NMR-based serum and urine metabolomic profile reveals suppression of mitochondrial pathways in experimental sepsis-associated acute kidney injury

Abstract

Sepsis-associated acute kidney injury (SA-AKI) is a significant problem in the critically ill that causes increased death. Emerging understanding of this disease implicates metabolic dysfunction in its pathophysiology. This study sought to identify specific metabolic pathways amenable to potential therapeutic intervention. Using a murine model of sepsis, blood and tissue samples were collected for assessment of systemic inflammation, kidney function, and renal injury. Nuclear magnetic resonance (NMR)-based metabolomics quantified dozens of metabolites in serum and urine that were subsequently submitted to pathway analysis. Kidney tissue gene expression analysis confirmed the implicated pathways. Septic mice had elevated circulating levels of inflammatory cytokines and increased levels of blood urea nitrogen and creatinine, indicating both systemic inflammation and poor kidney function. Renal tissue showed only mild histological evidence of injury in sepsis. NMR metabolomic analysis identified the involvement of mitochondrial pathways associated with branched-chain amino acid metabolism, fatty acid oxidation, and de novo NAD⁺ biosynthesis in SA-AKI. Renal cortical gene expression of enzymes associated with those pathways was predominantly suppressed. Renal cortical fatty acid oxidation rates were lower in septic mice with high inflammation, and this correlated with higher serum creatinine levels. Similar to humans, septic mice demonstrated renal dysfunction without significant tissue disruption, pointing to metabolic derangement as an important contributor to SA-AKI pathophysiology. Metabolism of branched-chain amino acid and fatty acids and NAD⁺ synthesis, which all center on mitochondrial function, appeared to be suppressed. Developing interventions to activate these pathways may provide new therapeutic opportunities for SA-AKI.

NEW & NOTEWORTHY NMR-based metabolomics revealed disruptions in branched-chain amino acid metabolism, fatty acid oxidation, and NAD⁺ synthesis in sepsis-associated acute kidney injury. These pathways represent essential processes for energy provision in renal tubular epithelial cells and may represent targetable mechanisms for therapeutic intervention.

INTRODUCTION

Sepsis-associated acute kidney injury (SA-AKI) represents the intersection of two multifaceted syndromes diagnosed using clinical criteria, Kidney Disease: Improving Global Outcomes (KDIGO) and Sepsis-3, where infection-induced organ failure results in impaired renal function (1, 2). This condition is highly prevalent in both critically ill children and adults. SA-AKI increases morbidity and mortality in those affected above what would be expected for either syndrome in isolation (3–5). Although intensive care unit caregivers use supportive treatments and technologies, no mechanistically targeted therapeutic interventions currently exist for SA-AKI. This necessitates the investment of time, effort, and resources to elucidate the pathophysiologies underlying this disorder.

Historically, SA-AKI was attributed to renal hypoperfusion and hypoxia, but a significant body of recent research points toward metabolic changes in the renal tubular epithelial cell (RTEC) at the root of the kidney dysfunction observed clinically. Gomez and Kellum (6–8) have proposed a model that explains SA-AKI as a consequence of an inflammation-induced metabolic shift away from fatty acid oxidation (FAO) and oxidative phosphorylation to aerobic glycolysis in RTECs. Teams led by Ronco and Parikh (9–11) have described significant changes in mitochondrial function in SA-AKI. Metabolites associated with cellular and mitochondrial metabolic processes, detectable in serum and urine, could serve as diagnostic biomarkers in identifying SA-AKI clinically and also as mechanistic biomarkers to identify specific pathophysiologies at play in the disease process.

In previous studies, we have used nuclear magnetic resonance (NMR)-based metabolomics to identify metabolic derangements in hypoxia and ischemia-reperfusion models of AKI (12, 13). Metabolites associated with energy production, amino acid conversion, and mitochondrial function were differentially affected. In this study, we undertook an evaluation of the metabolic changes induced in SA-AKI to identify metabolic processes at play in this condition potentially amenable to therapeutic intervention.

MATERIALS AND METHODS

Animal Experiments

All experiments were approved by the Cincinnati Children’s Hospital Medical Center Institutional Animal Care and Use Committee. Twenty-four male C57Black/6J mice were purchased from Jackson Laboratories (Stock No. 000664, Bar Harbor, ME) and acclimated in our vivarium for 7 days before experimentation. Male mice were used to avoid confounding of the sepsis model by female sex hormones that vary during the estrous cycle (14). Baseline urine was collected from each mouse using special metabolic cages. Subsequently, each mouse underwent sham or cecal ligation and puncture (CLP) surgeries to induce sepsis (n = 12 in each group) as previously described (15). Postoperatively, both sham and CLP mice received normal saline fluids, buprenorphine analgesia, and imipenem/cilastatin antibiotics every 12 h starting 2 h after surgery. All mice were fasted postoperatively to establish a similar metabolic state between conditions as septic mice exhibit anorexia. All mice had access to water. Urine was again collected using metabolic cages. At 24 h, mice were euthanized and tissue samples were collected for analysis. Histological sections obtained from sham and CLP mice were compared against those from mice subjected to bilateral ischemia-reperfusion injury (30-min renal artery clamp time), a well-established model of severe kidney injury previously described by our group, which served as a positive control (16).

Serum Blood Urea Nitrogen and Creatinine Measurements

Serum blood urea nitrogen (BUN) and enzymatic creatinine (EZCRE) levels were measured by Dimension clinical chemistry system RXL (Siemens, Malvern, PA).

Serum Cytokine Quantification

Serum cytokine concentrations [interleukin (IL)-1β, IL-6, IL-17, tumor necrosis factor-α (TNF-α), chemokine (C-X-C motif) ligand 1 (CXCL1)/keratinocyte-derived chemokine (KC), interferon-γ (IFN-γ), macrophage inflammatory protein-1α (MIP-1α), and IL-10] were measured using the MILLIPLEX MAP Mouse Cytokine/Chemokine Magnetic Bead Panel Multiplex Assay (Millipore/Sigma, Burlington, MA) and quantified on a Luminex 200 analyzer (Luminex, Austin, TX).

