Albuminuria, the HDL Proteome, and Coronary Artery Calcification in Type 1 Diabetes: the DCCT/EDIC Study

Abstract

Objective –

Albuminuria is an important risk factor for cardiovascular disease in diabetes. We determined whether albuminuria associates with alterations in the proteome of HDL of subjects with type 1 diabetes mellitus (T1DM), and whether those alterations associated with coronary artery calcification (CAC).

Approach and Results –

In a cross-sectional study of 191 subjects enrolled in the Diabetes Control and Complications Trial/Epidemiology of Diabetes Interventions and Complications Study (DCCT/EDIC), we used isotope dilution tandem-mass spectrometry to quantify 46 proteins in HDL. Stringent statistical analysis demonstrated that eight proteins associated with albuminuria. Two of those proteins, α−1-microglobulin/bikunin precursor (AMBP) and prostaglandin-H2 D-isomerase (PTGDS), strongly and positively associated with the albumin excretion rate (P<10⁻⁶). Furthermore, paraoxonase 1 (PON1) and PON3 levels in HDL strongly and negatively associated with the presence of coronary artery calcium, with odds ratios per 1-SD difference of 0.63 (0.43–0.92, 95% CI, P=0.018) for PON1 and 0.59 (0.40–0.87, 95% CI, P=0.0079) for PON3. Only one protein, PON1, associated with both albumin excretion rate and CAC.

Conclusions –

Our observations indicate that the HDL proteome is remodeled in T1DM subjects with albuminuria. Moreover, low concentrations of the anti-atherosclerotic protein PON1 in HDL associated with both albuminuria and CAC, raising the possibility that alterations in HDL’s protein cargo mediate in part the known association of albuminuria with cardiovascular risk in T1DM.

Graphical Abstract

Albuminuria is an independent risk factor for cardiovascular disease (CVD) in people with type 1 diabetes mellitus (T1DM), type 2 diabetes mellitus (T2DM), and in the general population^1–4. Albuminuria also increases the risk of progressing to end-stage renal disease (ESRD)^{5, 6}. Albuminuria may contribute to CVD risk in T2DM subjects by promoting an atherogenic lipid profile characterized by elevated triglycerides, low HDL cholesterol (HDL-C), and a shift in LDL to small, dense particles⁷. However, HDL-C levels are not consistently low in T1DM subjects despite their increase in CVD risk^8–10.

Clinical and epidemiological studies show a robust, inverse association of HDL-C levels with CVD risk^{11, 12}. However, several lines of evidence suggest that the association between HDL-C levels and CVD status is not a causal relationship and that elevating HDL-C is not necessarily therapeutic^{13, 14}. For example, a genetic mutation in SR-BI (SCARB1), a protein critically involved in HDL metabolism in the liver, greatly increases both HDL-C levels and atherosclerosis^{15, 16}. Collectively these observations indicate that HDL-C levels do not necessarily reflect HDL’s cardioprotective effects.

HDL metrics other than HDL-C are likely to mediate HDL’s cardioprotective effects. HDL carries a wide array of proteins linked to lipoprotein metabolism, inflammation, protease inhibition, and complement regulation, suggesting that this cargo contributes to HDL’s anti-inflammatory and anti-atherogenic properties¹⁷. We found that the HDL proteome is dramatically remodeled in ESRD patients¹⁸. It is unclear whether albuminuria in T1DM affects HDL’s protein cargo or whether alterations in HDL’s proteins associate with CVD in T1DM.

In the current study, we used targeted mass spectrometric proteomics to test the hypotheses that alternations in the HDL proteome associate with albuminuria in T1DM subjects. Because albuminuria is a risk factor for CVD, we also determined whether changes in the abundance of proteins in the HDL proteome associate with coronary artery calcification, which strongly predicts incident CVD risk¹⁹. Our observations support the proposal that certain HDL proteins may be markers–and perhaps mediators–of albuminuria and the increased cardiovascular risk associated with T1DM.

