Proteomics has been used to study type 2 diabetes, but the majority of available data are from White participants. Here, we extend prior work by analyzing a large cohort of self-identified African Americans in the Jackson Heart Study (n = 1,313). We found 325 proteins associated with incident diabetes after adjusting for age, sex, and sample batch (false discovery rate q < 0.05) measured using a single-stranded DNA aptamer affinity-based method on fasting plasma samples. A subset was independent of established markers of diabetes development pathways, such as adiposity, glycemia, and/or insulin resistance, suggesting potential novel biological processes associated with disease development. Thirty-six associations remained significant after additional adjustments for BMI, fasting plasma glucose, cholesterol levels, hypertension, statin use, and renal function. Twelve associations, including the top associations of complement factor H, formimidoyltransferase cyclodeaminase, serine/threonine–protein kinase 17B, and high-mobility group protein B1, were replicated in a meta-analysis of two self-identified White cohorts—the Framingham Heart Study and the Malmö Diet and Cancer Study—supporting the generalizability of these biomarkers. A selection of these diabetes-associated proteins also improved risk prediction. Thus, we uncovered both novel and broadly generalizable associations by studying a diverse population, providing a more complete understanding of the diabetes-associated proteome.

Although an estimated 1.5 million deaths worldwide in 2019 were attributed to type 2 diabetes (T2D) (1), the early identification and prevention of T2D in individuals at high risk for T2D could mitigate morbidity and mortality (24). Circulating biomarkers have a role in risk prediction, but biomarker discovery is challenging because of the differences in T2D burden across geographic regions (5) and racial/ethnic groups (6) for reasons yet to be appropriately and/or adequately explored. For example, in the U.S., African Americans (AAs) have persistently had higher T2D incidence and prevalence than White individuals (7,8), due in large part to inequities in the social, structural, and environmental determinants of health (9) and also the lack of the successful inclusion of these populations in studies on how to diagnose and treat diseases. Therefore, it is essential to include AAs in biomarker discovery.

Circulating proteins are attractive biomarkers because, as the end products of gene expression, proteins can provide insights into genetic contributions to disease (10) and implicate biological pathways through shared genetics (11). They also reflect posttranslational modifications caused by environmental and behavioral exposures. High-throughput proteomics has been used to study biomarkers of T2D in large community cohorts (1214), revealing novel associations with enzymes that are important in amino acid metabolism, such as 3-hydroxyisobutyryl-CoA hydrolase or aminoacylase-1, (13,14) or proteins that currently have less clear metabolic effects, such as WAP, Kazal, immunoglobulin, Kunitz, and NTR domain–containing protein 2 (WFIKKN2) (12,14). These studies, however, have largely focused on individuals who are White and of European descent.

The aim of this study is to extend our understanding of circulating proteins associated with the development of adult-onset diabetes, including their utility as prediction biomarkers in a large cohort of self-identified AAs via the Jackson Heart Study (JHS). Using the deep clinical phenotyping available in this cohort, we begin to parse out proteins that have unique or shared associations with known metabolic risk factors, which could uncover biological processes leading to diabetes. We then test their transferability across populations by comparing our findings with data from two large, self-identified White cohorts—the Framingham Heart Study (FHS) and the Malmö Diet and Cancer Study (MDCS) (14)—identifying robust protein diabetes biomarkers that are relevant across diverse human populations. Finally, we provide all of our data (to a nominal level of significance) to the scientific community to facilitate the study of how population diversity contributes to differences in observed associations.

Study Populations

The JHS is a community-based cohort of 5,306 self-identified AAs residing in Jackson, Mississippi (15). Diabetes status was assessed at three examinations (2000–2004, 2005–2008, and 2009–2013). Fasting baseline plasma samples from 2,143 participants were profiled; 399 were from a nested case-cohort study of incident coronary artery disease and the remaining were randomly selected from individuals with available plasma samples. At the first examination, 548 individuals had diabetes and, of the remaining 1,313 participants with complete follow-up data, 239 developed diabetes after a mean of 7.4 years. The FHS is a longitudinal community cohort of residents of Framingham, Massachusetts (16), with proteomic data available from 1,618 participants in Offspring, Exam 5 (1991–1995) of the study. Of these, 177 developed diabetes over a mean follow-up of 11.6 years. The MDCS is a longitudinal cohort study of a Swedish population (17), with proteomic data from 1,221 participants (1991–1996 enrollment), 272 of whom developed diabetes over a mean 12.9 years of follow-up (14). Supplementary Fig. 1 is a schematic of the different cohorts. Written consent was obtained from all study participants and study protocols were approved by the respective institutional review boards of each study site.

Clinical Variables and Outcome

Diabetes was defined in JHS as a fasting plasma glucose (FPG) level ≥ 126 mg/dL, hemoglobin A1c (HbA1c) ≥ 6.5%, diabetes diagnosis by a health care provider, or self-reported use of diabetes medication. Hypertension was defined as systolic blood pressure (SBP) >140 mmHg, diastolic blood pressure >90 mmHg, or self-reported use of blood pressure–lowering medications. Clinical measurements, including FPG, insulin, and lipids, were made using standard laboratory techniques as described elsewhere (18). BMI was calculated as weight (kg)/height squared (m2). HOMA for insulin resistance (HOMA-IR) was calculated using the following formula: fasting insulin (μIU/mL) × FPG (mg/dL)/22.5. Estimated glomerular filtration rate (eGFR) was calculated using the Chronic Kidney Disease Epidemiology Collaboration equation (19). In FHS, diabetes was defined as FPG ≥126 mg/dL or self-reported use of diabetes medications. In MDCS, diabetes at baseline was defined as FPG ≥126 mg/dL, self-reported diabetes diagnosis, or diabetes medication use. Incident cases were ascertained with the use of three national registries: the Malmö HbA1c registry, the Nationwide Swedish National Diabetes Register, and the regional Diabetes 2000 Register of the Scania region (20,21).

Proteomic Profiling

The SOMAscan proteomics platform, based on single-stranded DNA aptamers or “SOMAmers” with affinity for specific plasma proteins, was used for proteomics analysis, as described previously (14,2224). Briefly, the relative concentrations of target proteins in plasma were measured in a multiplex reaction with SOMAmers, including a photocleavage and washout step that removes nonspecific SOMAmer–protein pairs. The bound SOMAmers are then released from the target protein and quantified using a DNA oligo-array plate reader. Plasma samples were stored at −80°C. EDTA tubes were used for JHS and MDCS, and citrated tubes for FHS. JHS samples were run in three separate batches, and FHS and MDCS samples were run in two batches, with a total of 1,305 common aptamers measured in all three cohorts. Intra- and interplate variation in measurements were accounted for using normalization and scaling to controls included within each plate (Supplemental Materials). Five measured proteins had >98% missingness across JHS batches 2 and 3 and were excluded from the analysis; otherwise only leptin and adiponectin had missingness across the batches (<2%) (Supplementary Table 1). In the JHS data, the median intra-assay coefficient of variation (CV) was 2.7% and the median interassay CV was 8.0% across batches (Supplementary Table 2). Corresponding CVs were <4% and <7%, respectively, for FHS and <4% and <5%, respectively, for MDCS.