Histology and Immunohistochemistry

Paraformaldehyde-fixed, paraffin-embedded 5-μm-thick sections were stained with hematoxylin and eosin (H&E) and scored by a blinded reader for histopathological damage as previously described (17). Briefly, each parameter was assessed in 10 high-power fields (×40), and an average was determined for each section. The tested parameters included tubule dilatation, tubule cast formation, and tubule necrosis. Each parameter was scored on a scale of 0–4, ranging from none (0), mild (1), moderate (2), severe (3), to very severe (4). A total injury score was calculated from the sum of each individual parameter score for each subject. Sham and CLP samples were compared with five samples from mice subjected to ischemia-reperfusion injury (as positive controls). For immunohistochemistry, sections were deparaffinized and rehydrated. After antigen retrieval, sections were incubated overnight with antibodies against neutrophil gelatinase-associated lipocalin (NGAL/LCN2; Cat. No. orb11123, Biorbyt) and kidney injury molecule-1 (KIM-1/HAVCR1; Cat. No. ab78494, Abcam) followed by secondary antibody (ABC kit for NGAL and KIM-1, Vector Laboratories) for 30 min at room temperature and then incubated for 40 min with ABC reagent. After ABC reagent, peroxidase substrate solution (Cat. No. SK-410, Vector Laboratories) was applied. Finally, slides were counterstained with hematoxylin and mounted after dehydration.

Plasma and Urine NMR-Based Metabolomics

Plasma and urine samples were stored and prepared as previously described (12, 13, 18–25). Serum and urine samples were stored at −80°C before preparation for NMR analysis. Samples were thawed on ice, centrifuged at 1,000g for 5 min at 4°C using a 0.22-µm filter, and then reconstituted into buffer at final concentrations of 150 mM potassium phosphate, 1 mM trimethylsilyl propionate, 1 mM EDTA, and 0.1% sodium azide at 90% H₂O-10% D₂O. The pH of all samples was adjusted to 7.0. Six hundred microliters of each redissolved urine sample was transferred to a 5-mm NMR tube, and 200 μL of each serum sample was transferred to a 3-mm NMR tube for NMR analysis. NMR spectra were recorded at 298 K on a Bruker AVANCE III spectrometer operating at 850.1 MHz (Bruker, Billerica, MA). Standard one-dimensional (1-D) ¹H presaturation (ZGPR) and 1-D ¹H Carr-Purcell-Meiboom-Gill (CPMG) experiments were collected. The 1-D ¹H ZGPR spectrum was collected for each sample to ensure that the water suppression was sufficient. The 1-D ¹H CPMG NMR spectra were collected using a spectral width of 16 ppm and 64,000 points with 2.41-s acquisition time, 64 scans, 4 dummy scans, 3-s recycle delay, and receiver gain setting to 32. Two-dimensional ¹H-¹H total correlation spectroscopy (TOCSY) NMR experiments were performed using a spectral width of 14 ppm in both dimension with 2,000 points (F2) and 1,000 points (F1), 32 scans, 16 dummy scans, 1.5-s recycle time, and 60-ms mixing time. ¹H-¹³C heteronuclear single quantum coherence (HSQC) spectra were recoded with 96 scans and a 1.5-s recycle delay with 2,048 points (F2) and 256 points (F1). For the aliphatic ¹H-¹³C HSQC, spectral widths of 14 ppm (F2) and 80 ppm (F1), centered at 4.7 ppm (F2) and 45 ppm (F1), were set respectively with a J-coupling constant of 140 Hz. For the aromatic ¹H-¹³C HSQC, spectral widths of 14 ppm (F2) and 30 ppm (F1), centered at 4.7 ppm (F2) and 125 ppm (F1), were set, respectively, with a J-coupling constant of 165 Hz. All NMR spectra were processed, phase adjusted, baseline corrected, and referenced to 0.0 ppm using the internal trimethylsilyl propionate standard manually using Topspin 3.6.1 (Bruker BioSpin, Billerica, MA).

Identification and Quantification of Metabolites

All spectral resonances with visible differences were manually bucketed using AMIX v.3.9 software (Analysis of MIXtures software, Bruker Biospin). Unnormalized spectra were used for metabolite identification and quantification. The peak areas integrated from AMIX were exported for statistical significance analysis. Metabolite assignment was attempted for all bucketed peaks. The 1-D NMR peaks were identified by ChenomX Profile (https://www.chenomx.com/) using the Biological Magnetic Resonance Data Bank (BMRB) (26, 27) and the Human Metabolome Database (HMDB) (28–30). Two-dimensional NMR data were analyzed using COLMAR software (https://ccic.ohio-state.edu/) (31, 32) to aid in confirmation of the metabolite assignments. Where peak identification could be made, confidence rank in those assignments was expressed using the RANCM scheme as previously described (33).

Gene Expression Analysis

All reagents were purchased from Applied Biosystems/Thermo Fisher Scientific (Waltham, MA). Total RNA was extracted from snap-frozen kidney tissue (primarily the cortex) using TRIzol reagent according to the manufacturer’s instructions. cDNA was created from extracted RNA using the High-Capacity cDNA Reverse Transcription kit. Quantitative PCR gene expression analysis was subsequently conducted with TaqMan Gene Expression Assays using TaqMan Fast Universal PCR Master Mix on an Applied Biosystems QuantStudio 5 real-time PCR instrument.

FAO Rate Determination

The rate of renal cortical tissue FAO was measured using the Fatty Acid Oxidation Assay Kit from Assay Genie (Dublin, Ireland) according to the manufacturer’s specifications.

