Proteomic Identification of Urinary Biomarkers of Diabetic Nephropathy

Urine samples were collected from a total of 33 subjects with type 2 diabetes and from 9 healthy control subjects. In the diabetic group, 10 subjects had normoalbuminuria (urinary albumin <20 mg/g creatinine; group 1), 13 had microalbuminuria (20–200 mg/g creatinine; group 2), and 10 had macroalbuminuria (>200 mg/g creatinine; group 3). The subjects in group 3 exhibited creatinine clearance rates ≤50% of those of control subjects, and blood urea levels >60 mg/dl and serum creatinine levels >1.5 mg/dl. Control subjects (4 women and 5 men) were 45–69 years of age with BMIs between 19.67 and 27.64 kg/m². They had normal renal function and blood pressure, no proteinuria, normal urinary sediment, and no significant clinical events during the past 6 months, and none of them were taking medication. The clinical characteristics of control subjects and the patients in study groups 1, 2, and 3 are shown in Table 1. Informed consent was obtained from the subjects following the institutional review board guidelines for human subjects (Nizam’s Institute of Medical Sciences, Hyderabad, India).

All subjects had provided second voided clean-catch urine samples in the early morning after an overnight fast in a 5- to 10-ml volume in 15-ml conical tubes. These specimens were frozen and stored at −20°C until being shipped to Oregon Health and Science University for further analysis. Urine samples were centrifuged at 10,000g for 20 min at 4°C to remove cellular debris and nuclei. The supernatants were transferred to 15-ml Ultrafree 5K membrane concentrators (Millipore, Billerica, MA) and spun at 7,000g to reduce the volumes to ∼1 ml. Precipitates formed during the concentration step were removed by centrifugation at 7,000g for 15 min.

Urine samples were depleted of six major proteins (albumin, IgG, IgA, antitrypsin, transferrin, and haptoglobin) using the Agilent multiple affinity system (Agilent Technologies, Palo Alto, CA). Urine (1 ml) equilibrated in Agilent buffer A was processed using an Agilent immunoaffinity column (4.6 × 100-mm) attached to a Waters high-performance liquid chromatography system. Appropriate fractions were collected, buffer-exchanged with 10 mmol/l Tris, pH 8.4, and concentrated using 5,000 molecular weight cutoff filters. Protein concentration was determined using the Bio-Rad DC protein assay kit (Bio-Rad, Hercules, CA).

A total of 15 2-D-DIGE gels were run, each representing a pair of samples from control and diabetes subjects (5 each for diabetes without albuminuria, diabetes with microalbuminuria, and diabetes with macroalbuminuria). Urine proteins (50 μg) were labeled with CyDye DIGE Fluor minimal dye (Cy3 for control, Cy5 for diabetes, or Cy2 for reference [diabetes + control]; Amersham Biosciences, Piscataway, NJ) at a concentration of 100–400 pmol/l dye/50 μg protein, and all three labeled samples were multiplexed and resolved in one gel. Labeled proteins were purified by acetone precipitation and dissolved in isoelectric focusing buffer and rehydrated on to a 13-cm IPG strip (pH 4–7) for 12 h at room temperature. The IPG strip was then equilibrated with dithiothreitol equilibration buffer and iodoacetamide equilibration buffer for 15 min sequentially. Second-dimension 8–16% SDS-polyacrylamide gel electrophoresis was conducted at 80–90 V for 18 h.

Gels were scanned in a Typhoon 9400 scanner (Amersham Biosciences) using appropriate lasers and filters with photomultiplier tube voltage set between 550 and 600. Images in different channels (control and diabetes) were overlaid using pseudo-colors, and differences were visualized using ImageQuant software (Amersham Biosciences). 2-D gel image analysis to identify differentially abundant protein spots was performed using Phoretix 2D Evolution software (version 2005; Nonlinear USA, Durham, NC). A fixed area was selected from every gel, and a cross-stain analysis protocol was performed. Background subtraction was done using the “mode of nonspot” method, and images were wrapped to maximize spot matching. A ratiometric normalization algorithm was applied to account for potential concentration differences in protein labeling. Normalized protein spots in the Cy3 and Cy5 channels were compared with the internal standard (Cy2) to generate a ratio of relative amount. The statistical significance of differences in the intensity of protein spots was determined by t tests on the averaged gels for each group. Protein spots with a relative ratio >1.5 and a t test value of <0.05 were considered to be significantly differentially abundant.

