Urine complement proteins are associated with kidney disease progression of type 2 diabetes in Korean and American cohorts

Study patients

From March 2011 to June 2021, 93 patients with type 2 diabetes and biopsy-proven diabetic nephropathy were recruited at Seoul National University Hospital (SNUH), referred to as the SNUH-DN cohort. Of this group, 29 were excluded due to an estimated glomerular filtration rate (eGFR) <30 mL/min/1.73 m² (n = 25), combined type 1 and type 2 diabetes (n = 2), the presence of concomitant nondiabetic glomerular disease (n = 1), or the absence of glomeruli in biopsied kidney tissues (n = 1). The remaining 64 patients were included in the final analyses as the SNUH-DN cohort after conducting untargeted proteomics.

Additionally, this study incorporated a subset of patients with type 2 diabetes from the Chronic Renal Insufficiency Cohort (CRIC) study, termed CRIC-T2D, consisting of American patients with risk factors for cardiovascular disease, progression of chronic kidney disease, and mortality. A total of 3,939 patients aged 21–74 with an eGFR of 20–70 mL/min/1.73 m² were enrolled at seven clinical centers throughout the US from June 2003 to December 2008. CRIC over-sampled participants with diabetes, who constitute nearly half of the CRIC cohort. This study excluded participants with monogenic renal disease, liver cirrhosis, class III or IV heart failure, human immunodeficiency virus, cancer, autoimmune disease, polycystic kidney disease, pregnant women, recipients of organ or bone marrow transplants, and those receiving immunotherapy for primary renal disease or systemic vasculitis within the six months preceding recruitment or systemic chemotherapy.

Ethical considerations

This study was conducted in accordance with the ethical principles outlined in the Declaration of Helsinki and was approved by the institutional review boards (IRBs) of all participating institutions. For the SNUH-DN cohort, ethical approval was granted by the IRB at SNUH (no. H-2403-066-1519). The CRIC study was approved by the IRBs at all participating institutions and was conducted in accordance with the Declaration of Helsinki ^{16, 17}. Urine samples and data for the present study were obtained from the CRIC subset housed at the NIDDK Central Repository (R01 5R01DK104706). The use of de-identified CRIC samples was approved by the IRB at the University of California, Davis. The use of de-identified samples and data from the CRIC-T2D subgroup from the NIDDK Central Repository was approved by the Human Subjects Division, University of California, Davis. All participants provided written informed consent, having been fully informed of the study purpose, the procedures involved, potential risks, and their rights as study participants, including the right to withdraw at any point without consequence. The confidentiality of patient data was rigorously maintained throughout the study. All personal identifiers for SNUH-DN patients were removed, and data was stored securely to prevent unauthorized access. CRIC-T2D data and samples were de-identified at the NIDDK Central Repository.

Study outcomes

The primary outcome was the progression of DKD (named kidney disease progression) based on any of the following events: a doubling of baseline serum creatinine, a decline greater than 50% in the eGFR, or the onset of ESKD. The eGFR was calculated using the Chronic Kidney Disease Epidemiology Collaboration equation ¹⁸. ESKD was specifically defined as the need for kidney replacement therapy, including hemodialysis, peritoneal dialysis, or kidney transplantation.

Study variables

We collected comprehensive patient data encompassing demographic characteristics, such as age, sex, race, body mass index, diabetic duration, and comorbidities of hypertension, ischemic heart disease, cerebrovascular disease, retinopathy, and neuropathy. Laboratory findings included hemoglobin A1c, eGFR, and random urine protein-to-creatinine ratio (PCR) for the SNUH-DN cohort, while the CRIC-T2D cohort included the 24-hour urine protein amount.

We employed a detailed histopathologic classification system to evaluate kidney biopsies from patients diagnosed with diabetic nephropathy in the SNUH-DN cohort. This classification system follows the criteria described in the reference paper ¹⁹. Each biopsy was required to contain at least ten glomeruli for accurate assessment, excluding any incomplete glomeruli along the biopsy edges. The glomerular classification was segmented into five categories: class I, isolated glomerular basement membrane thickening; class IIA, mild mesangial expansion; class IIB, severe mesangial expansion; class III, nodular sclerosis (Kimmelstiel-Wilson lesions); and class IV, advanced diabetic glomerulosclerosis with more than 50% global glomerulosclerosis. The assessment of interstitial fibrosis and tubular atrophy (IFTA), interstitial inflammation, arterial hyalinosis, and arteriosclerosis was also quantified using a four-point or three-point scale according to the criteria.