Statistical Analysis

Statistical analyses were completed using R 3.6. Raw data files, including the complete set of observations, and analysis scripts are deposited on the Open Science Framework (https://doi.org/10.17605/osf.io/A5W4E). Throughout, group differences were evaluated with Welch’s t test, which allows for comparison of means between samples with unequal variance. Histology scores were compared with the Kruskal–Wallis test followed by post hoc Dunn’s pairwise contrasts with multiple comparison P values adjusted with the Benjamini–Hochberg method. One subject in the CLP cohort was removed from the analyses because it did not appear to develop sepsis, and another CLP subject had insufficient serum for NMR analysis. Three subjects (1 sham and 2 CLP) had insufficient postprocedure urine to analyze. For NMR and gene expression analyses, principal component analysis (PCA) was first used to assess for qualitative group differences on a dimensionally reduced plane and to identify outliers. Two extreme outliers, one each in the serum NMR and gene expression analyses, were excluded due to significant divergence from group localizations (Supplemental Fig. S1; all Supplemental material is available at https://doi.org/10.17605/osf.io/A5W4E). For the NMR analysis, partial least squares discriminant analysis (PLS-DA) was used to calculate Variable Importance in Projection (VIP) scores for each metabolite, which quantifies the relative importance of that metabolite’s contribution to the distinction between groups. Receiver operating characteristic area under the curve (ROC-AUC) values were calculated for each metabolite as an additional measure of its ability to distinguish sham from sepsis conditions. MetaboAnalyst and the MetaboAnalystR package were used for these analyses (34, 35). For the analysis of urinary metabolites, two-factor repeated-measures ANOVA was used with post hoc pairwise contrasts, determined a priori, to test for differences in metabolite abundance between baseline and sham conditions and between postoperative sham and CLP conditions. Urinary metabolite abundance was not normalized to urinary creatinine level or any other factor as none have been identified that effectively compensates for urine dilution in sepsis. Both creatinine production and excretion are highly variable in sepsis. Therefore, scaling metabolite measurements to creatinine risks confounding real differences in metabolite abundance. For the correlation analysis, Pearson coefficients were calculated between each analyte. Standard multivariate linear regression models were used to compare relationships between FAO rate and other study factors. A false discovery rate of 0.05 was used to correct for multiple comparisons within each analysis. An α value of 0.05 determined statistical significance.

RESULTS

Serum Markers of Kidney Failure and Inflammation

Serum creatinine more than doubled in CLP mice compared with sham controls (0.14 sham vs. 0.29 CLP, P = 0.002). BUN also increased dramatically in the septic cohort (43.2 sham vs. 104.1 CLP, P = 0.001; Fig. 1A). Inflammatory cytokines and chemokines (IL-1β, IL-6, TNF-α, CXCL1, MIP-1α, and IL-10) were significantly increased in CLP mice. Differences in serum IFN-γ and IL-17 trended toward statistical significance (Fig. 1B).

Figure 1.

Serum and tissue markers of inflammation and injury. A: serum blood urea nitrogren (BUN) and creatinine were elevated in cecal ligation and puncture (CLP)-operated, septic mice, indicating renal dysfunction. B: inflammatory cytokines were significantly elevated in septic mice. C: renal histology total injury scores were not significantly different between sham- and CLP-operated mice, although ischemia-reperfusion injury (IRI) positive controls differed from both. D: immunohistochemistry experiments of tissue markers of injury [neutrophil gelatinase-associated lipocalin (NGAL) and kidney injury molecule-1 (KIM-1)] differed remarkably only for mice subjected to IRI. Box and whisker plots display the interquartile range as the upper and lower boundaries of the box, with the middle line representing the median of the data. Whisker lines extend to the upper and lower values no more than 1.5 times the interquartile range. Individual data points are overlaid on the plot to further represent the spread of values. For A and B, *P < 0.05 by Welch’s t test corrected for multiple comparisons. For C, *P < 0.05 comparing IRI with CLP and #P < 0.05 comparing IRI with sham. For D, hematoxylin and eosin (H&E) and immunohistochemistry analyses were conducted in all study mice, and representative images of each group are shown here. n = 11−12 mice per sham or CLP groups; n = 5 mice as IRI controls. CXCL1, chemokine (C-X-C motif) ligand 1; IFNg, interferon-γ; IL-1b, interleukin-1β; MIP-1a, macrophage inflammatory protein-1α; TNFa, tumor necrosis factor-α.

Kidney Histology

Kidney tissue was stained for histological analysis with H&E and underwent immunohistochemistry for injury markers NGAL and KIM-1. Kidney sections from sham mice showed normal glomerular and tubular structures without signs of epithelial injury. Sections from CLP mice showed white blood cell infiltration into both glomeruli and tubules with interstitial inflammation. Tubules, however, exhibited signs of only minor epithelial injury that did not differ statistically when scored by a blinded reader (Fig. 1C). Compared with positive control sections obtained from mice subjected to renal ischemia, histological signs of damage in septic mice were far less dramatic. Sham kidneys had only trace NGAL and KIM-1 staining in the tubules and slightly more prominent KIM-1 staining in the glomerulus. Septic kidneys had mild NGAL tubular staining and moderate tubular KIM-1 staining. Ischemic control kidneys had intense NGAL and KIM-1 staining (Fig. 1D).