MATERIALS AND METHODS

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Subjects and experimental design.

The Epidemiology of Diabetes Interventions and Complications (EDIC) study²⁰ was an observational cohort study of T1DM subjects that followed the Diabetes Control and Complications Trial (DCCT). All subjects in our study were enrolled in an ancillary EDIC study of lipoproteins²¹ that involved subjects from all 28 clinical sites in the United States and Canada²¹. For this study, extra plasma was obtained every other year during EDIC years 3–12 (1997–2006). After an overnight fast, blood was collected into ice-cold tubes containing EDTA (6 mM final concentration). Plasma was prepared immediately by centrifugation (2500 g for 15 min) and frozen at −80°C until analysis²². All studies were approved by the Human Studies Committee at the University of Washington, which coordinated the lipoprotein ancillary study, and by the EDIC clinical centers.

We designed a cross-sectional study based on urinary albumin excretion, with oversampling of participants with microalbuminuria and macroalbuminuria. Urinary albumin excretion had been quantified every other year as albumin excretion rate (AER). Fasting blood and AER had been collected in the alternate years. We first selected all the stored plasma samples from annual visits at which the ancillary EDIC subjects had demonstrated macroalbuminuria (AER ≥300 mg/dL) at that visit as well as one year earlier and one year later. We then randomly sampled similar numbers of participants who demonstrated microalbuminuria (AER 30–299 mg/dL) or normoalbuminuria (AER <30 mg/dL)²³ during the annual visit and one year earlier and later. Subjects were not selected on the basis of gender, HbA1c, or other clinical criteria. When multiple time points were available for a given participant, we used the time point closest to EDIC year 7 or 8 (calendar year 2000–2002), when EDIC measured coronary artery calcification (CAC)²⁴. For each selected subject, we used the clinical characteristics obtained at the same visit as the plasma sample. The mean duration of diabetes at the time of sampling was >20 years in each group.

HDL isolation.

HDL (density 1.063–1.210 g/mL) was isolated by sequential ultracentrifugation from rapidly thawed plasma²⁵, using buffers supplemented with 100 μM diethylenetriaminepentaacetic acid (DTPA), 100 μM butylated hydroxytoluene (BHT), and a protease inhibitor cocktail (Sigma, St. Louis, MO). The protein concentration of HDL was determined using the Lowry assay (BioRad), with albumin as the standard. All samples were deidentified and analyzed in a blinded manner in HDL isolation, HDL proteomics, and statistical analysis.

Targeted quantification of HDL proteins by selected reaction monitoring (SRM) with ¹⁵N-labeled APOA1 as the internal standard.

Human [¹⁵N]APOA1 (>99% purity after isolation) was produced by a bacterial expression system²⁶. Following the addition of freshly prepared methionine (10 mM final concentration), HDL proteins (10 μg total proteins) and [¹⁵N]APOA1 (0.5 μg, as the internal standard) were reduced with dithiothreitol and alkylated with iodoacetamide. Tryptic digests of HDL were analyzed with a nano-LC-MS/MS Thermo TSQ Vantage coupled to a Waters nanoACQUITY UltraPerformance liquid chromatography system^{18, 25}. Resolution for Q1 and Q3 were 0.7 Da (full width at half maximum), and analyses were performed using a scheduled transition list generated by Skyline²⁷, an open source program for quantitative data processing and proteomic analysis. At least two peptides per protein and three or four SRM transitions of each peptide were used for the Skyline analysis^{18, 25}. The results were expressed as the ratio of the peak areas of each peptide and the [¹⁵N]APOA1 peptide [¹⁵N]THLAPYSDELR. To calculate the relative levels of each peptide, we set the average ratio of the peptide to the internal standard peptide in normoalbuminuric subjects to an arbitrary unit of one.

Statistical analysis.