Urine sample collection

In the SNUH-DN cohort, urine samples were collected at the time of kidney biopsy. Patients were instructed to provide a midstream clean-catch urine sample, which was immediately processed to minimize protein degradation. Upon collection, urine samples were centrifuged at 3000 rpm for 10 minutes to remove cellular debris and then aliquoted into cryovials. The aliquots were promptly frozen and stored at –80°C to preserve the integrity of the proteins until preparation for proteomics.

In the CRIC-T2D cohort, 24-hour urine samples were collected at home by study participants during their CRIC visit number 3. Participants were instructed to keep the samples at 4°C during collection. Collections that were less than 500 ml or collected for <22 hours or >26 hours were repeated. Accepted 24-hour collections were thoroughly mixed, and three 10 ml aliquots were obtained. The aliquots were shipped to the CRIC central laboratory on ice packs, where they were further aliquoted and stored at –80°C. A portion of these samples was shipped to the NIDDK Central Repository when they were sent to a study author (MA). An aliquot of these samples was then shipped to the Pacific Northwest National Laboratory on dry ice and were prepared for targeted proteomics.

Untargeted proteomics

Construction of a peptide spectral library

To create a peptide spectral library for using match between runs (MBRs) algorithm of maxquant, a pooling of urine proteins was performed ²³. We combined equal sub-aliquots of proteins, each containing 5 µg, to a total of 100 µg. This pooled sample was then subjected to a two-step filter-aided digestion process, consistent with the methods previously outlined ^{21, 24}. Following digestion, peptides were purified using Oasis HLB solid-phase extraction cartridges to ensure their readiness for high-performance liquid chromatography. For the library construction, we utilized 100 µg of these clean peptides and processed them through an Agilent 1260 bioinert HPLC system (Agilent, Santa Clara, CA) equipped with a standard analytical column (4.6 × 250 mm; 5-µm particle size). The peptides were fractionated using a high-pH reversed-phase method, employing a flow rate of 0.8 ml/min across a 60-minute gradient. Solvents included 15 mM ammonium hydroxide in water (solvent A) and 15 mM ammonium hydroxide in 90% acetonitrile (solvent B). We collected 96 separate fractions at 30-second intervals throughout a 48-minute period, which were then concatenated into 24 fractions in step-wise manner. Early-, middle-, and late-eluting peptides were pooled by mixing every 24^th original fraction for the proteome (e.g., combining fractions 1, 25, 49, and so on) ²⁵. These concatenated fractions were subsequently dried using a vacuum centrifuge and stored at –80°C, ready for analysis by liquid chromatography-tandem mass spectrometry (LC-MS/MS).

LC-MS/MS analytical procedure

LC-MS/MS analysis was conducted on a Q-Exactive Plus Quadrupole Orbitrap mass spectrometer (Thermo Fisher Scientific, Waltham, MA), which was interfaced with an Ultimate 3000 RSLC nano system (Dionex, Sunnyvale, CA) featuring a nanoelectrospray ion source. The process incorporated slight modifications from previously established protocols ^{20, 22}. Individual peptide fractions from urine samples underwent separation using a dual-column arrangement consisting of a C18 trap column (300 µm internal diameter × 0.5 cm, with 3 µm particles and 100 Å pore size) and a C18 analytical column (50 µm internal diameter × 50 cm, with 1.9 µm particles and 100 Å pore size). Prior to injection, dried peptide fractions were dissolved in solvent A (2% acetonitrile and 0.1% formic acid). The nano-LC system loaded the samples and applied a chromatographic gradient over 90 minutes, increasing from 8% to 30% of solvent B (100% acetonitrile with 0.1% formic acid). The setup operated under a spray voltage of 2.0 kV in positive ion mode, and the capillary heater was maintained at 320°C. Mass spectrometric data were collected in a data-dependent acquisition format, prioritizing the top 15 most abundant precursor ions. The Orbitrap detector analyzed precursor ions across a mass range from 300 to 1650 m/z, achieving a resolution of 70,000 at m/z 200. Fragmentation was performed via higher-energy collisional dissociation (HCD) at a resolution of 17,500 for m/z 200, using a normalized collision energy setting of 28. The system’s ion injection times were set to a maximum of 20 ms for survey scans and 120 ms for MS/MS scans.