Urine protein (1 mg) was subjected to 2-D DIGE and stained with Coomassie Blue R-250. Individual spots were cut from the gel, destained, and digested in-gel with trypsin for 24 h at 37°C. Tryptic peptides were extracted with 0.1% trifluoroacetic acid, purified using Zip-Tip_C18 pipette tips from Millipore, and analyzed on a Waters hybrid quadrupole time-of-flight mass spectrometer (Q-Tof-2) connected to a Waters CapLC. An MS/MS-MS survey method was used to acquire MS/MS-MS spectra. Masses of 400–1,500 Da were scanned for MS survey, and masses of 50–1900 Da were scanned for MS/MS. Data analysis was performed using the ProteinLynx Global Server (version 2.1; Waters) and by de novo sequencing using a PEAKS algorithm combined with the OpenSea alignment algorithm (version 1.3.1) (10). Peptides consisting of five or more amino acids were used and matched to either a nonredundant human IPI or the Swiss-Prot database to identify the corresponding proteins. Proteins with two or more peptides by both ProteinLynx (significance score >10.6) and OpenSea (significance score >100) scoring algorithms (6) were chosen.

To improve the detection of low-abundance proteins, we removed the six most abundant serum proteins (albumin, IgG, IgA, transferrin, haptoglobin, and antitrypsin). Because of the high levels of urinary protein in diabetic nephropathy, preparative 2-D gels were prepared from group 3 (macroalbuminuria) to identify proteins expressed in the urinary proteome. As shown in Fig. 1, 65 areas from the 2-D gels representing single or multiple spots potentially corresponding to the same protein were picked for identification. Analysis of MS/MS spectra by ProteinLynx, de novo sequencing algorithms, and a search of a nonredundant human IPI database identified 195 peptide MS/MS spectra that corresponded to 62 unique proteins (Table 2).

We classified identified proteins by function with DAVID (Database for Annotation, Visualization and Integrated Discovery; National Institute of Allergy and Infections, National Institutes of Health) and sorted by Gene Ontology (GO) Consortium terms (Table 2). Further sorting for relation to metabolism was also performed on the basis of Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway terms and the Bioinformatic Harvester Web site using SOURCE, STRING, and TrEMBL. The subgroups assigned were cell development, cell organization, defense response, metabolism, and signal transduction.

To enhance sensitivity, reproducibility, and detection over a wide dynamic range, we used a multiplex proteomic analysis approach to label proteins with fluorescent cyanine dyes (Cy3, Cy5, and Cy2; control, diabetes, and reference, respectively), and each pair of diabetic and control samples was resolved in the same gel. Five individual gels representing each group were analyzed to identify differentially abundant proteins. Spot quantification and statistical analysis (Phoretix 2D Evolution software; Nonlinear) of gels containing individual study group samples identified a total of 195 spots from all conditions. After background subtraction and ratiometric normalization, matched spots across all the gels were used for statistical analysis. By using the criteria of differential abundance >1.5-fold change and P < 0.05, 65 spots were classified as significantly differentially abundant.

Differential expression by relative abundance from the gel spots identified individual proteins whose levels differed between groups 3 and 1 and group 3 and control subjects (Tables 3 and 4) by this 2-D DIGE approach. To evaluate proteins exhibiting significant differences in abundance in different stages of diabetic nephropathy (normoalbuminuria versus microalbuminuria versus macroalbuminuria), we examined specific differentially abundant spots that were present in all three groups by spot volume quantification (Phoretix software) (Fig. 2). The relative values across three subgroups of albuminuria (groups 1, 2, and 3) representing normo-, micro-, and macroalbuminuria associated with diabetes are shown in Table 5. Whereas 11 proteins were progressively increased from normoalbuminuria to overt nephropathy, 4 proteins were progressively decreased with increasing disease severity.