Statistical Analysis

Proteomic data were standardized to a set of intraplate control samples and normalized using log transformation. To limit potential technical differences in signals across batches, samples were scaled to a mean of 0 and SD of 1 within each batch and then rank normalized across all batches. Linear regression was used to model protein associations with continuous baseline clinical traits, and logistic regression modeling was used for the dichotomous outcome of baseline diabetes status adjusted for age, sex, and batch within each study. Cox proportional hazards models were used to assess protein associations with incident diabetes. Models were adjusted for age, sex, and batch (model 1); additionally adjusted for BMI and FPG (model 2); and further adjusted for hypertension, triglyceride, and HDL cholesterol levels, statin use, and eGFR (model 3). To control for differences in social determinants of health, adjustments for income level and education level were used as proxies. A Benjamini–Hochberg false discovery rate (FDR) <5% was used to account for multiple comparisons (25).

For validation, the 323 proteins that were associated with incident diabetes at an FDR <0.05 in JHS models 1 and 3 were evaluated in FHS and MDCS. In previous FHS and MCDS analyses, protein levels were additionally transformed using inverse rank normalization and then regressed against plate to create plate-adjusted residuals to account for technical differences in the proteomics platform used during the FHS and MDCS profiling (14). Therefore, for the JHS and FHS and MDCS comparisons, JHS proteomic values were similarly transformed (FHS/MDCS). FHS and MDCS results were reported as a meta-analysis across the two cohorts. Differences in β estimates between JHS and the FHS/MDCS meta-analyzed protein associations with incident diabetes were assessed using the pnorm function in R statistical software (R Core Team, Vienna, Austria) on z-scores created from the differences in β estimate for each protein comparison.

The Harrell C statistic, Akaike information criterion, and Bayesian information criterion were calculated to determine the discrimination of these protein biomarkers in Cox proportional hazards prediction models that included clinical risk factors and protein predictors selected by elastic net regularization. Models included 1) a subset of protein predictors selected from all 1,305 proteins measured; 2) a subset of the 111 proteins associated with incident diabetes in JHS model 2 selected using elastic net regularization; and 3) a subset of the 36 proteins associated in the fully adjusted JHS model 3. Proteins were selected using elastic net regression with α and λ values tuned to optimize root mean square error using 10-fold cross-validation for each model. The clinical risk factors were derived from the FHS diabetes risk score, which included age; sex; BMI; SBP; levels of HDL cholesterol, triglycerides, and FPG; waist circumference; and parental history of diabetes (26). These same prediction models were tested in FHS. Bootstrapping was used to correct for predictive accuracy optimism (27) using the rms package in R.

All statistical analyses were conducted using Stata release 14 (StataCorp, College Station, TX) and R, version 3.6.3 (R Core Team) statistical software.

Data and Resource Availability

The data sets generated during and/or analyzed during this study have been uploaded to the JHS dbGaP repository under the JHS accession number phs000964. They are also available upon request from the respective study cohorts; requests can be facilitated by the corresponding author.

Baseline Characteristics of JHS Participants

Proteomic profiling was performed in baseline plasma samples from 2,143 JHS participants. Of these, 548 had prevalent diabetes (Table 1) and 239 developed diabetes over a median of 7.4 years. Baseline characteristics were similar among JHS participants without prevalent diabetes who were and were not included in the incident diabetes analysis (Supplementary Table 3), suggesting limited bias was introduced due to sampling. All individuals were self-identified AAs, and a majority were women. Individuals who had prevalent and incident diabetes were slightly older; had higher BMI and greater waist circumference; higher FPG, fasting insulin, LDL cholesterol, and triglyceride levels; greater prevalence of hypertension; slightly lower eGFRs; and were more likely to be taking a statin or blood pressure medications. Given our cohort age distribution and region-specific T2D incidence and prevalence rates, we assume that the majority of incident cases were of T2D. We did not have access, however, to islet autoantibodies or c-peptide levels and therefore could not exclude type 1 diabetes cases. Thus, we will refer to incident diabetes and not T2D as our primary outcome.

Table 1

Baseline clinical characteristics of participants from the JHS

Prevalent diabetes analysis cohort (n = 2,143)Incident diabetes analysis cohort (n = 1,313)
Proteomics cohortNoncasesCasesNoncasesCases
Patients, n 2,143 1,595 548 1,074 239 
Female patients, n (%) 1,312 (61) 963 (60) 349 (64) 655 (61) 144 (60) 
Age, years 55.3 (12.8) 53.8 (13.0) 59.7 (11.0) 53.0 (12.6) 55.0 (11.5) 
BMI, kg/m2 31.6 (7.2) 30.8 (7.0) 34.0 (7.4) 30.3 (6.6) 33.8 (7.5) 
Waist circumference, cm 100.9 (16.5) 98.5 (15.9) 107.9 (16.0) 97.2 (15.8) 105.1 (14.8) 
FPG, mg/dL 101.5 (36.2) 90.7 (8.8) 140.0 (61.4) 89.1 (7.8) 96.9 (9.8) 
Fasting plasma insulin, IU/mL 17.8 (12.7) 15.5 (8.7) 26.1 (19.5) 14.3 (7.3) 21.3 (11.7) 
HOMA-IR 3.5 (2.2) 3.5(2.2) — 3.2 (1.7) 5.1 (3.0) 
Fasting total cholesterol, mg/dL 200.0 (41.7) 199.0 (40.0) 203.5 (47.2) 198.2 (39.1) 201.2 (42.9) 
Fasting HDL cholesterol, mg/dL 51.7 (14.7) 52.3 (14.9) 49.3 (13.7) 53.0 (15.0) 49.0 (12.6) 
Fasting triglycerides, mg/dL 109.4 (82.1) 100.1 (60.5) 142.6 (127.5) 95.0 (54.2) 118.1 (66.6) 
Hypertension, n (%) 1,248 (58) 813 (51) 435 (79) 489 (46) 157 (66) 
SBP, mmHg 127.6 (16.6) 126.4 (16.4) 131.2 (16.6) 125.0 (15.2) 128.7 (17.4) 
Current smoker, n (%) 277 (13) 219 (14) 58 (11) 129 (12) 29 (12) 
eGFR, mL/min/1.73 m2 93.3 (23.0) 95.1 (21.3) 88.4 (26.7) 96.7 (20.6) 92.3 (19.3) 
Medication      
 Statin, n (%) 304 (14) 152 (10) 152 (28) 90 (8) 32 (14) 
 Blood pressure medication, n (%) 1,147 (54) 725 (46) 422 (78) 428 (40) 149 (63) 
 Diabetes oral medication, n (%) 264 (13) 0 (0) 264 (52) — — 
 Insulin, n (%) 147 (7) 0 (0) 147 (29) — — 
Prevalent diabetes analysis cohort (n = 2,143)Incident diabetes analysis cohort (n = 1,313)
Proteomics cohortNoncasesCasesNoncasesCases
Patients, n 2,143 1,595 548 1,074 239 
Female patients, n (%) 1,312 (61) 963 (60) 349 (64) 655 (61) 144 (60) 
Age, years 55.3 (12.8) 53.8 (13.0) 59.7 (11.0) 53.0 (12.6) 55.0 (11.5) 
BMI, kg/m2 31.6 (7.2) 30.8 (7.0) 34.0 (7.4) 30.3 (6.6) 33.8 (7.5) 
Waist circumference, cm 100.9 (16.5) 98.5 (15.9) 107.9 (16.0) 97.2 (15.8) 105.1 (14.8) 
FPG, mg/dL 101.5 (36.2) 90.7 (8.8) 140.0 (61.4) 89.1 (7.8) 96.9 (9.8) 
Fasting plasma insulin, IU/mL 17.8 (12.7) 15.5 (8.7) 26.1 (19.5) 14.3 (7.3) 21.3 (11.7) 
HOMA-IR 3.5 (2.2) 3.5(2.2) — 3.2 (1.7) 5.1 (3.0) 
Fasting total cholesterol, mg/dL 200.0 (41.7) 199.0 (40.0) 203.5 (47.2) 198.2 (39.1) 201.2 (42.9) 
Fasting HDL cholesterol, mg/dL 51.7 (14.7) 52.3 (14.9) 49.3 (13.7) 53.0 (15.0) 49.0 (12.6) 
Fasting triglycerides, mg/dL 109.4 (82.1) 100.1 (60.5) 142.6 (127.5) 95.0 (54.2) 118.1 (66.6) 
Hypertension, n (%) 1,248 (58) 813 (51) 435 (79) 489 (46) 157 (66) 
SBP, mmHg 127.6 (16.6) 126.4 (16.4) 131.2 (16.6) 125.0 (15.2) 128.7 (17.4) 
Current smoker, n (%) 277 (13) 219 (14) 58 (11) 129 (12) 29 (12) 
eGFR, mL/min/1.73 m2 93.3 (23.0) 95.1 (21.3) 88.4 (26.7) 96.7 (20.6) 92.3 (19.3) 
Medication      
 Statin, n (%) 304 (14) 152 (10) 152 (28) 90 (8) 32 (14) 
 Blood pressure medication, n (%) 1,147 (54) 725 (46) 422 (78) 428 (40) 149 (63) 
 Diabetes oral medication, n (%) 264 (13) 0 (0) 264 (52) — — 
 Insulin, n (%) 147 (7) 0 (0) 147 (29) — — 

