Proteomic Analysis Uncovers Multiprotein Signatures Associated with Early Diabetic Kidney Disease in Youth with Type 2 Diabetes Mellitus

Visual Abstract

Abstract

Key Points

• Proteomics analyses identified seven proteins predictive of time to development of albuminuria among youth with type 2 diabetes in the Treatment Options for Type 2 Diabetes in Adolescents and Youth cohort, 118 proteins predictive of time to development of hyperfiltration, and three proteins predictive of time to rapid eGFR decline.

• Seven proteins were predictive of all three outcomes (SEM4A, PSB3, dihydroxyphenylalanine decarboxylase, C1RL1, T132A, pyruvate carboxylase, and C1-esterase inhibitor) and have been implicated in immune regulatory mechanisms, metabolic dysregulation, proteostasis, and cellular signaling pathways.

• Elastic net Cox proportional hazards model identified distinct multiprotein signatures (38–68 proteins) of time to albuminuria, hyperfiltration, and rapid eGFR decline with concordance for models with clinical covariates and selected proteins between 0.81 and 0.96, whereas the concordance for models with clinical covariates only was between 0.56 and 0.63.

Background

The onset of diabetic kidney disease (DKD) in youth with type 2 diabetes (T2D) mellitus often occurs early, leading to complications in young adulthood. Risk biomarkers associated with the early onset of DKD are urgently needed in youth with T2D.

Methods

We conducted an in-depth analysis of 6596 proteins (SomaScan 7K) in 374 baseline plasma samples from the Treatment Options for Type 2 Diabetes in Adolescents and Youth study to identify multiprotein signatures associated with the onset of albuminuria (urine albumin-to-creatinine ratio ≥30 mg/g), a rapid decline in eGFR (annual eGFR decline >3 ml/min per 1.73 m² and/or ≥3.3% at two consecutive visits), and hyperfiltration (≥135 ml/min per 1.73 m² at two consecutive visits). Elastic net Cox regression with ten-fold cross-validation was applied to the top 100 proteins (ranked by P value) to identify multiprotein signatures of time to development of DKD outcomes.

Results

Participants in the Treatment Options for Type 2 Diabetes in Adolescents and Youth study (14±2 years, 63% female, 7±6 months diabetes duration) experienced high rates of early DKD: 43% developed albuminuria, 48% hyperfiltration, and 16% rapid eGFR decline. Increased levels of seven and three proteins were predictive of shorter time to develop albuminuria and rapid eGFR decline, respectively; 118 proteins predicted time to development of hyperfiltration. Elastic net Cox proportional hazards models identified multiprotein signatures of time to incident early DKD with concordance for models with clinical covariates and selected proteins between 0.81 and 0.96, whereas the concordance for models with clinical covariates only was between 0.56 and 0.63.

Conclusions

Our research sheds new light on proteomic changes early in the course of youth-onset T2D that associate with DKD. Proteomic analyses identified promising risk factors that predict DKD risk in youth with T2D and could deepen our understanding of DKD mechanisms and potential interventions.

Clinical Trial registry name and registration number:

NCT00081328.

Introduction

The burden of complications in individuals with youth-onset type 2 diabetes (T2D) mellitus is substantial, with most developing complications by young adulthood. Diabetic kidney disease (DKD) is particularly common. The Treatment Options for Type 2 Diabetes in Adolescents and Youth (TODAY2) study documented a 15-year cumulative incidence of 55% for moderate or severe albuminuria in youth with T2D transitioning to young adulthood.¹ In addition, we found that youth-onset T2D in American Indian individuals was associated with more severe structural kidney injury and higher risk of kidney failure and mortality, compared with adult-onset T2D.^2,3 Youth-onset T2D is characterized by higher degrees of insulin resistance and hyperinsulinemia, more rapid beta-cell failure, and enrichment of rare and common genetic variants related to obesity, monogenic diabetes, and beta-cell function compared with adult-onset T2D.⁴ In addition, growth and hormonal aspects unique to puberty and psychosocial aspects unique to adolescence all necessitate studies dedicated to youth. Despite the high incidence and severity of DKD in youth-onset T2D, current treatment regimens are suboptimal, and risk biomarkers associated with the early onset and rapid progression are needed.^5–7

Omics technologies, particularly proteomics, have significant potential to advance our understanding of mechanisms underlying DKD. Identifying specific proteins associated with youth-onset T2D may improve risk prediction, enhance understanding of pathophysiology, and facilitate development of new therapeutic targets.^8,9 Recent advancements in high-throughput proteomics platforms, such as SOMAScan's aptamer-based platform and Olink's proximity extension assays, have greatly facilitated the search for proteins relevant to DKD development.

