Differential Proteome Profiling Using iTRAQ in Microalbuminuric and Normoalbuminuric Type 2 Diabetic Patients

Diabetic nephropathy (DN) is a long-term complication of diabetes mellitus that leads to end-stage renal disease. Microalbuminuria is used for the early detection of diabetic renal damage, but such levels do not reflect the state of incipient DN precisely in type 2 diabetic patients because microalbuminuria develops in other diseases, necessitating more accurate biomarkers that detect incipient DN. Isobaric tags for relative and absolute quantification (iTRAQ) were used to identify urinary proteins that were differentially excreted in normoalbuminuric and microalbuminuric patients with type 2 diabetes where 710 and 196 proteins were identified and quantified, respectively. Some candidates were confirmed by 2-DE analysis, or validated by Western blot and multiple reaction monitoring (MRM). Specifically, some differentially expressed proteins were verified by MRM in urine from normoalbuminuric and microalbuminuric patients with type 2 diabetes, wherein alpha-1-antitrypsin, alpha-1-acid glycoprotein 1, and prostate stem cell antigen had excellent AUC values (0.849, 0.873, and 0.825, resp.). Moreover, we performed a multiplex assay using these biomarker candidates, resulting in a merged AUC value of 0.921. Although the differentially expressed proteins in this iTRAQ study require further validation in larger and categorized sample groups, they constitute baseline data on preliminary biomarker candidates that can be used to discover novel biomarkers for incipient DN.


Introduction
Diabetes mellitus is a chronic disease with potentially devastating complications. For example, diabetes mellitus is associated with macrovascular complications, such as cardiovascular and cerebrovascular diseases, and microvascular complications, including diabetic nephropathy (DN) and retinopathy [1]. DN is a long-term complication of diabetes that is caused by specific renal structural alterations, such as mesangium expansion due to the progressive accumulation of extracellular matrix (ECM), and by functional losses, such as elevated glomerular basement membrane (GBM) permeability [2]. DN occurs in 15% to 25% of type 1 diabetic patients and 30% to 40% of type 2 diabetic patients [3] and accounts for approximately one-half of all new cases of end-stage renal disease (ESRD). Furthermore, ESRD has a 5-year survival rate of only 21% [4]. Because the progression of ESRD in DN is irreversible, the early diagnosis of DN is necessary to prevent or delay progression to ESRD [5]. Microalbuminuria represents a potentially reversible incipient stage of nephropathy and is used as a noninvasive index for the detection of diabetic renal disease. Microalbuminuria is defined as a state in which abnormal amounts of albumin are excreted in urine (30-300 mg/24 h versus <30 mg/24 h in normoalbuminuria) [5,6].

Experimental Diabetes Research
The use of microalbuminuria to predict incipient DN, particularly in type 2 diabetic patients, is limited for several reasons [7]: the microalbuminuric state also predicts cardiovascular disease in diabetic and nondiabetic individuals [8,9], and it is associated with inflammation and hypertension [5]. Consequently, the likelihood of detecting nondiabetic renal disease or normal glomerular structure is observed with microalbuminuria patients [10]. Thus, more accurate biomarkers for incipient DN in type 2 diabetic patients are required that can differentiate incipient DN from other conditions in microalbuminuria patients, including cardiovascular disease, inflammation, and hypertension.
Recently, to compare DN patients with non-DN patients, proteomic technologies have been developed to identify urinary marker candidates that are associated with the development of DN. Various proteomic approaches have been used for this purpose, including 2-DE, 2-DE DIGE, and SELDI-TOF [5,11,12]. However, because many studies have focused on restricted sets of targeted proteins, alterations in comprehensive urinary protein profiles in type 2 diabetes have not been monitored. In particular, SELDI-TOF has been shown to be a valuable technology for urinary proteomic analysis, but the absolute identification of differentially excreted proteins remains challenging [13].
To scan a comprehensive differential proteome for preliminary DN candidate biomarkers, we used a 4-plex isobaric tag for relative and absolute quantification (iTRAQ, 4-plex), allowing us to identify and quantify proteins in up to 4 samples [14]. The advantages of iTRAQ include whole labeling of representative or pooled samples, comparatively high throughput, and retention of posttranslational modification (PTM) data; one of its shortcomings is that it cannot be applied easily to a large collection of individual clinical samples due to reagent cost and the required mass spectrometry effort [15]. To date, iTRAQ has been applied to a variety of sample sets, including E. coli, mammalian cells, yeast, plant cells, and human biological fluids [16][17][18][19][20][21][22].
Therefore, in this study, we used iTRAQ to identify and quantify differentially excreted urinary proteins in microalbuminuric versus normoalbuminuric type 2 diabetic patients and investigate the associations that would reflect the progress of DN. Afterward, those differentially excreted urinary proteins have been confirmed by 2-DE, followed by MALDI-TOF/TOF, or validated by Western blot and MRM.