Values are reported as mean (SD) unless otherwise specified.

Mean follow-up time for incident cases was 10.2 years.

eGFR was calculated using the Chronic Kidney Disease Epidemiology Collaboration equation.

Circulating Protein Associations With Incident Diabetes

Of 1,305 proteins measured, 325 were associated with incident diabetes in Cox models adjusted for age, sex, and batch (q < 0.05) (Fig. 1 and Supplementary Table 4). In model 2, which additionally adjusted for BMI and FPG level, 111 proteins were associated with disease onset (Supplementary Table 4). These findings included the expected positive association of leptin (hazard ratio [HR] 1.98 [95% CI 1.64–2.39]; q = 2.05 × 10−10) and inverse association of adiponectin (HR 0.60 [95% CI 0.52–0.69]; q = 1.92 × 10−10) in model 1 that were attenuated after adjusting for BMI. In model 3, which further adjusted for HDL cholesterol and triglyceride levels, eGFR, hypertension status, and statin medication use, 36 proteins were associated with incident disease (Fig. 2 and Supplementary Table 4). Histone 1H.2 (HIST1H1C) had the highest HR (1.52 [95% CI 1.239–1.78]; q = 4.19 × 10−4), whereas IGF-binding protein 2 had the lowest (HR 0.68 [95% CI 0.57–0.80]; q = 7.86 × 10−6). Aptamer specificity was confirmed for 23 of these proteins by strong correlations with protein concentrations measured by an orthogonal high-multiplex immunoassay (Olink) and/or genome-wide association study (GWAS) associations with genetic variants in the protein coding gene (Supplementary Table 5). Thirty proteins remained significantly associated with incident diabetes after adjusting for income and education levels as proxies for social determinants of health (Supplementary Table 6).

Figure 1

Protein associations with incident diabetes in JHS adjusted for age, sex, and batch. Volcano plot showing the HRs for incident diabetes from Cox proportional hazards models for every 1 SD increase in log-transformed circulating protein level adjusted for age, sex, and proteomic batch. Green dots represent proteins associated with a Bonferroni P < 3.83 × 10−5 correcting for the 1,305 aptamers measured. Purple dots represent proteins associated with FDR q < 0.05. Tabulated results of all Bonferroni and FDR significant proteins can be found in Supplementary Table 4.

Figure 1

Protein associations with incident diabetes in JHS adjusted for age, sex, and batch. Volcano plot showing the HRs for incident diabetes from Cox proportional hazards models for every 1 SD increase in log-transformed circulating protein level adjusted for age, sex, and proteomic batch. Green dots represent proteins associated with a Bonferroni P < 3.83 × 10−5 correcting for the 1,305 aptamers measured. Purple dots represent proteins associated with FDR q < 0.05. Tabulated results of all Bonferroni and FDR significant proteins can be found in Supplementary Table 4.

Close modal
Figure 2

Protein associations with incident diabetes in JHS with further adjustments for BMI, FPG, clinical cholesterol levels, hypertension, statin use, and renal function. Volcano plot showing the HRs for incident diabetes from Cox proportional hazards models for every 1 SD increase in log-transformed circulating protein level adjusted for age; sex; BMI; FPG, triglyceride, and HDL cholesterol levels; hypertension status; statin medication use; eGFR; and proteomic batch. Green dots represent proteins associated with a Bonferroni P < 3.83 × 10−5 correcting for the 1,305 aptamers measured. Purple dots represent proteins associated with FDR q < 0.05. Tabulated results of all Bonferroni and FDR significant proteins can be found in Supplementary Table 4.

Figure 2

Protein associations with incident diabetes in JHS with further adjustments for BMI, FPG, clinical cholesterol levels, hypertension, statin use, and renal function. Volcano plot showing the HRs for incident diabetes from Cox proportional hazards models for every 1 SD increase in log-transformed circulating protein level adjusted for age; sex; BMI; FPG, triglyceride, and HDL cholesterol levels; hypertension status; statin medication use; eGFR; and proteomic batch. Green dots represent proteins associated with a Bonferroni P < 3.83 × 10−5 correcting for the 1,305 aptamers measured. Purple dots represent proteins associated with FDR q < 0.05. Tabulated results of all Bonferroni and FDR significant proteins can be found in Supplementary Table 4.

Close modal

Circulating Protein Associations With Diabetes Clinical Risk Factors and Prevalent Diabetes

We next examined whether the 325 proteins associated with incident diabetes in model 1 reflected traditional diabetes risk factors such as obesity, glycemia, insulin resistance, and dyslipidemia, on the basis of their associations with BMI, FPG, HOMA-IR, and triglyceride levels (q < 0.05) (Fig. 3). Many were associated with at least one of these clinical traits, but we found 52 were not (q > 0.05) (Fig. 3 and Supplementary Table 7). These proteins could serve to highlight a biological process that may contribute to disease not traditionally associated with diabetes development. Another 29 of these proteins were not associated with eGFR (q < 0.05) (Table 2), which can influence circulating protein levels. There were 12 proteins associated with incident diabetes and BMI but not with HOMA-IR or eGFR (Supplementary Table 8). This group could highlight pathways of obesity-mediated changes to glucose homeostasis that are independent of insulin resistance. Eight proteins were only associated with incident diabetes and triglyceride levels—parsing out proteins that potentially link hypertriglyceridemia with diabetes that are independent of adiposity and insulin resistance (Supplementary Table 9). Of the 325 proteins associated with incident diabetes in JHS model 1, 145 were also associated with prevalent diabetes, and 136 had concordant directions of effect (q < 0.05) (Supplementary Table 10).