The primary aim of this study was to identify new circulating proteins predicting early DKD (defined as albuminuria, glomerular hyperfiltration, or rapid eGFR decline before eGFR <60 ml/min per 1.73 m²), including incident albuminuria, hyperfiltration, and rapid kidney function decline. We used the SOMAscan platform to measure circulating proteins in the Treatment Options for Type 2 Diabetes in Adolescents and Youth (TODAY) cohort of youth with T2D and identified proteins associated with DKD onset. In addition, our study aims to provide insights into proteomic mechanisms of early DKD in youth-onset T2D, which may facilitate development of targeted therapies and improve patient outcomes.

Research DESIGN and Methods

The TODAY protocol (ClinicalTrials.gov: NCT00081328) and primary outcome results were previously published.¹⁰ In brief, 699 youth with T2D diagnosed before age 18 years, with diabetes duration less than 2 years, body mass index (BMI) >85th percentile for age and sex, negative diabetes-associated antibodies (glutamic acid decarboxylase 65 and islet antigen 2 autoantibodies), and C-peptide >0.6 ng/ml were randomized at 15 diabetes centers across the United States to receive metformin alone, metformin plus rosiglitazone, or metformin plus an intensive lifestyle intervention program. TODAY participants were recruited over 4 years (2004–2009) and followed for a minimum of 2 years. The primary goal of TODAY (2004-2011 or study years 0–6) was to evaluate the effects of the three treatment arms on time-to-treatment failure defined as loss of glycemic control (hemoglobin A1c [HbA1c] ≥ 8% for six consecutive months or failure to wean from temporary insulin after acute metabolic decompensation).¹⁰

In 2011, 572 (82% of the original cohort) TODAY participants enrolled in the TODAY2 postintervention follow-up study. Between 2011 and 2014, TODAY2 participants no longer received randomized treatment, but continued to receive protocolized diabetes-related care with visits at 3-month intervals. From 2014 to 2020, 518 (74% of the original cohort) TODAY participants transitioned to community diabetes care but continued to be followed by the TODAY study group for annual observational visits. Each TODAY site had local Institutional Review Board approval, and participants and guardians provided informed consent/assent to participate. Continuing data analysis is approved by the Colorado Multiple Institutional Review Board.

The combined TODAY studies provided up to 15 years of longitudinal follow-up, with an average length of follow-up of 10.2 years since randomization. Characteristics of the cohort were nearly identical in all phases, as previously described.¹

Assessments for diabetes complications and comorbidities began during the TODAY trial and continued through TODAY2. Algorithms for classification accounted for differences in data collection frequency. Detailed methods of analysis and the definitions of complications and comorbidities have previously been published.¹ Standard laboratory assays were performed at Northwest Lipids Research Laboratory by methods previously described.¹¹

Definitions of DKD

Albuminuria

Urine samples were collected annually. Moderate albuminuria was defined as a ratio of urine albumin-to-creatinine 30–300 mg/g and severe albuminuria as ≥300 mg/g on at least two of three determinations within 3 months.¹² For these analyses, moderate and severe albuminuria were combined, and time to albuminuria was defined as the time to moderate and/or severe albuminuria.

GFR, Hyperfiltration, and Rapid GFR Decline

The full age spectrum combined serum creatinine (SCr) and cystatin C equation, which has been validated in children and adults,¹³ was used to calculate eGFR:¹⁴

$${\text{FAS}}_{\text{combi}}=\frac{107.3}{\alpha \times \frac{\text{Scr}}{Q_{crea}}+\left(1-\alpha \right)\times \frac{\text{ScysC}}{{\mathrm{Q}}_{\text{cysC}}}}$$

The full age spectrum equation is based on normalized SCr (SCr/Q), where Q_crea is the median SCr from healthy populations accounting for age and sex, and Q_cysC is 0.82 mg/L for age <70 years. The coefficient α in the denominator is a weighting factor for the normalized kidney biomarkers. We used α=0.5, making the denominator equal to the average of both normalized biomarkers.¹⁴ Hyperfiltration was defined as eGFR ≥135 ml/min per 1.73 m² at two consecutive visits. Rapid eGFR decline was defined as an annual eGFR decline >3 ml/min per 1.73 m² and/or ≥3.3% at two consecutive visits.^15,16

Selection of Samples for Proteomics

Of the 699 participants originally enrolled in the TODAY study, 22 individuals originally clinically diagnosed with T2D were later found to have monogenic diabetes mutations and were excluded from the present analyses (Figure 1). The study population was restricted to TODAY participants with plasma samples available for proteomic interrogation at baseline. A small subset of participants (n=27) underwent metabolic bariatric surgery and were censored at the time of surgery. The details of the power calculation are available in Supplemental Methods.

