Metabolomics window into the role of acute kidney injury after coronary artery bypass grafting in diabetic nephropathy progression

Definitions

T2DM was diagnosed via WHO criteria. Normoalbuminuria was defined as urinary albumin excretion rate (UAER) <30 mg/day; microalbuminuria was defined as UAER = 30–300 mg/day, and macroalbuminuria as UAER >300 mg/day. T2DM with microalbuminuria (DN-micro) was defined as early DN. T2DM with macroalbuminuria (DN-macro) was defined as overt DN. Postoperative AKI was defined as an absolute increase in serum creatinine by ≥0.3 mg/dL (26.5 µmol/L) from baseline within 48 h or ≥50% within 7 days after operation. The staging of AKI was evaluated according to the criteria proposed by Kidney disease: Improving global outcomes (KDIGO) classifications (Kellum, Lameire & Group, 2013). Anemia was defined as hemoglobin less than 110 g/L in females or 120 g/L in males. The definition of chronic kidney disease (CKD) followed the National Kidney Foundation-Kidney Disease Outcomes Quality Initiative (K/DOQI) guidelines, and the CKD stage was defined using patient baseline estimated glomerular filtration rate (eGFR) (National Kidney Foundation, 2002). eGFR was calculated using the Chronic Kidney Disease Epidemiology Collaboration equation (CKD-EPI).

Patient recruitment and urine collection

T2DM patients who received the elective off-pump CABG procedure between Jan 2018 and Aug 2019 at the Second Xiangya Hospital of Central South University were screened for study participation. The inclusion criteria were as follows: (1) T2DM with normoalbuminuria; (2) Age ranged from 30 to 75 years; (3) eGFR >60 mL/min/1.73 m². Exclusion criteria were as follows: (1) Diagnosis with type 1 diabetes mellitus (WHO criteria); (2) Other etiology of CKD; (3) AKI prior to surgery; (4) Acute and chronic inflammation of the urinary system or other systems; (5) Patients with significant coexisting illnesses likely to affect survival or combined with other systemic diseases (including but not limited to malignancy, severe heart failure, blood system diseases, autoimmune diseases, cancer patients, and AIDS); (6) Patients with insufficient clinical data; (7) Patients underwent emergent surgeries, reoperations, and combined surgeries. In total, 88 T2DM patients were enrolled, and 8 of them received the intra-aortic balloon pump (IABP) during operation. Off-pump CABG was performed using conventional techniques, and complete revascularization was achieved in all patients. The mean duration of operation was around 4 h. Preoperative urine samples collected on the day before operation were named A group. At about 24 h after operation, postoperative urine samples were collected from a Foley catheter in the ICU and named B group. According to whether AKI occurred after operation, patients were grouped in the AKI group (AKI, n = 44) or the non-AKI group (NAKI, n = 44). Every patient in each group supplied both pre-and postoperative urine samples; thus, the AKI group was further subdivided into the AKI-A group and AKI-B group. The NAKI group was subdivided into the NAKI-A group and NAKI-B group. There were no deaths or dialysis needed in the AKI or NAKI groups. Serum creatinine of patients in the AKI group decreased to baseline when discharged.

Urine samples were also collected from early DN (DN-micro, n = 46) and overt DN (DN-macro, n = 34) patients as the validation cohort between Feb 2019 and Aug 2019 in our hospital. The inclusion criteria were as follows: (1) Age ranged from 30 to 75 years; (2) eGFR >30 mL/min/1.73 m²; (3) Duration of T2DM >5 years. Exclusion criteria were as follows: (1) Other etiology of CKD; (2) AKI within three months; (3) Acute and chronic inflammation of the urinary system or other systems; (4) Patients with significant coexisting illnesses likely to affect survival or combined with other systemic diseases; (5) Patients with insufficient clinical data. The variables recorded for each participant included gender, age, body mass index (BMI), systolic blood pressure (SBP), diastolic blood pressure (DBP), diabetes duration, medical history, hemoglobin, albumin, fasting blood glucose (FBG), hemoglobin A1c (HbA1c), low-density lipoprotein (LDL), high-density lipoprotein (HDL), total cholesterol (CHOL), triglyceride (TG), serum creatinine (Scr), uric acid (UA), blood urea nitrogen (BUN), urinary NGAL, and UAER.