Urine Sample Preparation
. Type 2 diabetic subjects (age ≥ 40 years) with or without microalbuminuria who were patients at the Diabetes Center of Seoul National University Hospital, Seoul, Republic of Korea, were enrolled in 2006. Microalbuminuria patients were randomly selected out of these outpatients, whereas normoalbuminuric patients were selected to be matched to age, sex, body mass index (BMI), and DM duration with microalbuminuric patients.
Forty-three subjects with diabetic retinopathy and persistent microalbuminuria formed the microalbuminuria group (MA). Persistent microalbuminuria was defined as an albumin : creatinine ratio (ACR) between 30 and 300 mg/g in 2 urine samples that were taken over 3 months. The normoalbuminuria group (NA) comprised subjects who had no diabetic retinopathy, did not use angiotensin inhibitors or angiotensin receptor blockers that lowered albuminuria, and showed no microalbuminuria in their urine in the past year (urinary albumin < 30 mg/g creatinine). Forty-three subjects formed the NA group.
There were no significant differences in age, sex, body mass index, or diabetes mellitus duration between the 2 study groups. Subjects with hematuria, uncontrolled hypertension (blood pressure ≥ 140/90 mm Hg), uncontrolled hyperglycemia (glycated hemoglobin A1c ≥ 8.5%), urinary tract infection, acute febrile illness, congestive heart failure, or malignancy were excluded. Individuals who were receiving peroxisome proliferator-activated receptor gamma agonists were also excluded. Midstream urine of spot urine samples were collected in sterile 50-mL tubes that contained 50 μL 0.1 mM PMSF (serine protease inhibitor) and 500 μL 1 mM sodium azide from 86 patients and were stored at −80 • C until use. Informed consent was obtained from all subjects after obtaining approval for the study from the Institutional Review Board at Seoul National University Hospital.
Urine albumin and creatinine were measured in spot urine samples by immunoturbidimetric method using the TIA Micro Alb Kit (Nittobo, Tokyo, Japan) and enzymatic creatinine assay (Roche, Mannheim, Germany), respectively, on a Hitachi 7170 autoanalyzer (Hitachi, Tokyo, Japan).
For the iTRAQ and 2-DE experiments, pooled urine samples, based on average albumin-to-creatinine ratios, were used; the clinical characteristics of the study subjects are summarized in Table 1. Because the protein concentration of each urine sample varied widely, depending on the urine volume in the morning, equal amounts of total protein from each patient were pooled to prepare the urine samples (NA1-NA4 and MA1-MA4).
To prepare the protein samples, approximately 50 mL aliquots of normoalbuminuric and microalbuminuric urine were centrifuged at 3000 g for 30 min at 4 • C. Supernatants were filtered through a 0.22 μm MILLEX GP membrane (Millipore, Carrigtwohill, Cork, Ireland) and concentrated to 3 mL in an Amicon ultrafiltration cell (YM2, 3 kDa MW cutoff, Millipore). The concentrated urine samples were then desalted by dialysis twice using a Slide-A-Lyzer dialysis cassette kit (3.5 kDa, Pierce, Rockford, ILUSA) against 1000 volumes of distilled water, containing 0.1 mM PMSF (serine protease inhibitor) and 1 mM β-ME, at 4 • C. Proteins in the dialyzed urine were precipitated with 5 volumes of acetone for 4 hrs at −20 • C, and the resulting pellets were washed 3 times with cold acetone; the supernatants were discarded.

