1. Introduction
According to the 2019 Atlas of the International Diabetes Federation, approximately 463.0 million adults have type 1 (T1D) or type 2 diabetes (T2D) [
1,
2] Significant highest prevalence rates are observed in youth-onset T2DM mainly due to increased childhood obesity [
3]. Diabetic patients are at high risk of developing DKD, cardiovascular disease (CVD), neuropathy and retinopathy [
1].
The prevalence of DKD is increasing and is associated with a heavy societal and financial burden [
4] and therapeutic inertia, creating a major problem in DKD and diabetes treatment [
5]. DKD is characterized by altered glomerular filtration and proteinuria resulting in up to 50% of the patients with end stage kidney disease (ESKD) due to DKD [
6,
7]. Currently, early prevention and management of DKD remain limited. To date, the diagnosis of DKD is based on clinical features; however, the combination with a kidney biopsy is significant [
8]. DKD is characterized by kidney ultra-structural and morphological alterations such as mesangial expansion, nodular glomerular sclerosis, glomerular basement membrane (GBM) thickening and tubulointerstitial fibrosis [
9]. Kidney fibrosis is the downstream effect of activation due to hyperglycemia of several signaling pathways such as TGF-β [
10], JAK/STAT [
11] and Notch [
12] leading to oxidative stress. Endothelial glucocorticoid receptors (GRs) are crucial regulators of DKD fibrosis as demonstrated in a recent study where the loss of endothelial GRs in diabetic mice resulted in an increased Wnt/β-catenin pathway and decreased FAO and increased fibrinogenesis [
13]. However, endothelial SIRT3, a mitochondrial sirtuin, blocks kidney fibrosis through TGF-β/smad pathway regulation [
14]. Glomerular injury characterizes the early stages of DKD; thus, glomeruli are significant targets to investigate the molecular mechanisms of early DKD pathogenesis [
15].
Several biological processes relevant to DKD have been studied, such as mitochondrial dysfunction [
16], reactive oxygen species (ROS) production [
17], NADPH oxidase (NOX) activity [
18], podocyte apoptosis and autophagy [
19] leading to glomerular injury [
20]. Oxidative stress has been highlighted as a significant contributor to DKD and progression to ESKD [
21,
22,
23], being directly linked to podocyte damage, proteinuria, and tubulointerstitial fibrosis [
23]. Oxidative stress is triggered from changes in kidney lipid metabolism [
24] with kidney lipotoxicity and lipid accumulation being considered pathological hallmarks of DKD [
24,
25]. Glomerular lipid accumulation could lead to podocyte death and insulin resistance [
24,
25]. Despite this accumulated knowledge, the absence of efficient inhibitors for the progressive distortion of kidney structure and function reflects the gaps in our understanding of DKD pathogenesis [
26]. The most promising treatment of DKD currently is inhibition of sodium–glucose transporter 2 (SGLT2). SGLT2 inhibitors, such as dapagliflozin [
27], canagliflozin [
28] and empagliflozin [
29], were initially developed to lower blood glucose concentrations but also showed very favorable protective effects in DKD apparently independent of the blood glucose lowering effect. Other promising hypoglycemic drugs than SGLT2i are GLP-l receptor agonists [
18,
30] and DPP-4 inhibitors [
31] used in animal studies and clinical trials improving blood glucose control and reducing albuminuria [
32]. Another promising drug for DKD are the mineralocorticoid receptor antagonists (MRAs) diminishing the activation of kidney inflammation and fibrosis [
33]. Newly developed drugs targeting JAk/STAT, TGFb and PKC pathways and ROCK inhibitors show promising results in animal studies lowering albuminuria and/or kidney inflammation and fibrosis podocyte injury [
32].
Many mouse and rat models of T1D and T2D have been established in order to dissect diabetic and DKD pathogenesis [
34]. Several DKD studies use Ins2Akita mice as models of T1D since these mice have glomerular basement membrane thickening, increased albumin excretion, glomerulosclerosis and interstitial fibrosis, which mimic human DKD [
35]. Hyperglycemia in Ins2Akita mice is thought to induce oxidative stress, resulting in kidney injury [
36]. Additionally, the db/db mouse model is frequently used as a model of human T2D due to susceptibility to obesity, insulin resistance and T2D resulting from leptin deficiency and the development of progressive histological lesions in their kidneys [
34].