All urine samples were collected in sterile tubes (Corning Incorporated, Corning, NY, USA) and placed immediately on ice. Cell debris of samples was removed by centrifugation (2,000 rpm at 4 °C for 12 min). The supernatant was frozen at −80 °C for storage and subsequent extraction and analysis. All methods used in this study were approved by the Ethics Committee of the Second Xiangya Hospital, Central South University (IRB2017-S551). Written informed consent was collected from each patient before the study.

Chemicals

Methyl alcohol, acetonitrile, and ethyl alcohol were purchased from Merck Company, Germany. Milli-Q system (Millipore Corp., Bedford, MA, USA) ultrapure water was used throughout the study. Authentic standards were purchased from Sigma-Aldrich (St. Louis, MO, USA). Two standards of metabolites (trans-4-hydroxy-L-proline and azelaic acid) were purchased from Sigma-Aldrich. All chemicals were of analytical grade.

Urine preparation for metabolite extraction

The freeze-dried samples were crushed using a mixer mill (MM 400, Retsch) with a zirconia bead for 1.5 min at 30 Hz. Then, 100 mg of powder was extracted with 1.0 mL 70% aqueous methanol containing 0.1 mg/L lidocaine overnight at 4 °C. After 10,000 g centrifugation for 10 min, the supernatant was collected and filtered (SCAA-104, 0.22-µm pore size; ANPEL, Shanghai, China) prior to liquid chromatography–tandem mass spectrometry (LC–MS/MS) analysis. Quality Control (QC) samples were mixed by all samples to detect reproducibility of the whole experiment.

Untargeted metabolomics analysis of the pilot study

The extracted preoperative and postoperative urine samples from 12 patients were used in the pilot study. An LC-ESI-MS/MS system (UPLC, Shim-pack UFLC SHIMADZU CBM30A; MS/MS, Applied Biosystems 6500 QTRAP) were used to analyze the extracted compounds (Chen et al., 2013b). For analysis, 2- µL samples were injected onto a Waters ACQUITY UPLC HSS T3 C18 column (2.1 mm*100 mm, 1.8 µm) operated at 40 °C. The mobile phases used were 0.04% acidified water (Phase A) and 0.04% acidified acetonitrile (Phase B). Through the following gradient program process, compounds were effectively separated: 5% B at 0 min; 95% B at 11.0 min; 95% B at 12.0 min; 5% B at 12.1 min; and 5% B at 15.0 min. The flow rate was 0.4 mL/min. The parameters used in the analysis are as follows: 500 °C electrospray ionization (ESI) source temperature; 5500 V ion spray voltage (IS); 25 psi curtain gas (CUR); the collision-activated dissociation (CAD) was set the highest. QQQ scans were acquired as multiple reaction monitoring (MRM) experiments with optimized declustering potential (DP) and collision energy (CE) for each individual MRM transition. The m/z range was set between 50 and 1000.

Data filtering, peak detection, alignment, and calculations were performed using Analyst 1.6.1 software (AB Sciex). Signal/noise >10 peaks were checked manually. In-house software written in Perl was used to remove redundant signals. Accurate m/z for each Q1 was acquired to identify the metabolites. Total ion chromatograms (TICs) and extracted ion chromatograms (EICs or XICs) of QC samples were used to construct metabolite profiles of all samples. MassBank, KNApSAcK, HMDB (Wishart et al., 2013), MoTo DB, and METLIN (Zhu et al., 2013) were the internal databases and public databases used to identify metabolites. The m/z values, RT, and fragmentation patterns of metabolites were compared with standards.