Label-free quantitative data analysis

For the analysis of mass spectrometric data, we employed MaxQuant software (version 1.6.1.0) for label-free quantification ²⁶. MS/MS spectra were meticulously matched against the Human UniProt protein sequence database (as of December 2014, containing 88,657 entries) via the Andromeda search engine ²⁷. The analysis parameters were carefully selected, with a precursor ion tolerance set at 6 ppm for total protein analysis and an MS/MS ion tolerance of 20 ppm. We designated carbamido-methylation of cysteine as a fixed modification, while N-acetylation of proteins and methionine oxidation was considered variable modifications. The digestion parameter was configured for full tryptic digestion with allowances for up to two missed cleavages per peptide, which had to be at least six amino acids long. We rigorously maintained a false discovery rate (FDR) of 1% across peptide, protein, and modification levels to ensure high data integrity. Additionally, to enhance quantification accuracy across samples, we utilized a strategy of matching between runs anchored by the peptide library constructed from pooled urine samples. The Intensity Based Absolute Quantification (iBAQ) algorithm, a component of the MaxQuant suite, facilitated the estimation of protein abundances ²⁸. In iBAQ, raw intensities were normalized by the count of theoretical peptides, rendering them reflective of the proteins’ molar quantities.

Targeted proteomics

Sample processing and digestion

Urine samples were processed and underwent LC-MS/MS, as described previously ²⁹. Briefly, urine samples were thawed on ice, run through a 10 kDa prewashed (50 mM AMBIC, i.e., NH₄HCO₃, pH 8.0) Amicon ultrafiltration column, and underwent buffer exchange with AMBIC twice. The final urine protein concentration was determined using the BCA protein assay. Urea was added to the urine sample to a final concentration of 8 M, followed by reduction with 500mM DTT to a final concentration of 10 mM, brief sonication, and incubation at 37°C for 1 hour while shaking. Alkylation was performed by adding IAA to a final concentration of 40 mM and incubation for one hour at 37°C in the dark while shaking. The sample was then diluted ten-fold with digestion buffer, and trypsin was added to a protein/trypsin ratio of 50:1 (w/w), followed by incubation at 37°C for 3 hours. Digestion was stopped by adding TFA to a final concentration of 0.1%. The tryptic digest was separated using SPE C18 columns, preconditioned using methanol and then SPE conditioning buffer, and washed with SPE washing buffer. Peptides were eluted from the SPE C18 columns with 1ml SPE elution buffer and, dried under a reduced vacuum using a speed vac and redissolved in MS-grade water. The peptide concentration was determined using the BCA protein assay, and peptide samples were stored at –80°C until further use. Heavy isotope-labeled peptide standards were mixed in a single 1 ml stock solution, having been reconstituted at 1000 fmol/ml per peptide in 0.1% TFA in water. The digested peptide mix and the heavy peptide stock were combined while shaking and advanced to LC-SRM analysis.

LC-SRM analysis and quantification

The urine sample/heavy peptide mixes were analyzed using the final transition list, derived from the orbitrap HCD MS/MS, and optimized in one typical urine sample. LC-SRM was performed using a nanoACQUITY UPLC system coupled online to a TSQ Vantage triple quadrupole mass spectrometer. The peak area ratios (PAR) of endogenous transitions and heavy internal standard transitions represent the molar ratios between the amounts of endogenous and heavy internal standard peptides. The PARs were calculated using Skyline software ³⁰. The data was imported into a Skyline file, peak boundaries for each peptide were manually verified for each dataset, and detection and optimal transition (for best peak area ratio, PAR) were confirmed for each peptide. Protein concentration was calculated using the equation below. where L/H is PAR of endogenous (L) to heavy isotope-labeled internal standard peptide (H), C_heavy, LC-SRM is the molar concentration of heavy internal standard peptides (fmol/μl) in the final LC-SRM solutions; MW is the molecular weight of targeted protein (Da or g/mol); C_{digest, LC-SRM} is the mass concentration of total protein digest (μg/μl) in the final LC-SRM solutions; and C_protein is mass concentration of total protein (μg/μl) in the 400 μl concentrated retentate from 10 ml of original urine.