To shed more light onto the molecular mechanisms of early DKD pathogenesis and progression, our study targeted the comprehensive molecular characterization of kidney tissue compartments at different developmental time points from the widely used Ins2Akita model of DKD with the subsequent validation of findings in the db/db model and in human kidney biopsies. We performed a high-resolution, quantitative mass spectrometry (MS)-based proteomics analysis of kidney glomeruli or cortex from the mouse models. Our results highlight a conserved (throughout T1D and T2D models and humans) downregulation of peroxisomal function and their cross-talk with mitochondria in early and late DKD, opening up new avenues for DKD therapy.
3. Discussion
This study aimed to investigate proteomic changes associated with early DKD and its progression. Proteomic data were generated from two well-characterized mouse models (Ins2Akita) of diabetes T1D and (db/db) of T2D. Our study was mainly focused on glomerular proteins that are consistently changed in T1D animals in early and late DKD. The expression of these glomerular proteins was further investigated in the kidney cortex proteome of db/db mice. Multiple consistent changes were observed in proteins involved in cholesterol biosynthesis, mitochondrial respiratory chain function, peroxisomal function and amino acid metabolism.
The observed elevated levels of mitochondrial enzymes are in agreement with the existing literature: GLS (glutaminase kidney isoform) catalyzes the first reaction in the kidney catabolism of glutamine [
43]. Associations of incident prediabetes or T2D with higher levels of glutamate were reported previously [
44,
45,
46,
47,
48]. GLDC (glycine dehydrogenase-decarboxylating) and AMT (aminomethyltransferase) participate in mitochondrial glycine cleavage in the kidney [
49]. Low levels of glycine are related to diabetes and could potentially predict future T2D [
48,
50,
51,
52]; they may also reflect glycine utilization towards glutathione production to counteract oxidative stress [
53]; and/or an increased uptake of glycine by insulin-resistant tissues to support gluconeogenesis [
54]. GCAT (2-amino-3-ketobutyrate coenzyme A ligase) participates in the degradation of L-threonine to glycine. Interestingly, decreased threonine levels are reported in diabetes [
55]. AASS (alpha-aminoadipic semialdehyde synthase) catalyzes the first two steps in lysine degradation. Of note, lysine levels in the plasma and serum of T2D patients are lower in comparison to controls [
45]. Mitochondrial enzymes involved in branched-chain amino acid catabolism are elevated in glomeruli of late DKD in agreement with previous studies [
56].
The observed changes in peroxisomal proteins, the most prominent observed alteration in our study, are in general agreement with earlier reports [
22,
57], which suggested the decreased expression of key peroxisomal enzymes and regulators of fatty acid oxidation (FAO) in CKD or DKD compared to healthy kidneys.
Peroxisomal enzymes shorten the long chain of very long-chain fatty acids (VLCFA), which are then oxidized to acetyl-CoA by -acyl-CoA oxidase (ACOX) [
58], and subsequently converted to acyl-carnitine by the carnitine octanoyltransferase (CROT) [
59]. Interestingly, the measurements of CAT, ACOX and CROT in the kidney of db/db mice revealed significantly reduced levels (to approximately 2/3) of these peroxisomal enzymes in comparison to controls [
60] in line with our results, which is potentially related to the well-established accumulation of lipids in the kidneys of diabetic humans and experimental animals [
61,
62].
EHHADH (enoyl-CoA hydratase and 3-hydroxyacyl CoA dehydrogenase) [
63] was recently shown to oxidize medium- and long-chain fatty acids [
64,
65]. In line with our findings, decreased mRNA levels of EHHADH were detected in human DKD glomeruli tissues [
66].