Labeling with iTRAQ Reagents.
Aliquots of 100 μg of protein were reduced, alkylated, digested, and labeled according to the manufacturer's instructions (Applied Biosystems, Foster City, CA, USA). Briefly, 1 μL of denaturant (2% SDS) and 1 μL of reducing reagent (50 mM tris- [2-carboxyethyl] phosphine) were added to each sample and incubated for 1 hr at 60 • C. Each sample was allowed to cool at room temperature, and 1 μL of cysteine blocking reagent (200 mM methyl methanethiosulfonate (MMTS) in isopropanol) was In this study, 3 iTRAQ experiments were performed. The detailed iTRAQ labeling strategy is summarized for the specified NA/MA urine samples in Figure 1 and Table 1; iTRAQ Experiments 1, 2, and 3 were performed for labeling (a) and (b), (c) and (d), and (e), respectively. Each normoalbuminuric peptide was labeled with iTRAQ reagents 114, 115, and 116, and the microalbuminuric peptide was labeled with iTRAQ reagents 115 and 117 ( Figure 1). The 2 sample sets (microalbuminuric and normoalbuminuric) were combined and dried. To analyze the proteome quantitatively using iTRAQ labeling, we determined the labeling efficiency, as described [23]; the number of possible labeling sites (the N-termini of all peptides and lysine side chains) in 21,610 peptides were compared manually with that of completely labeled sites, represented by the Pro GroupTM Algorithm in ProteinPilot.

Strong Cation Exchange Chromatographic Fractionation.
iTRAQ-labeled samples were subjected to LC-MS/MS at the National Instrumentation Center for Environmental Management, Seoul National University, and fractionated using strong cation exchange (SCX) chromatography, as follows. Dried samples were reconstituted in 500 μL of buffer A (25% v/v acetonitrile (ACN) and 5 mM ammonium formate, adjusted to pH 2.7 with formic acid) and loaded onto a PolySULFOETHYL A column (4.6 mm id × 100 mm, 5 μm, 200Å; PolyLC, Columbia, MD, USA) in a HP1100 series HPLC (Agilent Technologies, Palo Alto, CA, USA). The column was equilibrated for 5 min in buffer A, and the peptides were eluted using a gradient of 0% to 30% buffer B (25% v/v ACN and 1 M ammonium formate [pH 3] with formic acid) over 80 min and 30% to 90% buffer B for 40 min at a flow rate of 0.7 mL/min. Absorbance was monitored at 280 nm, and the fractions were collected every 2 min after injection.

LC-MS/MS Analysis.
Fractions were reconstituted in solvent A and injected into an LC-ESI-MS/MS system. LC-MS/MS was performed using an integrated system, which consisted of an autosampler switching pump and a micropump (Tempo Nano LC system; Applied Biosystems) with a hybrid quadrupole-TOF LC-MS/MS spectrometer (QStar Elite; Applied Biosystems) that was equipped with a nanoelectrospray ionization source (Applied Biosystems) and fitted with a 10 μm fused silica emitter tip (New Objective, Woburn, MA, USA).
For data acquisition, the mass spectrometer was set in the positive ion mode at a selected mass range of 400-1600 m/z for a 1 sec TOF-MS survey scan to detect precursor ions. The 5 most abundant peptides (count >20) with charge states of +2 to +4 were selected to perform the information-dependent acquisition (IDA) of MS/MS data. Once selected, the precursor ions were dynamically excluded for 60 sec at a mass tolerance of 100 ppm.

Data
Analysis. Data file processing, protein identification, and relative abundance quantification were performed using ProteinPilot v.2.0.1 (Applied Biosystems; MDS-Sciex, Concord, Canada) and the Paragon algorithm [24]. Database searches were performed against the Celera human database (human KBMS 5.0, 2005-03-02; a total of 187,748 entries provided by Applied Biosystems). The search parameters used were: a peptide and fragment ion mass tolerance of 0.2 Da; 1 missed trypsin cleavage; fixed cysteine modification by MMTS; variable oxidation of methionine; and iTRAQ labeling of the N-termini of peptides and lysine side chain residues.
The confidence threshold for protein identification was an unused ProtScore >1.3 (95% confidence interval). Pro-teinPilot computes a percentage confidence that reflects the probability that a hit is a false positive; thus, at the 95% confidence level, the false positive identification rate is approximately 5% [24,25]. Although this program automatically accepts all peptides that have a confidence level >1%, only proteins with at least 1 peptide that had a confidence level >95% were initially recorded. At these low confidence levels, peptides do not identify a single protein by themselves but, rather, support protein identification in the presence of other peptides [24,25]. Quantification results were reported only when the error factor (EF) was <2, which indicates a standard deviation of quantification <20%.
2.6. GO Ontology Analysis. The "biological process" and "molecular function" classifications were analyzed using PANTHER ID numbers (http://www.pantherdb.org/), provided in the ProteinPilot output when a Celera human database is used. To construct a graphical representation of differentially excreted proteins, MultiExperiment Viewer (Version 4.3) was used, allowing us to generate a "heatmap" of differentially excreted proteomes (http://www.tm4.org/mev/).