Figure 3

Proteins that share associations with incident diabetes and baseline measurements of select diabetes risk factors. The number, listed in bold, of the 1,305 protein profiles that were associated with BMI, HOMA-IR, and FPG and triglyceride levels (q < 0.05) and their overlap with the 325 proteins associated with incident diabetes in model 1 (age, sex, and batch adjusted at q < 0.05, demarcated in red) are visualized. Numbers in parentheses represent the clinical trait the protein is associated with 1) incident diabetes, 2) BMI, 3) FPG, 4) HOMA-IR, and 5) triglycerides. A total of 124 proteins were associated with incident T2D and BMI, HOMA-IR, and FPG and triglyceride levels. Fifty-two were only associated with incident T2D and not with the other clinical traits.

Figure 3

Proteins that share associations with incident diabetes and baseline measurements of select diabetes risk factors. The number, listed in bold, of the 1,305 protein profiles that were associated with BMI, HOMA-IR, and FPG and triglyceride levels (q < 0.05) and their overlap with the 325 proteins associated with incident diabetes in model 1 (age, sex, and batch adjusted at q < 0.05, demarcated in red) are visualized. Numbers in parentheses represent the clinical trait the protein is associated with 1) incident diabetes, 2) BMI, 3) FPG, 4) HOMA-IR, and 5) triglycerides. A total of 124 proteins were associated with incident T2D and BMI, HOMA-IR, and FPG and triglyceride levels. Fifty-two were only associated with incident T2D and not with the other clinical traits.

Close modal
Table 2

Proteins associations with diabetes that are not associated with measures of adiposity, glycemia, insulin resistance, triglycerides, and eGFR

Protein nameIncident T2D
HR (95% CI)PFDR q
VEGF-C 0.74 (0.64–0.85) 2.47 × 10−5 4.13 × 10−4 
Angiopoietin-1 0.77 (0.68–0.87) 4.61 × 10−5 7.07 × 10−4 
CHIP 0.78 (0.69–0.88) 1.01 × 10−4 1.25 × 10−3 
LRRT1 1.27 (1.12–1.43) 2.09 × 10−4 2.31 × 10−3 
PAFAH β subunit 0.77 (0.67–0.89) 3.24 × 10−4 3.31 × 10−3 
IL-2 sRa 1.26 (1.11–1.43) 4.41 × 10−4 4.30 × 10−3 
ERBB4 1.22 (1.09–1.37) 5.98 × 10−4 5.38 × 10−3 
PFD5 0.80 (0.70–0.91) 7.84 × 10−4 6.73 × 10−3 
kallikrein 5 1.24 (1.09–1.41) 8.12 × 10−4 6.74 × 10−3 
IMB1 0.80 (0.70–0.91) 9.07 × 10−4 7.26 × 10−3 
Cyclophilin F 0.80 (0.70–0.91) 1.09 × 10−3 8.44 × 10−3 
Gro-a 0.80 (0.71–0.92) 1.12 × 10−3 8.53 × 10−3 
EG-VEGF 1.22 (1.08–1.38) 1.32 × 10−3 9.78 × 10−3 
ALK-1 0.80 (0.70–0.92) 1.66 × 10−3 1.14 × 10−2 
PolyUbiquitin K48 1.23 (1.08–1.39) 1.71 × 10−3 1.16 × 10−2 
Caspase-3 0.81 (0.71–0.93) 1.99 × 10−3 1.29 × 10−2 
Fas ligand, soluble 0.82 (0.72–0.93) 2.48 × 10−3 1.55 × 10−2 
EP15R 0.83 (0.73–0.94) 2.75 × 10−3 1.65 × 10−2 
PRKACA 0.83 (0.73–0.94) 3.11 × 10−3 1.84 × 10−2 
GRB2 adapter protein 0.83 (0.73–0.94) 3.42 × 10−3 1.95 × 10−2 
a-Synuclein 0.82 (0.72–0.94) 3.77 × 10−3 2.07 × 10−2 
Cofilin-1 0.82 (0.71–0.95) 6.32 × 10−3 3.08 × 10−2 
Carbonic anhydrase XIII 0.83 (0.72–0.95) 7.90 × 10−3 3.67 × 10−2 
GPVI 0.83 (0.73–0.95) 8.22 × 10−3 3.74 × 10−2 
IL-4 sR 1.17 (1.04–1.32) 9.03 × 10−3 3.98 × 10−2 
NSE 0.85 (0.75–0.96) 9.59 × 10−3 4.09 × 10−2 
I-309 1.19 (1.04–1.35) 9.75 × 10−3 4.14 × 10−2 
Sorting nexin 4 0.84 (0.74–0.96) 9.96 × 10−3 4.21 × 10−2 
ATPO 0.83 (0.72–0.96) 1.23 × 10−2 4.93 × 10−2 
Protein nameIncident T2D
HR (95% CI)PFDR q
VEGF-C 0.74 (0.64–0.85) 2.47 × 10−5 4.13 × 10−4 
Angiopoietin-1 0.77 (0.68–0.87) 4.61 × 10−5 7.07 × 10−4 
CHIP 0.78 (0.69–0.88) 1.01 × 10−4 1.25 × 10−3 
LRRT1 1.27 (1.12–1.43) 2.09 × 10−4 2.31 × 10−3 
PAFAH β subunit 0.77 (0.67–0.89) 3.24 × 10−4 3.31 × 10−3 
IL-2 sRa 1.26 (1.11–1.43) 4.41 × 10−4 4.30 × 10−3 
ERBB4 1.22 (1.09–1.37) 5.98 × 10−4 5.38 × 10−3 
PFD5 0.80 (0.70–0.91) 7.84 × 10−4 6.73 × 10−3 
kallikrein 5 1.24 (1.09–1.41) 8.12 × 10−4 6.74 × 10−3 
IMB1 0.80 (0.70–0.91) 9.07 × 10−4 7.26 × 10−3 
Cyclophilin F 0.80 (0.70–0.91) 1.09 × 10−3 8.44 × 10−3 
Gro-a 0.80 (0.71–0.92) 1.12 × 10−3 8.53 × 10−3 
EG-VEGF 1.22 (1.08–1.38) 1.32 × 10−3 9.78 × 10−3 
ALK-1 0.80 (0.70–0.92) 1.66 × 10−3 1.14 × 10−2 
PolyUbiquitin K48 1.23 (1.08–1.39) 1.71 × 10−3 1.16 × 10−2 
Caspase-3 0.81 (0.71–0.93) 1.99 × 10−3 1.29 × 10−2 
Fas ligand, soluble 0.82 (0.72–0.93) 2.48 × 10−3 1.55 × 10−2 
EP15R 0.83 (0.73–0.94) 2.75 × 10−3 1.65 × 10−2 
PRKACA 0.83 (0.73–0.94) 3.11 × 10−3 1.84 × 10−2 
GRB2 adapter protein 0.83 (0.73–0.94) 3.42 × 10−3 1.95 × 10−2 
a-Synuclein 0.82 (0.72–0.94) 3.77 × 10−3 2.07 × 10−2 
Cofilin-1 0.82 (0.71–0.95) 6.32 × 10−3 3.08 × 10−2 
Carbonic anhydrase XIII 0.83 (0.72–0.95) 7.90 × 10−3 3.67 × 10−2 
GPVI 0.83 (0.73–0.95) 8.22 × 10−3 3.74 × 10−2 
IL-4 sR 1.17 (1.04–1.32) 9.03 × 10−3 3.98 × 10−2 
NSE 0.85 (0.75–0.96) 9.59 × 10−3 4.09 × 10−2 
I-309 1.19 (1.04–1.35) 9.75 × 10−3 4.14 × 10−2 
Sorting nexin 4 0.84 (0.74–0.96) 9.96 × 10−3 4.21 × 10−2 
ATPO 0.83 (0.72–0.96) 1.23 × 10−2 4.93 × 10−2 