Our study also indicated decreased levels of AGPS in mouse proteomics and in human IHC analyses of DKD tissues, with AGPS levels decreasing with DKD progression. AGPS is the main peroxisomal enzyme mediating (bio)synthesis of plasmalogens [
67], which act as antioxidants [
68] and bile acids. Low levels of plasmalogens have been earlier observed in both T1D [
69,
70] and T2D [
71].
Decreased levels of PIPOX (peroxisomal sarcosine oxidase) were observed in our diabetic mice. PIPOX lowers pipecolate accumulation through oxidation and increases synthesis of glutaryl-CoA [
72,
73]. In line with our findings, previous studies detected increased levels of pipecolate in T1D mice compared to healthy controls [
61]. Further, decreased mRNA levels of PIPOX were also reported in human DKD glomerular tissues [
66].
Our study indicated decreased levels of AMACR, which is involved in the bile acid biosynthesis pathway ([
74] in DKD mouse and human kidney tissues. This finding may be linked to the earlier observed impairment of bile acid synthesis in T2D [
75].
Decreased levels of NUDT19 were also observed in our diabetic mice as well as in human DKD tissues. NUDT19 degrades and regulates CoA in the kidneys [
76], thus, regulating the peroxisomal CoA pool and b-oxidation [
77].
Interestingly, in the cortex of db/db mice of late DKD, three additional downregulated peroxisomal proteins were detected: CAT, EPHX2 (bifunctional epoxide hydrolase 2) and DAO (D-amino acid oxidase; also known as DAAO). CAT is an antioxidant enzyme [
42] whose expression levels also decreased with DKD progression in our human IHC analyses. CAT has also been found downregulated in kidneys and serum from STZ (T2D induced) rodents [
78]. Interestingly, aberrant catalase activity has been found to increase mitochondrial oxidative stress in kidney proximal tubules [
79]. EPHX2 (an antioxidant enzyme) protein and mRNA levels, in accordance with our study, were found decreased in the kidneys of streptozotocin (STZ)-induced diabetic mice [
80] as well as rodents of progressive kidney disease [
81]. Finally, DAO (participating in amino acid degradation [
82]) expression was also decreased in the kidney of DKD alloxan-diabetic rats, in line with our results [
83].
5. Materials and Methods
5.1. Animals
The mouse models C57BL/6-Ins2Akita/J (Ins2Akita-T1D) and BKS.Cg-+Leprdb/+Leprdb/OlaHsd (dbdb-T2D) were used. In the glomerular proteome study 4 groups of animals were included, 2-month-old Ins2Akita (INS2) (
n = 8) and respective controls (WT2;
n = 7); and 4-month-old Ins2Akita (INS4) (
n = 8) and respective controls (WT4,
n = 8) [
15]. For the kidney cortex proteome study, db/db mice were used. The former included 6-month-old db/db (
n = 3) and respective controls db/dm: (
n = 5).
5.2. Isolation of Glomeruli
Isolation of glomeruli was conducted as previously described [
15]. In brief, the aorta of anesthetized mice was catheterized and perfused with 40 mL of a Dynabeads M-450 Tosylactivated suspension (4.5 µm diameter, Dynal A.S., Oslo, Norway) at 2 × 10
6 beads/mL followed by a perfusion of cold PBS (15 mL). The kidneys were pressed through a cell strainer (70 µm) and washed with cold PBS (20 mL). After centrifugation (200×
g for 5 min at 4 °C), the pellet was resuspended in PBS (2 mL) in an Eppendorf tube and Dynabead-loaded glomeruli were pelleted using a concentrator (Dynal A.S., Oslo, Norway). The pellet was washed with PBS (5 × 1 mL) and resuspended in 100 µL PBS, resulting in the enriched glomerular suspension (~ 4000 per kidney). This protocol allows to obtain a relatively high purity level of isolated glomeruli based on microscopy observation and enrichment in glomerular-specific genes [
15].
5.3. Murine Kidney Histology
Kidney histology was performed as described in Ins2Akita mice [
15] by Klein et al., 2020, and in db/db mice (
Figure S1). In brief, kidney lesions were assessed by specific (immuno) histological evaluation of the kidney structure (PAS) and fibrosis (glomerulo and tubulointerstitial fibrosis, Masson-trichrome and Sirius red staining, collagen III staining). Quantifications resulted from the analysis of at least 50 glomeruli.