Targeted metabolomics analysis of the validation study

Urine samples collected from 38 NAKI patients (24 h after operation), 38 AKI patients (24 h after operation), 46 early DN patients, and 34 overt DN patients were used in the validation study. Each urine sample was thawed on ice. Methanol was added to urine sample (3:1, v/v), then vortex-mixed for 30 s and centrifuged at 12,000 rpm for 15 min at 4 °C. The supernatant was collected and placed at −40 °C for 1 h. After centrifugation (12,000 rpm for 15 min at 4 °C), supernatant was placed in the vial for analysis. Standard stock solutions of trans-4-hydroxy-L-proline and azelaic acid were prepared in water/methanol (1:1, v/v) at the concentration of 1 mg/mL. A series of trans-4-hydroxy-L-proline standards was prepared at the following concentration levels:10, 20, 50, 100, 200 and 500 ng/mL. A series of azelaic acid standards was prepared at the following concentration levels: 20, 50, 100, 200, 500, 1000, and 2000 ng/mL. Gradient diluted standards were used to establish standard curves.

Quantitative analysis was performed using an LC-ESI-MS/MS system (Waters Acquity UPLC; MS/MS: Applied Biosystems API 5500 QQQ-MS). 4- µL prepared samples were applied to the ACQUITY UPLC BEH T3 column (100*2.1 mm, 1.7 µm) at 40 °C. The mobile phases were composed of solvent A (water with 0.02% formic acid) and solvent B (acetonitrile), and the following gradient elution conditions were used with a flow rate of 0.35 mL/min: 5% B at 0 min; 10% B at 1.0 min; 70% B at 2.0 min; 90% B at 3 min; 90% B at 4.0 min; 10% B at 4.01 min; 10% B at 5.0 min. The following parameters were set for the mass spectrometer: temperature 550 °C, ion spray voltage 5500 V, CUR 35 arb, CAD 7 arb, ionsource GS1 55 arb, and ionsource GS2 55 arb. According to the above conditions, the standard solutions were sampled separately. Retention time of trans-4-hydroxy-L-proline was 0.68 min, and azelaic acid was 2.53 min. Standard curves were generated by plotting peak areas of the mass chromatogram (y) versus the concentration (x) and fitting to a linear regression (Trans-4-hydroxy-L-proline, y = 136.32x-686.72, R2=0.9997; azelaic acid, y = 6888.3x+311038, R2=0.9967). The identification of target metabolites in urine sample was based on the spectral pattern and retention time. Trans-4-hydroxy-L-proline and azelaic acid were quantified based on their peak areas and standard curves.

The methods were validated based on parameters specified for bioanalytical methods according to U.S. Food and Drug Administration document and other related guidelines (Gonzalez et al., 2014). Selectivity, limit of detection (LOD) and limit of quantification (LOQ), linearity, precision and accuracy were evaluated. The LOD was defined as the concentration at a signal-to-noise ratio (S/N) of ≥ 3, and the LOQ was defined at S/N ≥ 10 with precision lower than 20%. Each sample was analyzed in biological replicates of three and technical replicates of three in this experiment.

Data analysis

SPSS 23.0 software (SPSS Inc., Chicago, IL, USA) and GraphPad Prism 6.0 software were used for statistical analysis. Continuous variables were expressed as the means ± SD and were compared using a Student’s t test, a Wilcoxon test, one-way ANOVA or receiver operating characteristic (ROC) curves analysis. Categorical data were expressed as percentages (%) where appropriate and analyzed using Chi-square tests or Fisher’s exact test. Statistical significance was considered if two-sided p < 0.05. Multivariate principal component analysis (PCA), partial least squares discriminant analysis (PLS-DA), orthogonal projection to latent structures- discriminant analysis (OPLS-DA), loadings plot and T-test, and univariate ANOVA were carried out to determine the differential metabolites between groups. A variable importance in projection (VIP) value of OPLS-DA model was applied to rank the metabolites that best distinguished between groups. Metabolites between two groups were considered different when the p-value of the T-test <0.05 and the variable importance in projection (VIP) ≥ 1. Metabolites were mapped to the Kyoto Encyclopedia of Genes and Genomes (KEGG) database to determine important metabolic pathways and enrichment analysis. The calculated p-value was validated through false discovery rate (FDR) correction, taking FDR ≤ 0.05 as a threshold.

