protective effect of metformin on Pb-induced nephrotoxicity might be attribuable to the reduction of d-LA excretion. Further studies should clarify the relationships between biomarkers and disease stage. The effects of different doses of metformin on metabolic pathways are worth studying. Animal experiment, sample preparation and histopathology. Eight-week-old male Wistar rats (n = 30) were purchased from BioLASCO Taiwan Co., Ltd. (Taipei, Taiwan) and housed in the laboratory animal centre (Taipei Medical University). The animal experiment was conducted according to previous studies 13, 33 . After 1 week of acclimation, rats were randomly divided into five groups: control (C), Pb only (Pb), Pb plus low dose of metformin (50 mg/kg/d) (Pb + ML), Pb plus high dose of metformin (100 mg/kg/d) (Pb + MH), and high dose of metformin only (MH). Pb (250 ppm) was given through drinking water, and metformin was given orally for 28 consecutive days. All animal experiments had been reviewed and approved by the Institutional Animal Care and Use Committee or Panel (Permit Number: LAC-2014-0286) to minimize pain and discomfort.

Chemicals and reagents.
At the end of the experiment, twenty-four-hour urine was collected and centrifuged at 100 × g at 4 °C for 1 min. The supernatants were collected and stored at −80 °C. After that, rats were anaesthetized. Blood was collected and placed at room temperature at least 30 minutes for clotting, followed by centrifuging at 1200 × g for 10 minutes at 4 °C. Subsequently, sera were collected and stored at −80 °C.
Kidney and liver tissues were collected, immersed in 10% neutralized formalin and embedded in paraffin, followed by sectioning at 3 μm thickness. Haematoxylin and eosin (H&E) staining was used for investigating renal and hepatic morphology and was analysed in accordance with previous studies 13,74 . Clinical chemistry. Serum blood urea nitrogen (BUN), serum creatinine, alanine aminotransferase (ALT), aspartate aminotransferase (AST) were analysed by the Laboratory Animal Center of Taipei Medical University using VetTest TM (IDEXX Laboratories, Inc., Westbrook, Maine, United States).
Determination of urinary creatinine was described in previous studies 13,27,28 . In brief, 20 μL of 2 mM cimetidine (as internal standard), 100 μL of urine and 400 μL of ACN were mixed together by vortexing for 1 minute, following by centrifuging at 875 × g for 15 minutes at 4 °C. The supernatants were collected and 10 μL was introduced into an HPLC-UV system at 234 nm wavelength. TSKgel ODS-80Ts QA (4.6 × 250 mm I.D., 5 μm) (TOSOH, Tokyo, Japan) was used. The mobile phase consisted of an aqueous phase (30 mM sodium lauryl sulfate and 100 mM potassium dihydrogen phosphate (adjusted to pH 3.0 using 85% o-phosphoric acid)) and an organic phase (ACN) (ratio of aqueous and organic phase = 12/7, v/v). The flow rate was set at 0.7 mL/min. Urinary proteins were determined by the Bradford method 75 . In brief, 240 μL of urine was added to 60 μL of reagent dye and mixed well. Two hundred microliters of the mixture was transferred to a 96-well plate, and its absorbance was measured at 590 nm with a plate reader.
NMR data processing. NMR spectra were processed using Bruker TOPSPIN v3.2 (Bruker Biospin, Germany). The spectra were subjected to 1 Hz exponential broadening and zero-filled to 64 k points prior to Fourier transformation. Next, baseline and phase were manually corrected with the chemical shift referenced to TSP (δ 0.00 ppm). Due to an imperfect water signal and urea resonance, the chemical shifts between δ4.38 and 6.30 ppm were eliminated.
Metabolite identification and pathway analysis. On the basis of multivariate analysis, the chemical shifts of 1D-NMR and 2D-NMR spectra were integrated into peak lists for the following assay. The lists of 1D-NMR were introduced into MetaboHunter for identification 77 . The shift tolerance was set at ±0.03 ppm. The lists of 2D-NMR spectra were introduced into MetaboMiner for identification 78 . The shift tolerance was set at ±0.03 ppm (for 1 H) and ±0.1 ppm (for 13 C). Metabolites were identified with the Human Metabolome Database and Madison Metabolomics Consortium Database. The metabolites that existed in urine were enrolled in our study for further analysis.
All significant metabolites were enrolled, and pathway analysis was carried out using Metaboanalyst 3.5. After uploading the identified compounds, those pathways with impact value >0.2 and p value < 0.05 were regarded as the most important pathways.
Quantitative approach for estimating the potential biomarkers in urine . N-Acetyl-β-d-glucosaminidase (NAG). Urinary NAG was measured as described in previous study 28

5-Aminolevulinic acid (ALA).