Differences in relative protein expression by albuminuria status determined by SRM analysis were evaluated by ANOVA of the group means. We also performed multiple linear regression analyses for percent changes of protein levels in HDL per doubling in AER with or without adjustment for clinical characteristics including age, sex, DCCT treatment group, use of renin-angiotensin-aldosterone system inhibitors (RAASi), duration of diabetes, use of lipid-lowering medications, smoking, body mass index (BMI), and mean percentage of hemoglobin A1c (HbA1c%). To account for multiple comparison testing of HDL proteins, we used the Benjamini-Hochberg method with a 10% false discovery rate, i.e. based on the corrected or adjusted p-values (q-values), only proteins with a q-value less than 0.10 were considered significant. We tested associations of HDL proteins with the presence of CAC by using logistic regression analysis. The logistic models were adjusted for potential confounders as described above and also for log(AER). Odds ratios (ORs) are reported per analyte SD. Correlation analysis with continuous variables used Pearson’s coefficient. All statistical analyses were performed in R, a free software environment for statistical computing and graphics.

RESULTS

The clinical characteristics of the 191 subjects we studied are listed in Table 1. The three groups (normoalbuminuria, microalbuminuria, macroalbuminuria) had similar mean ages, percentages of smokers, BMIs, and HDL-C levels. LDL-C levels were slightly lower and the duration of diabetes was slightly longer in the normoalbuminuria group. In contrast, the groups differed significantly in gender distribution. Also, median levels of triglycerides differed significantly among the three groups (higher in the macroalbuminuria group), as did estimated glomerular filtration rates (eGFR) (lower in macroalbuminuria subjects) and HbA1c levels (lower in the normoalbuminuria subjects). There were no differences in HbA1c, albuminuria, or coronary artery calcium between the EDIC subjects, whose glucose levels had been tightly controlled, and the DCCT subjects, whose levels had been controlled less tightly.

Table 1.

Clinical characteristics of subjects with T1DM by albumin excretion rate (AER) status.

Covariate	Normal AER (< 30 mg/day)	Microalbuminuria (30 – < 300 mg/day)	Macroalbuminuria (≥ 300 mg/day)	P-value
Number of subjects	67	64	60
Age (years)	42.0 (37.0–48.0)	44.0 (35.0–48.2)	43.5 (36.2–49.0)	0.45
Gender (Female)	39 (58)	16 (25)	10 (17)	<0.0001
Caucasian	67 (100)	64 (100)	60 (100)
DCCT treatment
Conventional glucose control	39 (58)	41 (64)	48 (80)	0.01
Intensive glucose control	28 (42)	23 (36)	12 (20)	0.01
BMI (kg/m2)	27.6 (24.6–30.7)	27.0 (24.8–29.8)	28.3 (25.2–31.2)	0.34
Smoker	17 (25)	12 (19)	21 (35)	0.24
Blood pressure med(s)	41 (61)	44 (69)	48 (80)
Lipid-lowering med(s)	46 (69)	49 (77)	53 (88)	0.008
Statin	18 (27)	3 (5)	18 (30)	0.0006
RAASi
ACEi	42 (63)	49 (77)	53 (88)	0.001
ARB	10 (15)	0 (0)	9 (15)
Duration of DM (years)	26.1 (22.1–30.0)	20.0 (17.3–24.3)	22.0 (18.4–24.3)	0.04
eGFR (CKD-EPI) (mL/min/1.73m2)	106.0 (93.8–113.4)	106.5 (95.3–114.5)	84.1 (57.1–106.5)	<0.0001
HbA1c (%) (mmol/mol)	7.7 (7.0–8.8) 61 (54–73)	8.4 (7.7–9.4) 68 (61–79)	8.6 (7.8–9.6) 70 (62–82)	0.006
SBP (mmHg)	117.0 (111.5–124.8)	124.0 (114.5–132.5)	132.0 (125.0–144.0)	0.0001
DBP (mmHg)	73.0 (69.0–78.5)	78.0 (72.0–82.5)	80.0 (71.5–86.5)	0.02
HDL-C (mg/dL)	50.0 (43.0–62.0)	49.0 (41.8–59.0)	50.0 (42.0–59.5)	0.55
LDL-C (mg/dL)	103.0 (86.0–118.0)	110.0 (86.2–134.2)	110.0 (84.0–135.5)	0.03
Total Cholesterol (mg/dL)	173.0 (157.0–188.0)	183.5 (153.8–203.0)	187.0 (154.0–221.5)	0.004
Triglycerides (mg/dL)	69.0 (53.0–117.0)	85.0 (54.8–111.0)	117.0 (70.0–180.5)	0.002
AER	12.0 (9.3–16.4)	69.5 (50.7–97.0)	758.8 (559.4–1497.2)	<0.0001