Protein associations with incident T2D were modeled using Cox proportional hazards adjusted for age, sex, and batch. Associations with continuous clinical traits were modeled using linear regression models adjusted for age, sex, and batch. Proteins listed are those associated with incident T2D with q < 0.05 and not associated with BMI, HOMA-IR, triglyceride and FPG levels, and eGFR defined as q > 0.05 (these nonsignificant associations are listed in Supplementary Table 7).

Generalizability of Incident Diabetes Associations Across Diverse Populations

To determine the generalizability of our findings, we compared incident diabetes-associated proteins in AA individuals from the JHS with meta-analyzed results from the FHS and MDCS cohorts (14). A total of 1,618 individuals (n = 177 incident diabetes cases over a mean of 11.6 years) and 1,221 (n = 272 cases over a mean of 12.9 years) were profiled from FHS and MDCS, respectively. Individuals from both cohorts self-identified as White. Fewer FHS and MDCS participants were women, and the participants were older and had lower BMI, FPG levels, and HOMA-IR in case and control participants compared with the JHS cohort, and more had hypertension in MDCS (Supplementary Table 11). Of the 325 proteins associated with incident diabetes in JHS model 1 (q < 0.05), 196 were also associated with incident diabetes in the FHS/MDCS meta-analyses (P < 0.05), adjusting for the same covariables (Supplementary Table 12). Of the 111 proteins associated in JHS model 2, 41 replicated (Supplementary Table 13) and of the 36 proteins associated in JHS model 3, 12 replicated (Fig. 4 and Supplementary Table 14).

Figure 4

Diabetes-associated proteins in the JHS that replicated in an FHS/MDCS meta-analysis. Volcano plot shows the HRs for incident diabetes from Cox proportional hazards models for a 1-SD increase in log-transformed, inverse rank normalized, and plate regressed residualized level of circulating protein levels in a meta-analysis of FHS/MDCS in age; sex; BMI; FPG, triglyceride, and HDL cholesterol levels; eGFR; hypertension; statin use; and batch-adjusted models. The 36 proteins that were associated with incident diabetes in JHS at an FDR q < 0.05 were included in this plot. Green dots represent proteins that were significant at P < 0.05 in the FHS/MDCS meta-analysis in JHS.

Figure 4

Diabetes-associated proteins in the JHS that replicated in an FHS/MDCS meta-analysis. Volcano plot shows the HRs for incident diabetes from Cox proportional hazards models for a 1-SD increase in log-transformed, inverse rank normalized, and plate regressed residualized level of circulating protein levels in a meta-analysis of FHS/MDCS in age; sex; BMI; FPG, triglyceride, and HDL cholesterol levels; eGFR; hypertension; statin use; and batch-adjusted models. The 36 proteins that were associated with incident diabetes in JHS at an FDR q < 0.05 were included in this plot. Green dots represent proteins that were significant at P < 0.05 in the FHS/MDCS meta-analysis in JHS.

Close modal

In exploratory analyses, we found four proteins with different directions of effect in incident diabetes–protein associations. Plasma kallikrein (KLKB1; HR 1.30 [95% CI 1.15–1.48]; q = 2.14 × 10−3) and protein S (PROS1; HR 1.30 [95% CI 1.13–1.50]; q = 9.44 × 10−3) were positively associated; importin subunit β-1 (KPNB1, 0.80 [95% CI 0.71–0.91]; q = 2.03 × 10−2) was inversely associated in JHS, but was not associated with incident disease in FHS/MDCS (Supplementary Fig. 2). Creatinine kinase M-type protein (CKB/CKM; 0.82 [95% CI 0.71–0.93]; q = 1.18 × 10−2) was inversely associated with incident diabetes in the FHS/MDCS meta-analysis but had a positive association in JHS, albeit with a q value >0.05 (1.25 [95% CI 1.06–1.46]; q = 6.99 × 10−2). These proteins demonstrate that, although a majority of associations are generalizable to most populations, there may be findings that are unique to specific subpopulations.

Proteins as Clinical Diabetes Prediction Biomarkers

Elastic net regularization was used to select protein predictors for inclusion in a clinical diabetes risk score (which included age; sex; BMI; waist circumference; SBP; levels of FPG, HDL, and triglycerides; and family history of diabetes) (Supplementary Table 15) (26). The clinical risk model alone had a Harrell C statistic of 0.73 (95% CI 0.69–0.77) in JHS (Fig. 5 and Supplementary Table 16). This increased to 0.79 (95% CI 0.76–0.92) with the addition of the 14 protein predictors selected using elastic net regression on all 1,305 circulating proteins profiled and to 0.81 (95% CI 0.78–0.84) with the 28 proteins selected using elastic net from the 36 JHS model 3 diabetes-associated proteins. The C statistic further increased to 0.82 (95% CI 0.79–0.85) with the inclusion of 34 proteins selected using elastic net from the 111 JHS model 2 diabetes-associated proteins. After correcting for optimism (27), there was a decrease of 0.02 in the C statistic to 0.77 with the 14-protein model, a decrease by 0.03 (C = 0.77) for the 28-protein model, and a decrease of 0.03 (C = 0.79) for the 34-protein model. In FHS, the model with only clinical risk factors had a higher C statistic value of 0.89 (95% CI 0.85–0.92), which was expected given that the score was derived in this cohort (26). Incremental improvements in model discrimination were again demonstrated with the inclusion of circulating proteins, especially with the 34-protein model (C = 0.93; 95% CI 0.90–0.96) (Fig. 5 and Supplementary Table 16) that was different from the base clinical prediction model at P = 7.60 × 10−5 using a likelihood ratio test. The receiver operating characteristic curve and associated area under the curve of these two models in FHS are shown in Supplementary Fig. 3. The C statistic of the 14-protein was 0.90 (95% CI 0.87–0.93) and was 0.91 (95% CI 0.88–0.94) for the 28-protein model.

Figure 5

Discrimination of clinical diabetes prediction models using proteins. C statistics for prediction models of incident diabetes, including clinical risk factors alone and elastic net–selected proteins in the JHS and FHS are visualized. Clinical risk factors included were age; sex; BMI; SBP; HDL cholesterol, triglyceride, and FPG levels; and family history of diabetes. Elastic net regularization was used to select 1) 14 proteins from the 1,305 measured; 2) 28 of the 36 diabetes-associated proteins in JHS model 3; and 3) 34 of the 111 diabetes-associated proteins in JHS model 2. DM, diabetes mellitus.