Figure 1.

Study design and participants included in analyses. TODAY, Treatment Options for Type 2 Diabetes in Adolescents and Youth.

SOMAScan

Plasma protein concentrations were measured in samples collected at baseline using the SOMAscan 7K Proteomic platform (SomaLogic, Inc.) at Washington University, St. Louis, Missouri. Internal controls were run with each sample and were normalized for intraplate and interplate variation.¹⁷ The SOMAScan 7K platform comprises 7604 Aptamers corresponding to 6596 human proteins.

Statistical Analysis

Descriptive statistics include means and standard deviations for normally distributed continuous variables, percentiles for continuous variables with skewed distributions, and counts and percentages for categorical variables. Proteins were natural log transformed and scaled by SD (i.e., each protein measurement was divided by the SD for that protein in the sample) before analysis. Cox proportional hazards models, adjusted for baseline HbA1c, log-transformed triglycerides, systolic BP, and estimated insulin sensitivity, were used to evaluate whether proteins predict the time to develop incident albuminuria, hyperfiltration, and rapid eGFR decline, and the hazard ratio per SD is reported. Participants with albuminuria, hyperfiltration, or rapid eGFR decline at baseline were excluded from the respective analyses. Covariates were selected based on their strong associations with early DKD in prior TODAY analyses.¹⁸ We performed a sensitivity analysis adjusting for randomized intervention during TODAY. As an exploratory analysis, we evaluated the association of proteins and DKD outcomes after stratifying by race and ethnicity (Hispanic, non-Hispanic Black, non-Hispanic White); for the stratified analyses, the other group was excluded because of small sample size (n=14). P values were adjusted to control the false discovery rate at 5% (q values).

To develop multiprotein signatures of time to development of each DKD outcome, elastic net Cox regression was applied to the top 100 proteins from the adjusted Cox proportional hazards models, ranked by q value. Optimal values of α and λ, the elastic net hyperparameters, were estimated using ten-fold cross-validation by minimizing the cross-validation error. Final models selected by elastic net were estimated using multivariable Cox proportional hazards regression. Concordance¹⁹ of the multiprotein model was compared with (1) a model containing the same proteins adjusted for HbA1c, log-transformed triglycerides, systolic BP, and estimated insulin sensitivity; (2) a model containing the same proteins adjusted for HbA1c, log-transformed triglycerides, systolic BP, estimated insulin sensitivity, sex, and baseline BMI; (3) a model containing only HbA1c, log-transformed triglycerides, systolic BP, and estimated insulin sensitivity; and (4) a model containing only HbA1c, log-transformed triglycerides, systolic BP, estimated insulin sensitivity, sex, and baseline BMI. To evaluate the parsimony of the multiprotein models, we further compared the concordance of the above models to models with the top protein for each outcome and to models with the seven proteins that were associated with all three DKD outcomes. Pathway analyses using the results of the univariable Cox models were performed using Ingenuity Pathway Analysis.²⁰ Statistical analyses were conducted using R software version 4.2.2 (R Core Team, Vienna) and Ingenuity Pathway Analysis (QIAGEN, Hilden).

Data and Resource Availability

Datasets generated during and/or analyzed in this study are available from the corresponding author on reasonable request.

Results

Clinical Characteristics

This analysis included data from 374 participants from the TODAY study with proteomic data (Figure 1). Participants were age 14.0±2.0 years on average, had diabetes duration of 7±6 months, were 63% female, had a mean HbA1c of 6.0±0.8% and BMI 35±8 kg/m², and were ethnically diverse (42% Hispanic, 34% non-Hispanic Black, 20% non-Hispanic White, and 4% other race and ethnicity, Table 1). The baseline prevalence of albuminuria was 8%. By the end of follow-up in the TODAY2, 43% of participants developed albuminuria, and the prevalence of hyperfiltration and rapid eGFR decline were 48% and 16%, respectively. The 374 participants in this study were similar to the entire cohort in terms of baseline characteristics.²¹ Only six study participants reported taking sodium-glucose cotransporter-2 inhibitors during TODAY2; data on glucagon-like peptide-1 receptor agonist medications were not available.

Table 1.