Serum NMR Metabolomics

Analysis of NMR spectra collected from serum quantified the intensity of 73 different peaks in each sample. Metabolite assignments were successfully made for 32 of these. Using only identified features, PCA showed good separation between sham and CLP conditions, indicating that the serum metabolome differed significantly (Fig. 2A). The relative abundance of each metabolite was compared between sham and CLP conditions. Importantly, metabolites that accumulate in renal failure (creatinine, allantoic acid, and arabinose) were elevated in the CLP cohort (Fig. 2B). Although mice in both conditions were fasted, metabolites associated with ketogenesis and branched-chain amino acid (BCAA) metabolism had predominantly lower abundance in the CLP group (Fig. 2C). Lactate levels were also lower in CLP mice, which is consistent with our prior observations using this sepsis model (Fig. 2D) (15). In addition, metabolites associated with glycerophospholipid metabolism were affected (Fig. 2E). These data were submitted to MetaboAnalyst for pathway analysis, which confirmed that changes in pathways involving glycerophospholipid; glycine, serine, and threonine; ketone; and BCAA metabolism distinguished septic mice from sham controls (Fig. 2F). The glycerophospholipid and glycine, serine, and threonine pathways are both related to choline metabolism. PLS-DA analysis computed VIP scores for each metabolite and indicated that glycerophosphocholine, choline, acetoacetic acid, and lactate were some of the metabolites that most contributed to differentiating the two conditions (Fig. 2G). Interestingly, 2,3-butanediol, a metabolite of possible microbial origin, received the highest VIP score. Finally, ROC-AUC values were computed for each metabolite (Fig. 3 and Supplemental Table S1). A heatmap illustrating relative serum metabolite abundance is provided in Supplemental Fig. S2.

Figure 2.

Serum metabolomics analysis. A: principal component analysis showing separation of sham and cecal ligation and puncture (CLP) groups. B: metabolites associated with renal failure had increased abundance in sepsis. C: metabolites associated with branched-chain amino acid (BCAA) and ketone production were predominantly reduced in sepsis. D: lactate quantities were reduced in sepsis. E: metabolites associated with glycerophospholipid metabolism were altered in sepsis. F: pathway analysis identifing multiple pathways that differed in septic mice based on serum metabolite levels. G: VIP scores, calculated by PLS-DA analysis of serum metabolomics, ranked metabolites by relative contribution to group difference. The top 15 features are shown. A description of the box and whisker plots is included in Fig. 1. *P < 0.05 by Welch’s t test corrected for multiple comparisons. n = 10 or 11 mice/group.

Figure 3.

Serum metabolite ranking and metabolic process assignment. Shown are serum metabolites identified by NMR ranked by their priority score, which was calculated as the product of the fold change and negative log of the false discovery rate (FDR)-adjusted P value. Metabolite (ppm) indicates the NMR feature assignment. Rank is the confidence of the assignment given that feature. Integrated peak intensities are shown for each feature with their associated fold change and FDR-adjusted P values. VIP scores were computed in PLS-DA analysis, and the receiver operating characteristic area under the curve (ROC AUC) values are reported. These data are available in tabular format in Supplemental Table S1. BCAA, branched-chain amino acid; CLP, cecal ligation and puncture.

Urine NMR Metabolomics

Paired urine samples were collected from each mouse, pre- and postoperatively, and peak intensity for 110 NMR features was quantified in each urine sample. Metabolite assignments were successfully made for 78 of these peaks. Two separate analyses were conducted: 1) baseline versus sham mice to assess for urinary metabolomic changes attributable to the stress of surgery and 2) sham versus CLP mice to assess for changes specifically attributable to sepsis. Regarding the comparison of greatest biological interest, sham versus CLP, PCA showed some overlap between the two conditions, indicating the presence of variability in the urine metabolome within the groups (Fig. 4A). Comparing the urinary abundance of each metabolite between the two conditions offers a clearer picture of metabolic processes influenced by sepsis. Metabolites associated with BCAA metabolism had decreased urinary abundance in CLP mice (Fig. 4B). Quinolinic acid, a metabolite related to NAD⁺ metabolism, was increased (Fig. 4C). Interestingly, urinary glucose levels were increased in septic mice compared with sham mice even though CLP mice are known to be more hypoglycemic than their sham counterparts (Fig. 4D). Abundance of metabolites related to ketogenesis was lower in CLP mice compared with sham mice (Fig. 4E). Urinary levels of ascorbic acid and metabolites related to coenzyme A (CoA) metabolism were decreased in CLP mice (Fig. 4F). MetaboAnalyst identified pathways related to BCAA, ketone, CoA, and NAD⁺ metabolism as significantly altered in sepsis (Fig. 4G). PLS-DA VIP scores ranked metabolites associated with BCAA and ketone metabolism highest in distinguishing CLP urine samples from sham (Fig. 4H). Again, ROC-AUC values were computed for each metabolite (Fig. 5 and Supplemental Table S2). A heatmap illustrating relative urine metabolite abundance for this comparison is provided in Supplemental Fig. S3.

Figure 4.

Urine metabolomics analysis. A: principal component analysis showing separation of pre- and postoperative urine samples in both sham and cecal ligation and puncture (CLP) groups. B: metabolites associated with branched-chain amino acid (BCAA) metabolism had decreased abundance in sepsis. C: quinolinic acid, a metabolite related to NAD⁺ synthesis, had increased abundance in septic urine compared with sham. D: urinary glucose levels were also increased in sepsis. E: metabolites associated with ketogenesis were decreased in sepsis. F: metabolic cofactors had reduced urine concentrations after CLP. G: pathway analysis identifying multiple pathways that differed in septic mice based on urine metabolite levels. H: VIP scores, calculated by PLS-DA analysis of urine metabolomics, ranked metabolites by relative contribution to group difference. Shown are the top 15 features. A description of the box and whisker plots is included in Fig. 1. Two-way ANOVA with repeated measures was run first, followed by post hoc pairwise contrasts on metabolites with a significant ANOVA P value. *P < 0.05, sham vs. CLP; #P < 0.05, pre- vs. postoperative on post hoc contrasts corrected for multiple comparisons. n = 10 or 12 mice/group.

Figure 5.

Urine metabolite ranking and metabolic process assignment. Shown are urine metabolite differences comparing postoperative sham and cecal ligation and puncture (CLP) mice. Metabolites identified by NMR were ranked by their priority score, which was calculated as the product of the fold change and negative log of the false discovery rate (FDR)-adjusted P value. Metabolite (ppm) indicates the NMR feature assignment. Rank is the confidence of the assignment given that feature. Integrated peak intensities are shown for each feature with their associated fold change and FDR-adjusted P values. VIP scores were computed in PLS-DA analysis, and the receiver operating characteristic area under the curve (ROC AUC) values are reported. These data are available in tabular format in Supplemental Table S2. BCAA, branched-chain amino acid.