Candidate Validation Using Multiple Reaction Monitor-
ing. In addition to Western blot, multiple reaction monitoring (MRM) was performed to verify the candidate biomarkers using 9 NA and 14 MA urine samples from the urine sample groups (NA1-NA4 and MA1-MA4, resp.) ( Figure 1 and Table 1). In our MRM experiment [28], triple quadrupole linear ion trap MS (4000 Qtrap, coupled with a nano Tempo MDLC, Applied Biosystems) was performed; the detailed procedure is previously described [28]. Data were processed using the MultiQuant program (Applied Biosystems, version 1.0), and each peak area of the transitions was normalized to an input internal standard (Q1/Q3 transitions at 542.3/636.3 m/z for beta-galactosidase peptide) [28]. In the statistical analysis, receiver operating characteristic (ROC) curves and interactive plots were generated using Medicalc (MedCalc Software, Mariakerke, Belgium, version 10.0.1.0).   Figure 2: Correlation between the 2 technical replicates and determination of the cutoff value for significant fold changes. (a) Plots of iTRAQ ratios for two technical replicates. Forty-four proteins were commonly observed from technical replicate 1 (labeled 115/114), and 26 proteins were commonly observed from technical replicate 2 (labeled 117/116). These differentially excreted proteins (P value < 0.05, more than two unique peptides: >95%) were plotted in the linear dynamic range. The technical variations yielded a correlation coefficient of r 2 = 0.9527 and r 2 = 0.9178 between iTRAQ experiments 1 and 2, respectively. (b) The % variations for the common proteins from the two technical replicates. The 44 and 26 common proteins from the 2 technical replicates were used as inputs to calculate % variations. The vertical axis represents the number of proteins, and the horizontal axis denotes % variation. Ninety percent of the proteins fell within 25% of the respective experimental variation. Thus, we considered a fold-change of >1.25 or <0.80, a meaningful cutoff that represented actual differences in the iTRAQ experiments. labeled (a), replicate 2 was labeled (d), and replicate 3 was labeled (e)); 2 technical replicates (technical replicate 1 was labeled (a) and (c), replicate 2 was labeled (b) and (d)) were generated from normoalbuminuric and microalbuminuric urine ( Figure 1). The 3 biological replicates were used to profile and quantitate the urinary proteome; the 2 technical replicates were used solely to determine the cutoff for significant fold-changes.

Identification of Urinary
Seven hundred ten proteins were identified from 21,610 peptides of the 3 combined biological replicates at a minimum confidence level of 95% (unused ProtScore > 1.3). Of the proteins that were identified by iTRAQ, 27% comprised 1-peptide proteins; 14% was 2-peptide proteins; 8% was 3peptide proteins; 5% was 4-peptide proteins; 46% comprised proteins that had 5 or more peptides. In our iTRAQ experiment, 83 proteins (unused ProtScore > 1.3) were common to all 3 biological replicates at a minimum confidence level of 95%, using 3 different pooled urine samples.

Determination of Cutoff for Significant Fold-Change in iTRAQ Experiments.
To generate the quantitative proteome using iTRAQ labeling, we first determined the labeling efficiency, which exceeded 98% (data not shown). Next, the cutoff for significant fold-change was determined based on the 2 technical replicates ((a) and (c) of iTRAQ experiment 1, The technical variations for the 115/114 and 117/116 reporter ions, calculated using the ratios of the 44 and 26 commonly observed proteins between the 2 technical replicates, were r 2 = 0.9527 and r 2 = 0.9178, respectively ( Figure 2(a)). Accordingly, 90% of the commonly observed in the technical replicates fell within 25% of the respective experimental variation (Figure 2(b)). Therefore, we set fold-change thresholds of >1.25 or <0.80 to identify true differences between the expression of 115/114 and 116/117 reporter ions, as described in [23].