Entries are median (interquartile range) for continuous covariates and N (%) for categorical covariates. RAASi, renin-angiotensin-aldosterone system inhibitor; ACEi, angiotensin converting enzyme inhibitor; ARB angiotensin II receptor blocker. P-values were obtained by ANOVA analysis of the differences among group means.

SRM analysis revealed marked differences in the abundance of multiple HDL proteins in T1DM subjects with albuminuria.

To test whether albuminuria and CVD associate with abnormalities in HDL’s protein cargo in T1DM, we used tandem MS analysis with targeted SRM¹⁸ and isotope dilution to quantify relative levels of 46 proteins in HDL from the three groups of subjects. We selected these proteins because they were reproducibly detected in preliminary shotgun proteomics analysis^{17, 18} of TIDM subjects with normoalbuminuria, microalbuminuria, and macroalbuminuria (10 in each group, data not shown). All 46 proteins were detected in HDL isolated by ultracentrifugation from each of the 191 subjects in this study. Differences in protein expression were initially evaluated by a test for trend without adjustment for clinical characteristics but with adjustment for multiple comparison (Benjamin-Hochberg adjusted p-value < 0.10). This analysis revealed that 14 proteins were differentially expressed in subjects with micro- or macro-albuminuria, as compared with subjects with normoalbuminuria (Supplemental Table I).

Seven proteins (AMBP, B2M, CST3, LPA, PSAP, PTGDS, and RBP4) were significantly increased in HDL isolated from subjects with micro- and macro-albuminuria, as compared with those with normoalbuminuria. It is noteworthy that five of the seven proteins, AMBP, B2M, CST3, PTGDS, and RBP4, were increased markedly in HDL of ESRD subjects in a previous study¹⁸. In contrast, seven proteins (APOA1, APOA2, APOE, PCYOX1, PON1, PON3, and SAA4) were significantly reduced in HDL isolated from subjects with micro- and macro-albuminuria compared with those with normoalbuminuria. Representative examples of proteins that were markedly altered in the HDL of subjects with albuminuria are shown in Fig. 1. These observations demonstrated that albuminuria in T1DM subjects is associated with abnormalities in HDL protein composition.

Figure 1. SRM analysis of HDL proteins in normo-, micro- and macroalbuminuria subjects.

Box plots of AMBP (A), PTGDS (B), and PON1 (C) in HDL isolated from normo-, micro- and macroalbuminuria subjects. Peptides were quantified as the integrated peak area relative to that of [¹⁵N]APOA1. The box plots show the distribution of the data (median, interquartile ranges), while the dots represent individual data points. P-values are relative to the normoalbuminuria group. AU, arbitrary units.

HDL proteins associate with continuous albumin excretion rate (AER).

We used the absolute values of albumin excretion rate (i.e. the continuous AER) instead of albuminuria status (i.e. normo-, micro-, and macroalbuminuria) to investigate further the relationship between the levels of HDL proteins and albuminuria. We first used a linear regression model to test the association between HDL proteins and continuous AER without adjustment for clinical characteristics. When we adjusted for multiple comparison with the Benjamini-Hochberg method, 13 proteins remained significantly associated with continuous AER. Next, we tested the association between HDL proteins and continuous AER with a multivariable linear regression model with adjustment for clinical characteristics (Supplemental Table II). After adjusting for differences in gender distribution and other potential confounders and controlling for multiple comparisons, 8 proteins were associated with continuous AER (Table 2). Two of those proteins, AMBP and PTGDS (Fig. 2), exhibited large percentage differences per doubling of the continuous AER (≥10%) that were highly significant (P<0.0002). Taken together, these observations indicate that the relative abundance of certain proteins in HDL of subjects with albuminuria differed markedly from that of subjects with normoalbuminuria.