The pattern of urinary metabolite and metabolic pathway changes in the baseline versus sham comparison largely reflected those described in the sham versus CLP comparison, but to a lesser degree. This might explain some of the overlap between the sham and CLP conditions on PCA. One notable exception was that some of the metabolites associated with BCAA (α-ketoisovaleric acid and 3-methyl-2-oxovaleric acid) were dramatically increased in sham mice compared with baseline control urine samples (Fig. 4B). The decreased abundance of these metabolites noted in the sham versus CLP comparison was reduced from these higher levels. Similar to previous comparisons, MetaboAnalyst pathway analysis was conducted (Supplemental Fig. S4A), and PLS-DA VIP scores (Supplemental Fig. S4B) and ROC-AUC values were calculated (Supplemental Fig. S5 and Supplemental Table S3). A heatmap illustrating relative urine metabolite abundance for this comparison is provided in Supplemental Fig. S6.

Kidney Gene Expression Analysis

Because metabolites associated with BCAA, ketone production, and NAD⁺ pathways differed significantly between CLP and sham controls, we assessed gene expression of enzymes associated with these pathways. Kidney cortical expression of the inflammatory cytokine Il-6 (Il6) and renal injury markers NGAL (Lcn2) and KIM-1 (Havcr1) were significantly upregulated, indicating that the tissue itself was affected by sepsis (Fig. 6A).

Figure 6.

Gene expression analysis of kidney tissue. A: kidney markers of inflammation and injury were upregulated in septic mice. B and C: expression of enzymes involved with branched-chain amino acid (BCAA) metabolism showing the reduction in the Dbt subunit of branched-chain ketoacid dehydrogenase (BCKDH) as well as its regulatory enzymes branched-chain keto acid dehydrogenase kinase (Bckdk) and protein phosphatase, Mg²⁺/Mn²⁺ dependent 1 K (Ppm1k). D and E: enzymes associated with fatty acid oxidation (FAO) upon which ketogenesis depends were reduced in sepsis. The regulatory genes mammalian target of rapamycin (Mtor) and hypoxia-inducible factor-1α (Hif1a), which promote enhanced glycolysis, were upregulated. See text for relevant abbreviations. A description of the box and whisker plots is included in Fig. 1. *P < 0.05 by Welch’s t test corrected for multiple comparisons. n = 11 mice per group. Havcr1, kidney injury molecule; Il6, interleukin-6; Lcn2, neutrophil gelatinase-associated lipocalin.

The first step in BCAA metabolism involves deamination by a BCAA aminotransferase (BCAT) enzyme, of which two forms exist: cytosolic BCAT1 and mitochondrial BCAT2. Bcat2 expression levels were lower in CLP mice. The mitochondrial branched-chain ketoacid dehydrogenase (BCKDH) enzyme complex mediates the rate-limiting, irreversible step in BCAA catabolism. Expression of only one of its four components, Dbt, which encodes the critical E2 subunits of the enzyme complex, was lower in CLP mice. Phosphorylation inhibits and dephosphorylation activates BCKDH. Expression of both enzymes mediating these effects [branched-chain keto acid dehydrogenase kinase (Bckdk) and protein phosphatase, Mg²⁺/Mn²⁺ dependent 1 K (Ppm1k), respectively] was reduced in sepsis (Fig. 6, B and C).

Ketogenesis is dependent on mitochondrial fatty acid β-oxidation (FAO). Expression of genes associated with that pathway [peroxisome proliferator-activated receptor (PPAR)-γ coactivator-1α (Ppargc1, PGC-1α), PPAR-α (Ppara), and medium-chain acyl-CoA dehydrogenase (Acadm)] were decreased in CLP kidneys. Expression of regulatory genes upstream of the β-oxidation pathway, hypoxia-inducible factor-1α (HIF-1α, Hif1α) and mammalian target of rapamycin (mTOR, Mtor), were upregulated, whereas no change was observed in the levels of AMP-activated protein kinase (AMPK, Prkaa1) (Fig. 6, D and E).

Many enzymes associated with the de novo NAD⁺ synthesis pathway were altered in their expression. Indoleamine 2,3-dioxygenase 2 (Ido2), which catalyzes the rate-limiting first step in the pathway, as well as most other pathway enzymes [arylformamidase (Amfid), kynureninase (Kynu), 3-hydroxyanthranilate 3,4-dioxygenase (Haao), and quinolinate phosphoribosyltransferase (Qprt)] were downregulated in CLP kidneys. Expression of Ido1, which activates in inflammation, showed a trend toward increased expression. Kynurenine 3-monooxygenase (Kmo) expression was the same between groups. 2-Amino 3-carboxymuconate 6-semialdehyde decarboxylase (Acmsd), an enzyme that siphons substrate away from NAD⁺ synthesis, also had reduced expression in sepsis-effected kidneys. Nicotinamide phosphoribosyltransferase (Nampt), the rate-limiting enzyme in the NAD⁺ salvage pathway, showed no expression differences (Fig. 7).

Figure 7.

Gene expression of enzymes associated with NAD⁺ metabolism. Expression of the majority of genes associated with de novo NAD⁺ synthesis was downregulated in sepsis. Expression of nicotinamide phosphoribosyltransferase (Nampt), the rate-limiting step in the NAD⁺ salvage pathway, was not different between conditions. See text for relevant abbreviations. A description of the box and whisker plots is included in Fig. 1. *P < 0.05. n = 11 mice/group.

Correlation Analysis

Tissue gene expression was first correlated to serum markers of kidney function and inflammation (Fig. 8). Serum and tissue markers of injury and inflammation were highly correlated. With rare exceptions, tissue expression of genes associated with BCAA, fatty acid metabolism, and NAD⁺ biosynthesis correlated inversely to markers of injury and inflammation where a statistically significant relationship was identified. Thus, as serum and tissue mediators of inflammation increased, expression of enzymes associated with these three metabolic pathways decreased. Notably, several correlations involving genes associated with FAO lost significance after adjustment for multiple comparisons, but P values of <0.10 indicated strong trends of likely biological significance.