Differential Proteomes between Microalbuminuria and
Normoalbuminuria. In our iTRAQ study, we obtained diverse biomarker candidates from 3 pooled biological NA/MA Experimental Diabetes Research 7

Tr T T ansfe f f r pr
Signaling m

Cell adhesi
The others The others Selected regulatory molecule (13) Figure 3: Comprehensive functional annotation of the differentially excreted proteome. The 196 quantitated urinary proteins were annotated for the "biological process" (x-axis) and "molecular function" subcategories (y-axis) in a heatmap. The "biological process" and "molecular function" categories comprised 8 and 13 subcategories, respectively. The 196 quantitated urinary proteins are individually assigned into the "biological process" and "molecular function" subcategories.
urine samples (each pooled NA or MA urine sample consisted of 9 individual urine specimens; thus, the 3 pooled NA, and 3 pooled MA urine samples comprised 54 different individual urine samples). Further, biomarker candidates were confirmed and validated by 2-DE, Western blot, and MRM.
To analyze urinary proteomes in normoalbuminuria and microalbuminuria subjects, 3 biological replicates were generated, wherein 196 proteins met the following criteria: P value < 0.05, EF < 2, more than 2 unique peptides with >95% confidence level, and protein expression >1.25 or <0.80 for all reporter ions; 99 and 97 proteins were upregulated and downregulated, respectively (Appendix A).
These proteins were further analyzed by differential proteomic expression. All quantified proteins were classified into "biological process" and "molecular function" subcategories using the PANTHER classification program, allowing us to analyze phenotypic features and molecular functions between microalbuminuria and normoalbuminuria ( Figure 3). Moreover, to visualize the comprehensive functional annotations graphically, such as in a heatmap, the 196 proteins  Figure 4: Functional distribution of differentially excreted proteins in microalbuminuria versus normoalbuminuria. Functional classification of differentially excreted proteins into (a) "biological process" and (b) "molecular function." Only 5 major subcategories for "biological process" and "molecular function" are shown; each subcategory is presented as the percentage of up-and down-regulated proteins.
were first categorized by "biological process," and a second dimension was added to coordinate the "molecular function" subcategories ( Figure 3). The 196 proteins constituted a preliminary list of biomarkers; Figure 3 shows the expression ratios of the iTRAQ dataset and differentially excreted proteins in microalbuminuria versus normoalbuminuria urine.

Differentially Excreted Urinary Proteins
Are Associated with Pathogenic Status. One hundred ninety-six tentative biomarker candidates were differentially expressed, based on the iTRAQ data, and they were characterized biologically according to "biological process" and "molecular function." Moreover, in a detailed association study of diabetic nephropathy and differentially excreted proteins using references and databases, we prioritized 10 candidates, of the 196 differentially excreted proteins (Figure 3), that were associated with pathogenic status, such as glomerular and tubular dysfunction and other types of diseases. Accordingly, these proteins were classified into categories of pathogenesis in Table 2.
Moreover, several differentially excreted proteins that were related to tubular dysfunction, such as vitamin D-binding protein (VDBP) and alpha-1-acid glycoprotein 1 precursor (AGP1) were selected for further validation. VDBP and AGP1 were upregulated in microalbuminuria versus normoalbuminuria (Table 2).
In addition, FABP (fatty acid-binding protein) and PSCA (prostate stem cell antigen) correlate with other types of disease, and PSCA was selected for further validation. In this iTRAQ experiment, FABP was downregulated, whereas PSCA expression increased in the microalbuminuric versus normoalbuminuric urinary proteome ( Table 2).

MRM Validation for 7 Selected Biomarker Candidates.
To verify the 7 biomarker candidates (TF, CP, A1AT, HP, VDBP, AGP1, and PSCA), MRM was performed using 9 individual normoalbuminuric and 14 microalbuminuric samples ( Figure 1 and Table 1). The peak area for each Q1/Q3 transition (Table 4) for the candidates was first normalized to the peak area of beta-galactosidase that was spiked with 50 fmol as the internal standard and compared between microalbuminuric versus normoalbuminuric samples.
In the interactive plots, TF, CP, A1AT, VDBP, AGP1, and PSCA were upregulated in microalbuminuric versus normoalbuminuric urine, whereas HP was down-regulated. TF, CP, A1AT, VDBP, AGP1, and PSCA had the same excretion patterns by iTRAQ and western blot; conversely, HP had the opposite excretion pattern.