Table 2.

Adjusted associations of HDL proteins with log(AER).

Protein	% Difference per doubling in AER (95% CI)	p-value
AMBP	10 (5, 15)	9.11E–05
PTGDS	30 (14, 45)	0.00015
APOA2	−2 (−3, −1)	0.0023
APOE	−4 (−7, −1)	0.0092
CFD	−3 (−6, −1)	0.0099
PSAP	4 (1, 7)	0.014
SAA4	−2 (−4, 0)	0.016
PON1	−3 (−6, −1)	0.017

P-values are from a linear regression model that adjusts for age, sex, DCCT treatment group, use of RAASi, duration of DM, use of lipid-lowering medications, smoking, BMI, and HbA1c. HDL proteins that are significant when controlling the Benjamini-Hochberg false discovery rate (q-values) at 10% are shown (Supplemental Table II).

Figure 2. Volcano plot of adjusted associations of continuous log(AER) with each HDL protein (percent difference per doubling in AER).

For each protein, p-values are plotted versus the percent change in HDL protein levels per doubling in AER from a model that adjusts for age, sex, DCCT treatment group, use of RAASi, duration of diabetes, use of lipid-lowering medications, smoking, BMI, and HbA1c. Proteins that were increased with the doubling in AER are displayed to the right of the value 0 on the x axis (the vertical dashed line), while decreased proteins are to the left. The horizontal dotted line indicates the significance threshold using Benjamini-Hochberg false discovery rate of 10%.

Certain HDL proteins correlate highly with both eGFR and albuminuria while others correlate only with albuminuria.

We next determined the correlation between HDL proteins, albuminuria and GFR, quantified as log transformed AER and eGFR, respectively (Supplemental Table III). AMBP, PSAP and PTGDS strongly and negatively correlated with eGFR (Table 3; r=−0.47, −0.29 and −0.43, respectively). All three proteins also correlated strongly and positively with log(AER) (Table 3; r=0.46, 0.22 and 0.36, respectively). In contrast, APOE, SAA4 and PON1 exhibited significant negative correlations with log(AER) (r=−0.28, −0.20 and −0.22, respectively) but did not correlate significantly with eGFR (Table 3; r<0.12). These observations suggest that certain HDL proteins (AMBP, PSAP, PTGDS) are markers for both eGFR and albuminuria. In contrast, other proteins (APOE, SAA4, PON1) correlate only with albumin excretion.

Table 3.

Correlations (r) of HDL proteins with log(AER) and eGFR.

Protein	r log(AER)	P-value	Protein	r eGFR	P-value
AMBP	0.46	1.3E–11	AMBP	−0.47	6.0E-12
PTGDS	0.36	2.9E–07	PTGDS	−0.43	8.9E-10
APOE	−0.28	0.0001	APOE	0.06	0.41
PSAP	0.22	0.0024	PSAP	−0.29	6.4E-05
SAA4	−0.20	0.0059	SAA4	0.11	0.15
PON1	−0.22	0.0024	PON1	0.01	0.88

Pearson’s coefficients (r) and p-values are from correlation analysis between the levels of HDL proteins and log(AER) or eGFR. Only HDL proteins that are significant when controlling the Benjamini-Hochberg false discovery rate (q-values) at 10% are shown (Supplemental Table II).

PON1 and PON3 in HDL strongly and negatively associate with coronary artery calcification.

