Decreased Renal Gluconeogenesis Is a Hallmark of Chronic Kidney Disease

Significance Statement

The ability to produce glucose from nonhexose precursors is a main metabolic function of renal proximal tubule (PT) cells. PT cells adapt metabolically during CKD, but little is known about gluconeogenesis in chronically injured PT cells. Our study demonstrates the progressive loss of gluconeogenesis enzymes in animal models and in CKD patients in parallel to global change in metabolic pathway expression and activation of injury pathways. This alteration is not only due to loss of PT cells but has systemic repercussions on glucose and lactate levels in experimental and human CKD. In retrospective human studies, gluconeogenesis downregulation predicted CKD progression. This work provides new evidence for metabolic regulation during CKD and the functional effect.

Visual Abstract

Abstract

Introduction

CKD is associated with alterations of tubular function. Renal gluconeogenesis is responsible for 40% of systemic gluconeogenesis during fasting, but how and why CKD affects this process and the repercussions of such regulation are unknown.

Methods

We used data on the renal gluconeogenic pathway from more than 200 renal biopsies performed on CKD patients and from 43 kidney allograft patients, and studied three mouse models, of proteinuric CKD (POD-ATTAC), of ischemic CKD, and of unilateral urinary tract obstruction. We analyzed a cohort of patients who benefitted from renal catheterization and a retrospective cohort of patients hospitalized in the intensive care unit.

Results

Renal biopsies of CKD and kidney allograft patients revealed a stage-dependent decrease in the renal gluconeogenic pathway. Two animal models of CKD and one model of kidney fibrosis confirm gluconeogenic downregulation in injured proximal tubule cells. This shift resulted in an alteration of renal glucose production and lactate clearance during an exogenous lactate load. The isolated perfused kidney technique in animal models and renal venous catheterization in CKD patients confirmed decreased renal glucose production and lactate clearance. In CKD patients hospitalized in the intensive care unit, systemic alterations of glucose and lactate levels were more prevalent and associated with increased mortality and a worse renal prognosis at follow-up. Decreased expression of the gluconeogenesis pathway and its regulators predicted faster histologic progression of kidney disease in kidney allograft biopsies.

Conclusion

Renal gluconeogenic function is impaired in CKD. Altered renal gluconeogenesis leads to systemic metabolic changes with a decrease in glucose and increase in lactate level, and is associated with a worse renal prognosis.

Kidney function is traditionally determined by GFR but the renal tubule compartment is increasingly recognized as a key player in the pathophysiology and prognosis of CKD.^1–4 Modifications of tubule metabolism are currently considered as hallmarks of kidney disease. During CKD, decrease of fatty acid oxidation (FAO) and NAD⁺ content are well described in proximal tubule (PT) cells and FAO loss is prognostic of the evolution of kidney disease.^5–11 The estrogen related receptor α (ESRRA) and the peroxisome proliferator-activated receptor α (PPARα) have been implicated in FAO regulation during CKD.¹² We have shown that during AKI, in addition to NAD⁺ deficiency, loss of renal gluconeogenesis was a key metabolic phenotype with major systemic repercussions.^10,13,14 Whether renal gluconeogenesis is affected during CKD is unknown. Although alteration of renal gluconeogenesis during CKD has been suspected to participate in the phenotype of “burnt-out diabetes,” it has never been demonstrated.¹⁵

The kidney and the liver are the main organs able to produce glucose from nonhexose precursors through the process of gluconeogenesis.^13,16,17 Gluconeogenesis encompasses a series of enzymatic reactions that are usually regulated in opposition to glycolysis.¹³ Gluconeogenesis is classically regulated by hepatocyte nuclear factor 4 α (HNF4α),^18,19 PPARα,²⁰ and FOXO1.²¹ PT cells are the only cell type able to perform this process in the cortical kidney area, harboring the expression of rate-limiting enzymes such as fructose-1,6-bisphosphatase (FBP1), phosphoenolpyruvate carboxykinase 1 (PCK1), or glucose-6-phosphatase (G6PC) and using lactate as the main substrate.²²

Renal gluconeogenesis appears particularly relevant in acute conditions, because it is also regulated by pH and stress hormones, and provides 40% of systemic gluconeogenesis in fasting conditions.^17,23 In diabetic patients, the ACCORD study demonstrated that hypoglycemia risk was associated with mortality, and that the risk was highest in patients with low renal function.^24,25 In-hospital hypoglycemia has also been associated with CKD, in both diabetic and nondiabetic patients, implying that factors other than drug accumulation play a role in CKD-related hypoglycemia risk.²⁶ In addition to glucose production, renal gluconeogenesis also implies precursor clearance. The main substrate for gluconeogenesis in the kidney is lactate, followed by glutamine and glycerol.²² The kidney is thus considered as a main systemic lactate sink.²⁷ Hyperlactatemia is a well-known factor for mortality in the intensive care setting, and abnormal lactate clearance is an important risk factor for mortality in acute disease.²⁸

In this study, we investigate the regulation of renal gluconeogenesis and its consequences during CKD.

Methods

Animal Studies

All animal studies were approved by the Institutional Ethical Committee of Animal Care in Geneva and cantonal authorities, Institutional Animal Care and Use Committees at the University of Southern California, for the C57BL6/J (Six2TGC Rosa26rtTA pTREH2-GFP) mouse model, and Institute of Oncology Research in Bellinzona, Switzerland, for the C57BL6/J Pten ^pc−/−. The corresponding animal authorization numbers are GE-70/19, GE-181/19, and TI-04/2017. Mice were housed at 20°C with free access to food and water. For proteinuric mouse model POD-ATTAC, 8-week-old male POD-ATTAC mice on FVB or C57BL/6J background were used for the proteinuric progressive tubulo-interstitial fibrosis model. Dimerizer (Clontech Laboratories, Inc.) was prepared according to the product datasheet and mice were injected once with 0.5 µg/g or five times with 0.2 µg/g of dimerizer and euthanized respectively after 7, 14, or 28 days. Control mice were either littermate mice without the transgene and injected with dimerizer. For the unilateral ureteral obstruction (UUO) mouse model, 8-week-old male mice (C57BL/6J) were anesthetized with isoflurane and underwent unilateral ligature of the ureter as described previously.²⁹ Euthanasia and organ collection were performed 7 days after the surgery. The contralateral nonobstructed kidney was used as a control.

Single-Nucleus RNA Sequencing Analysis

Gene Expression Omnibus (GEO) datasets were downloaded from http://www.ncbi.nlm.nih.gov/geo/. GSE139107 included a sum of 24 renal samples characterized by single-nucleus RNA sequencing (snRNA-seq) as previously published.³⁰ GSE151167 included five samples characterized by snRNA-seq as previously described and reported in this paper.¹⁴ The data processing was executed by running the script on the Ente Ospedaliero Cantonale server, Switzerland. Seurat v3.2.0 in R v4 was used for downstream analyses, including normalization, scaling, and clustering of nuclei. First, we analyzed each dataset separately and excluded nuclei with <150 or >8000 genes detected. We also excluded nuclei with a relatively high percentage of unique molecular identifiers mapped to mitochondrial genes (>1) and ribosomal genes (>1, for normal kidney sample; and >2, all other samples). We performed curated doublet removal based on known lineage-specific markers.

The samples from different datasets were integrated to avoid batch effect using Seurat standard work flow split by dataset resulting in a total of 64,316 renal nuclei from n=33 samples, see Supplemental Table 1 for a detailed composition. Following ScaleData, RunPCA, FindNeighbors, and FindCluster at a resolution of 0.5 were performed. FindAllMarkers generated the list of genes differentially expressed in each cluster compared with all other cells, within the major subgroups defined (nephron cells, collecting duct, other cells) based on the Wilcoxon rank sum test and limiting the analysis with a cutoff for minimum log fold change difference (0.3) and minimum cells with expression (0.3) (see Supplemental Table 2). Cluster reassignment was performed based on manual review of lineage-specific marker expression.