Characteristics of the treatment options for type 2 diabetes in adolescents and youth cohort

Characteristic	TODAY Cohort (N=374)
Age, yr	14 (2)
Sex, %
Female	63
Male	37
Race and ethnicity, %
Hispanic	42
Non-Hispanic Black	34
Non-Hispanic White	20
Other	4
Diabetes duration, mo	7 (6)
BMI, kg/m2	35 (8)
HbA1c, %	6.0 (0.8)
Treatment arm, %
Metformin alone	33
Metformin+rosiglitazone	33
Metformin+intensive lifestyle	34
UACR, mg/g	6 (4–13)
Prevalence of end point at the start of TODAY, %
Albuminuria (≥30 mg/g)	8
Hyperfiltration	13
Rapid eGFR decline	1
Prevalence of endpoint at the end of TODAY2, %
Albuminuria (≥30 mg/g)	43
Hyperfiltration	48
Rapid eGFR decline	16
Systolic BP, mm Hg	113 (11)
Diastolic BP, mm Hg	67 (8)
Mean arterial pressure, mm Hg	82 (9)
Hypertension at the start of TODAY, %	22
Hypertension at the end of TODAY2, %	61
eGFR, ml/min per 1.73 m2a	156 (36)
SCr	0.6 (0.1)
Serum cystatin-C	0.8 (0.1)
Estimated insulin sensitivity, ml/μUb	0.05 (0.04)
Triglycerides, mg/dl	96 (67–140)

Descriptive statistics are mean (SD), percentage, or median (25th–75th percentile). BMI, body mass index; FAS, full age spectrum; HbA1c, hemoglobin A1c; mm Hg, millimeters of mercury; SCr, serum creatinine; TODAY, Treatment Options for type 2 Diabetes in Adolescents and Youth; TODAY2, Treatment Options for Type 2 Diabetes in Adolescents and Youth; UACR, urine albumin-to-creatinine ratio.

^aeGFR by full age spectrum.

^bInsulin sensitivity estimated as 1/fasting insulin.

Individual Proteins Associated with Incident Markers of DKD

In adjusted Cox proportional hazards models, increased levels of seven proteins, including neural epidermal growth factor-like 1 and FAM189A2 (CI061, a recently discovered ITCH E3 ubiquitin ligase activator), were predictive of shorter time to development of albuminuria (Figure 2 and Supplemental Table 1). One-hundred and eighteen proteins predicted time to development of hyperfiltration (Figure 2 and Supplemental Table 2), and increased levels of three proteins (zona pellucida-like domain-containing protein 1, ZPLD1 [HR, 1.60; 95% confidence interval (CI), 1.33 to 1.92; q value=0.003]; resistin-like beta, resistin-like molecule beta [HR, 1.37; 95% CI, 1.19 to 1.56; q value=0.02]; and clustered mitochondria protein homolog, clusterin [HR, 1.51; 95% CI, 1.26 to 1.82; q value=0.03]) predicted shorter time to development of rapid eGFR decline (Figure 2 and Supplemental Table 3). Overlap between DKD outcomes in the nominally (P < 0.05) significant markers was relatively small (Figure 3), with the most overlap between albuminuria and hyperfiltration. Seven proteins were predictive of all three outcomes (SEM4A, PSB3, dihydroxyphenylalanine decarboxylase, C1RL1, T132A, pyruvate carboxylase, and C1-esterase inhibitor). The results of the sensitivity analysis adjusted for the TODAY randomized intervention did not meaningfully change any findings (data not shown).

Figure 2.

Volcano plots summarizing the association of proteins with outcomes in TODAY, using adjusted Cox proportional hazards models. Each point represents a protein. The x axis is the HR of the association per SD of the protein, and the y axis is the negative of the base 10 log-transformed P value, such that proteins with larger effect size are further to the right and left on the x axis and proteins with more significant association are higher on the y axis. Unadjusted P values < 0.05 are shown in blue, and the horizontal dashed line represents P = 0.05. CLU, clusterin; HR, hazard ratio; IGFBP-6, IGF-binding protein 6; MYOC, myocilin; NELL1, neural epidermal growth factor-like 1; PEDF, pigment epithelium-derived factor; RELM-beta, resistin-like molecule beta.

Figure 3.

Venn diagram showing the degree of overlap between proteins with nominally significant (P < 0.05) associations with time to albuminuria, hyperfiltration, and rapid eGFR decline.