Using a logistic regression model, we next determined if levels of HDL proteins associated with coronary artery calcification (CAC). In this model, in addition to age, sex, DCCT treatment group, use of renin- angiotensin-aldosterone system inhibitors, duration of diabetes, use of lipid-lowering medications, smoking, BMI, and HbA1c, we also adjusted for log(AER). This analysis revealed that three HDL proteins, PON1, PON3 and LCAT were negatively associated with CVD, defined as CAC>0 (Supplemental Table IV). For PON1 and PON3, the odds ratios per 1-SD change were 0.63 and 0.59 with a p-value of 0.018 and 0.0079, respectively (Fig. 3). The mean levels of PON1 and PON3 in HDL isolated from subjects with coronary artery calcification were 18% and 24% lower, respectively, than those from subjects without calcification. The significant differences for PON1 and PON3 persisted after controlling for levels of HDL-C, LDL-C, and triglycerides. Moreover, LDL-C, HDL-C and triglycerides did not associate with coronary artery calcification in the fully adjusted model (Fig. 3). In contrast, the difference in LCAT levels was no longer significant after controlling for levels of HDL-C, LDL-C, and triglycerides. Although AMBP and PTGDS were strongly and positively associated with albuminuria, these two proteins were not associated with CAC>0 (Fig. 3).

Figure 3. PON1 and PON3 in HDL associate with the presence of CAC in a multiple regression model.

Odds ratios are per analyte SD. Estimates are from a multiple regression model that additionally adjusts for age, sex, DCCT treatment group, use of RAASi, duration of diabetes, use of lipid-lowering medications, smoking, BMI, HbA1c, and continuous log(AER).

To determine whether PON1 enzymatic activity correlates with PON1 mass (assessed by isotope dilution MS/MS), we quantified PON1 activity with two substrates—phenyl acetate (PA) and 4-(chloromethyl)phenyl acetate (CMPA)²⁸—in 238 subjects participating in the Coronary Artery Calcification in Type 1 Diabetes (CACTI) study²⁹. We found a positive correlation between PON1 mass and PON1 activity with each assay (PA, r=0.47, P<10⁻¹³; CMPA, r=0.31, P<10⁻⁶). These observations support the proposal that PON1 mass correlates strongly with PON1 enzymatic activity. In contrast, we found no correlation of PON1 mass or activity with cholesterol efflux capacity (measured with the Rothblatt/Rader method³⁰), suggesting that PON1 mass and/or activity are not major modulators of HDL’s ability to promote cholesterol efflux from macrophages.

DISCUSSION

In the current study, we used targeted MS/MS with isotope dilution to quantify 46 proteins in HDL isolated from subjects in the DCCT/EDIC study to determine whether albuminuria associated with alterations in HDL’s protein cargo. The relative levels of eight proteins in HDL associated significantly with continuous albumin excretion rate in a model corrected for multiple comparisons and clinically relevant characteristics (Fig. 4). Two of the proteins, AMBP and PTGDS, were very strongly and positively associated with albuminuria. Because albuminuria is an independent risk factor for CVD in both diabetic and non-diabetic subjects^{4, 31}, we also determined whether changes in HDL protein levels associate with coronary artery calcification (CAC>0), a strong predictor of CVD risk¹⁹. We identified two proteins, PON1 and PON3 that strongly and negatively associated with coronary artery calcification (Fig. 4). Only one protein, PON1, associated with both albumin excretion rate and CVD. Taken together, our observations indicate that the HDL proteome is remodeled in albuminuria, that low levels of PON1 and PON3 associate with coronary artery calcification, and that increased CVD risk in T1DM subjects with albuminuria may be mediated in part by decreased PON1 levels in HDL.

Figure 4. Venn diagram of HDL proteins, albuminuria and CVD.

Of the 46 proteins quantified in HDL isolated from T1DM subjects with or without albuminuria, eight proteins were significantly associated with albuminuria with adjustment for clinical characteristics and controlling the Benjamini-Hochberg false discovery rate at 10%. Two of those HDL proteins were negatively associated with logit (CAC>0) with adjustment for clinical characteristics and controlling the HDL-C, LDL-C, and triglycerides levels.