Secondary Seurat analyses on 39,910 PT cells used the SubsetData function to create new r-objects from cohorts of primary analysis. For data visualization we used RunUMAP and FeaturePlot from Seurat and dittoHeatmap from dittoSeq package. Gluconeogenesis score enrichment was performed using the escape package v1 with the R-MMU-70263 gene set from the Reactome pathway database.

RNA Sequencing of Murine Kidneys

RNA quantification was performed from control and POD-ATTAC kidney mice with a Qubit fluorimeter (Thermo Fisher Scientific) and RNA integrity assessed with a Bioanalyzer (Agilent Technologies). The TruSeq mRNA stranded kit from Illumina was used for library preparation with 700 ng of total RNA as input. Library molarity and quality were assessed with the Qubit and TapeStation using a DNA high sensitivity chip (Agilent Technologies). Libraries were pooled at 2 nM and loaded for clustering on a single-read Illumina flow cell for an average of 25 million reads per sample. Reads of 100 bases were generated using the TruSeq SBS chemistry on an Illumina HiSeq 4000 sequencer. After quality control, reads were mapped to the GRCm38 genome using TopHat software. Table of counts were thus generated with HTSeq. Downstream analyses were performed using edgeR. Raw gene counts were normalized by the trimmed mean of M-values (TMM) method using edgeR package and counts were expressed in counts per million. Comparisons of gene expression were performed using a gene-wise negative binomial linear model with the quasi-likelihood test. Values are expressed as TMM-normalized counts per million.

Western Blot Analysis

Western blot using lysates from renal cortex were performed with the following antibodies: rabbit polyclonal FBP1 (1:1000) and rabbit polyclonal PCK1 (1:1000) (Supplemental Table 3). Cortex area was dissected and lyzed in imidazole buffer (imidazole 25 mM, sucrose 300 mM, Triton X-100 1%, sodium fluoride 50 mM, and sodium orthovanadate 2 mM) supplemented with protease inhibitors (Roche), triturated, and centrifuged. A total of 30 µg of proteins were mixed with sample buffer (Tris-HCl 50 mM, glycerol 10%, sodium dodecyl sulfate 1%, bromophenol blue 0.01%, and 2-mercaptoethanol 10 mM) and heated for 5 minutes at 95°C. Each sample was applied equally to 10% bis-acrylamide gel at 100 V for 1 hour then transferred to nitrocellulose membrane (GE Healthcare, Life Sciences) during 1 hour at 100 V. The membrane was briefly washed in Tris-buffered saline (TBS)-Tween (Tris-HCl 50 mM, sodium chloride 150 mM, and Tween 20 0.1%) and allowed to block at room temperature in 5% milk-TBS-Tween during 1 hour. After washing three times in TBS-Tween, the membrane was blotted with primary antibody diluted in 5% milk-TBS-Tween overnight at 4°C. Then, the membrane was washed three times in TBS-Tween and incubated for 1 hour with goat anti-rabbit horseradish peroxidase secondary antibody (1:5000). To detect protein expression, ECL detection reagent WesternBright Quantum (Advansta) was applied to the membrane for 5 minutes and chemiluminescence was detected with PXi gel imaging system (Syngene). ImageJ software (National Institutes of Health) was used to quantify band density and protein expressions were normalized to Ponceau S staining. Results are expressed as fold change in protein expression compared with the control/sham samples.

Histologic Analyses

After mice euthanasia, kidneys were fixed immediately in 4% paraformaldehyde overnight, then washed three times in PBS solution. After that, they were embedded in paraffin and cut in 4- to 5-μM longitudinal sections, which were mounted on glass slides.

To quantify fibrosis, a representative slide of the kidney in its median part was chosen and Sirius Red staining was performed. Briefly, the slides were deparaffinized in xylene substitute Neo-Clear (Merck Millipore) and rehydrated in successive ethanol baths (100%, 90%, and 70%). After, hematoxylin was processed during 20 minutes, rinsed with water, then stained with Sirius Red (Sirius Red F3B 0.1% in picric acid, saturated aqueous solution) for 50 minutes. After that, the slides were quickly rinsed successively with acidified water (acetic acid 0.5% in water), ethanol 100%, and Neo-Clear. After staining, the slides were scanned with an Axioscan image scanner (Zeiss) at 20× magnification. To quantify kidney fibrosis, the cortical area was defined and analyzed with Definiens Tissue Phenomics software.

For FBP1 and PCK1 immunostaining, samples were deparaffinized and rehydrated as previously described.³¹ Antigen retrieving was performed with citric acid buffer at pH 8 (10 mM). The samples were blocked 1 hour with 5% BSA (Calbiochem) and incubated overnight with the following primary antibodies: anti-FBP1 (rabbit, 1:100) or anti-PCK1 (rabbit, 1:100) (Supplemental Table 3). Then, the samples were incubated with secondary goat anti-rabbit horseradish peroxidase (1:5000) for 1 hour at room temperature and DAB revelation was performed using Dako kit (Dako). A hematoxylin counterstain was applied and the slides were scanned with Axioscan image scanner.

Real-Time Quantitative PCR

Total RNA was extracted from renal cortex area with TRIzol reagent (Invitrogen) according to the manufacturer’s instructions. After RNA dosage using Nanodrop (Thermo Fisher Scientific), 1 μg of total RNA was reverse transcribed using qScript cDNA supermix (Quantabio). cDNA was used to perform real-time quantitative PCR (RT-qPCR) in triplicate using PowerUp SYBR Green Master Mix (Applied Biosystems) and StepOnePlus Real-Time PCR System (Applied Biosystems) or a QuantStudio 5 Real-Time PCR System (Thermo Fisher Scientific). The 2^{−δ δCT} method was used to analyze the relative changes in gene expression levels. Primers used in RT-qPCR are listed in Supplemental Table 3.

Transcutaneous GFR Measurement

To measure GFR, FITC-sinistrin clearance was used. Briefly, mice were anesthetized with isoflurane then shaved on the right flank. A mini camera was attached on the flank and a solution of FITC-sinistrin (Fresenius Kabi) injected in the tail at 0.35 g/kg after 2 minutes of basal recording. After 90 minutes of recording, the mice were anesthetized with isoflurane and the camera was removed. The recording of FITC-sinistrin was analyzed with MPD Lab software (Mannheim Pharma and Diagnostics) and GFR was calculated using the appropriate formula.³² GFR was then expressed in milliliters per minute per kilogram of body weight and normalized to control.

FBP1 and PCK1 Enzymatic Activity Assay

FBP1 and PCK1 enzymatic activity from renal cortex was assessed using Fructose-1,6-Bisphosphatase Activity Assay Kit (Biovision) and Phosphoenolpyruvate Carboxykinase 1 Activity Assay Kit (Biovision), respectively (Supplemental Table 3). Renal cortical protein content was measured by Bradford protein assay. Each measure was performed in duplicate with a control background and results were expressed as specific activity in milli units per milligram.

Mouse PT Cells Isolation

Primary PT cells from mice were obtained as described previously.³³ Briefly, mice were euthanized and their kidneys harvested. Whole kidneys were roughly chopped and mechanically dissociated using the gentleMACS cell dissociator (Miltenyi Biotec). Tubular-enriched fractions were obtained using anti-prominin-1 microbead-conjugated antibodies and autoMACS cell separator (Miltenyi Biotec). Tubular-enriched fractions were immediately lyzed with TRIzol reagent and total mRNA was extracted according to the manufacturer’s instructions. Quality control was performed by RT-qPCR in order to verify the enrichment in PT cells in the isolated fraction (Supplemental Figure 4).

Glucose Response to Lactate Tolerance Test

To perform glucose response as previously described,³⁴ POD-ATTAC mice (C57BL/6J) were starved for 16 hours before the experiment and then injected intraperitoneally with 10% lactate solution at 1.5 g/kg diluted in PBS. Glucose levels were measured with a glucometer (Bayer) and a lactate meter (Nova Biomedical) in tail vein sampling before and 15, 30, 60, 90, and 120 minutes after intraperitoneal injection (Supplemental Table 3). The glucose course over time was summarized as area under the curve and compared among groups using a t test. A self-starting nonlinear asymptotic regression model was used to determine lactate clearance. Group effect was assessed by an F test for nonlinear regression model.