Seven proteins were upregulated >1.5-fold (P < 0.05) in diabetic nephropathy with macroalbuminuria when compared with diabetes without albuminuria. Normalized volumes of these upregulated proteins were also progressively increased across three categories of diabetic patients with normo-, micro-, and macroalbuminuria, indicating their positive association with disease progression (Table 5). α_1B-Glycoprotein showed the greatest overall increase (7.0-fold), followed by zinc-α₂-glycoprotein (5.9-fold), α₂-HS-glycoprotein (4.7-fold), vitamin D–binding protein (VDBP) (4.8-fold), calgranulin B (3.9-fold), α₁-antitrypsin (A1AT) (2.9-fold), and hemopexin (2.4-fold). When compared with control subjects, VDBP exhibited the greatest increase (11.1-fold), followed by zinc-α₂-glycoprotein (6.0-fold), α₂-HS-glycoprotein precursor (2.3-fold), and A1AT (2.2-fold). A preliminary analysis of urine VDBP levels using immunoassays verified a significant difference between healthy subjects (0.025 ± 0.002 μg/μl, n = 5) and macroalbuminuric patients (75 ± 10 μg/μl, n = 5) (data not shown).

Four proteins were downregulated <1.5-fold (P value < 0.05) in diabetic nephropathy with macroalbuminuria compared with diabetes without albuminuria. Transthyretin was highly downregulated (4.3-fold), followed by apolipoprotein (apo) A-I (3.2-fold), α₁-microglobulin/bikunin precursor (AMBP) (1.61-fold), and plasma retinol-binding protein (1.52-fold). The reduction in the normalized volumes of plasma retinol-binding protein and AMBP across the control and three diabetic groups was consistent with worsening diabetic renal disease (Table 5). Compared with control subjects, AMBP was the least abundant in macroalbuminuria versus control (5.06-fold), followed by plasma retinol-binding protein (2.56-fold) and apoA-I (2.53-fold).

To date, 2-D gel electrophoresis–based methods have been most successful in identifying individual disease-specific protein biomarkers. 2-D gel electrophoresis and protein identification using LC-MS/MS allowed us to identify 62 unique proteins expressed in urine (Table 2); ∼65 distinct urinary protein spots associated with these 62 individual gene products unique to diabetic nephropathy were mapped. Further subgrouping of their annotated functions revealed the presence of transport proteins, proteases and protein inhibitors, glycoproteins, growth factors, extracellular proteins, and complement factors and immunoglobulins.

With the multiple proteomic comparisons used in this study, we identified distinct sets of proteins that were differentially abundant in individuals with diabetic nephropathy compared with those without diabetic nephropathy or control subjects. Seven urine proteins exhibited high relative abundance in diabetic nephropathy (macroalbuminuria) compared with diabetes without albuminuria, and their differential abundance was consistent across groups of nephropathy. Four urine proteins were of low relative abundance in macroalbuminuria compared with diabetes without albuminuria, and the differential abundance of two of them was consistent across three groups of nephropathy. It is likely that further studies of these potential markers will facilitate better understanding of the mechanism of diabetic nephropathy and provide new avenues for therapy.