The PON gene family has 3 members, PON1, PON2, and PON3, that share 80% to 90% sequence identity at the amino acid level among mammalian species³². PON1 and PON3 are exclusively carried by circulating HDL^33–35, while PON2 is an intracellular protein that localizes in the membrane fraction of cells³⁶. Importantly, PON1 deficient mice and mice deficient in both PON1 and APOE developed atherosclerosis faster than control mice fed with atherogenic diet^{37, 38}. Overexpression of PON1 protected APOE-deficient mice from atherosclerosis^{39, 40}. Because PON1 is strongly anti-atherogenic in mouse models of hypercholesterolemia^{37, 40}, and the levels of PON1 activity in HDL associate negatively with CVD in multiple human studies^41–45, these observations suggest that albuminuria might promote pro-atherogenic changes in the HDL of T1DM subjects.

PON1 polymorphisms have been widely investigated for their possible involvement in atherosclerosis⁴⁶, and epidemiological studies indicate genetic associations between PON1 and the risk of CVD in some populations of subjects^{47, 48}. However, the role of PON1 genetic polymorphisms in the development of CVD remains controversial^{46, 49}. In contrast, levels of PON1 activity associate with CVD in non-diabetic populations in multiple studies^41–45. We found that PON1 mass as assessed by MS/MS correlated strongly with PON1 enzymatic activity in a second cohort of T1DM subjects enrolled in the CACTI study. In contrast, PON1 mass and activity did not correlate with macrophage cholesterol efflux capacity. A previous study of T1DM subjects suggested that low levels of PON3, but not PON1, associate with established atherosclerosis²¹. However, the small number of subjects in that study (28 cases and 28 controls) may have limited the ability to detect a significant association of PON1 with atherosclerosis. Together with our observations, these findings suggest that PON1 and/or PON3 concentration in HDL are more predictive of the risk of atherosclerotic coronary artery disease than are PON1 genotypes. It will be important to quantify PON1 mass and activity as well as other proposed metrics of HDL’s cardioprotective effects in future studies of incident CVD risk in T1DM subjects.

Previous studies demonstrated that APOE associates positively with CVD risk in females with high levels of HDL cholesterol and C-reactive protein but not in males⁵⁰. In an analysis adjusted for multiple comparisons, we also found that APOE associated with CAC in females but not in males in our cohort. However, that association lost significance after further adjustment for clinical variables, perhaps because there were relatively few females in the groups of albuminuric subjects. It is important to note that albuminuria is more common in males than females with T1DM⁵¹.

We used a logistic regression model to determine if any of the proteins whose HDL levels were abnormal in the T1DM subjects with albuminuria associated with prevalent coronary artery disease as assessed by CAC>0. After adjustment for clinical covariates that associate with coronary artery disease, only one protein, PON1, associated strongly and negatively both with CAC>0 and with log(AER). In contrast, PON3 associated strongly and negatively with calcification but not with log(AER). Importantly, these observations were independent of HDL-C, LDL-C, and triglycerides, demonstrating that alterations in HDL’s protein cargo can be dissociated from traditional lipid risk factors for atherosclerosis.

We also found that LCAT associated significantly with coronary artery calcification in the model that adjusted for multiple clinically relevant covariates. However, this association lost significance when we also controlled for LDL-C, HDL-C, and triglyceride levels. These observations suggest that the association of LCAT with coronary artery calcification reflects lipid abnormalities in T1DM subjects.