Proteomics data processing and analysis

The proteomic data obtained using label-free quantification of urine proteins were initially processed using the Seurat package (version 5.0.3) ³¹. These values were normalized for urine dilution by indexing (division) by urine creatinine, and then log2 transformed for further normalization using the NormalizeData function from the Seurat package. Finally, for each identified protein, the normalized data was scaled using the ScaleData function from the Seurat. Unsupervised clustering techniques were utilized to group proteins based on their expression patterns, and the FindNeighbor function and RunUmap function were used to create cluster groups and to draw UMAP, respectively. Gene names related to each protein were identified using the UniProt accession numbers ³². Differentially expressed proteins between rapid and slow progressors were identified (FindMarkers function). Using the UniProt database, proteins with well-documented gene associations were selected for pathway analysis, and differentially represented pathways were identified using WikiPathways ³³.

Complement scores

The complement score was devised to evaluate the prevalence of complement pathway proteins in urinary proteomics data. We identified the proteins involved in the complement pathway by referencing the Kyoto Encyclopedia of Genes and Genomes database ³⁴. The list of complement proteins used for scoring is presented in Supplementary Table 2. The average protein expression of denoted genes was calculated based on the assumption of previous studies ^35–37. Additionally, we formalized the distinction between activators and inhibitors. For each patient, this score was calculated by taking the mean of the normalized abundance of identified complement proteins, adjusted for their functional roles: inhibitory proteins were assigned a modifier of –1, and activating proteins a modifier of 1. The formula used was: where i denotes the sample number, and j is the individual complement protein. The continuous variable complement score was binarized into low vs. high complement score bins by the median value.

Statistical analysis

Descriptive statistical analyses were conducted on patient information. For continuous variables with a normal distribution, data are presented as the mean ± standard deviation, and for those without a normal distribution, as medians with interquartile ranges (IQR). The Kolmogorov-Smirnov test was used to assess the normality of the data distribution. Categorical variables are shown as percentages. Comparison of categorical variables was performed using either the chi-square test or Fisher’s exact test, while continuous variables were compared using either Student’s t-test or the Mann-Whitney U test, depending on their distribution. When comparing multiple groups for continuous variables, analysis of variance (ANOVA) was used, with the groups treated as ordinal variables. Pairwise comparisons were conducted with Bonferroni correction.

Survival analysis was employed to investigate the association between variables and the risk of kidney disease progression. The hazard ratio (HR) and corresponding 95% confidence intervals were calculated using the Cox proportional hazards regression model. Variables included in the model were selected based on clinical relevance, such as age, sex, body mass index, diabetic duration, and baseline kidney function. Adjusted survival curves were then generated to visualize the differences in adjusted cumulative survival between the two groups using the adjustedCurves package ³⁸. Additionally, penalized spline regression was utilized to further elucidate the relationship between complement scores and kidney disease progression. All statistical analyses were conducted using the R program (version 4.2.2). A two-sided P value of less than 0.05 was considered statistically significant.

Baseline characteristics in the SNUH-DN cohort

This cohort had a mean age of 53.3 ± 11.1 years, with 71.9% male subjects. Hypertension was present in 92.2% of the subjects, and the median duration of diabetes was 10 years (IQR, 4– 14 years). Diabetic retinopathy and neuropathy were observed in 54.7% and 29.7% of all patients, respectively. The median hemoglobin A1c was 7.2% (IQR, 6.6–8.4), while the median PCR and eGFR values were 3.1 g/g (IQR, 1.7–7.0) and 55 mL/min/1.73 m² (IQR 44–75), respectively. Details of histopathologic features are provided in Table 1.