α_1B-Glycoprotein contains five Ig-like V-type (immunoglobulin-like) domains and is homologous to the immunoglobulin supergene family, suggesting its possible role in the autoimmunity involved in nephropathy. Zinc-α₂-glycoprotein was the second most abundant urinary protein in diabetic nephropathy. This protein stimulates lipid degradation in adipocytes and may also bind polyunsaturated fatty acids. An earlier study of type 2 diabetes described zinc-α₂-glycoprotein and three other proteins, α₁-acid glycoprotein, α₁-microglobulin, and IgG, as specific markers for diabetic nephropathy (9). α₂-HS-glycoprotein precursor (fetuin A) is an inflammation-related calcium-regulatory glycoprotein that acts as a systemic calcification inhibitor. Both chronic inflammation and uremia may contribute to exhausting fetuin A release in the late stages of kidney disease. Deficiencies of calcification inhibitors such as fetuin A are relevant to uncontrolled vascular calcification and may offer potential for future therapeutic approaches (11). Hemopexin (β_1B-glycoprotein) binds heme and transports it to the liver for breakdown and iron recovery, after which the free hemopexin returns to the circulation. Hemopexin levels have been described to be higher in type 2 diabetes, its relationship with C-reactive protein is lost in both type 1 and type 2 diabetes, and its levels are independently determined by triacylglycerol and the diabetic state (12). The hemopexin molecule can potentially act as a toxic protease, leading in the rat to proteinuria and glomerular alterations (13). Although there were no direct references to its possible effects in humans with diabetes or nephropathy, because diabetes is an inflammatory condition associated with iron abnormalities, it can be postulated that hemopexin is altered in diabetes.

VDBP is a multifunctional protein found in body fluids and on the surface of many cell types. In plasma, it carries vitamin D sterols and prevents polymerization of actin by binding its monomers. VDBP also associates with membrane-bound immunoglobulin on the surface of B lymphocytes and with IgG Fc receptors on the membranes of T lymphocytes, suggesting its possible role in the immunopathogenesis and progression of the disease. Furthermore, an association of VDBP with type 1 and type 2 diabetes, glucose intolerance, hypoinsulinemia/hyperinsulinemia, and/or insulin resistance and pregnancy associated-diabetes was reported in Dogrib Indians (14), American Anglos and Hispanics (15), Pima Indians (16), Japanese (17,18), Russians (19), Polynesian Island populations (20), Alsacian French (21,22), Bantu (23), and Finnish populations (24). VDBP was also found in the vitreous in diabetic macular edema, along with pigment epithelium-derived factor, apoA-4, apoA-1, trip-11, and plasma retinol-binding protein (RBP). These chemical mediators in the posterior vitreous may play a role in the pathogenesis of diabetic macular edema (25).

Calgranulin B is expressed by macrophages in inflamed tissues and is an inhibitor of protein kinases. Differences in the isoforms and abundance of several urine proteins, including calgranulin B, inter-α-trypsin inhibitor, prothrombin fragment 1, and CD59, were known to be associated with stone formation (26). There have been no previous reports on their association with diabetes or other renal diseases. A1AT is an inhibitor of serine proteases, and its primary target is elastase, but it also has a moderate affinity for plasmin and thrombin. The serum levels of A1AT and a α₁-acid glycoprotein, as well as their glycosylated protein fractions, were reported to be significantly greater in sera from patients with diabetic nephropathy compared with healthy adults. Marked linear deposition of these proteins in the glomerular or dermal vascular walls was also observed in the same patients (27), linking them to diabetic nephropathy. Whereas the association between A1AT deficiency and glomerulonephritis has been reported only sporadically, A1AT deficiency was more frequently linked to lung emphysema with or without hepatic cirrhosis (28).

Transthyretin (prealbumin) is a thyroid hormone-binding protein, which transports a small part of thyroxine from the bloodstream to the brain. About 40% of plasma transthyretin circulates in a tight complex with plasma RBP. Transthyretin was reported as a better and suitable marker for nutrition assessment in patients with chronic renal failure (29). Elevated plasma RBP in insulin-resistant humans with obesity and type-2 diabetes was known to induce hepatic expression of the gluconeogenic enzyme phosphoenolpyruvate carboxykinase and impairs insulin signaling in muscle (30). It is possible that low RBP-4 in diabetic nephropathy increases insulin sensitivity and causes spontaneous hypoglycemia, because low RBP-4 is potentially hypoglycemic.