Kidney Perfusion

After 28 days of proteinuric induction, POD-ATTAC mice were anesthetized with pentobarbital injection (Inresa) and euthanized. To isolate the kidney, the superior mesenteric artery, celiac artery, abdominal artery, and inferior vena cava were ligatured. Two catheters were inserted in the vein and aorta, respectively, corresponding to output and input flow. Kidney perfusion solution was based on Prismasol 4 enriched with hematocrit 10%, vitamins, energetic substrates, and BSA for a concentration of 20 g/L. The detailed composition is provided in Supplemental Table 4. The perfusion solution was oxygenated with an O₂/CO₂ (95%/5%) mix at a flow rate of 2 L/min through a membrane oxygenator (Medos Hilite). The oxygen flow rate was adjusted into the oxygenator to maintain the oxygen partial pressure in the perfusion solution around 30 kPa, whereas the oxygen partial pressure in the renal venous was around 16 kPa. The perfusion solution was perfused in the kidney from aorta to inferior vena cava at 700 µl/min. After 20 minutes of kidney flushing, the perfusion flow rate was reduced to 350 µl/min and samples were collected through the two catheters at 60 minutes after flushing. HCO₃, glucose, lactate, and VO₂ were measured with ABL90 Flex analyzer (Radiometer).

Renal Glucose Release

Renal glucose release was measured in isolated perfused kidneys. As kidneys are both consumers and producers of glucose, we used the isotopic dilution technique, aiming at measuring the renal dilution of labeled glucose by the nonlabeled produced glucose. Deuterated glucose (D2-glucose) isotope was preferred because it cannot be recycled from gluconeogenesis and only at a very low rate from glycogenolysis.³⁵ D2-glucose was added to the kidney perfusion solution at a 1:1 ratio with native glucose, for a final concentration of 6 mmol/L. For D2-glucose quantification, samples were first prepared by the addition of ¹³C-glucose for internal control. Enrichment in D2-glucose was performed by LC-HRMS on an HPLC Agilent 1290 with DAD connected to Agilent Q-TOF 6538. HPLC was carried out on an Agilent Poroshell 120 HILIC-Z (100 × 2.1 mm ID, 2.7 µm) column connected to an Agilent Infinity 1290 HPLC. The solvent system was the following: A, 25 mM of ammonium formate and 0.1% formic acid in H₂O; and B, acetonitrile. The gradient program began and stayed with 97% B during 1 minute, then ramped to 60% B at 13 minutes, increased to 30% B in 2 minutes, held at 30% during 2 minutes, returned to the initial conditions, and kept constant for 3 minutes. The flow rate was 0.5 ml/min and the injection volume was 4 µl. All compounds’ responses were measured in electrospray ionization (ESI) and calibrated externally. The electrospray ionization gas temperature was 300°C, Vcap was set at 3000 V, drying gas was set at 12 L/min, and nebulizer gas at 30 psig. The fragmentor was set at 135 V. The HRMS spectrum was registered at 2 Hz in the mass range of 100–1200 m/z with internal calibration. The LC flux was sent to the mass spectrometer between 2 and 7 minutes and to the trash at the end of the time. MassHunter software was used for data processing. The isotopic abundances of the monoisotopic ions m/z 179 (glucose), 181 (D2-glucose) and 185 (¹³C-glucose) were extracted from the glucose peak at the retention time of 5.05 min corresponding to the retention time of ¹³C-glucose. Ratio calculations were performed after subtracting the natural abundance of M+2 from glucose.

The renal glucose release (RGR) was calculated as previously described¹⁴:

$$\begin{array}{l}\mathit{\text{RGR}}=\mathit{\text{RBF}}\cdot \left\{\left(\left[\mathit{\text{Aglucose}}\right]-\left[\mathit{\text{Vglucose}}\right]\right)+\left(\left[\mathit{\text{AGlucose}}\right]\cdot \mathit{\text{fx}}\right)\right\}\\ \hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\text{with}\hspace{0.17em}\mathit{\text{fx}}=\frac{\left[\mathit{\text{AD2glucose}}\right]-\left[\mathit{\text{VD2glucose}}\right]}{\left[\mathit{\text{AD2glucose}}\right]}\end{array}$$

where RBF is the renal blood flow, [Aglucose] is the renal arteria concentration of glucose, [Vglucose] is the renal venous concentration of glucose, [AD2glucose] is the renal arteria concentration of D2-glucose, and [VD2glucose] is the renal venous concentration of D2-glucose.

Comparisons were performed using a linear mixed model for each group with a random intercept defined for each animal. The negative fx values that were encountered were set to 0 and included in the calculations to avoid introducing a bias.

Fibrosis Progression

The relationship between fibrosis progression and mRNA expression was fitted using a robust linear model. Fibrosis progression was defined as the difference between the Banff interstitial fibrosis and tubular atrophy (IFTA) score (ci+ct) at 3 and 12 months after transplantation.

eGFR Evolution

eGFR evolution over time for CKD patients was fitted using a linear mixed model with a random intercept defined for each patient and interaction term between time and group of interest, without specified correlation structure.

Gene Set Enrichment Analysis

Gene set enrichment analysis was conducted using RNA sequencing data from kidney allograft recipients and classified as recovery or CKD pattern. Glycolytic and gluconeogenic gene sets from Reactome pathway database were used.

Renal Flux

Lactate and glucose net renal flux (RnetFlux) was calculated as follows:

$$\mathit{\text{RnetFlux}}\left(\mu \text{mol}/\text{min per kg}\right)=(\mathit{\text{venous concentration}}\times \mathit{\text{RaBF}}–\mathit{\text{arterial concentration}}\times \mathit{\text{RvBF}})/\mathit{\text{BodyWeight}}$$

with

$$\mathit{RaBF}\left(\text{ml}/\text{min}\right)=2\times \left(\mathit{\text{unilateral renal vein blood flow}}\right)+\mathit{\text{urine flow}}$$

and

$$\mathit{RvBF}\left(\text{ml}/\text{min}\right)=2\times \left(\mathit{\text{unilateral renal vein blood flow}}\right)$$

where RaBF is the renal arterial blood flow and RvBF is the renal venous blood flow.

Human Studies

For the intensive care unit (ICU) cohort, we reanalyzed our retrospective cohort previously described.¹⁴ Briefly, all patients admitted to the ICU of the Geneva University Hospitals between January 2007 and December 2018 and older than 18 years were included. We thus classified every patient in a metabolic pattern, defined as the most represented metabolism status during the whole ICU stay. Metabolism status was defined according to the glucose and lactate levels, in five groups: baseline (lactate levels below median and with glucose levels between the 25th and the 50th percentile); impaired metabolism (lactate levels above the median with glucose level below the 75th percentile); isolated low glucose level (lactate levels below median with glucose levels below the 75th percentile); isolated high glucose level (lactate levels below median with glucose levels above the 25th percentile); and stress response (lactate levels above median and glucose levels above the 75th percentile). Patients with low stable renal function were defined as patients with eGFR (creatinine CKD-EPI equation) at ICU admission below 60 ml/min per 1.73 m², stable during the whole ICU stay (i.e., maximal variation <10%). CKD patients were defined as patients with eGFR (creatinine CKD-EPI equation) at ICU admission below 60 ml/min per 1.73 m², stable during the whole ICU stay (i.e., maximal variation <10%) and with an eGFR 3 months after ICU discharge <60 ml/min per 1.73 m². A propensity score for low stable renal function was estimated with a nonparsimonious logistic regression. The variables included were age, admission reason, weight, insulin, dexmedetomidine, bilirubin, feeding, GOT (glutamic-oxaloacetic transaminase), GPT (glutamic pyruvic transaminase), sex, metformin, dobutamine, propofol, epinephrine, and NE infusion. Values at ICU admission were used. Patients with or without low stable renal function were thus matched according to the logit of the propensity score. We used matching strategy based on nearest neighbor matching without replacement (controls cannot be used more than once) using calipers of width equal to 0.1 SD. An absolute standardized difference <10% was considered to support the assumption of balance between groups. Mortality analyses were performed and differences between groups were assessed using conditional logistic regression stratified on matched patients.