Urinary ALA was determined according to a previously published method 79 . In brief, 10 μL of urine with 90 μL of 10% formaldehyde, and 700 μL of reagent (contained acetylacetone, ethanol and ddH 2 O = 1.5: 1: 7.5, v/v/v) were mixed and placed into boiling water for 15 minutes, then cooled down on ice. Then, 20 μL of ALA-derivatives were separated on a TSK-GEL ODS-Ts (4.6 × 150 mm I.D., 5 μm) with HPLC with a fluorescence detector at 370 nm excitation and 460 nm emission wavelength. The column oven and flow rate were set at 40 °C and 0.7 mL/min, respectively. The mobile phase was composed of methanol:water:acetic acid (50:50:1, v/v/v). d-and l-Lactate (d-and l-LA). d-LA and l-LA were measured in our laboratory using a well-established column-switching system 13,[27][28][29]80 . In brief, urine (20 μL) and internal standard (10 μL, 1 mM propionic acid) were mixed with ACN (170 μL), and the mixture was centrifuged at 1000 × g for 10 minutes at 4 °C. The supernatants (100 μL) were derivatized with 100 μL of 8 mM NBD-PZ, and 50 μL of 280 mM DPDS and TPP at 30 °C for 3 hours. After that, 0.1% trifluoroacetic acid (aq) (100 μL) was added to terminate the reaction. The resultant solution was eluted through the MonoSpin TM SCX cartridge (GL Science Inc., Tokyo, Japan) eliminate excesses derivatization reagents 81 . The eluent (20 μL) was injected into the column-switching system to quantify d-and l-lactate in urine. The 1 st dimension of total lactate was separated on an ODS column (250 mm × 4.6 mm ID; 5 μm particle size; Biosil Chemical Co. Ltd., Taipei, Taiwan). The mobile phase was composed of 12% ACN, 20% MeOH and 68% ddH 2 O (v/v/v). The flow rate was set at 0.7 mL/min, and the oven was set at 30 °C. The 2 nd dimension of chiral separation was performed on a Chiralpak AD-RH column (150 mm × 4.6 mm ID; 5 μm particle size; Daicel Co., Osaka, Japan) at room temperature. The mobile phase was composed of 60% ACN and 40% ddH 2 O (v/v) and the flow rate was 0.3 mL/min. The 1 st and 2 nd dimensions were detected at an excitation and emission wavelength of 491 and 547 nm, respectively.
Guanidinoacetic acid (GAA), hippuric acid (HA) and indoleacetic acid (IAA). Urinary GAA, HA and IAA were quantified with an Agilent 6470 triple quadrupole coupled with an Agilent 1260 Infinity II Quaternary Pump LC system (Waldbronn, Germany) according to a published article with some modifications 56 . In brief, 40 μL of urine was mixed with 10 μL of sulbactam (SB) (5 μg/mL, internal standard) and 160 μL ACN for deproteinization. After mixing for 3 minutes, the mixture was centrifuged at 16,400 × g for 8 minutes. The supernatant (1 μL) was introduced to liquid chromatography coupled with tandem-mass spectrometry (LC-MS/MS) and separated on a CAPCELL PAK C18 MGII (2.0 × 150 mm I.D., 5 μm) column. The mobile phase consisted of (A) water with 0.1% formic acid and (B) ACN with 0.1% formic acid, and the flow rate was 0.3 mL/min. The gradient was as follows: 0-2.5 min 80% A, 2.6-8 min 30% A, 8.1-14 min 0% A (for washing column), and 14.1-20 min 80% A (for column equilibration). The column oven was set at 30 °C. For multiple reaction monitoring (MRM), a quantifier and a qualifier were used to determine GAA, HA and IAA as shown in Table 2. Chromatogram of these metabolites was showed in Supplementary Fig. S3.
Multivariate and univariate statistical analysis. NMR spectra were integrated into peak lists and then introduced into the Metaboanalyst 3.5 website for multivariate analysis 82 . Data processing was according to user guideline of Metaboanalyst 3.5, and all data was normalized by sum of the spectrum followed auto-scaling.
Principal component analysis (PCA) was performed to examine the intrinsic outliers of the data set. The PCA score plot was used to reveal observations lying outside the 0.95 Hotelling's T2 ellipse (strong outliers), and the loading plot is used to explain the pattern of score plot. Partial least squares-discriminant analysis (PLS-DA) was used to find the best discriminant function model. The PLS-DA loading plot described the contribution of metabolites. However, PLS-DA analysis can be marred by a 'data overfitting' problem. Leave-one-out-cross validation (LOOCV) and the permutation test (n = 2000) were used to validate the PLS-DA model. In LOOCV, Q 2 and R 2 mean the robustness and reliance of the PLS-DA model, respectively. The difference between R 2 and Q 2 values was used as the indicator to determine if PLS-DA was overfitted, and it ideally is less than 0.3 39 . If not, permutation test should be used to validate the multivariate analysis model. The VIP score is used to estimate the important metabolites in a PLS-DA model. Among the identified metabolites, both VIP score (>1) and p-value (<0.05) were used to identify the potential biomarkers.
All quantitative data are expressed as the mean ± SD. They were compared using one-way ANOVA with the post hoc LSD test, or the two-tailed Students' t-test. If the p value was less than 0.05, it represented a significant difference.  Table 2. Multiple reaction monitoring (MRM) parameters of GAA, HA and IAA. Guanidinoacetic acid (GAA); hippuric acid (HA); indoleacetic acid (IAA); sulbactam (SB); cell accelerator voltage (CAV); collision energy (CE). For mass spectrometry, dry gas temperature, dry gas flow, nebulizer, sheath gas temperature, sheath gas flow, capillary voltage, EMV voltage and nozzle voltage were set at 300 °C, 5 L/min, 35 psi, 250 °C, 11 L/ min, 4000 V for positive and −3500 V for negative mode, 250 V (both positive and negative mode) and 500 V, respectively.