Figure 5

Discrimination of clinical diabetes prediction models using proteins. C statistics for prediction models of incident diabetes, including clinical risk factors alone and elastic net–selected proteins in the JHS and FHS are visualized. Clinical risk factors included were age; sex; BMI; SBP; HDL cholesterol, triglyceride, and FPG levels; and family history of diabetes. Elastic net regularization was used to select 1) 14 proteins from the 1,305 measured; 2) 28 of the 36 diabetes-associated proteins in JHS model 3; and 3) 34 of the 111 diabetes-associated proteins in JHS model 2. DM, diabetes mellitus.

Close modal

In this study, we used DNA aptamer–based technology to identify 325 circulating proteins associated with incident diabetes in the JHS, a self-identified AA cohort, after adjusting for age, sex, and batch (Fig. 1), with a mean 7.4 years of follow-up that could shed light on biological processes involved in the development of diabetes. Many of these proteins report on known development pathways such as obesity and insulin resistance, but we determined a subset is uniquely associated with incident diabetes independent of established markers of diabetes disease pathways (Supplementary Table 7). Furthermore, 36 proteins remained significantly associated in our fully adjusted model, which further adjusts for BMI; and levels of FPG, HDL cholesterol, and triglycerides; hypertension status; statin use; and eGFR (Fig. 2 and Supplementary Table 4), 12 of which replicated in two large, self-identified White community cohorts, the FHS and MDCS, confirming that they are relevant biomarkers across diverse populations. It is important to note, however, that the inclusion of self-identified AAs in our discovery cohort also allowed us to illuminate a number of associations that, to our knowledge, were not previously found, supporting the need for diversity in discovery populations. Finally, we demonstrated that the addition of these diabetes associated proteins adds value to clinical diabetes prediction models, stressing the additive knowledge provided by these new protein associations.

The aptamer-based technologies we used have been used to study diabetes in five other White cohorts: the Ages Gene/Environment Susceptibility (AGES)-Reykjavik Study (12); the Cooperative Health Research in the Region of Augsburg (KORA) cohort and the Nord-Trøndelag Health Study (HUNT3) (13); and a meta-analysis of FHS and MDCS (14). Of the 325 proteins associated with incident diabetes in JHS model 1 and the 36 proteins in model 3, 196 and 12, respectively, were validated in the FHS/MDCS meta-analyses adjusting for the same covariables (P < 0.05) (Fig. 4 and Supplementary Tables 1214). Ten associations that were significant in JHS model 1 and were seen in AGES, KORA, and/or HUNT3, including inverse associations with adiponectin (1214), histone-lysine N-methyltransferase EHMT2, iduronate 2-sulfase (12,14), WFIKKN2 (12,14), and NT-3 growth factor receptor (12,14). Positive associations included afamin (1214), aminoacylase-1 (12,14), growth hormone receptor, reticulon-4 receptor, and the proto-oncogene tyrosine-protein kinase receptor Ret (12,14). Associations to a nominal P level of significance were available in AGES-Reykjavik (12) and FHS/MDCS (14), and to an FDR < 0.05 in KORA and HUNT (13). Given that these associations have been found in at least four large human cohorts with >5,000 individuals, they are strong candidates for additional mechanistic studies.

Of the 12 proteins associated with incident diabetes in JHS model 3 that replicated in FHS/MDCS, 10 were positively associated with incident diabetes (Supplementary Table 14) including HIST1H1C, serine/threonine-protein kinase 17B, DNA topoisomerase 1 (TOP1), interleukin-18 receptor 1 (IL18R1), completement factor H (CFH), high-mobility group protein B1 (HMGB1), formimidoyltransferase cyclodeaminase (FTCD), matrilysin (MMP7), sialic acid–binding Ig-like lectin 7 (SIGLEC7), and IGF-binding protein 4. Glutathione S-transferase P (GSTP1) and ubiquitin carboxyl-terminal hydrolase isozyme L1 (UCHL1) were inversely associated. Some proteins were moderately correlated with each other (Supplementary Fig. 3), but the median Spearman ρ was only 0.13 across the 12 proteins. Given the successful replication of these associations across cohorts that span different races, ethnicities, and geographies, these are also excellent candidates to prioritize for experimental studies to test for casual relationships.

Of these proteins, only CFH, a component of the alternative complement pathway, has been positively associated with incident diabetes previously, in a small Chinese cohort (28). Missense mutations in GSTP1 reduce the protein’s ability to protect against oxidative stress and increases cardiovascular disease risk in individuals with T2D (29), but they has not been linked with incident T2D. In mouse and cell models, overexpression of HMGB1 induces inflammation (30), UCHL1 protects human β-cells from amyloid toxicity (31), and HIST1H1C (32) and IL18R1 (33) are associated with diabetic retinopathy. We now demonstrate that circulating levels of these proteins precede the development of diabetes in humans. SIGLEC7 is an adhesion-signaling molecule in the same family as SIGLEC1, which has been linked to islet cell autoimmunity (34) with decreased expression in the islet cells of individuals with diabetes. Our data suggest there may be a compensatory increase in SIGLEC7 levels several years prior to the development of clinical disease. We also found that IGFB4, which is in the same family as IGFB2, a protein inversely associated with incident diabetes (35), is associated with incident disease. There is little in the literature that links the TOP1, FTCD, and MMP7 proteins to diabetes, making them ideal candidates for additional study.

In addition to these broadly generalizable protein biomarkers, we identified an additional 129 proteins in JHS model 1 and 24 proteins in model 3 that were not found in the FHS/MDCS meta-analysis, AGES-Reykjavik Study, KORA, or HUNT3—cohorts which have a more homogenous racial/ethnic makeup. Thus, the inclusion of a diverse cohort has enabled us to develop a more comprehensive understanding of the diabetes proteome. We did note four protein associations that appeared to differ between the JHS and FHS/MDCS (including KLKB1, PROS1, KPNB1, and CKB/CKM) that could reflect cohort differences, including variation in environmental exposures, dietary patterns, and social determinants of health between these populations. Only PROS1 remained associated with diabetes after adjusting for education level and household income as rudimentary proxies for social determinants of health. Replication of these findings in other multiethnic cohorts is needed to confirm these associations and further clarify which of these key factors contribute to these differences.

Fifty-two proteins associated with incident diabetes were not associated with baseline BMI, HOMA-IR, FPG levels, or triglyceride levels, which are known diabetes risk factors (Supplementary Table 7). Twenty-nine proteins remained significant after adjusting for eGFR (Table 2). On the basis of UniProt Knowledgebase (36) functional annotations (Supplementary Table 17), the 29 proteins are broadly involved in growth and development, especially in angiogenesis, enzymatic reactions, protein processing, inflammation, cell cycle regulation, and neurotransmission and nervous system organization. Obesity and insulin resistance have consistently been associated with procoagulant factors and increased risk for thromboembolism (3739) and may be involved in cross talk with inflammatory pathways (40). Dysregulation of angiogenesis is seen after the development of diabetes (41) but may also play a role in pancreatic β-cell survival (42). These findings indicate that inflammatory and procoagulopathic states may precede the development of diabetes and may be associated with diabetes independent of adiposity and insulin resistance.