AMBP contains both α₁-microglobulin and inter-α-trypsin inhibitor light chain (bikunin). Bikunin is an important anti-inflammatory substance to modulate inflammatory events. Whether decreases in these protein levels affect immunocompetence in diabetic renal disease remains unstudied. The other component of AMBP protein, α₁-microglobulin, in urine was directly related to progressive albuminuria in Chinese, Malays, and Asian Indians with type 2 diabetes (31). This is in contrast to the decreasing levels of this protein observed in our study. Urinary α₁-microglobulin indicates proximal tubular dysfunction and could be a useful biomarker for the early detection of nephropathy in diabetic subjects in addition to albuminuria, which indicates glomerular dysfunction.

ApoA-I, acting as a cofactor for the lecithin cholesterol acyltransferase, participates in the reverse transport of cholesterol from tissues to the liver for excretion. ApoA1, as well as apoB/A1, was included among several nontraditional cardiovascular risk factors in the progression to pediatric metabolic syndrome (32), but there have been no reported references to its association with diabetes or nephropathy.

In summary, this study differs from previous proteomics studies of urinary proteins in diabetic nephropathy in several aspects. Earlier analyses have been focused on kidney-derived proteins in a particular mouse model of type 1 diabetes (4) or have described peptide patterns in human type 1 and type 2 diabetic urine, in which the identity of a only small fraction of the source proteins was identified, and which were, in those cases, high-abundance proteins (6–8). One other recent study assessed a limited number of candidate biomarkers by 2-D gel electrophoresis and MS and found a small number of proteins that were upregulated in diabetic urine (9). In the current study, on the other hand, we used a more robust 2-D DIGE approach coupled with LC-MS/MS to produce a comprehensive list of proteins in human urine and identified a significant number that exhibited either increased or decreased abundance in the presence of diabetic nephropathy, including several whose levels correlated with increasing degree of albuminuria.

Seven potential upregulated urinary biomarkers for diabetic nephropathy found in this study were α_1B-glycoprotein, zinc-α₂-glycoprotein, α₂-HS-glycoprotein, VDBP, calgranulin B, A1AT, and hemopexin. The detection of a significant number of glycoproteins in diabetic urine could potentially be representative of the hyperglycosylated state of these proteins in the serum proteome. This is the first report of an elevated transport protein (VDBP) and two defense-response proteins (A1AT and calgranulin B) in the urine in diabetic nephropathy.

A limitation of the current study, as well as of the other studies of the human urinary proteome in diabetic nephropathy cited above, is that the candidate biomarkers and peptide patterns described could be indicative of nephropathy per se, rather than being uniquely associated with diabetic nephropathy. Distinguishing between these alternatives would require large-scale analyses in light of the wide spectrum of nondiabetic causes of nephropathy (33–37). However, because these biomarkers have been derived from a well-characterized patient population and would be used in a similarly characterized population, this issue does not detract from the potential utility or possible mechanistic significance of the candidate proteins.

The identification of novel protein biomarkers of diabetic nephropathy represents an important step forward in advancing our understanding of the physiologic perturbances that lead to ESRD. Further studies of these proteins may elucidate the mechanisms and pathways that underlie the pathogenesis of diabetic nephropathy, thus facilitating the design of effective therapeutic agents. Characterization and identification of the differentially abundant proteins seen in this study can also provide the basis for development of robust, noninvasive, multianalyte assessment of diabetic nephropathy. We acknowledge that our findings should be considered preliminary until validated in a larger cohort and that to be clinically significant, the utility of these markers in practice must be better than that of currently available tests. However, to reverse the alarming trend of the steady rise in the rate of ESRD in the U.S., innovative treatment strategies based on the reliable identification of diabetic patients at high risk for nephropathy must be developed.

This study was supported, in part, by the Department of Endocrinology and Metabolism, Nizam’s Institute of Medical Sciences, and by ProteoGenix, Inc.

The authors thank their colleagues at Nizam’s Institute of Medical Sciences and the Oregon Health and Science University for technical and informatics support and Dr. Rakesh Mittal, Senior Deputy Director General, Indian Council of Medical Research, for assistance in gaining approvals from the Ethics Committee, Nizam’s Institute of Medical Sciences; Ministry of Health and Family Welfare, Government of India; and the Ministry of Science and Technology, Government of India.