In integrated pathway analysis, pathways related to liver X receptor/retinoid X receptor activation were shared between the albuminuria and hyperfiltration outcomes, but otherwise, pathways associated with each DKD outcome were relatively distinct (Supplemental Figure 1). Multiple biosynthesis pathways (retinoate, glycerophospholipid, retinol, and arginine) were upregulated in participants with shorter time to albuminuria. Hepatic fibrosis/hepatic stellate cell activation was the most significantly enriched pathway in participants with hyperfiltration; however, this pathway was not clearly upregulated or downregulated. Regulation of IGF transport and uptake by IGF-binding proteins was also upregulated in hyperfiltration. Pathways related to eukaryotic translation (elongation, termination, and initiation) were significantly upregulated in rapid eGFR decline, as were nonsense mediated decay and class 1 MHC–mediated antigen processing and presentation.

Although our study was not powered to test for differences in association of proteins and DKD outcomes among racial/ethnic groups, we performed an exploratory analysis repeating the adjusted Cox models after stratifying by race and ethnicity (Hispanic, non-Hispanic Black, non-Hispanic White, Supplemental Tables 4–6, Supplemental Tables 7–9, and Supplemental Tables 10–12). Many of the significant proteins in the racial/ethnic groups were distinct, and, surprisingly, in some cases there were more significant associations after stratification (Supplemental Figure 2). For example, there were 536 nominally significant associations between proteins and time to albuminuria when the entire cohort was pooled. After stratification, there were 566, 1161, and 292 nominally significant proteins in the Hispanic, non-Hispanic Black, and non-Hispanic White participants, respectively.

Multivariable Elastic Net Models of Time to Incident DKD

Elastic net Cox proportional hazards regression using the top 100 proteins for each outcome identified multiprotein signatures of time to incident DKD. Proteins retained by the elastic net algorithm for each DKD outcome are shown in Supplemental Tables 13–15. To evaluate the performance of the multiprotein models, we calculated the concordance statistic, which is the analog in a time-to-event setting to the area under the receiver operating characteristic curve for a binary outcome. The concordance statistic is the proportion of observations which are correctly ordered by the model in terms of the time to the outcome. Concordance for models with covariates (HbA1c, log-transformed triglycerides, systolic BP, estimated insulin sensitivity) was 0.62 (95% CI, 0.57 to 0.66) for albuminuria, 0.56 (0.51 to 0.62) for hyperfiltration, and 0.63 (0.55 to 0.71) for rapid eGFR decline (Figure 4). The protein-only model selected by elastic net regression for time to albuminuria retained 43 proteins and had a concordance of 0.82 (0.79 to 0.86). Thirty-eight proteins were retained in the model for time to hyperfiltration, with a concordance of 0.80 (0.77 to 0.84). The model for time to rapid eGFR decline included 68 proteins, with a concordance of 0.93 (0.89 to 0.96). Combined models with proteins and covariates had higher concordance than either the covariate-only model or the protein-only model, with a concordance of 0.84 (0.80 to 0.87) for albuminuria, 0.81 (0.78 to 0.85) for hyperfiltration, and 0.96 (0.93 to 0.98) for rapid eGFR decline. Taken together, these data suggest that the proteins were more predictive of early DKD than the clinical covariates.

Figure 4.

Concordance of multivariable Cox proportional hazards models for DKD outcomes. Models compared are (1) covariates only (HbA1c, log-transformed triglycerides, systolic BP, and estimated insulin sensitivity), (2) proteins only, with proteins selected by elastic net regression, and (3) proteins and covariates. CI, confidence interval; DKD, diabetic kidney disease; HbA1c, hemoglobin A1c.

To further characterize the predictive performance of the multiprotein models, we repeated the same elastic net analysis but added BMI and participant sex. Apart from the covariate-only model for hyperfiltration, for which the concordance improved from 0.56 (0.51 to 0.62) to 0.61 (0.56 to 0.66), the concordance of the models containing BMI and sex did not demonstrate significant improvements (Supplemental Figure 3). We also compared the protein-only models selected by elastic net with models with only the top protein (ranked by P value) and a model with the seven proteins that were significantly associated with all three outcomes. These models did not perform as well as the models with all proteins selected by elastic net (Supplemental Figure 4).

Discussion

We undertook a comprehensive proteomic analysis in youth-onset T2D, identifying novel circulating proteins linked to risk of developing DKD. Given the participants' short diabetes duration, we were able to pinpoint very early perturbations in the proteome associated with early DKD outcomes. The multiprotein signature predicted early DKD more effectively than key clinical covariates such as HbA1c, triglycerides, systolic BP, estimated insulin sensitivity, sex, and BMI, suggesting a significant potential to enhance risk stratification. Although the exact mechanisms remain unclear, DKD is associated with cellular repair, tissue remodeling, perturbed metabolism, and kidney fibrosis. The detected proteins are proposed to be involved in these processes. These findings highlight the potential value of proteomics to refine risk stratification, enable early detection, and guide therapeutic strategies for youth-onset T2D.