Finally, we demonstrate that these protein biomarkers of diabetes improve disease prediction beyond traditional clinical risk factors (Fig. 5). Model discrimination improved with the addition of 34 elastic net–selected proteins from the 111 diabetes-associated proteins in JHS model 2 (Harrell C = 0.73 and 0.82, respectively), with similar improvements in prediction model discrimination when tested in FHS (Fig. 5). Although the inclusion of a large number of protein biomarkers may be inefficient for diabetes diagnosis given the availability of HbA1c, these findings demonstrate the additive value these proteins provide that is likely orthogonal to known clinical risk factors.

Although the strength of our analysis is the use of large and deeply phenotyped cohorts, we do note some limitations. First, although we demonstrate ample support for protein-incident disease associations, we are unable to establish causality. Mendelian randomization studies, although outside the scope of the present study, represent an important next step in this regard. Also, in our cross-sectional associations with clinical risk factors, the data were less powered to detect heterogeneity among associations and, therefore, some of the proteins we have highlighted as “uniquely” associated with diabetes independent of BMI, FPG triglyceride levels, and eGFR may actually be associated with these covariables. However, given the inclusion of UCHL1 (31) and IL2RA (43,44) in this group of proteins that have been linked with plausible pathways independent of adiposity and insulin resistance to T2D, we believe that these proteins are an appropriate subgroup to nominate for additional mechanistic study. Finally, our replication cohort was limited to White individuals, and although these results support the broad generalizability of our findings, additional replication in other multiethnic cohorts that include White individuals, AAs, and participants of other diverse ancestries are nevertheless warranted.

Conclusion

We have identified 325 circulating proteins associated with incident diabetes over a decade prior to the disease diagnosis in a large, self-identified AA cohort after adjusting for age, sex, and batch. Of these, 196 proteins replicated in a meta-analysis of two large White cohorts, confirming these are strong diabetes biomarkers in diverse human populations. Furthermore, we identified an addition 129 potentially novel diabetes biomarkers with the inclusion of a diverse cohort in proteomics, offering new insights into diabetes pathophysiology. A selection of these proteins also increased discrimination in diabetes prediction models, both in the JHS training cohort and in the FHS testing cohort. Our analyses motivate proteomic studies in increasingly diverse populations to understand how important disease factors, both including and beyond biological contributors, converge to cause these perceived proteomic differences.

This article contains supplementary material online at https://doi.org/10.2337/figshare.21830886.

Acknowledgments. The authors thank all the study participants in and staff working on the Jackson Heart Study for their important contributions.

Funding. Research reported here was supported by the National Institute of Diabetes and Digestive and Kidney Diseases (grant K23DK127073 to Z.-Z.C.), the National Heart, Lung, and Blood Institute (NHLBI) (grant R01HL133870, R01HL132320, and R01AG063507), and the Trans-Omics for Precision Medicine program of the NHLBI (grant HHSN268201600034I) to R.E.G. The Jackson Heart Study is supported and conducted in collaboration with Jackson State University (HHSN268201800013I), Tougaloo College (HHSN268201800014I), the Mississippi State Department of Health (HHSN268201800015I/HHSN26800001), and the University of Mississippi Medical Center (HHSN268201800010I, HHSN268201800011I, and HHSN268201800012I) contracts from the NHLBI and the National Institute for Minority Health and Health Disparities.

The views expressed in this manuscript are those of the authors and do not necessarily represent the views of the National Heart, Lung, and Blood Institute, the National Institutes of Health, or the U.S. Department of Health and Human Services.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. Z.-Z.C. designed the study, led the data analysis, and drafted the manuscript. L.A.F. and D.S. performed the proteomics profiling of the cohorts. Z.-Z.C., Y.G., M.J.K., S.D., and M.M. performed the statistical analyses and contributed to the drafting of the manuscript. U.A.T., D.E.C., D.N., M.D.B., and J.M.R. provided critical feedback during the drafting and revisions of the manuscript. A.C., J.G.W., and R.E.G. provided mentorship and critical feedback at all stages of the project, including during the drafting of the manuscript. R.E.G. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity and accuracy of the data and data analysis.