For the cardiac surgery (renal arterio-venous measurement) cohort, renal vein catheterization was performed in postcardiac surgery patients or in patients with heart failure with renal impairment as previously described.^36,37 After placement of the catheter in the left renal vein, renal blood flow was measured by the retrograde thermodilution technique. For measurement of GFR, an intravenous priming dose of the filtration marker, 51Cr-EDTA, was given, followed by infusion at a constant rate individualized to body weight and serum creatinine. Arterial and renal vein blood samples were obtained twice at a 30-minute interval. The study was performed according to the Declaration of Helsinki principles and was approved by the Human Ethic Committee of the University of Gothenburg.

For renal allograft RNA sequencing database, RNA sequencing analyses were performed in kidney allograft recipients as previously described.³⁸ Briefly, 42 post-transplantation patients with protocol biopsies were enrolled at the University Hospitals of Leuven. For each of them, a protocol biopsy was performed at four different time points: before implantation, after reperfusion, and 3 and 12 months after transplantation. To mitigate the sample-specific effects, raw gene counts were normalized by the TMM method using edgeR package and counts were expressed in counts per million. Comparisons of gene expression were performed using a gene-wise negative binomial linear model with the quasi-likelihood test. Values are expressed as TMM-normalized counts per million. Transcriptomic profiles of 3- and 12-month biopsies were classified in recovery, early transition, and CKD patterns as previously described.³⁸ Raw data are available at GEO accession GSE126805. A correlation matrix using the Spearman coefficient was carried out, with TMM-normalized and log transformed counts per million.

Human kidney biopsy microarray data were issued from the European Renal cDNA Bank–Kröner-Fresenius Biopsy Bank (CKD: GSE99340; living donor: GSE32591, GSE35489, and GSE37463).^39,40 Kidney biopsies, RNA isolation, preparation, and microarray analysis were performed as described previously.³¹ Biopsies from different kidney diseases were collected (cadaveric donor, tumor nephrectomy, diabetic nephropathy, thin basement disease, minimal change disease, hypertensive nephropathy, IgA nephropathy, focal segmental glomerulosclerosis, membranous nephropathy, lupus nephritis, and ANCA-vasculitis) and grouped by CKD stages (CKD1–5) using the CKD-EPI equation (CKD1: n=56; CKD2: n=46; CKD3: n=37; CKD4: n=26; CKD5: n=10). Healthy living donors’ biopsies were used as controls (living donor, n=42). The significance analysis of microarrays method was applied using TiGR (MeV, Version 4.8.1) to identify differentially expressed genes. A q-value <5% was considered to be statistically significant.

Human kidney biopsies were issued from the tissue bank of the Service of Pathology, University Hospital of Geneva. Biopsies from different kidneys were collected. Twelve samples were included for this study with diverse kidney disease and various degrees of renal function and fibrosis. Standard analyses were performed in all samples and fibrosis was graded by an experimented renal pathologist. Kidney biopsies were fixed in formaldehyde and embedded in paraffin, then sections (3 μm thick) of each kidney biopsy were incubated with a polyclonal rabbit anti-FBP1 antibody (1:100) as described previously.⁴¹ Slides were digitized and evaluated by two nephrologists in a blinded fashion. The 12 biopsies were ranked by FBP1 staining intensity from the least intense to the most intense. Each patient gave informed consent before enrollment in the study. The institutional ethical committee board approved the clinical protocol (CEREH number 03-081). The research was performed according to the Declaration of Helsinki principles.

Cell Culture

The renal cortical cell isolation protocol was adapted from Legouis et al.¹⁴ Briefly, the renal cortex area was dissected from the kidney and chopped manually, then dissociated with gentleMACS dissociator (Miltenyi Biotec). The renal cortex mixture was rinsed and filtered through a 0.22-µm filter, then centrifuged at 2000 × g for 5 minutes. The supernatant was removed and replaced by fresh medium DMEM/F12 (GIBCO). Cortical cells were exposed with albumin (60 mg/ml), TGFβ (25 ng/ml), LPS (100 ng/ml), or EGF (100 ng/ml) for 12 hours (Supplemental Table 3).

Data Availability

The accession number for the snRNA-seq data reported in this paper is GSE163863. The public datasets used are from GEO with the accession numbers GSE151167 and GSE139107. The accession numbers for the human kidney biopsy microarray (Kröner-Fresenius Biopsy Bank) are CKD: GSE99340; and living donor: GSE32591, GSE35489, and GSE37463. The accession number for the renal allograft RNA sequencing database is GSE126805. The RNA sequencing of kidney cortex from POD-ATTAC mice is available upon request from 10.5281/zenodo.4450133.

Statistical Analyses

Data were as mean ± 95% confidence interval estimated via bootstrapping. They were compared by t tests or Mann–Whitney tests depending on their class and their distribution.

A P value <0.05 was considered significant. All P values were two-tailed and, if applicable, adjusted for multiple comparisons using Benjamini–Hochberg correction. Software are listed in Supplemental Table 3.

Results

Alterations of Gluconeogenesis Is an Early Feature of CKD

To determine whether gluconeogenesis is regulated during CKD, we analyzed a microarray dataset from tubulo-interstitial microdissections of biopsies issued from the European Renal cDNA Bank–Kröner-Fresenius Biopsy Bank at different stages of kidney disease (CKD1: n=56; CKD2: n=46; CKD3: n=37; CKD4: n=26; CKD5: n=10) and compared them to control living donors (n=42). The gluconeogenic transcripts (PC, PCK1, FBP1, and G6PC) were downregulated according to CKD stages, whereas glycolysis genes (PKM and HK1) showed opposite regulation. These changes were occurring in parallel to the known downregulation of FAO genes (ACAA1, ACACB, ACOX1, and ACOX2). Gluconeogenesis is known to be regulated by the transcription factor HNF4α, and other key metabolic regulators, such as PPARα and FOXO1. HNF4α and PPARα genes displayed a stage-specific downregulation during CKD, in coherence with gluconeogenesis regulation, whereas some classic markers of kidney damage (MYC, NFKB1, TGFβ1) were increased (Figure 1A and Supplemental Figure 1A). Thus, loss of the gluconeogenesis pathway at the mRNA level is a key metabolic feature of progressive CKD.

Figure 1.

Alterations of the gluconeogenesis pathway in CKD. (A) Analysis of gluconeogenic (PC, FBP1, PCK1, G6PC), glycolytic (HK1, PKM), fatty acid oxydation (ACAA1, ACACB, ACOX1 ACOX2) and key regulator (HNF4α, PPARα, FOXO1, PGC1α) genes in the Affymetrix microarray expression dataset obtained in the European Renal cDNA Kröner-Fresenius Biopsy Bank, sorted by CKD stage (CKD1, n=56; CKD2, n=46; CKD3, n=37; CKD4, n=26; CKD5, n=10). Biopsies from kidney donors (n=42) are used as controls. *q<0.05. (B) Representative immunostaining of FBP1 in 12 kidney biopsies classified according to fibrosis level by an experienced renal pathologist (fibrosis <20%, n=4; between 40% and 50%, n=3; and >70%, n=4). (C) Association between eGFR by creatinine CKD-EPI equation and ranking of FBP1 expression established blindly by two nephrologists in kidney biopsies from 12 CKD patients (fibrosis <20%, n=4; between 40% and 50%, n=3; and >70%, n=4), fitted through linear regression. (D) Gene set enrichment analysis. Top panel displays the enrichment score (ES), defined as the highest deviation from zero in each gene set. A positive ES indicates gene set enrichment whereas a negative ES indicates gene set depletion. The middle panel shows the position of each gene from the gene sets in the ranked list of genes. The bottom panel shows the value of the ranking metric, based on the log fold change between kidney allograft biopsies classified as CKD (n=27) compared with those classified as recovery (n=23). A positive value indicates an increase in gene expression and a negative value indicates a decrease in gene expression in the CKD group compared with the recovery group. Gluconeogenesis (gray) and glycolysis (orange) pathways from the Reactome database were used as reference gene sets. The input metric was the log fold change of gene expression. Adjusted P value for gluconeogenesis is P=0.0015 and for glycolysis is P=0.109. (E) Pearson correlation between metabolic regulators (HNF4α, PPARα, FOXO1, ESRRA) and gluconeogenic (PCK1, FBP1, PC, G6PC) or glycolytic enzyme (PKM, HK1) gene expression in kidney allograft biopsies classified as CKD (n=27). (F) Pearson correlation coefficient between the expression of different biologic pathways and FBP1 and PCK1 expression as assessed by RNA sequencing in biopsies from a cohort of post-transplant in patients with CKD profile (n=27). (G) mRNA expression of gluconeogenic (Pcx, Fbp1, G6pc, Pck1) and glycolytic (Hk1, Pkm) genes after 12 hours of TGFβ (25 ng/ml), albumin (60 mg/ml), or combined, with and without TGFβ inhibitor (SB-431542) (25 µM) treatment in primary culture of cortical cells. n=7, *p<0.05. Results are presented with error bars showing mean ± bootstrapped confidence intervals.