Identifying risk biomarkers of DKD in youth-onset T2D and understanding the underlying pathophysiology is a necessary first step to designing effective and efficient trials with new medical interventions.^10,22,23 The prospective longitudinal data in TODAY2 provide an excellent foundation to test novel protein biomarkers of DKD in youth-onset T2D. This study, to our knowledge, is the most comprehensive proteomic analysis to date in youth-onset T2D. We identified multiprotein signatures predicting time to albuminuria, rapid eGFR decline, and hyperfiltration over 15 years in youth with T2D, outperforming clinical covariates previously shown to predict early DKD in this population. The top proteins have been implicated in cell cycle, repair, inflammation, axon guidance, chemotaxis, and endocytosis, which may play a significant role in the mechanisms contributing to early DKD in youth with T2D. Proteomic analyses performed in the Joslin Kidney Study and in American Indian individuals identified 11 circulating proteins associated with a higher risk of kidney failure in adults with T2D, involved in inflammation, axon guidance, and cell signaling.²⁴ We identified overlap in pathways, but not necessarily the same proteins, possibly because of the difference in the number of proteins measured (6596 versus 1129).

Our study identified seven novel proteins that overlapped in their association with time to development of hyperfiltration, albuminuria, and rapid eGFR decline, markers of early DKD. These proteins highlight the complex interplay of immune regulatory mechanisms, metabolic dysregulation, proteostasis, and cellular signaling pathways in DKD progression. SEMA4A, C1RL1, and C1-esterase inhibitor play pivotal roles in the processes by which immune modulation and inflammation contribute to tissue damage and fibrosis in DKD.^25–27 PSMB3's critical function in maintaining protein homeostasis under stress conditions like diabetes suggests its importance for preserving kidney function.²⁸ The activities of metabolic enzymes dihydroxyphenylalanine decarboxylase and pyruvate carboxylase implicate these enzymes in energy production and glucose homeostasis pathways fundamental to DKD etiology.^29,30 T132A's role in modulating cellular signaling and maintaining cellular integrity may be disrupted in response to hyperglycemic stress, contributing to structural modifications leading to albuminuria and compromised filtration capacity.^31,32

There are several notable strengths in our study, including the utilization of the SOMAscan platform, which enabled the simultaneous analysis of many proteins in a robust and reproducible manner. This comprehensive approach allowed us to uncover novel proteins associated with early DKD in youth with T2D. Furthermore, the large sample size of youth-onset T2D, the diverse, national cohort, and the longitudinal nature of the TODAY study provided valuable insights into the temporal relationship between proteomic features at baseline and the development of early DKD.

However, there are some limitations that must be considered. First, we used surrogate measures (e.g., eGFR and albuminuria) to define early DKD because of the lack of kidney biopsy data. Second, although the SOMAscan platform enables the analysis of many proteins, it does not cover the entire human proteome. Consequently, additional proteins or pathways relevant to DKD may not be detected in our analysis. Furthermore, the observational nature of this analysis limits our ability to establish causality between the identified proteomic changes and DKD development. In addition, although our study identified several proteins associated with DKD, the precise mechanisms by which these proteins contribute to disease pathogenesis remain to be elucidated. Further experimental studies, such as in vitro and murine models, are required to investigate the functional roles of these proteins in DKD development and progression. In addition, these analyses do not provide a direct comparison of proteins related to DKD in youth- and adult-onset T2D. Finally, owing to the uniqueness of the TODAY study, we were unable to perform validation in an independent external cohort, which is a major limitation to use of these biomarkers in research or in the clinical setting. Our findings need to be replicated in independent longitudinal cohorts of youth with T2D to ensure generalizability of our results. There are smaller cohorts of youth with T2D with proteomics data from the same platform, but to our knowledge, there are no other studies with participants with similar diabetes duration, similar length of follow-up, or availability of adjudicated clinical end points over 15 years of follow-up. To avoid bias in the comparison with a validation cohort, it is important that the participants have similar length of diabetes duration, be at comparable stages of DKD, and have outcome data that can be meaningfully compared (e.g., a random urine albumin-to-creatinine ratio value may not be comparable with time to development of albuminuria confirmed on ≥2 of three occasions). Furthermore, we contend that because of the differences in the phenotypic severity in youth-onset T2D compared with adult-onset T2D, validation should first be performed in a cohort of youth-onset T2D because there is evidence to suggest that the pathophysiology of the two diseases may differ in important aspects.^4,33,34 This approach ensures that any validation efforts are most relevant to the population studied and can provide the most meaningful insights into the progression of DKD in youth with T2D.