1.
World Health Organization
.
Definition and diagnosis of diabetes mellitus and intermediate hyperglycaemia [Internet]
.
World Health Organization website. Accessed 23 September 2019. Available from https://apps.who.int/iris/handle/10665/43588
2.
Knowler
WC
,
Barrett-Connor
E
,
Fowler
SE
, et al.;
Diabetes Prevention Program Research Group
.
Reduction in the incidence of type 2 diabetes with lifestyle intervention or metformin
.
N Engl J Med
2002
;
346
:
393
403
3.
Pan
XR
,
Li
GW
,
Hu
YH
, et al
.
Effects of diet and exercise in preventing NIDDM in people with impaired glucose tolerance. The Da Qing IGT and Diabetes Study
.
Diabetes Care
1997
;
20
:
537
544
4.
Tuomilehto
J
,
Lindström
J
,
Eriksson
JG
, et al.;
Finnish Diabetes Prevention Study Group
.
Prevention of type 2 diabetes mellitus by changes in lifestyle among subjects with impaired glucose tolerance
.
N Engl J Med
2001
;
344
:
1343
1350
5.
World Health Organization Global Report on Diabetes
.
Geneva, Switzerland
,
World Health Organization
;
2016
6.
Centers for Disease Control and Prevention
.
National Diabetes Statistics Report, 2020
.
7.
Brancati
FL
,
Kao
WHL
,
Folsom
AR
,
Watson
RL
,
Szklo
M
.
Incident type 2 diabetes mellitus in African American and white adults: the Atherosclerosis Risk in Communities Study
.
JAMA
2000
;
283
:
2253
2259
8.
Bancks
MP
,
Kershaw
K
,
Carson
AP
,
Gordon-Larsen
P
,
Schreiner
PJ
,
Carnethon
MR
.
Association of modifiable risk factors in young adulthood with racial disparity in incident type 2 diabetes during middle adulthood
.
JAMA
2017
;
318
:
2457
2465
9.
Hill-Briggs
F
,
Adler
NE
,
Berkowitz
SA
, et al
.
Social determinants of health and diabetes: a scientific review
.
Diabetes Care
2020
;
44
:
258
279
10.
Sun
BB
,
Maranville
JC
,
Peters
JE
, et al
.
Genomic atlas of the human plasma proteome
.
Nature
2018
;
558
:
73
79
11.
Emilsson
V
,
Ilkov
M
,
Lamb
JR
, et al
.
Co-regulatory networks of human serum proteins link genetics to disease
.
Science
2018
;
361
:
769
773
12.
Gudmundsdottir
V
,
Zaghlool
SB
,
Emilsson
V
, et al
.
Circulating protein signatures and causal candidates for type 2 diabetes
.
Diabetes
2020
;
69
:
1843
1853
13.
Elhadad
MA
,
Jonasson
C
,
Huth
C
, et al
.
Deciphering the plasma proteome of type 2 diabetes
.
Diabetes
2020
;
69
:
2766
2778
14.
Ngo
D
,
Benson
MD
,
Long
JZ
, et al
.
Proteomic profiling reveals biomarkers and pathways in type 2 diabetes risk
.
JCI Insight
2021
;
6
:
144392
15.
Sempos
CT
,
Bild
DE
,
Manolio
TA
.
Overview of the Jackson Heart Study: a study of cardiovascular diseases in African American men and women
.
Am J Med Sci
1999
;
317
:
142
146
16.
Mahmood
SS
,
Levy
D
,
Vasan
RS
,
Wang
TJ
.
The Framingham Heart Study and the epidemiology of cardiovascular disease: a historical perspective
.
Lancet
2014
;
383
:
999
1008
17.
Berglund
G
,
Elmstähl
S
,
Janzon
L
,
Larsson
SA
.
The Malmo Diet and Cancer Study. Design and feasibility
.
J Intern Med
1993
;
233
:
45
51
18.
Carpenter
MA
,
Crow
R
,
Steffes
M
, et al
.
Laboratory, reading center, and coordinating center data management methods in the Jackson Heart Study
.
Am J Med Sci
2004
;
328
:
131
144
19.
Levey
AS
,
Stevens
LA
,
Schmid
CH
, et al.;
CKD-EPI (Chronic Kidney Disease Epidemiology Collaboration)
.
A new equation to estimate glomerular filtration rate
.
Ann Intern Med
2009
;
150
:
604
612
20.
Cederholm
J
,
Eeg-Olofsson
K
,
Eliasson
B
,
Zethelius
B
,
Nilsson
PM
;
Swedish National Diabetes Register
.
Risk prediction of cardiovascular disease in type 2 diabetes: a risk equation from the Swedish National Diabetes Register
.
Diabetes Care
2008
;
31
:
2038
2043
21.
Lindholm
E
,
Agardh
E
,
Tuomi
T
,
Groop
L
,
Agardh
CD
.
Classifying diabetes according to the new WHO clinical stages
.
Eur J Epidemiol
2001
;
17
:
983
989
22.
Candia
J
,
Cheung
F
,
Kotliarov
Y
, et al
.
Assessment of variability in the SOMAscan assay
.
Sci Rep
2017
;
7
:
14248
23.
Ngo
D
,
Sinha
S
,
Shen
D
, et al
.
Aptamer-based proteomic profiling reveals novel candidate biomarkers and pathways in cardiovascular disease
.
Circulation
2016
;
134
:
270
285
24.
Jacob
J
,
Ngo
D
,
Finkel
N
, et al
.
Application of large-scale aptamer-based proteomic profiling to planned myocardial infarctions
.
Circulation
2018
;
137
:
1270
1277
25.
Benjamini
Y
,
Hochberg
Y
.
Controlling the false discovery rate: a practical and powerful approach to multiple testing
.
J R Stat Soc B
1995
;
57
:
289
300
26.
Wilson
PWF
,
Meigs
JB
,
Sullivan
L
,
Fox
CS
,
Nathan
DM
,
D’Agostino
RB
Sr
.
Prediction of incident diabetes mellitus in middle-aged adults: the Framingham Offspring Study
.
Arch Intern Med
2007
;
167
:
1068
1074
27.
Iba
K
,
Shinozaki
T
,
Maruo
K
,
Noma
H
.
Re-evaluation of the comparative effectiveness of bootstrap-based optimism correction methods in the development of multivariable clinical prediction models
.
BMC Med Res Methodol
2021
;
21
:
9
28.
Gou
W
,
Yue
L
,
Tang
XY
, et al
.
Circulating proteome and progression of type 2 diabetes
.
J Clin Endocrinol Metab
2022
;
107
:
1616
1625
29.
Doney
ASF
,
Lee
S
,
Leese
GP
,
Morris
AD
,
Palmer
CNA
.
Increased cardiovascular morbidity and mortality in type 2 diabetes is associated with the glutathione S transferase theta-null genotype: a Go-DARTS study
.
Circulation
2005
;
111
:
2927
2934
30.
Park
JS
,
Svetkauskaite
D
,
He
Q
, et al
.
Involvement of toll-like receptors 2 and 4 in cellular activation by high mobility group box 1 protein
.
J Biol Chem
2004
;
279
:
7370
7377
31.
Costes
S
,
Gurlo
T
,
Rivera
JF
,
Butler
PC
.
UCHL1 deficiency exacerbates human islet amyloid polypeptide toxicity in β-cells: evidence of interplay between the ubiquitin/proteasome system and autophagy
.
Autophagy
2014
;
10
:
1004
1014
32.
Wang
W
,
Wang
Q
,
Wan
D
, et al
.
Histone HIST1H1C/H1.2 regulates autophagy in the development of diabetic retinopathy
.
Autophagy
2017
;
13
:
941
954
33.
Wang
J
,
Sun
C
,
Gerdes
N
, et al
.
Interleukin 18 function is mediated by the interleukin 18 receptor and Na-Cl co-transporter
.
Nat Med
2015
;
21
:
820
826
34.
Ferreira
RC
,
Guo
H
,
Coulson
RMR
, et al
.
A type I interferon transcriptional signature precedes autoimmunity in children genetically at risk for type 1 diabetes
.
Diabetes
2014
;
63
:
2538
2550
35.
Wittenbecher
C
,
Ouni
M
,
Kuxhaus
O
, et al
.
Insulin-like growth factor binding protein 2 (IGFBP-2) and the risk of developing type 2 diabetes
.
Diabetes
2019
;
68
:
188
197
36.
UniProt [Internet]
.
Accessed 12 January 2022. Available from https://www.uniprot.org/
37.
Godsland
IF
,
Crook
D
,
Proudler
AJ
,
Stevenson
JC
.
Hemostatic risk factors and insulin sensitivity, regional body fat distribution, and the metabolic syndrome
.
J Clin Endocrinol Metab
2005
;
90
:
190
197
38.
Eichinger
S
,
Hron
G
,
Bialonczyk
C
, et al
.
Overweight, obesity, and the risk of recurrent venous thromboembolism
.
Arch Intern Med
2008
;
168
:
1678
1683
39.
Stewart
LK
,
Kline
JA
.
Metabolic syndrome increases risk of venous thromboembolism recurrence after acute deep vein thrombosis
.
Blood Adv
2020
;
4
:
127
135
40.
Morange
PE
,
Alessi
MC
.
Thrombosis in central obesity and metabolic syndrome: mechanisms and epidemiology
.
Thromb Haemost
2013
;
110
:
669
680
41.
Cheng
R
,
Ma
JX
.
Angiogenesis in diabetes and obesity
.
Rev Endocr Metab Disord
2015
;
16
:
67
75
42.
Qian
J
,
Tao
D
,
Shan
X
,
Xiao
X
,
Chen
C
.
Role of angiogenesis in beta-cell epithelial-mesenchymal transition in chronic pancreatitis-induced diabetes
.
Lab Invest
2022
;
102
:
290
297
43.
Lowe
CE
,
Cooper
JD
,
Brusko
T
, et al
.
Large-scale genetic fine mapping and genotype-phenotype associations implicate polymorphism in the IL2RA region in type 1 diabetes
.
Nat Genet
2007
;
39
:
1074
1082
44.
Dwyer
CJ
,
Ward
NC
,
Pugliese
A
,
Malek
TR
.
Promoting immune regulation in type 1 diabetes using low-dose interleukin-2
.
Curr Diab Rep
2016
;
16
:
46
Readers may use this article as long as the work is properly cited, the use is educational and not for profit, and the work is not altered. More information is available at https://www.diabetesjournals.org/journals/pages/license.