5.4. Biochemical Analysis
Albumin concentration in urine was determined using the AlbuWell kit (WAK-Chemie Medical GmbH, Steinbach, Germany). Creatine concentration was determined using the Jaffe method. Glucose concentration in blood was determined using a glucometer.
5.5. Sample Preparation for Proteomics
Sample preparation was performed as previously described [
84]. Briefly, samples were homogenized in lysis buffer (7 M Urea, 2 M Thiourea, 4% CHAPS and 1% DTE) and processed with the GeLC–MS method [
85]. Ten micrograms of each sample were loaded in SDS-PAGE (5% stacking, 12% separating) and the electrophoresis was stopped when the samples entered the separating gel. A fixation step (30% methanol, 10% acetic acid) for 30 min was performed and the gels were washed with water (3 × 5 min washes) prior to colloidal Coomassie Blue staining (overnight). Another series of water washes (3 × 5 min washes) was performed to remove the excess stain. All bands were excised from the gel and sliced into small pieces (1–2 mm). Gel pieces were destained with destain solution (40% acetonitrile, 50 mM NH
4HCO
3) followed by reduction (10 mM DTE in 100 mM NH
4HCO
3) for 20 min RT and alkylation (54 mM iodoacetamide in 100 mM NH
4HCO
3) for 20 min RT in the dark. A series of washes was performed for 20 min at RT with the following buffers: 100 mM NH
4HCO
3, destaining solution (40% acetonitrile, 50 mM NH
4HCO
3), ultra-pure water. Gel pieces were dried in a Speed Vac and trypsinized overnight in the dark at RT. For the trypsinization process, 600 ng trypsin per sample was utilized (trypsin stock solution: 10 ng/μL in 10 mM NH
4HCO
3, pH 8.5). After trypsinization, the peptide extraction was performed with subsequent incubations of the gel pieces with the following buffers: 50 mM NH
4HCO
3 for 15 min RT, 5% formic acid, 50% acetonitrile for 15 min RT (the latter was repeated once). Extracted peptides were eluted in a final volume of 600 μL and cleaned with 0.22 μm PVDF filters (Merck, Darmstadt, Germany). After cleaning, the tryptic peptides were placed in a Speed Vac to dry. Dried peptides were resuspended in mobile phase A (0.1% formic acid, pH 3.5) and subjected to LC–MS/MS analysis.
5.6. LC–MS/MS Analysis
A Dionex Ultimate 3000 UHPLC system coupled with the high-resolution nano-ESI Orbitrap-Elite mass spectrometer (Thermo Fisher Scientific, Waltham, MA, USA) was utilized for the LC–MS/MS analysis. Each sample was resuspended in 10 μL mobile phase A. Injection volume for the LC–MS/MS analysis was 5 μL. Samples were loaded on an Acclaim PepMap 100, 100 μm × 2 cm C18, 5 μm, 100 Ȧ trapping column with the ulPickUp Injection mode at a flow rate of 5 μL/min. Peptides were separated in an Acclaim PepMap RSLC, 75 μm × 50 cm, nanoViper, C18, 2 μm, 100 Ȧ column retrofitted to a PicoTip emitter. Mobile phase A (aqueous: 0.1% formic acid, pH 3.5) and B (organic: 100% acetonitrile, 0.1% formic acid) were used for a multi-step gradient elution. The peptides were eluted with a 240 min LC gradient starting from 2% B and rising to 80% B with a flow rate of 300 nL/min and a column temperature at 35 °C. Gaseous phase transition of the separated peptides was achieved with positive ion electrospray ionization applying a voltage of 2.5 kV. In every MS survey scan, the top 10 most abundant multiply charged precursor ions (m/z 300–2200) with an intensity threshold of 500 counts were selected and fragmented with the HCD method. Mass resolution was 60,000 in MS and 15,000 in MS/MS. Normalized collision energy was set to 33 and already targeted precursors were dynamically excluded for further isolation and activation for 45 sec with 5 ppm mass tolerance.