Differentially Excreted Proteomes between Microalbuminuria and Normoalbuminuria.
To identify and quantify proteins that were associated with diabetic nephropathy in microalbuminuric and normoalbuminuric urine, we used relative quantitative proteomic techniques, such as iTRAQ, 2-DE, Western blot, and MRM. In our iTRAQ experiment, 710 urinary proteins were identified at a >95% confidence level, of which 196 were differentially excreted by >1.25   [29]. Using this database, we can easily determine whether a biomarker candidate has been identified by another group for the same disease and evaluate its disease specificity. Thirty-six of the 196 quantified proteins from our iTRAQ experiment were registered in the UPB database; the remaining 160 proteins were not listed.  Total protein (100 μg) samples were loaded onto IPG strips (pH 4-7, nonlinear) for IEF and separated in the second dimension on a 12% polyacrylamide gel. The horizontal and vertical axes represent pI and molecular weight, respectively. The arrowed numbers denote for differentially excreted proteins and correspond to the proteins in Table 3. Subsequently, 196 differentially excreted proteins yielded 10 preliminary biomarker candidates for further validation.
The "molecular function" subcategories accounted for 196 differentially excreted proteins, of which "immunity" represented 25 of the quantitated proteins-the largest component ( Figure 3). This proportion reflects the increased inflammatory reactions and higher vascular lesion counts in kidneys during the development of diabetic nephropathy.
Through detailed association searches between diabetic nephropathy and the 196 differentially excreted proteins using relevant databases and references, we identified and classified several biomarker candidates that were associated with pathogenic status, such as glomerular and tubular dysfunction and other types of disease (Table 2).
Consequently, 10 proteins were selected for preliminary validation studies such as 2-DE, Western blot, and MRM.  In the ROC curves, the solid lines represent the corresponding score in sensitivity (x-axis) and 100-specificity (y-axis). In the interactive plots, the y-axis indicates the normalized concentration of the target protein against the spiked internal standard (50 fmol of beta-galactosidase peptide). Sens. and Spec. represent the sensitivity and specificity for the target proteins, respectively. The AUC values are shown inside the ROC curves.
For example, 3 proteins (HSPG, FABP, and MASP2) that were identified by 2-DE had the same pattern of differential excretion in the 2-DE and iTRAQ experiments ( Table 2). Six differentially excreted proteins (TF, CP, A1AT, VDBP, AGP1, and HP) were validated by Western blot, for which the patterns of excretion were consistent with the iTRAQ results (Table 2).
Recently, an optimized quantitative proteomic strategy in urinary proteomic analysis was proposed for urine biomarker discovery using a small set of samples [21]. According to this report, the initial amount of proteins that is analyzed and the precipitation method (methanol precipitation) of the urine proteins are critical. Notably, our preparation methods for urinary proteomes approximated this optimized  protocol, although we used acetone precipitation instead of methanol precipitation. Nevertheless, an advantage of our study was that we used a large collection of urine samples from 86 diabetic patients to perform 3 iTRAQ experiments, including 3 biological replicates and 2 technical replicates, resulting in more reliable statistical significance.