To further analyze the regulation at the protein level, we performed immunostaining for the gluconeogenesis rate-limiting enzyme FBP1 in 12 random kidney biopsies from CKD patients with kidney disease of various origins. We confirmed a global decrease of the protein expression in kidneys with chronic injury as assessed by renal fibrosis (Supplemental Table 5), mostly in atrophic tubules. The ranking of the intensity in FBP1 staining in biopsies performed blindly correlated negatively with patients’ eGFR (Figure 1, B and C) and to a lower extent with fibrosis (Supplemental Figure 1B).

Kidney allograft recipients are a different group of patients usually harboring CKD in whom we wanted to confirm our observations. RNA sequencing data from kidney allograft biopsies, sampled at 3 and 12 months after transplantation, were classified as recovery (n=23), early transition to CKD (n=22), and CKD (n=27), according to their transcriptomic pattern.³⁸ In coherence with the data in native kidney disease, kidney allograft recipient biopsies harboring features of CKD displayed decreased enrichment in the gluconeogenic pathway whereas the glycolytic pathway was enriched (Figure 1D). In the transplant biopsies, we studied the association between the expression of key genes involved in gluconeogenesis and classically described regulatory transcription factors. As expected, HNF4α and PPARα expression was strongly and positively associated with gluconeogenesis gene expression (Figure 1E).

In order to determine further the upstream mechanism of gluconeogenesis inhibition during CKD, we studied different pathways associated with FBP1 and PCK1 expression in biopsies from kidney allograft patients. FBP1 and PCK1 expression were negatively associated with several cellular stress pathways such as PI3K, TGFβ, MAPK, and EGFR (Figure 1F). The TGFβ profibrotic cytokine is a key factor in CKD evolution and decreased FAO in kidney tubule cells,⁶ downregulates HNF4α,⁴² and is expressed in kidney transplant biopsies transitioning to CKD (Supplemental Figure 1C). To determine a potential role of stress pathways in gluconeogenesis regulation, we exposed primary cells isolated from mouse kidney cortex with LPS (100 ng/ml) or EGF (100 ng/ml) for 12 hours (Supplemental Figure 1D). Analysis of gluconeogenic genes (Pcx, Pck1, Fbp1, and G6pc) did not show any regulation at the mRNA level. On the other hand, incubation with 25 ng/ml of TGFβ or 60 mg/ml of albumin during 12 hours induced downregulation of Pcx, Pck1, Fbp1, and G6pc transcripts. Nevertheless, the association of albumin and TGFβ was not additive in gluconeogenesis gene downregulation. Treatment with TGFβ inhibitors reversed this effect, suggesting that the downregulation of gluconeogenesis genes is TGFβ-dependent (Figure 1G). These findings are consistent with gluconeogenesis loss being an early feature of CKD, partly driven by TGFβ.

Reduced Gluconeogenesis Is a General Feature of Experimental CKD

To better evaluate the regulation and systemic consequences of gluconeogenesis during CKD, we investigated renal glucose metabolism in two animal models of renal injury: a model of glomerular damage leading to severe proteinuria⁴³ (POD-ATTAC) and a model of renal fibrosis induced by UUO. POD-ATTAC mouse model mimics chronic glomerulopathies such as focal segmental glomerulosclerosis and the chronic toxicity of proteinuria. After 28 days of proteinuria, POD-ATTAC mice developed alterations of kidney function, which were characterized by a significant decrease of GFR and an increase of kidney fibrosis (Supplemental Figure 2, A–C), with persistent albuminuria (Supplemental Table 6). Proinflammatory (Myc, Ccl2, and Havcr1) and profibrotic (Tgfβ1 and Fn1) genes were induced (Supplemental Figure 2D).

In order to analyze gene expression between POD-ATTAC and control mice, we performed RNA sequencing of the kidney cortex area. The multidimensional scaling using a gluconeogenic/glycolytic gene subset showed two well-clustered POD-ATTAC and control phenotypes (Supplemental Figure 2E). RNA sequencing analysis showed a downregulation of gluconeogenic transcripts (Pcx, Pck1, Fbp1, and G6pc) and an upregulation of glycolytic transcripts (Pkm and Hk1) in diseased animals compared with healthy animals (Figure 2, A and B). Key gluconeogenesis regulators such as HNF4α and PPARα were downregulated during disease, potentially participating in gluconeogenesis decrease. This observation was also confirmed by RT-qPCR in POD-ATTAC mice at 7, 14, and 28 days postinduction (Supplemental Figure 2F). In order to ascertain that the downregulation was not only a consequence of the loss of tubule mass, we repeated RT-qPCR after enrichment for PT cells as previously described.^14,33 The PT cell-enriched solution displayed as expected a decrease in aquaporin 2 and ENaC expression, with an increase in megalin and cubulin expression (Supplemental Figure 3). With this method, we could show that the downregulation was still observed at the cellular level (Figure 2C) and therefore independent of tubular mass decrease.

Figure 2.

Gluconeogenesis enzyme expression is severely impaired in two models of CKD. RNA sequencing data from controls (CTL, n=5) and 28 days POD-ATTAC (POD, n=6) mice. (A) Volcano plot representation showing log fold change (log FC) in mRNA expression among groups according to the log10 of the P value (log10Pvalue). The cutoff is α=0.05. (B) Heatmap displaying mRNA levels of genes of interest in CTL (gray) and POD (orange) animals. (C) mRNA levels of classic genes implicated in the gluconeogenesis (Pcx, Fbp1, Pck1, G6pc) and glycolysis (Hk1, Pkm) pathways and key regulators (Hnf4α, Pparα, Foxo1, Esrra) in PT cells isolated from POD-ATTAC mice at 28 days. (CTL, n=5) and (POD, n=5). *P<0.05. (D) Representative immunoblots of kidney cortex for FBP1 and PCK1 in controls (CTL) and 28 days POD-ATTAC (POD) animals. Loading control corresponds to Ponceau S staining. (E) Protein quantification by Western blotting of FBP1 and PCK1 proteins in the kidney cortex of controls (CTL, n=12) and 28 days POD-ATTAC (POD, n=13) mice. **P<0.01, ****P<0.0001. (F) Representative immunostaining of kidney cortex for FBP1 and PCK1 in controls (CTL) and 28 days POD-ATTAC (POD). (G) mRNA levels of classic genes implicated in the gluconeogenesis (Pcx, Fbp1, Pck1, G6pc) and glycolysis (Hk1, Pkm) pathways in the UUO mouse model at 7 days. Sham-contralateral (SHAM, n=7) and obstructed kidneys (UUO, n=7). *P<0.05. (H) Representative immunoblot of FBP1 and PCK1 proteins in UUO mouse model. Loading control corresponds to Ponceau S staining. (I) Protein quantification by Western blotting of FBP1 and PCK1 proteins in UUO mouse model. Sham-contralateral (SHAM, n=7) and 7 days obstructed kidneys (UUO, n=7). *p<0.5, ****p<0.01. (J) Representative immunostaining of kidney cortex for FBP1 and PCK1 in sham-contralateral (SHAM) and 7 days UUO. (K) FBP1 and PCK1 enzymatic activity measured in the kidney cortex in 28 days POD-ATTAC (POD, n=6) and control (CTL, n=6). *P<0.05. (L) FBP1 and PCK1 enzymatic activity in sham-contralateral (SHAM, n=6) and 7 days after UUO (UUO, n=6). **p<0.01. Results are presented with error bars showing mean ± bootstrapped confidence intervals.