Figure 8.

Correlation matrix for serum biomarker and gene expression analyses. This matrix illustrates the relationships between serum biomarkers of inflammation and injury and kidney gene expression of enzymes associated with the metabolic pathways implicated in the NMR-based metabolomic analyses. The top half of the matrix shows blue circles to indicate positive correlation, whereas red indicates negative correlations. The bottom half shows Pearson correlation coefficients. Only those correlations that were significant after P value adjustment for multiple comparisons are displayed. A full correlation matrix including all metabolites identified by NMR is provided in Supplemental Fig. S7.

Gene expression of HIF-1α and mTOR correlated positively with several markers of injury and inflammation, but AMPK was not. Furthermore, HIF-1α showed strong inverse correlations with NAD⁺ biosynthetic enzymes. mTOR had similar inverse correlations with some of the enzymes, but of lower magnitude. IDO1, known to be induced by inflammatory stimuli, had many strong, direct correlations with inflammatory mediators.

We next conducted a large correlation analysis between serum markers and tissue gene expression and serum and urine metabolites quantified by NMR (Supplemental Fig. S7). The most striking feature of this analysis was the concordant, inverse correlation between markers of injury and inflammation and the great majority of urinary metabolites. The dramatic exception to this pattern was urinary quinolinic acid levels, which increased in conjunction with kidney injury and inflammation. Interestingly, although quinolinic acid comprises one of the steps in the NAD⁺ biosynthetic pathway, urinary levels were related to only one enzyme (KYNU) in that pathway with an inverse correlation.

Urinary glucose levels correlated directly with serum and tissue markers of injury (serum creatinine, BUN, and kidney NGAL and KIM-1 expression).

Nine serum metabolites measured by NMR exhibited strong, direct correlations with serum markers of both injury and inflammation and HIF-1α and mTOR expression (2,3-butanediol, 3-hydroxybutyric acid, allantoic acid, arabinose, carnitine, creatinine, glycerophosphocholine, glycine, and N,N-dimethylaniline). Most of these metabolites correlated inversely with genes related to BCAA, fatty acid, and NAD⁺ metabolism. More correlations existed with the NAD⁺ pathway than others. Serum metabolites associated with BCAA and fatty acid metabolism (acetoacetic acid, 2-hydroxy-3-methylbutyric acid, hydroxyisocaproic acid, and 2-methyl-3-ketovaleric acid) showed few significant correlations with enzymes in their respective pathways except in some for PGC-1α, Hif-1α, and mTOR.

Serum choline levels showed strong inverse correlations with markers of injury and inflammation and levels of glycerophosphocholine. Some of the most significant correlations between gene expression and metabolites identified by NMR are shown in Table 1.

Table 1.

Gene expression, protein, and metabolite correlations of greatest significance

Row	Column	Correlation Coefficient	P Value	FDR P Value	Priority Score
Bckdk	Ppargc1	0.951	1.26 e−12	1.92 e−10	21.262
Creatinine	s_AllantoicAcid	0.957	1.22 e−11	1.13 e−09	19.713
Ppargc1	Qprt	0.934	2.58 e−11	2.17 e−09	18.635
IL-1b	s_N,N-Dimethylaniline	0.937	4.09 e−10	2.07 e−08	16.578
Afmid	Haao	0.914	4.36 e−10	2.15 e−08	16.14
MIP-1a	s_N,N-Dimethylaniline	0.927	1.65 e−09	6.64 e−08	15.314
Bcat2	Ppm1k	0.906	1.10 e−09	4.72 e−08	15.287
TNFa	s_N,N-Dimethylaniline	0.926	1.75 e−09	6.95 e−08	15.264
IL-1b	s_Carnitine	0.925	1.89 e−09	7.38 e−08	15.198
Creatinine	s_Glycine	0.925	1.96 e−09	7.55 e−08	15.172
IL-10	s_N,N-Dimethylaniline	0.923	2.55 e−09	9.27 e−08	14.947
CXCL1	s_N,N-Dimethylaniline	0.923	2.57 e−09	9.30 e−08	14.943
IL-1b	s_3-HydroxybutyricAcid	0.918	4.36 e−09	1.48 e−07	14.443
Lcn2	Hif1a	0.892	5.00 e−09	1.67 e−07	13.916
Bckdk	Ppm1k	0.887	7.81 e−09	2.47 e−07	13.493
Bcat2	Ppargc1	0.885	8.98 e−09	2.79 e−07	13.364
Ido 2	Haao	0.869	3.70 e−08	9.23 e−07	12.071
Ppm1k	Acadm	0.858	8.59 e−08	1.91 e−06	11.291
Bckdk	Qprt	0.856	9.34 e−08	2.05 e−06	11.217
Lcn2	Havcr1	0.850	1.44 e−07	2.90 e−06	10.842
Bckdhb	Dld	0.849	1.52 e−07	3.02 e−06	10.797
IL-6	s_N,N-Dimethylaniline	0.872	2.63 e−07	4.64 e−06	10.706
Creatinine	s_Glycerophosphocholine	0.870	2.89 e−07	5.00 e−06	10.626
CXCL1	s_Glycine	0.870	3.09 e−07	5.32 e−06	10.56
Dbt	Acadm	0.842	2.42 e−07	4.32 e−06	10.407

Italicized genes: mRNA expression. Nonitalicized names: serum analyte (cytokine, blood urea nitrogen, and creatinine). “s_” prefix: serum metabolite identified by NMR. “u_” prefix: urine metabolite identified by NMR. The priority score was calculated as the product of the correlation coefficient and negative log of the false discovery rate (FDR)-adjusted P value.