We previously demonstrated that AMBP and PTGDS are markedly elevated in HDL of ESRD patients on dialysis¹⁸. In the current study, AMBP and PTGDS strongly and positively correlated/associated with AER and strongly and negatively correlated/associated with eGFR, which is consistent with the observation that EDIC subjects albuminuria were more likely to exhibit a decrease in eGFR²⁴. Proteolytic processing of AMBP generates alpha-1-microglobulin and bikunin. Alpha-1-microglobulin is a member of the superfamily of lipocalin transport proteins that are implicated in the regulation of inflammatory pathways⁵². PTGDS is a glutathione-independent prostaglandin D synthase. It catalyzes the conversion of prostaglandin H2 to prostaglandin D2, which regulates smooth muscle contraction and platelet aggregation. Inflammation, altered smooth muscle contraction, and platelet aggregation are implicated in renal injury, raising the possibility that the HDL-associated AMBP and PTGDS contribute to the pathogenesis of renal disease. In future studies, it will be important to determine if elevated levels of AMBP and PTGDS in HDL predict incident albuminuria and/or progressive renal disease in T1DM subjects.

Previous studies have shown that elevated plasma levels of inflammatory makers predict an increased risk of progressive nephropathy in T1DM⁵³. Moreover, levels of the acute phase proteins SAA1 and SAA2 are elevated in HDL from subjects with ESRD^{18, 54, 55}, and elevated levels of SAA1/2 during inflammation impair HDL’s cholesterol efflux capacity in both humans and mice⁵⁶. However, we found that levels of SAA1 and SAA2 in HDL were similar in subjects with normoalbuminuria and albuminuria and that they failed to associate with coronary artery calcification, suggesting that inflammation-mediated changes in HDL’s protein cargo were not major contributors to altered HDL protein levels, kidney disease, or atherosclerosis in the T1DM subjects in our study.

Strengths of our study include a well-validated targeted proteomic approach to quantifying HDL proteins, the large number of T1DM subjects with varying degrees of albuminuria, stringent statistical analysis, and the similar clinical characteristics of the case and control subjects. A limitation is our study’s cross-sectional design, which could not reveal whether the associations of the HDL proteome with coronary artery calcification and albuminuria were causal. Our study also contained a relatively small number of females with albuminuria, which might have limited our ability to identify gender-dependent markers of CVD. It will therefore be critical to determine whether alterations to the HDL proteome predict the future onset of CVD and kidney disease in T1DM subjects.

In conclusion, we demonstrated that multiple HDL proteins are altered in T1DM subjects with albuminuria. We also found that two HDL proteins associate with prevalent CVD as assessed by the presence of coronary artery calcification and that this association was independent of lipid levels. However, only PON1 associated with both albuminuria and CVD, raising the possibility that increased CVD risk in T1DM subjects with albuminuria is mediated in part by alterations in HDL’s content of PON1. Our observations support the proposal that HDL’s protein cargo can serve as a marker—and perhaps mediator—of kidney disease and risk of atherosclerosis.

HIGHLIGHTS.

• The HDL proteome is remodeled in T1DM subjects with albuminuria.

• AMBP, PSAP and PTGDS in HDL all strongly and positively correlated with log(AER); they also strongly and negatively correlated with eGFR.

• PON1 and PON3 levels in HDL strongly and negatively associate with atherosclerosis in T1DM but only PON1 associate with both atherosclerosis and renal function.

• Alterations in HDL’s protein cargo might mediate in part the known association of albuminuria with cardiovascular risk in T1DM.

Abbreviations:

AER

albumin excretion rate

AMBP

α−1-microglobulin/bikunin precursor

APOA1

apolipoprotein A-I

CAC

coronary artery calcium

CVD

cardiovascular disease

DCCT

Diabetes Control and Complications Trial

DM

diabetes mellitus

EDIC

Epidemiology of Diabetes Interventions and Complications Study

eGFR

estimated glomerular filtration rate

ESRD

end-stage renal disease

HDL

high-density lipoprotein

HDL-C

HDL cholesterol

MS/MS

tandem mass spectrometric analysis

LDL

low-density lipoprotein

PON1

serum paraoxonase/arylesterase 1

PON3

serum paraoxonase/arylesterase 3

PONs

paraoxonases

PTGDS

prostaglandin-H2 D-isomerase

RAASi

renin-angiotensin-aldosterone system inhibitors

SRM

selective reaction monitoring

T1DM

type 1 diabetes mellitus