Analysis of FBP1 and PCK1 protein expression in the kidney cortex by Western blot confirmed a downregulation of these two enzymes in POD-ATTAC mice compared with controls (Figure 2, D and E). This downregulation was also observed by immunohistochemistry, showing an absence of staining in fibrotic areas and a decreased staining in injured or atrophic tubules for these two enzymes (Figure 2F). Similarly, in the UUO model (Supplemental Figure 2, G, H and J), gluconeogenesis transcripts (Pcx, Pck1, Fbp1, and G6pc) were decreased in the damaged kidney cortex compared with the healthy kidney 7 days after surgery (Figure 2G). Western blots and immunostaining for PCK1 and FBP1 confirmed the decrease at the protein level (Figure 2, H–J).

Modifications of PCK1 mRNA expression are usually well correlated to its function.⁴⁴ To confirm the functional effect of the downregulation of the gluconeogenesis pathway, we measured FBP1 and PCK1 enzymatic activity in the kidney cortex of our two models. Both enzymes displayed a significantly decreased specific activity in the two CKD models (Figure 2, K and L).

Loss of Gluconeogenesis Is an Early Feature of CKD and Occurs at the PT Cell Level

To confirm our data in a third model of CKD and to show the cell specificity of the regulation, we further investigated the transition from AKI to chronic kidney injury at single-cell resolution. We combined multiple snRNA-seq data⁴⁵ covering the early phase (4, 12, 48, 64, and 96 hours), the late phase (14 days, 28 days, and 6 weeks) after ischemia-reperfusion injury, and appropriate controls (no injury or sham surgery). This model displays some degree of albuminuria as described previously.⁴⁶ The good integration of the datasets obtained in different laboratories was consistent with strong experimental reproducibility (Supplemental Figure 4).

All expected cell types were identified by the expression of established markers (Figure 3, A and B and Supplemental Table 1). For the specific analysis on gluconeogenesis, we focused on PT cells, because renal gluconeogenesis occurs in this part of the nephron.¹⁴ After quality control, the PT cell cluster consisted of 39,910 cells. The presence of segment-specific markers (e.g., Slc5a12, Slc13a3) was exploited to identify the three segments of the PT. Moreover, we found a large population of PT cells showing reduced levels of classic differentiation markers and the expression of known injury markers (e.g., Havcr1, Vcam1). Despite evidence for heterogeneity within this population, we globally considered this cluster in order to define a general classification of PT cells into “injured” and “noninjured” (Figure 3, C and D). In the early phase, 80% of the cells displayed a transcriptional profile of tubular injury (Figure 3D). In the late phase, the majority of the cells showed a noninjured transcriptional profile, similar to the cells obtained in controls. However, 26% of the PT cells remained in an altered cell state, characterized by the sustained expression of injury markers (Figure 3, C and D).

Figure 3.

snRNA-seq analyses of mouse IRI kidney identify the persistence of injured PT cells with altered metabolism in chronic phase. (A) UMAP plot of all integrated datasets identifies the different cellular component of the kidney. PT, proximal tubule cells; DTL, descending limb of loop of Henle; ATL, thin ascending limb of loop of Henle; TAL and TAL2, thick ascending limb of loop of Henle; POD, podocytes; DCT, distal convoluted tubule; CNT, connecting tubule; ICA, type A intercalated cells of collecting duct; ICB, type B intercalated cells of collecting duct; PC, principle cells; EC, endothelial cells. (B) Heat map of marker genes for all renal cell types identified categorized into tree broad cell categories. Each cell type is represented by the top five genes ranked by average log fold change of a Wilcoxon rank sum test between one cell cluster versus all other cells. Each column represents the average expression per sample hierarchically grouped by cell type and phase. Gene expression values are normalized from 0 to 1 across rows within each cell category. (C) Feature plot shows expression of differentiated PT (Slc5a12, Slc13a3, Scl16a9) and injury markers (Havcr1, Vcam1, Gdf15) of the total PT cells. (D) UMAP plot highlighting uninjured and injured PT populations across phases. Total includes PT cells in control, early, and late phases, whereas early and late phases include only the samples from the respective indicated phase. (E) Heat map shows expression of gluconeogenic (Pcx, Pck1, Fbp1, G6pc), glycolytic (Pkm, Hk1), and key regulator (Hnf4α, Pparα, Foxo1, Esrra) genes in uninjured and injured PT cells in early and late phases after injury. Each column represents the average expression per cell state (uninjured and injured) in each phase (control, early, late) hierarchically grouped by cell state and phase. Gene expression values are normalized from 0 to 1 across rows.

Overall, we obtained 56.1% of noninjured PT cells in controls, 12.8% in the early phase after ischemia-reperfusion injury (IRI), and 42.6% in the late phase. Injured PT cells represented respectively 6.6%, 51.6%, and 15%, whereas non-PT cells accounted for 37.4%, 35.7%, and 42.5% (Supplemental Figure 5). A comparison of the different PT cell populations focused on glycolysis and gluconeogenesis indicated that the early reduction of key gluconeogenesis enzymes (Pcx, Pck1, Fbp1, and G6pc) in injured cells persisted during the late phase in injured cells. This effect was much less pronounced in cells maintaining a noninjured transcriptional profile in the early and in the late phase (Figure 3E). Conversely, key enzymes of glycolysis (Pkm and Hk1) were increased in injured cells in the early and in the late phase (Figure 3E). These findings show regulation of gluconeogenesis pathway expression at the PT cell level, independently of modifications of cell mass. In addition, the proportion of PT cells was not modified by injury at the late stage (Supplemental Figure 5).

Functional Relevance of Altered Gluconeogenesis during CKD

To assess the systemic effect of impaired renal gluconeogenesis in CKD, we performed a lactate tolerance test after a 12-hour fasting period in control and CKD mice.⁴⁷ POD-ATTAC mice with reduced renal function exhibited lower glucose increase after intraperitoneal sodium lactate load (Figure 4, A and B) and decreased lactate clearance (Figure 4C). Liver mRNA expression of the gluconeogenic and glycolytic enzyme transcripts was unaffected by CKD (Supplemental Figure 2I), suggesting that this effect was not an indirect effect of CKD on hepatic function.

Figure 4.

Functional effect of gluconeogenesis alteration in CKD mouse models. Lactate tolerance test in control (CTL, gray, n=8) and 28-days POD-ATTAC (POD, orange, n=5) mice after an acute intraperitoneal sodium lactate load (1.5 g/kg) in fasting animals. (A) Change in blood glucose levels (expressed as fold change from baseline time point in each group) over time. Dots indicate mean and vertical bars represent bootstrapped 95% confidence intervals. (B) Area under the curve (AUC) of glucose level course according to GFR assessed by sinistrin clearance. Regression line is fitted with a logarithmic model. (C) Change in blood lactate levels (expressed as fold change from baseline in each group) and fitted with a nonlinear self-starting asymptotic regression model ex vivo perfused kidney. (D) Schematic representation of the ex vivo kidney perfusion set-up. A perfusion solution is oxygenated with 95% O₂ and 5% CO₂ with a flow of 2 L/min through a membrane oxygenator. The kidney is isolated from the main circulation and perfused through the renal artery at 350 µl/min flow after a 20-minute wash at 700 µl/min flow, allowing the calculation of renal arterio-venous metabolite flux. (E–H) Renal arterio-venous flux of bicarbonate (HCO₃⁻), glucose, lactate, and renal oxygen consumption (VO₂) in control (CTL, n=4) and 28 days POD-ATTAC (POD, n=6) kidneys. *P<0.05, **P<0.01. (I) Renal glucose release and uptake in controls (CTL, n=10) and 28 days POD-ATTAC (POD, n=6) kidneys. *P<0.05. Results are presented with error bars showing mean ± bootstrapped confidence intervals.