Considering these limitations, it will be important to identify a suitable validation cohort. It is likely that stored specimens from a cohort with similar outcome assessment and diabetes duration will need to be assayed using the SOMAscan 7K platform. Future research should also elucidate specific mechanisms linking the identified proteins and DKD in youth-onset T2D. Investigating these associations within established DKD pathways, like perturbed metabolism, inflammation, cellular repair, and fibrosis, could provide insights into the proteins' roles in DKD development and progression. Our group is actively integrating blood proteomic data with tissue-specific protein and transcript information from kidney biopsies to better understand the mechanistic foundations of the highlighted pathways.

Comparing the multiprotein signatures associated with DKD in youth-onset T2D identified in our study with those reported in adult-onset T2D is an important future direction. Although some overlap exists between our findings and biomarkers reported in adults, e.g., cystatin C and beta-2 microglobulin,^9,35 our study also revealed several unique proteins not previously associated with DKD in adult-onset T2D. Given the recent evidence of distinct genetic and biological differences between youth-onset and adult-onset T2D,⁴ it is crucial to understand the similarities and differences in their DKD biomarker profiles. Future research should focus on conducting comparative analyses of protein biomarkers in these two patient populations, investigating the biological pathways associated with the overlapping and unique biomarkers, and exploring their potential as therapeutic targets or prognostic tools. This knowledge will be essential for developing targeted interventions.

In summary, our study has significantly advanced understanding of the proteomic changes associated with DKD in youth with T2D. Identifying novel proteins provide avenues for biomarker development to aid risk stratification, enrich clinical trials, and improve the understanding of DKD mechanisms. Early identification of those at high risk of DKD, particularly in young persons with T2D, is paramount. Future research can build on these findings to improve risk stratification, early detection, and personalized therapeutic approaches. Translating these findings into clinical practice demands rigorous validation across diverse cohorts, demonstrating consistent reliability, standardized assays, and economic feasibility. Determining optimal application, including frequency of testing, is essential. Collaborative efforts between researchers, clinicians, and industry partners will be key to advancing these promising findings from the research realm to real-world clinical settings, aiming for early detection and personalized management of DKD in affected youth.

Supplementary Material

cjasn-19-1603-s001.pdf ^{(1.7MB, pdf)}

cjasn-19-1603-s002.pdf ^{(2.1MB, pdf)}

cjasn-19-1603-s003.xlsx ^{(7.6MB, xlsx)}

cjasn-19-1603-s004.xlsx ^{(7.6MB, xlsx)}

cjasn-19-1603-s005.xlsx ^{(7.7MB, xlsx)}

cjasn-19-1603-s006.xlsx ^{(7.5MB, xlsx)}

Disclosures

Disclosure forms, as provided by each author, are available with the online version of the article at http://links.lww.com/CJN/C63.

Funding

P. Bjornstad: National Institute of Diabetes and Digestive and Kidney Diseases (U01 DK61242). L. Pyle: American Diabetes Association (7-23-ICTSTDY-08).

Author Contributions

Conceptualization: Petter Bjornstad, Laura Pyle.

Data curation: Ye Ji Choi, Laura Pyle, Timothy Vigers.

Formal analysis: Petter Bjornstad, Ye Ji Choi, Laura Pyle, Timothy Vigers.

Funding acquisition: Petter Bjornstad, Laura Pyle.

Investigation: Fida Bacha, Petter Bjornstad, Frank Brosius, Jesse A. Goodrich, Hiddo Heerspink, Elvira Isganaitis, Matthias Kretzler, Anita Layton, Jane Lynch, Kristen J. Nadeau, Phoom Narongkiatikhun, Robert G. Nelson, Laura Pyle, Pierre J. Saulnier, Kumar Sharma, Kalie L. Tommerdahl, Jeanie B. Tryggestad, Timothy Vigers, Sushrut Waikar, Ian de Boer, Daniel van Raalte.

Methodology: Ye Ji Choi, Laura Pyle, Timothy Vigers.

Project administration: Petter Bjornstad, Laura Pyle.

Resources: Petter Bjornstad, Laura Pyle.

Software: Ye Ji Choi, Laura Pyle, Timothy Vigers.

Supervision: Petter Bjornstad, Laura Pyle.

Validation: Petter Bjornstad, Ye Ji Choi, Laura Pyle, Timothy Vigers.

Visualization: Petter Bjornstad, Ye Ji Choi, Laura Pyle, Timothy Vigers.

Writing – original draft: Petter Bjornstad, Laura Pyle.