5.7. MS Data Processing
Raw files were analyzed with the Proteome Discoverer 1.4 software package (Thermo Fisher Scientific, Waltham, USA), using the SEQUEST search engine and the UniProt mouse (
Mus musculus) reviewed database, downloaded on 22 November 2019, including 16.935 entries. Stringent criteria were used for protein identification and relative quantification as previously established in our lab [
86]. Cysteine carbamidomethylation was used as the fixed modifications and methionine oxidation as the variable modifications. Two missed cleavage sites were allowed, and the precursor and fragment mass tolerance were set at 10 ppm and 0.05 Da, respectively. The False Discovery Rate was 1% (peptide level). Label free quantification analysis was performed considering the precursor ion area values that were exported from the total ion chromatogram as defined by the Proteome Discoverer 1.4 software package.
Output files from the Proteome Discoverer were processed with an in-house script in the R environment for statistical computing (version 4.0.3) as follows: Protein lists were concatenated into a master table. Raw protein intensities for each individual sample were subjected to normalization according to X’ = X/Sum(Xi) ∗ 106 and only proteins present in at least 55% of the samples in at least one group were further selected for downstream statistical analysis.
5.8. Functional Analysis
Pathway annotation was performed with the ClueGO plug-in 3.7.2 (Cytoscape) using the REACTOME pathway database (updated on 17 February 2020). The same GlueGO plug-in was used for the biological function annotation with the addition of the GO-Biological Process EBI-Uniprot GOA database (updated on 17 February 2020). Statistically significant pathways corrected for multiple testing (Benjamini–Hochberg (BH) corrected p-value ≤ 0.05, two-sided hypergeometric test) were further considered. Results were interpreted based on biological relevance by utilizing only the leading term from each group.
5.9. Investigation through Transcriptomics Data Analysis
Nephroseq (
www.nephroseq.org, accessed on 30 November 2021) was employed for the investigation of the expression of the shortlisted mitochondrial and peroxisomal proteins in existing mouse and human transcriptomics datasets. The list of proteins was uploaded in Nephroseq v4 in the form of EntrezGene IDs. DKD datasets selection was held after application of the filters: Primary Filters > Group > Diabetic nephropathy. The corresponding gene expression was searched in seven available DKD mouse and human datasets observed after filtering, on comparison of DKD vs. Healthy Living Donor groups (
Table S1). Only significantly deregulated genes (
p < 0.05) were extracted, and their differential expression was compared with the mitochondrial and peroxisomal deregulated proteins.
5.10. Clinical Material
Within a period of eight years (2013–2020) a total number of 100 kidney biopsies diagnosed as DKD were retrieved from the human Renal Biopsies archive of the 1st Department of Pathology of Athens (National and Kapodistrian University of Athens, Medical School, Greece). All cases were initially classified based on their glomerular lesions into the four classes of DKD (I, IIa/b, III and IV) according to Tervaert et al. [
87]. Interstitial fibrosis and tubular atrophy (IFTA), as well as interstitial inflammatory infiltration and vascular lesions (arteriolar hyalinosis, arteriosclerosis), were also studied and scored from 0 to 9 according to Tervaert et al. [
87] in order to assess the severity and chronicity of DKD.
Information on gender, age, serum creatinine value and albuminuria levels was collected, and the estimated Glomerular Filtration Rate (eGFR) was calculated using the online National Kidney Foundation GFR Calculator [
88]. Combining eGFR and albuminuria levels, the cases were classified to 5 clinical stages (G1, G2, G3a/b, G4 and G5–A1, A2 and A3) of chronic kidney disease (CKD) due to diabetes mellitus (DM) according to KDIGO Guidelines [
89]. All cases were classified as A3 since albuminuria levels were always >0.3 g/24 h.