Participant characteristics

This study was carried out in two stages, as shown in Fig. 1. Clinical characteristics of all study patients are summarized in Table 1. A group of 88 T2DM inpatients, who underwent the elective off-pump CABG procedure, were enrolled in our study. The mean age of the study population was 59.15 ± 8.76 years, and 50% were male. The mean eGFR was 92.28 ± 17.67 mL/min 1.73 m². The mean diabetes duration was 10.00 ± 5.12 years. Urine samples were collected from these patients before and 24 h after CABG. Among the 44 patients in the AKI group, 70.5% were in AKI stage 1, 22.7% were in AKI stage 2, and only 6.8% were in AKI stage 3. Six patients from the AKI group and six patients from the NAKI group were chosen as the pilot cohort. The preoperative clinical characteristics of two groups were matched. After operation, as expected, Scr and uNGAL in the AKI group were significantly higher than that in the NAKI group (Scr, p = 0.001; uNGAL, p = 0.002). The untargeted metabolomics analysis on the pilot cohort was aimed to identify biomarkers of postoperative AKI among T2DM patients. Two significantly altered metabolites found in untargeted metabolomics were quantified in the validation cohort by targeted metabolomics analysis. Postoperative urine samples from 38 NAKI patients and 38 AKI patients, urine samples from 46 early DN patients and 34 overt DN patients were used in the validation phase. Among the validation cohort, around 20% of T2DM patients receiving the CABG procedure had diabetic retinopathy, while more than 60% of DN patients had diabetic retinopathy. In addition, 26% early DN patients and 29.4% overt DN patients had a history of ischemic heart disease. Age, gender, diabetes duration, BMI, FBG, and HbA1c were not significantly different between the four groups. SBP, TG, UAER and number of patients with anemia were significantly higher in the DN-macro group, compared with the other three (p < 0.001). Renal function (BUN, Scr, eGFR) of the AKI group was the worst, followed by DN-macro, DN-micro, and NAKI groups (p < 0.001).

Results of untargeted metabolomics in the pilot cohort

Metabolic profiling of preoperative samples and postoperative samples

The metabolic profiling of the A group and B group was evaluated to explore the postoperative differential metabolites among DM. The PCA score plot indicated the separation between the A group and B group in the discovery set (Fig. 3A). The OPLS-DA model was constructed to obtain better discrimination between the above groups. As shown in Fig. 3B, each code represents a tested urine sample. The loading plot of OPLS-DA clearly separated tested samples into two blocks according to their metabolic profiles of different groups. These results suggest that the biochemical metabolites in urine changed significantly after the CABG procedure among DM patients. According to the p-value of the T-test, combined with VIP values of the OPLS-DA model, 61 metabolites were statistically distinguishable between the two groups (R2X = 0.407, R2Y = 0.908, Q2 = 0.836) and are shown in Supplementary Table In Fig. 3C, heat map analysis shows significantly different metabolites in the urine samples of the A and B groups, with 25 being presented at higher levels and 36 at lower levels in the B group compared with A group. In total, 22 pathways of KEGG classification were identified from 61 differential metabolites. KEGG pathway annotation showed that differentially expressed metabolites in DM patients after CABG were enriched in carbohydrate and amino acid metabolisms (Fig. 3D). The functional mechanism of selected metabolomics was based on the pathway enrichment analysis using KEGG. The top 20 pathways are shown in Fig. 3E. The top five metabolic pathways with the highest p-values in the metabolic pathway enrichment analysis were phenylalanine metabolism, galactose metabolism, citrate cycle, prolactin signaling pathway, and proximal tubule bicarbonate reclamation.