Writing – review & editing: Fida Bacha, Frank Brosius, Ye Ji Choi, Jesse A. Goodrich, Hiddo Heerspink, Elvira Isganaitis, Matthias Kretzler, Anita Layton, Jane Lynch, Kristen J. Nadeau, Phoom Narongkiatikhun, Robert G. Nelson, Laura Pyle, Pierre J. Saulnier, Kumar Sharma, Kalie L Tommerdahl, Jeanie B. Tryggestad, Timothy Vigers, Sushrut Waikar, Ian de Boer, Daniel van Raalte.

Data Sharing Statement

De-identified clinical data, protocols, and data dictionaries are publicly available through the National Institute of Diabetes and Digestive and Kidney Diseases Central Repository.

Supplemental Material

This article contains the following supplemental material online at http://links.lww.com/CJN/C62, http://links.lww.com/CJN/C83, http://links.lww.com/CJN/C84, http://links.lww.com/CJN/C85, http://links.lww.com/CJN/C86

Supplemental Methods

Supplemental Figure 1. Top ten pathways (ranked by P value) by Ingenuity Pathway Analysis for (A) time to albuminuria, (B) time to hyperfiltration, and (C) time to rapid eGFR decline.

Supplemental Figure 2. Venn diagram showing the degree of overlap between nominally significant (P < 0.05) proteins in Hispanic, non-Hispanic White, and non-Hispanic Black participants for (A) time to albuminuria, (B) time to hyperfiltration, and (C) time to rapid eGFR decline.

Supplemental Figure 3. Concordance of multivariable Cox proportional hazards models for DKD outcomes, in sensitivity analyses adding participant sex and body mass index (BMI). Models compared are: (1) covariates only (HbA1c, log-transformed triglycerides, systolic blood pressure, estimated insulin sensitivity, sex, and BMI) (2) proteins only, with proteins selected by elastic net regression, and (3) proteins and covariates.

Supplemental Figure 4. Concordance of multivariable Cox proportional hazards models for DKD outcomes. Models compared are: (1) proteins only, with proteins selected by elastic net regression, (2) a model with only the top protein (by p-value) from univariable Cox proportional hazards models and (3) a “shared protein” model with the seven proteins that were significantly associated (after FDR adjustment) with all 3 DKD outcomes in univariable Cox proportional hazards models.

Supplemental Table 1. Hazard ratios for time to albuminuria, per SD of each protein, from univariable Cox proportional hazards models.

Supplemental Table 2. Hazard ratios for time to hyperfiltration, per SD of each protein, from univariable Cox proportional hazards models.

Supplemental Table 3. Hazard ratios for time to rapid eGFR decline, per SD of each protein, from univariable Cox proportional hazards models.

Supplemental Table 4. Hazard ratios for time to albuminuria among Hispanic participants, per SD of each protein, from univariable Cox proportional hazards models.

Supplemental Table 5. Hazard ratios for time to hyperfiltration among Hispanic participants, per SD of each protein, from univariable Cox proportional hazards models.

Supplemental Table 6. Hazard ratios for time to rapid eGFR decline among Hispanic participants, per SD of each protein, from univariable Cox proportional hazards models.

Supplemental Table 7. Hazard ratios for time to albuminuria among non-Hispanic Black participants, per SD of each protein, from univariable Cox proportional hazards models.

Supplemental Table 8. Hazard ratios for time to hyperfiltration among non-Hispanic Black participants, per SD of each protein, from univariable Cox proportional hazards models.

Supplemental Table 9. Hazard ratios for time to rapid eGFR decline among non-Hispanic Black participants, per SD of each protein, from univariable Cox proportional hazards models.

Supplemental Table 10. Hazard ratios for time to albuminuria among non-Hispanic White participants, per SD of each protein, from univariable Cox proportional hazards models.

Supplemental Table 11. Hazard ratios for time to hyperfiltration among non-Hispanic White participants, per SD of each protein, from univariable Cox proportional hazards models.

Supplemental Table 12. Hazard ratios for time to rapid eGFR decline among non-Hispanic White participants, per SD of each protein, from univariable Cox proportional hazards models.

Supplemental Table 13. Hazard ratios for time to albuminuria, per SD of each protein, for proteins retained by elastic net Cox proportional hazards regression.

Supplemental Table 14. Hazard ratios for time to hyperfiltration, per SD of each protein, for proteins retained by elastic net Cox proportional hazards regression.

Supplemental Table 15. Hazard ratios for time to rapid eGFR decline, per SD of each protein, for proteins retained by elastic net Cox proportional hazards regression.