We further developed an ex vivo isolated perfused kidney model. The kidney was isolated from general circulation and perfused with a human red blood cell-enriched solution at 700 μl/min for 20 minutes, then 350 μl/min for 60 minutes, oxygenated with 95% O₂ and 5% CO₂ through an oxygenator membrane (Figure 4D and Supplemental Table 4). This allowed us to directly measure renal arterio-venous flux of lactate, glucose, and bicarbonate, as well as renal oxygen consumption, in control and POD-ATTAC mice (Figure 4, E–H). Kidneys isolated from POD-ATTAC mice harboring advanced kidney disease displayed increased lactate release and glucose uptake compared with control kidneys after 60 minutes of perfusion. Thus, diseased kidneys display impaired lactate clearance and decreased glucose production.

Finally, using the same model, we assessed renal gluconeogenesis via the renal arterio-venous D2-glucose dilution technique. Because glucose comes from various sources, the use of isotope-labeled glucose is the only way to measure the production of glucose by an organ. This technique assesses the dilution of labeled glucose by the kidney-produced glucose, measuring the delta between the renal artery and the renal vein in our ex vivo isolated perfused kidney settings, knowing that the labeled glucose is not recycled by gluconeogenesis. Using this technique, we were able to show that measured glucose production by the kidney was decreased in CKD kidneys (Figure 4I).

Alteration of Gluconeogenesis during Human CKD Has Functional Repercussions

To study the functional repercussions in a clinical context, we took advantage of a cohort of patients with CKD undergoing renal venous catheterization in the context of chronic heart failure with renal impairment, or cardiac surgery.³⁶ Those who experienced AKI after the procedure were excluded from this analysis to avoid potential biases. Consistent with the in vivo and ex vivo results, patients with lower baseline eGFR exhibited a higher renal lactate release and a higher glucose uptake (Figure 5A), thus confirming a functional alteration of renal gluconeogenesis in patients with CKD.

Figure 5.

Gluconeogenesis impairment is correlated with kidney function and fibrosis in human CKD. (A) Renal lactate and glucose flux in a mixed cohort of patients undergoing cardiac surgery and patients with congestive heart failure, stratified by eGFR (estimated using creatinine CKD-EPI equation). eGFR>100 ml/min per 1.73 m², n=6; eGFR <100 and >70 ml/min per 1.73 m², n=62; eGFR <70 and >40 ml/min per 1.73 m², n=34; eGFR <40 ml/min per 1.73 m², n=6. Results are expressed in micromoles per minute per kilogram. (B) Individual eGFR trajectories of CKD patients admitted to ICU and displaying during their ICU an impaired metabolism (n=130) or a baseline (n=105) metabolic profile, fitting by linear regression. P=0.012. General model was fitted using a linear mixed model and plotted as a plain line. Results are presented with error bars showing mean ± bootstrapped confidence intervals. (C) Odds ratio for fibrosis progression assessed by the 3 and 12 months difference in the Banff Lesion Score IFTA (ci+ct) relative to transcript levels of key regulator (PPARα, HNF4α, PGC1α, ESRRA, FOXO1, blue), gluconeogenic (PC, FBP1, PCK1, G6PC, dark blue), and glycolytic (HK1, PKM, orange) genes in 42 biopsies from allograft kidney recipients sampled 3 months after transplant. The line represents a 95% confidence interval.

To evaluate whether these modifications could have a systemic effect, we took advantage of a large cohort of ICU patients including 24,273 patients with 661,517 available coupled glucose and lactate measurements over 10 years.¹⁴ This cohort was instrumental for more functional analysis focused on potential associations among impaired gluconeogenesis, kidney function, and outcome. To avoid potential biases related to AKI, we first defined CKD as eGFR at ICU admission <60 ml/min per 1.73 m², remaining stable during the ICU stay (<10% variation) and still <60 ml/min per 1.73 m² 3 months after the ICU stay, thus fulfilling the widely used Kidney Disease Improving Global Outcomes criteria for CKD.⁴⁸ As previously described,¹⁴ we defined five metabolic profiles according to glucose and lactate levels: baseline, isolated high glucose levels, isolated low glucose levels, stress response (high lactate/high glucose levels), and impaired metabolism (high lactate/low glucose levels). The “impaired metabolic profile” is compatible with an alteration of systemic gluconeogenesis. CKD patients with a predominance of impaired metabolism pattern (high lactate/low glucose) displayed a more rapid decline of renal function at distance from the ICU stay compared with CKD patients with a baseline metabolic profile, independently of their baseline renal function (Figure 5B). In a second analysis including ICU mortality, we had to modify our initial definition of CKD and discard the 3-month follow-up. We thus included patients with an eGFR <60 ml/min per 1.73 m² with a stable eGFR during their ICU stay (<10% variation), calling this group “low stable eGFR.” Patients with “normal stable eGFR” (>60 ml/min per 1.73 m² with <10% variation during the stay) displayed a physiologic positive linear relationship between glucose and lactate levels during the ICU stay, whereas patients with “low stable eGFR” exhibited a U-shaped relationship with a paradoxical increase in lactate levels at low glucose levels, as well as global higher lactate levels (Supplemental Figure 6A). This suggested that impaired renal gluconeogenesis associated with CKD might become functionally relevant at the systemic level under stress conditions. To alleviate confounding factors, we performed a propensity score matching involving 2494 patients in each low and normal stable eGFR group. In this matched population, the impaired metabolism pattern was observed more frequently in “low stable eGFR” patients and was strongly associated with ICU mortality (Supplemental Figure 6, B–D). Thus, impaired gluconeogenesis during CKD is associated with mortality under stress conditions. These data are consistent with our previous observation during AKI and indicate that low renal function, either acute or chronic, implies alterations to the metabolic function of the kidney, which translates into systemic changes. This effect may be more preeminent during acute illness, such as during an ICU stay.

We showed that metabolic alterations in CKD patients during an ICU stay is associated with a more rapid eGFR decline at follow-up, suggesting that gluconeogenesis impairment is linked to a worse renal prognosis. To verify this at the transcriptional level in the kidneys, we took advantage of our kidney allograft recipient cohort with serial protocol biopsies and looked at the association between 3-month mRNA expression of gluconeogenic and glycolytic genes and key regulators and fibrosis evolution assessed by the difference in the Banff IFTA score (ci+ct) between 3 and 12 months. Gluconeogenic gene expression at 3 months, including PCK1 and FBP1 mRNA, was negatively associated with a more rapid progression of IFTA, whereas the association was not significant for glycolytic genes (Figure 5C). Loss of HNF4α and PPARα expression was also correlated with enhanced fibrosis progression. This finding confirms that early loss of gluconeogenesis is associated with a worse renal prognosis.

Discussion

In this study, we show that reduced gluconeogenesis is a common feature of different types of CKD, in human and three experimental models (ischemic, proteinuric, and unilateral fibrosis). We further show that loss of gluconeogenesis is an early process in PT cells, paralleling CKD severity. The major regulation observed at the tissue level was reproduced at the cellular level, as assessed by the single-cell analysis of late stage IRI and isolated PT cells from POD-ATTAC mice. Therefore, this regulation appears independent or additive to tubular mass variation.

Experimentally, decreased renal gluconeogenesis induced by CKD led to systemic metabolic alterations during an exogenous lactate load and in the isolated perfused kidney setting. In patients, CKD was associated with a functional defect in lactate clearance and glucose production as assessed by renal catheterization. Patients with low renal function hospitalized in the ICU displayed a higher frequency of metabolic alteration, and this was associated with elevated in-hospital mortality risk and faster kidney disease progression, confirmed by an association between gluconeogenesis pathway loss and histologic fibrosis progression in biopsies of allograft patients.