Differentially Excreted Proteome and Glomerular
Dysfunction. Glomerular dysfunction is caused by GBM thickening and mesangial expansion due to ECM accumulation [2], and several proteins, such as TF, CP, MASP2, A1AT, HP, and HSPG, were associated with glomerular dysfunction.
Transferrin-to-creatinine and ceruloplasmin-to-creatinine ratios are known to reflect changes in renal hemodynamics, and these ratios are significantly higher microalbuminuric patient than normoalbuminuric patients [30]. This result is caused by elevated intraglomerular hydraulic pressure, which leads to the development of diabetic glomerulosclerosis [30][31][32]. Moreover, TF and CP were listed in the UPB database, showing upregulation. TF is associated with diabetic nephropathy, normoalbuminuric type 2 diabetes, kidney calculi, and ureteropelvic junction obstruction, whereas CP is linked to diabetic nephropathy and normoalbuminuric type 2 diabetes. In our iTRAQ and 2-DE, TF and CP urinary proteins were commonly upregulated in the comparison of microalbuminuric and normoalbuminuric urinary proteome ( Table 2).
MASP2 is a serum protease that activates the complement cascade, which regulates the maintenance of glomerular permeability and the pathogenesis of focal segmental glomerulosclerosis [33]. MASP2 was not listed in the UPB database, and in this iTRAQ and 2-DE, this protein was commonly downregulated in comparison of microalbuminuric and normoalbuminuric urinary proteome (Table 2).
A1AT is a serine protease inhibitor, which prevents neutrophil elastase by degrading ECM proteins, which maintains vascular elasticity and glomerular integrity [13,34]. In a previous study, A1AT excretion was elevated in microalbuminuria, which can cause matrix molecules to accumulate [35]. A1AT was also listed in the UPB database, showing upregulation; it is associated with diabetic nephropathy, severe acute pancreatitis, kidney calculi, nephrotic syndrome, and ureteropelvic junction obstruction. In our iTRAQ and Western blot, A1AT was also upregulated in the comparison of microalbuminuric and normoalbuminuric urinary proteome (Table 2). Haptoglobin (HP) is an acute phase protein that binds to free hemoglobin (Hb) with the highest affinity. Formation of the Hb-HP complex prevents the loss of renal iron and oxidative damage that are driven by free Hb. Specifically, HP mediates complement-dependent podocyte damage [36]. Thus, complement activation results in the release of proteases, oxidants, and growth factors, damaging the functional integrity of the GBM. HP, which was downregulated and is associated with diabetic nephropathy and type 2 diabetes mellitus, appeared in the UPB database. However, in our iTRAQ and Western blot, HP was commonly upregulated in the comparison of microalbuminuric and normoalbuminuric urinary proteome ( Table 2).
HSPG is present in the basement membrane of every vascularized organ, including the GBM. The highly negatively charged side chains of HSPG are important determinants for the charge-selective permeability of the GBM [37]. Under hyperglycemic conditions, the loss of HSPG from the GBM alters the charge-selective properties of the glomerular capillary, causing increased filtration of negatively charged albumin [37,38]. HSPG is not registered in the UPB database, and in our iTRAQ and 2-DE, HSPG was commonly downregulated in the comparison of microalbuminuric and normoalbuminuric urinary proteome ( Table 2).

Differentially Excreted Proteome and Tubular Dysfunction and Other Types of Diseases.
Low-molecular-weight proteins, such as VDBP and AGP1, are associated with kidney tubular dysfunction and were commonly upregulated in the iTRAQ and Western blot ( Table 2). Unlike high-molecular-weight proteins, they are filtered in the glomerulus on the basis of charge selectivity and pore size of the GBM and are reabsorbed into proximal renal tubules under normal conditions        [38]. If reabsorption is impaired, however, these proteins can be overexcreted into urine.
VDBP is a multifunctional serum glycoprotein and is an important mediator in immunopathogenesis because it is associated with immunoglobulin receptors on the surfaces of B-and T-lymphocytes [12]. Moreover, VDBP binds to circulating vitamin D metabolites with high affinity [39], and VDBP-bound 25-hydroxyvitamin D3 crosses the GBM and is reabsorbed by proximal tubular cells in a megalin-dependent manner, suggesting that it controls renal uptake and the activation of metabolites [4,40]. VDBP, also listed in the UPB database, was upregulated and is associated with diabetic nephropathy and Dents disease. In our iTRAQ and Western blot, VDBP was commonly upregulated in the comparison of microalbuminuric and normoalbuminuric urinary proteome (Table 2).
AGP1 is synthesized in response to systemic tissue injury, inflammation, and infection, like most other acute phase proteins [7,41]. Further, increased levels of AGP1 reflect elevated levels of cytokines, such as interleukin-1 (IL-1), IL-6, and tumor necrosis factor alpha (TNF-α), which are associated with type 2 diabetes [7]. A study has demonstrated that AGP1 is an indicator of tubular disorder in multiple myeloma [42], and another has shown that serum and urinary AGP1 levels are elevated in type 2 diabetic patients with kidney disorders [5,7]. AGP1 was listed in the UPB database, showing upregulation, and is associated with diabetic nephropathy, diabetic kidney disorder, preeclampsia, and acute appendicitis. In our iTRAQ and Western blot, AGP1 was also commonly upregulated in the comparison of microalbuminuric and normoalbuminuric urinary proteome ( Table 2).
FABP (fatty acid-binding protein: FABP) correlated with other types of diseases [43] and was also chosen for further validation. FABP was not listed in the UPB database, and in this iTRAQ and 2-DE, FABP was commonly downregulated in the comparison of microalbuminuric and normoalbuminuric urinary proteome (Table 2).
PSCA is highly expressed in the prostate and, to a lesser extent, in the bladder, placenta, colon, kidney, and stomach. Moreover, it is upregulated in prostate cancer and is detected in cancers of the bladder and pancreas [44]. PSCA was not listed in the UPB database but was upregulated in the microalbuminuric versus normoalbuminuric urinary proteome in  our iTRAQ experiment (Table 2). No relationship between prostate stem cell antigen and diabetic nephropathy has been reported.