Renal Cortical FAO Rate

Univariate analysis of FAO rates did not reveal an absolute difference between sham and CLP mice (Fig. 9A). However, when circulating cytokines were considered in the analysis as a surrogate of illness severity, it was evident that FAO rates decreased as systemic inflammation increased in septic mice (regression coefficient = −0.093, P = 0.0002; Fig. 9B). This pattern was not observed in sham mice (regression coefficient = −0.018, P = 0.617). When we assessed the relationship between renal cortical FAO rate and serum creatinine levels, we observed a strong interaction between the CLP condition and FAO rate (interaction coefficient = −0.314, P = 0.0018; Fig. 9C), indicating that creatinine rose as tissue FAO decreased in septic mice. Plotting these four factors in one figure revealed important relationships: as the severity of sepsis increased, renal cortical FAO decreased, which was associated with increased creatinine (Fig. 9D). Renal function deteriorates with low kidney FAO in the setting of high inflammation. Supporting this conclusion, we observed strong positive associations between renal cortical FAO rates and tissue gene expression of proteins associated with FOA in CLP mice but not sham mice (interaction P values all <0.05; Fig. 9E).

Figure 9.

Renal cortical fatty acid oxidation (FAO) analysis. A: no absolute differences in tissue FAO rates were observed between sham and cecal ligation and puncture (CLP) mice by univariate analysis. B: incorporating inflammatory cytokine levels in a multivariate analysis revealed a striking negative correlation between kidney FAO rates and inflammation in CLP mice. C: creatinine levels rose in septic mice as FAO rates fell. D: creatinine levels plotted as a function of kidney FAO rate and serum IL-6 levels revealed dramatic differences between sham and CLP mice. E: in septic mice, tissue expression of genes associated with FAO predicted FAO rates. n = 12 mice/group.

DISCUSSION

We have shown that CLP-operated, septic mice develop AKI, as demonstrated by elevated serum BUN and creatinine levels. In addition, serum NMR metabolomic analysis confirmed decreased renal function in septic mice by showing elevation of metabolites normally cleared by the kidney. Circulating mediators of inflammation are also remarkably increased, demonstrating that septic mice have a generalized, systemic inflammatory condition.

As has been previously described in numerous animal and human studies (36–41), we observed relatively bland kidney histology septic mice. We noted that few structural changes on H&E staining and immunohistochemistry for NGAL and KIM-1 showed minimal staining in CLP kidneys. Compared with our previous research using ischemia-reperfusion and hypoxia models (12, 13), which showed dramatic tissue expression of these protein injury markers, our current findings were particularly underwhelming. We found upregulation of NGAL and KIM-1 gene expression in septic kidney tissue, although this was not observed in our immunohistochemistry staining. This may be due to post-transcriptional regulation of protein production, longer time needed between onset of sepsis and tissue expression of protein products, or just that the baseline expression of these markers is so low in healthy tissue that any expression results in a dramatically elevated relative quantification (RQ) even though the absolute mRNA count and resultant protein expression are still relatively lower than other, more severe kinds of kidney injury. We favor this latter explanation.

Importantly, kidney function in sepsis decreased even though tissue and cellular integrity was largely maintained. Gomez and Kellum (6, 8) have articulated an elegant hypothesis, supported by a substantial body of literature, that RTECs undergo metabolic reprogramming in sepsis, temporarily surrendering advanced functions to preserve viability in the face of physiological instability. Our findings support this understanding of SA-AKI pathophysiology. NMR analysis revealed significant perturbations in multiple metabolic cofactors. Serum and urine abundance of ascorbic acid, carnitine, and metabolites associated with CoA were significantly different in samples collected from septic versus sham mice. In addition, important metabolic pathways integral to mitochondrial function were specifically implicated.

Both serum and urine metabolomics identify altered BCAA metabolism in sepsis. Inflammatory signals associated with sepsis and other critical stressors induce breakdown of skeletal muscle to release amino acids, which can be used to construct new proteins necessary for cellular and organismal stress responses or metabolized for energy. Some amino acids are converted by the liver to glucose or ketones for use by other tissues, whereas BCAAs can be oxidized directly in the mitochondria for ATP production. This occurs primarily in muscle, but brain, liver, kidney, and adipose tissues can also use BCAAs in this way (42). Serum and urine levels of most metabolites associated with BCAA catabolism were decreased in sepsis. Importantly, levels of urinary metabolites were greatly increased in sham mice compared with baseline, indicating that BCAA metabolism is increased in that fasted condition but curtailed in sepsis. Kidney expression of mitochondrial BCAA catabolic enzymes was downregulated, signifying that the kidney relies more on non-BCAA substrates for energy provision in sepsis.

Most serum and urine metabolites associated with ketogenesis, which depends on mitochondrial FAO, are decreased in sepsis. Consistent with this finding, genes associated with fatty acid metabolism (Ppara and Acadm) and mitochondrial function (Ppargc1), where these reactions take place, have decreased expression in sepsis. We and others have previously described this (15, 43). It is important to note that the expression of upstream regulatory proteins HIF-1α and mTOR are increased. These proteins control many cellular functions in the stress state and influence cellular metabolism to shift away from fatty acid toward glucose oxidation and suppress mitochondrial function (44, 45). Our observation that renal cortical FAO rates decrease in septic mice as serum markers of inflammation increase and that these changes associate with increased creatinine levels denote a dependence of renal function on maintenance of FAO. Interestingly, septic mice who maintained lower creatinine levels had higher rates of FAO than even sham mice, indicating that extra energy was required from lipid substrate to maintain adequate kidney function in the physiologically stressful sepsis condition. The heart, like RTECs, relies on FAO as a primary energy source, and we have likewise observed an increase in cardiac FAO in early, mild sepsis (15). It is apparent that as the severity of sepsis increases, FAO rate and renal function decrease together.