Proteomic clustering in the SNUH-DN cohort

Unsupervised clustering of the iBAQ-normalized urine protein abundances from the SNUH-DN cohort identified two distinct clusters, designated as clusters 1 and 2 (Figure 1A). Heatmap of iBAQ-normalized urine proteins that were abundant and scarce in clusters 0 and 1 are illustrated in Supplementary Figures 1A and B, respectively. These data showed that of the 1,877 identified proteins, 71 and 85 were significantly abundant and scarce in cluster 2, compared to cluster 1, after the Bonferroni adjustment, respectively. Many of the abundant proteins mapped to the complement pathway, either complement activators such as complement factor D (CFD), component 2 (C2), component 8 beta chain (C8B), component 6 (C6), and component 5 (C5) ³⁹, or complement inhibitors such as complement factor H (CFH) and serpin family G member 1 (SERPING1) ^{40, 41}. Conversely, among the scarce proteins, CD55 and CD59 act as complex inhibitors ⁴², while the protein C receptor (PROCR) functions indirectly as a complement inhibitor ⁴³.

• Download figure
• Open in new tab

Figure 1. Unsupervised clustering based on untargeted proteomics in the SNUH-DN cohort.

A. Uniform manifold approximation and projection (UMAP) plot with two distinct clusters (clusters 1 and 2). B. Multivariable-adjusted cumulative incidence of kidney disease progression according to clusters, after adjustment for age, sex, body mass index, hypertension, diabetic duration, urine protein-to-creatine ratio, and eGFR. C. Sankey plot illustrating the relationships among glomerular classes, clusters, and kidney disease progression within one year.

Kidney disease progression within one year after biopsy occurred in 6 out of 24 subjects (25%) with urine proteins in cluster 2, and 2 out of 40 subjects (5%) in cluster 1. Compared to subjects in cluster 1, those in cluster 2 had a 4.62-fold (95% CI, 1.86–11.50) greater hazard of DKD progression in multivariable time-to-event (Cox proportional hazard) regression, adjusted for age, sex, body mass index, hypertension, diabetic duration, PCR, and eGFR (Table 2). The adjusted cumulative incidence of DKD progression displayed more rapid kidney disease progression in cluster 2 than in cluster 1 (Figure 1B). These results indicate that the unsupervised clustering of urine proteins was able to segregate subjects into two clusters, displaying slower and more rapid DKD progression. Notably, the accelerated DKD progression in cluster 2, compared to cluster 1, was associated with high glomerulosclerosis scores in the paired kidney biopsies from these subjects (Figure 1C).

Different enrichment in complement proteins

Pathway analysis identified a notable enrichment of complement and related pathways in cluster 2 (Figure 2A and Supplementary Table 3). The volcano plot of differentially abundant proteins highlighted that complement activators, including CFD, component 3 (C3), C5, C6, and C8B, and some complement inhibitors, including CFH, complement factor I (CFI), and clusterin (CLU), were prominently ranked in cluster 2. Conversely, other complement inhibitors (CD55 and CD59) were scarce in cluster 2 (Figure 2B). Detailed information regarding the proteins related to the complement pathway is included in Supplementary Table 2. The heatmap with the normalized abundance of complement proteins depicted that complement pathway-related proteins were more prominently altered in cluster 2 than in cluster 1 (Figure 2C).

• Download figure
• Open in new tab

Figure 2. Complement pathway and kidney disease progression in the SNUH-DN cohort.

A. Pathway analysis of the differentially abundant proteins in cluster 2 compared to cluster 1. B. Volcano plot of the differentially abundant proteins in cluster 2 compared to cluster 1. The dashed line denotes the threshold of significance. Red and blue dots represent proteins increased and decreased in cluster 2, respectively, compared to cluster 1. C. Heatmap of protein abundance for urinary complement components according to clustering and complement function.

Complement score and kidney disease progression in the SNUH-DN cohort

To comprehensively assess the effect of complement proteins, we calculated a complement score that considers their expression levels and modifies their contribution by their roles in pathway activation (activators or inhibitors). The overall median complement score for all samples was 0.28 (IQR, 0.08–0.62), with a range from –0.25 to 0.92. Cluster 2 had a median complement score of 0.69 (IQR, 0.51–0.76), while cluster 1 had a median score of 0.14 (IQR, 0.01–0.28) (P <0.001) (Figure 3A). Based on the median score of 0.28 from all samples, we divided the patients into high (>0.28) and low (≤0.28) complement score groups (Supplementary Figure 2).

• Download figure
• Open in new tab

Figure 3. Complement score and kidney disease progression in the SNUH-DN cohort.