Validation of Differentially Excreted Proteins Using MRM.
For the MRM experiments, we used a bacterial beta galactosidase peptide as the internal standard for relative quatitation [28]. Seven preliminary biomarker candidates (TF, CP, A1AT, VDBP, AGP1, HP, and PSCA) were confirmed in 9 normoalbuminuric and 14 microalbuminuric urine samples by MRM (Figure 7). In the interactive plots, TF, CP, A1AT, VDBP, AGP1, and PSCA were preferentially excreted in microalbuminuria versus normoalbuminuria, whereas HP was downregulated. TF, CP, A1AT, VDBP, and AGP1 had the same pattern of excretion in the iTRAQ and western analysis, and HP had the opposite pattern between the MRM and Western analysis. HP consists of α-chain (amino acid sequence: 19-160) and βchain (aminoacid sequence: 162-406), which are connected by disulfide bridges. In the MRM experiment, the transition (amino acid sequence: 203-215) in the β-chains was used for the relative quantitation of HP. In contrast, portions of both the α-chain and β-chain were used in the iTRAQ quantitation (sequence coverage: 31.0%), and an antibody that targeted a sequence in the α-chain was used for the Western blot analysis. It is conceivable that these disparate targets resulted in contradictory patterns between iTRAQ, Western blot, and MRM. Regardless of the methods or targets, reproducible patterns must be obtained with each method.
Although our results require further validation in a larger collection of urine samples that contains various control samples, it appears that A1AT, AGP1, and PSCA are excellent biomarker candidates, with AUC values > 0.8; combining the candidates improved the AUC value of 0.921. Accordingly, the other differentially expressed proteins from our iTRAQ experiment in Appendix A are potential candidates for further validation in obtaining DN biomarkers.

Conclusions
Microalbuminuria is used as a noninvasive index for the detection of diabetic renal disease. Yet, more specific and accurate biomarkers for DN are required, particularly in type 2 diabetic patients, due to several reasons, including nonspecific detection in nondiabetic renal disease, cardiovascular disease, inflammation, and hypertension. In our iTRAQ experiments, 710 urinary proteins were identified at a >95% confidence level, of which 196 were differentially excreted by >1.25 or <0.80-99 and 97 proteins were up-and down-regulated, respectively (Appendix A). We prioritized 196 proteins to select preliminary biomarker candidates by characterizing them with regard to "biological process" and "molecular function" and associating them with pathogenesis. Consequently, 10 proteins were selected. To confirm and validate these candidates, 2-DE, Western blot, and MRM were performed. Based on the MRM results, A1AT, AGP1, and PSCA, which had AUC values > 0.8, are good biomarker candidates, and we improved the AUC value to 0.921 on combining the 3 proteins.
Further validation studies on other differentially excreted proteins might contribute to a greater understanding of the mechanism of renal dysfunction and its association with the pathogenesis of DN, facilitating the development of better biomarkers for DN.

A. Differentially Excreted Urinary Proteome in Microalbumiuric versus Normoalbuminuric Urine
For more details, see Table 5.

B. Mascot Search and ROC Curves for Each Transition of MRM
To verify the MRM validation, we analyzed minimum of 2 transitions for each biomarker candidate protein.