The NAD⁺ biosynthetic pathway has received significant attention recently due to its association with various forms of AKI (10, 46–48). We found that urinary quinolinic acid, an NAD⁺ precursor, was elevated in septic mice. Poyan Mehr et al. (49) observed elevation of urinary quinolinate levels in both a mouse ischemic AKI model and a human study of postoperative kidney injury, indicating that this metabolite serves as an important marker of NAD⁺ biosynthetic disruption. We confirmed this observation by showing that a majority of genes related to NAD⁺ metabolism are downregulated in sepsis. The one enzyme that showed a trend toward increased expression in the kidney, Ido1, is expressed predominantly in activated leukocytes to stoke inflammation (50). Ido2, the isoform primarily expressed in the kidney, was one of the most downregulated genes assayed.

Taken together, these findings show that vital mitochondrial metabolic pathways related to renal tubular energy production are disrupted in sepsis and that these changes are associated with alteration of kidney function. Importantly, at least some of these changes are mediated by alterations in gene expression, indicating their role in a coordinated cellular response to sepsis. The gene expression changes observed likely localize to the RTEC. Tissue submitted for gene expression analysis was carefully cut from the renal cortex, which is primarily populated by cells that comprise the metabolically active proximal and distal convoluted tubules. Although some infiltrating leukocytes were present on histological evaluation of septic kidney tissue, we did not observe significant numbers, decreasing the likelihood that changes in gene expression are attributable to a substantially altered tissue cellular composition.

Although kidney expression of all genes in each of these pathways did not correlate exactly with changes in relevant serum and urine metabolites, we saw important, consistent changes in each pathway. Considering that BCAA, fatty acid, and NAD⁺ metabolism occur in many other tissues outside the kidney, differences between metabolite abundance and renal gene expression are expected. In addition, mRNA abundance of constituent enzymes does not directly indicate metabolic flux through any specific pathway due to translational regulation of gene products and post-translational modification of enzymatic activity. Further studies to measure these variables directly in kidney tissue are necessary.

We found that the overwhelming majority of urine metabolites had decreased abundance in sepsis. Although unmeasured, the urine volumes were lower than those produced by sham mice and appeared darker. We have previously observed this in our other models in kidney injury (12, 13). We ascribe the low urine volumes to decreased glomerular filtration rate, likely due to sepsis-related afferent arteriolar vasoconstriction and efferent arteriolar vasodilation mediated by a stress response designed to reduce solute load delivery to the nephron that would subsequently require significant energy expenditure to reabsorb (5, 7). In addition, we previously observed that intraluminal casts may obstruct some flow through the nephron, resulting in decreased urine output (12).

Globally, low levels of urinary metabolites indicate reduced concentrating activity of RTECs. The findings that urinary glucose levels are increased in septic mice known to be hypoglycemic and that levels of urinary creatinine, a compound actively secreted into the nephron, are low illustrate that tubular transport functions are reduced in sepsis. These findings support the concept that renal energy expenditure is decreased as part of a coordinated stress response in sepsis.

Metabolites of potential bacterial origin also had differential abundance between sepsis and sham conditions. Serum 2,3-butanediol levels, a metabolite associated with coliform bacteria that reside in the gut (51) and likely contribute to the polymicrobial sepsis induced by CLP, received the highest ranking VIP score on PLS-DA analysis, indicating preeminent importance in distinguishing septic mice from sham mice. Indole and methylindole, also putative bacterial metabolites (52, 53), were less significantly altered.

The limitations of this study include those inherent to any animal model system, principally the translatability of finding to human disease. Mouse models of sepsis have received particular attention in the past (54). There are, however, great similarities between mammalian model pathophysiology and human disease with many examples of successful knowledge transfer and no other platform currently allows for the analysis of complex, intact biological systems (55, 56). Furthermore, we have developed a clinically relevant model of polymicrobial sepsis using fluids, antibiotics, and analgesics (43). We will also seek to confirm and advance the findings reported here in a future study of human samples.

Another limitation is fundamental to the nature of NMR-based metabolomics. After sample NMR spectra are collected, metabolite identification must be manually assigned to NMR features, and there is always a degree of uncertainty in those assignments. Many features remain unidentified even after careful review. To address this, we have given a confidence rank to each of our assignments that indicates a relative degree of certainty (33). Ultimately, we will conduct future experiments to confirm biochemical identities of centrally important metabolites.

Conclusions

We have shown in a mouse model of polymicrobial sepsis that sepsis induces significant changes in renal function directly related to inflammatory signaling. NMR metabolomic analysis of serum and urine samples showed that central metabolic cofactors and pathways related to mitochondrial function are implicated in these differences. Gene expression of enzymes related to multiple pathways important for cellular energy provision is suppressed. Specifically, we show that increased inflammation predicts decreased renal cortical FAO and reduced kidney function in septic mice. BCAA metabolism, FAO, and de novo NAD⁺ synthesis are potentially targetable pathways for future metabolic interventions in SA-AKI.

SUPPLEMENTAL DATA

Raw data, analysis scripts, and Supplmental Material are available on the Open Science Framework at https://doi.org/10.17605/osf.io/A5W4E.

GRANTS

This work was supported by National Institute of Health Grant 1K08HL133377 (to S.W.S.) and Grant P50DK096418 (to P.D.).

DISCLOSURES

P.D. is a co-inventor on submitted patents for the use of NGAL as a biomarker of kidney injury. P.D. is on the medical advisory board of BioPorto, Reata, Dicerna. None of the other authors has any conflicts of interest, financial or otherwise, to disclose.

AUTHOR CONTRIBUTIONS

S.W.S., P.D., and M.A.K. conceived and designed research; S.W.S., S.X., L.B., Q.M., A.K., and P.L. performed experiments; S.W.S., S.X., Q.M., P.D., and M.A.K. analyzed data; S.W.S., S.X., P.D., and M.A.K. interpreted results of experiments; S.W.S. prepared figures; S.W.S. drafted manuscript; S.W.S., S.X., P.D., and M.A.K. edited and revised manuscript; S.W.S., S.X., L.B., Q.M., A.K., P.L., P.D., and M.A.K. approved final version of manuscript.