A. Bar plot of complement score in clusters 1 and 2. ***P <0.001. B. Multivariable adjusted cumulative incidence of kidney disease progression according to complement score groups after adjustment for age, sex, body mass index, hypertension, diabetic duration, urine protein-to-creatinine ratio, and eGFR. C. Penalized spline regression according to complement score after adjusting for age, sex, body mass index, hypertension, diabetic duration, urine protein-to-creatinine ratio, and eGFR.

Patients with high complement scores had a 2.41-fold (1.07–5.40) higher risk of kidney disease progression than those with low scores, after adjustment for demographic and clinical variables (Table 3). Multivariable-adjusted cumulative incidence of DKD progression also demonstrated a significant difference between the high and low complement groups (Figure 3B). Notably, in multivariable-adjusted spline regression, the association between the complement score and the hazard of DKD progression was linear (Figure 3C).

To eliminate the potential contribution of the urine creatinine denominator to the outcome, a sensitivity analysis was conducted using the urine abundance of proteins without normalization for urine creatinine. These analyses yielded the same results, with cluster 2 and the high complement score group both significantly associated with rapid DKD progression, compared to cluster 1 and the low complement score group, respectively (Supplementary Figure 3).

Histopathologic correlation with complement scores in SNUH-DN cohort

In the SNUH-DN cohort, which included paired urine and kidney tissues from the same individuals, urine complement scores were significantly correlated with three of five features used to classify DKD: glomerular lesions, interstitial fibrosis and tubular atrophy (IFTA), and arteriolar hyalinosis. The correlation between complement score and interstitial inflammation did not reach statistical significance. Urine complement score was not correlated with arteriosclerosis in this sample set (Supplementary Figure 4).

Baseline characteristics in the CRIC-T2D cohort

The association between urinary complement proteins and kidney disease progression was examined in an independent cohort: the CRIC-T2D. The CRIC-T2D cohort (Table 4), comprising 282 patients, was a majority (62%) male group with a mean age of 60.8 ± 7.5 years. Of these patients, 51.8% were non-Hispanic black and 30.1% were non-Hispanic white. Hypertension and cardiovascular disease were identified in 93.6% and 46.5% of the participants, respectively. The median 24-hour urine protein and eGFR were 0.48 g (IQR, 0.10–1.87) and 42 mL/min/1.73 m² (IQR, 37–49), respectively. The median follow-up duration was 9.7 years (IQR, 6.3–12.0).

Complement Score in the CRIC-T2D cohort

In the CRIC-T2D cohort, targeted urinary proteomics was used to quantify 12 complement proteins, including complement activators such as factor B (CFB), C3, component 4 (C4), C5, C6, component 7 (C7), component 8, alpha chain (C8A), and component 9 (C9), and inhibitors such as CD59, CFI, clusterin, and CFH-related protein 2 (Supplementary Table 2 and Supplementary Figure 5). The complement score was calculated for each participant as described in the Complement scores section. The median complement score for all participants was 1.07 (IQR, 0.42–2.10), ranging from –0.76 to 3.04. Based on this median value, the participants were divided into groups with high (>1.07) and low (≤1.07) complement scores (Figure 4A). The median complement scores for the high and low groups were 2.16 (IQR, 1.63–2.45) and 0.41 (IQR, 0.17–0.68), respectively. The group with high scores had a 2.51-fold (95% CI, 1.76–3.57) higher risk of kidney disease progression than the low score group, after adjustment for age, sex, body mass index, hypertension, ethnicity and race, 24-hour urine protein, and eGFR (Figure 4B and Table 5). Multivariable-adjusted spline regression revealed a linear association between the complement score and the hazard of DKD progression (Figure 4C).

• Download figure
• Open in new tab

Figure 4. Complement score and kidney disease progression in the CRIC-T2D cohort.

A. Bar plot of complement score in high and low score groups. ***P <0.001. B. Multivariable-adjusted cumulative incidence of kidney disease progression according to complement score groups after adjustment for age, sex, body mass index, race, hypertension, 24-hour urine protein, and eGFR. C. Penalized spline regression according to complement score after adjusting for age, sex, body mass index, race, hypertension, 24-hour urine protein, and eGFR.