We first show in a large human cohort of renal biopsies that gluconeogenesis decreases during CKD, whereas glycolysis increases in PT cells. This PT cell metabolism shift adds to the increased glycolytic ability of fibroblasts to globally modify kidney metabolism.^49–52 This pattern was confirmed in kidney allograft patients and in three models of CKD in mice. The downregulation was not only related to a loss of tubular mass, because it was observed in isolated tubular cells in a proteinuric model of CKD and in single-cell analysis of ischemic nephropathy. Loss of the expression of gluconeogenesis genes and proteins resulted in decreased functional ability of the key enzymes of the pathway. Changes in the expression of key gluconeogenesis regulators, such as HNF4α and PPARα, are probably involved in the decreased regulation of the pathway. HNF4α deserves specific attention because it was recently described as a key marker of terminal PT cell differentiation⁵³ and was positively associated with gluconeogenesis in all studied models and in human disease. Its loss may signal early PT phenotype change during CKD, therefore explaining why gluconeogenesis loss is an early and consistent phenomenon. TGFβ has already been described as an important regulator of FAO and HNF4α expression.^6,42 We confirm a role of this cytokine in the downregulation of tubular gluconeogenesis. Thus, the reduction of expression and function of rate-limiting enzymes of gluconeogenesis is a general feature of CKD that may be related to key cytokines such as TGFβ. Gluconeogenesis is fed mainly by lactate conversion to pyruvate in the healthy kidney. Consistently, lower glycemia and decreased lactate clearance were observed in CKD animals submitted to an exogenous lactate load. The contribution of the kidney to this process was confirmed by the isolated perfused kidney technique and the absence of gluconeogenesis regulation in the liver. During renal venous catheterization, CKD patients also displayed an alteration in renal glucose and lactate handling. Altogether, we confirm an impairment of kidney metabolism with a decrease in gluconeogenesis and an increase in glycolysis in kidney PT cells, leading to local and systemic metabolic abnormalities under stress conditions.

Patients with low renal function admitted to the ICU displayed a higher rate of metabolic impairment defined as a relative rise in lactate level and fall in glucose level, and a higher mortality risk. In surviving patients, those harboring systemic metabolic alterations compatible with more pronounced alterations of the renal gluconeogenic function displayed a faster loss of renal function over time. In kidney biopsies from allograft patients, loss of the gluconeogenesis pathway was associated with a worse evolution of kidney fibrosis and tubular atrophy at follow-up, whereas loss of glycolysis was not. The association between impaired gluconeogenesis during CKD and mortality in stress conditions may partly rely on the alteration of lactate clearance by the kidney, because this factor has often been related to mortality in the ICU.^28,54 In addition, higher hypoglycemic risk has been associated with CKD in hospitalized patients, and may likely contribute to the enhanced mortality observed.^26,55 Whether the correlation of altered renal gluconeogenesis with structural lesion progression and renal function loss is merely an association, or whether modification of gluconeogenic enzyme expression alters the course of kidney disease deserves further investigation. Interestingly, the nephroprotective anti-SLGT2 therapeutic drug class has been demonstrated to enhance gluconeogenesis.⁵⁶

The role of gluconeogenesis as an important function of the kidney is often neglected. We show here that modulation of PT cell metabolism during CKD demonstrates cell-specific inhibition of renal gluconeogenesis, a crucial pathway to maintain systemic homeostasis under stress conditions. We further observe in humans that the kidney is less efficient in glucose production and lactate clearance during CKD. This dysregulation is associated with a faster evolution of kidney fibrosis. It also has systemic consequences, because low renal function leads to an alteration of the systemic metabolic profile in patients hospitalized for an acute illness, associated with mortality risk. Alterations of renal gluconeogenesis may participate in the higher hypoglycemic risk of CKD patients with diabetes, precluding optimal treatment of this condition. These observations confirm that the kidney is a key metabolic organ, with the role unmasked under acute stress or hypoglycemic conditions.

Disclosures

S. de Seigneux reports honoraria from Otsuka, Pfizer, AstraZeneca, and Bayer. P. Martin reports consultancy agreements with Alexion and Otsuka; honoraria from Alexion, Otsuka, and Vifor; scientific advisor or membership with Amgen and Otsuka; and other interests/relationships with Fondation AGIR. All remaining authors have nothing to disclose.

Funding

S. de Seigneux is supported by grants from the Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung (SNSF PP00P3-187186/1 and 320030_204187), the Jules Thorn Foundation, and NCCR Kidney.ch. D. Legouis is supported by a Young Researcher Grant from Hôpitaux Universitaires de Genève (PRD 5-2020-I) and by a grant from the Fondation Ernst et Lucie Schmidheiny. A. Faivre is the recipient of a grant from the Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung (323530_191224) and from the Fondation Centre de Recherches Médicales Carlos et Elsie de Reuter.

Author Contributions

S. de Seigneux and D. Legouis conceptualized the study; L. Berchtold, S. de Seigneux, A. Faivre, M. Fernandez, C. Heckenmeyer, D. Legouis, S.-E. Ricksten, A. Rinaldi, and T. Verissimo were responsible for investigation; S. de Seigneux was responsible for funding acquisition; S. de Seigneux was responsible for project administration; S. de Seigneux, P.-Y. Martin, and J. Pugin were responsible for visualization; P. Cippà, S. de Seigneux, D. Legouis, P.-Y. Martin, and J. Pugin were responsible for supervision; P. Cippà, C. Cohen, V. Delitsikou, A. Faivre, D. Legouis, M. Lindenmeyer, S. Moll, M. Naesens, S.-E. Ricksten, A. Rinaldi, J. Rutkowski, and T. Verissimo were responsible for resources; A. Rinaldi was responsible for software; P. Cippà, C. Cohen, D. Legouis, M. Lindenmeyer, S. Moll, and M. Naesens were responsible for methodology; C. Cohen, D. Dalga, V. Delitsikou, A. Faivre, M. Fernandez, K. Haupt, C. Heckenmeyer, M. Lindenmeyer, F. Merlier, M. Naesens, S.-E. Ricksten, A. Rinaldi, T. Verissimo, and C. Veyrat-Durebex were responsible for data curation; L. Berchtold, P. Cippà, S. de Seigneux, A. Faivre, K. Haupt, D. Legouis, F. Merlier, S. Moll, M. Naesens, S.-E. Ricksten, A. Rinaldi, and T. Verissimo, were responsible for formal analysis; S. de Seigneux, A. Faivre, D. Legouis, P.-Y. Martin, S. Moll, J. Pugin, and T. Verissimo were responsible for validation; S. de Seigneux and D. Legouis wrote the original draft; and L. Berchtold, P. Cippà, D. Dalga, A. Faivre, and T. Verissimo reviewed and edited the manuscript.

Supplemental Material

This article contains the following supplemental material online at http://jasn.asnjournals.org/lookup/suppl/doi:10.1681/ASN.2021050680/-/DCSupplemental.

Supplemental Figure 1. Gene expression of kidney damage markers in CKD.

Supplemental Figure 2. Alterations of kidney function in POD-ATTAC and UUO mice.

Supplemental Figure 3. Mouse PT cells isolation quality control.

Supplemental Figure 4. Datasets integration.

Supplemental Figure 5. Proportion of PT and non-PT cells in controls, early and late phase after ischemia-reperfusion injury.

Supplemental Figure 6. Metabolic profiles in ICU cohort.

Supplemental Table 1. Detailed composition of the integrated snRNA-seq of IRI mouse model datasets.

Supplemental Table 2. Names of the segment-specific markers used for clustering and top 10 genes expressed by cluster.

Supplemental Table 3. Resources table.

Supplemental Table 4. Detailed composition of the perfusion solution used for the isolated perfused kidney experiment.

Supplemental Table 5. Estimation of fibrosis degree in 12 kidney biopsies from patients with various severity of kidney disease. The estimation was made by an experienced renal pathologist according to standard practice.

Supplemental Table 6. Albuminuria and albumin-creatinine ratios for POD-ATTAC mice. Controls (CTL), n=5 and POD-ATTAC (POD), n=7.