Screening differential metabolites related to postoperative AKI in DM

We analyzed another two sets of pairwise data to identify AKI-related metabolites among DM patients: (i) AKI-B vs. AKI-A, (ii) AKI-B vs. NAKI-B. Two OPLS-DA models were constructed for these two comparisons. Figure 4A shows a clear separation between AKI-B and AKI-A. In total, 41 metabolites were differentially expressed in the AKI-B, compared with the AKI-A group; of these, 18 were up-regulated and 23 were down-regulated (Fig. 4B). Figure 4C shows a clear separation between AKI-B and NAKI-B. Moreover, 35 metabolites were differentially expressed in the AKI-B group, compared with the NAKI-B group. Of these, 23 were up-regulated and 12 were down-regulated (Fig. 4D).

The differential metabolites between AKI-B and AKI-A groups represent the postoperative AKI-related metabolites. Because there were 41 metabolites in these data sets, it was particularly important to select the metabolites with the greatest association with postoperative AKI among DM patients. The data of B vs. A provided us with information about operation-related differential metabolites. The data of AKI-B vs. NAKI-B provided us with information about AKI-related differential metabolites. Through Venn analysis, the postoperative AKI-related metabolites (the data of AKI-B vs. AKI-A) were compared with operation-related metabolites (the data of B vs. A) and AKI-related metabolites (the data of AKI-B vs. NAKI-B) to further reduce the number of postoperative AKI-related metabolites. Based on the results of Venn analysis, the nine most significant markers of postoperative AKI were found. The results show that the five up-regulated metabolites included trans-4-hydroxy-L-proline, uridine triphosphate, p-aminobenzoate, caffeic acid, and adrenochrome. The relative intensities of these metabolites in the A and B groups are shown in Figs. 5A–5F. The four down-regulated metabolites included δ-valerolactam, L-norleucine, 5′-deoxy-5′-(methylthio) adenosine, and azelaic acid. The relative intensities of these metabolites in the A and B groups are shown in Figs. 5G–5K. These nine overlapping metabolites shown in Table 2 were mainly distributed in the functional categories of amino acid, nucleotide, and organic acid metabolomics.

Validation of potential markers

The discovered metabolite markers trans-4-hydroxy-L-proline and azelaic acid, which have the maximum absolute fold change value among the nine overlapping metabolites, were further quantified in the validation cohort by targeted metabolomics analysis. To better understand the possible role of these two metabolites in the acute and chronic development of renal injury among T2DM patients, NAKI (n = 38), AKI (n = 38), DN-micro (n = 46), and DN-macro (n = 34) were chosen as the validation group. The results are shown in Fig. 6. The absolute trans-4-hydroxy-L-proline concentration increased gradually from the NAKI group to the AKI, DN-micro and DN-macro groups (p < 0.0001) (Fig. 6A). On the basis of quantitative analysis results, ROC curves were applied to investigate the clinical diagnostic potentials of these metabolites. An AUC value approaching 1.0 indicates a better diagnostic effectiveness. As shown in Figs. 6B–6C, the AUC of trans-4-hydroxy-L-proline between AKI and DN-micro groups was 0.722 (Fig. 6B, P = 0.0005, 95% CI [0.612–0.833]). The AUC of trans-4-hydroxy-L-proline between AKI and DN-macro groups was 0.783 (Fig. 6C, P < 0.0001, 95% CI [0.669–0.898]). These data suggest that the abundances of urinary trans-4-hydroxy-L-proline increased from acute renal damage to chronic renal damage in DM. For azelaic acid, the highest level was found in the NAKI and DN-micro groups, followed by the DN-macro group. The lowest level was found in the AKI group (p < 0.0001, Fig. 6D). The azelaic acid concentrations in the validation cohort were positively correlated with eGFR levels (p < 0.0001), and the Pearson correlation coefficient was 0.539 (Fig. 6E). As shown in Figs. 6F–6G, the AUC of azelaic acid between NAKI and AKI groups was 0.860 (Fig. 6F, P < 0.0001, 95% CI [0.770–0.949]). The AUC of azelaic acid between DN-micro and DN-macro groups was 0.754 (Fig. 6G, P = 0.0001, 95% CI [0.648–0.860]). These data suggest that the abundances of urinary azelaic acid decreased with acute or chronic deterioration of renal function in DM.