Out of the 100 cases, 16 were selected in order to assess the immunohistochemical expression of the peroxisomal proteins Nucleoside diphosphate-linked moiety X motif 19 (NUDT19), alpha-methylacyl-CoA racemase (AMACR), peroxisomal catalase (CAT) and alkyldihydroxyacetonephosphate synthase (AGPS) in human renal tissues with DKD.
The selected cases met the following criteria: (1) Absence of coexisting non diabetic kidney disease, (2) Adequate representation of the histopathological lesions of DKD, reflecting the 4 histological classes of DKD and (3) kidney function deterioration, as it is reflected in the CKD stage, equivalent to DKD lesion progression.
Additionally, normal kidney tissue showing no signs of DKD, CKD or other kidney pathology obtained from radical nephrectomy specimens were used as controls.
The main clinicopathological parameters of the 16 cases are shown in
Table S2.
5.11. Immunohistochemistry of Human DKD Specimens
Immunohistochemical detection of the examined proteins was performed on 4-μm-thick formalin-fixed paraffin sections which underwent overnight heating at 37 °C, deparaffinization, rehydration and antigen retrieval using an automated module (PT Link, Dako) for 20 min at 96 °C with the reagents EnVision FLEX Target Retrieval Solution High pH (50×) (Dako, DAKO EnVision kit, DAKO, Carpinteria, CA) for CAT and EnVision FLEX Target Retrieval Solution Low pH (50×) (Dako) for NUDT19, AGPS and AMACR. To block endogenous peroxidase activity, 0.3% hydrogen peroxide in Tris-buffered saline (TBS) was applied for 15 min. Sections were rinsed with TBS and normal horse serum was applied for 20 min to prevent non-specific antibody binding. This step was followed by overnight incubation of the sections at 4 °C with the primary antibodies: anti-AMACR (rabbit polyclonal) (Atlas Antibodies, Stockholm, Sweden) at a dilution 1:2000, anti-AGPS (rabbit polyclonal) (Atlas Antibodies, Sweden) at a dilution 1:50, anti-CAT (rabbit polyclonal) (Atlas Antibodies, Sweden) at a dilution 1:3000 and anti-NUDT19 (EPR13162-63) (rabbit monoclonal) (abcam, Cambridge, UK) at a dilution 1:100. For visualization, a two-step technique (polymer, HRP-conjugated; Vector Laboratories, Burlingame, CA, USA) was used with diaminobenzidine as a chromogen. Haematoxylin was used to counterstain the sections.
5.12. Evaluation of Immunohistochemistry
Qualitative immunohistochemical evaluation was assessed by two independent observers (HG and DP) blinded to clinical data. The pattern, topography, extension and intensity of staining of all the antibodies under study in both the controls and DKD cases were analyzed. A rough comparison of the expression of each marker between the controls and DKD cases, as well as between cases of different DKD classes was also performed.
Staining intensity was also further quantified using ImageJ software as previously described [
85]. Optical density was normalized over the unstained tissue and mean intensity values were estimated. Statistical significance among multiple groups (stages) was confirmed with ANOVA analysis and further pair-wise comparisons were performed with the Student’s
t-test. Values of
p < 0.05 were considered as statistically significant.
5.13. Statistical Analysis of Proteomics Data
Statistical significance of continuous variables was defined at
p < 0.05 with the non-parametric Mann–Whitney test. Proteins with
p value ≤ 0.05 and ratio ≥ 1.5 (upregulated) or ≤0.67 (downregulated) were considered statistically significant and differentially expressed. Dotplots for the mouse physiopathologic characterization were created with functionality from the ggplot2 and ggpubr R packages. Depicted statistical comparisons correspond to independent Μann–Whitney tests. For the correlation analysis, Spearman’s correlation was utilized to assess the relationships of our data to one external proteomics dataset [
37], using the means of the normalized protein intensity across the samples after transforming them to the natural logarithmic scale. A heatmap was created with the gplot package, after Z-scaling of the normalized protein intensities. Euclidean distance and Ward’s hierarchical method (option: ward.D2) were selected for the clustering of both rows and columns. Graphing and statistical analysis were performed in the RStudio environment (R version 4.0.3).