Proteomics of high-density lipoprotein subfractions and subclinical atherosclerosis in type 1 diabetes mellitus: a case–control study

Type 1 diabetes (T1D) represents 5% to 10% of all diabetes cases [1] with an overall cardiovascular disease (CVD) risk—mainly manifested by coronary heart disease, cerebrovascular and peripheral artery disease—increased up to eight times in comparison to people without diabetes [2]. Good glycemic control in individuals with T1D is associated with a lower risk of microvascular complications and cardiovascular disease. Glycated hemoglobin (HbA1c) has been used as a standard measure for long-term (two to three months) glucose control, and serum fructosamine can be used to assess the glycemic control over the past two to three weeks [3]. The pathophysiology underlying CVD in T1D is not well understood, but its prevalence is related to increased HbA1c levels, diabetes duration, age, sex, and possibly to race and ethnicity. Besides, abnormal vascular function with increased artery calcification, endothelial dysfunction, and cardiovascular autonomic neuropathy (CAN) increase the susceptibility to macrovascular atherosclerotic disease [4].

Plasma lipids and lipoproteins are not classically altered in T1D as usually observed in type 2 diabetes (T2D) [5]. Nonetheless, changes in lipoprotein functionality are considered as putative contributors to CVD risk. HDL cholesterol (HDLc) plasma levels that are inversely related to CVD risk are normal or even higher in T1D [4, 6]. In this sense, alterations in HDL particle functionality ascribed to chemical modifications by glycation, oxidation, and autoantibodies and in its lipidomics, and proteomics have been described as pro-atherosclerotic by impairing its antiatherogenic properties [6, 7]. HDL are heterogeneous particles varying from 7 to 10 nm with different subpopulations. Pre-beta nascent HDL and lipid poor apolipoprotein A-I (apoA-I) remove excess cholesterol from peripheral cells, including arterial macrophages, by interacting with ATP binding cassette transporter A-1 (ABCA-1) in the first step of the reverse cholesterol transport (RCT). After esterification by the lecithin cholesterol acyltransferase, larger HDL particles are formed also removing cell cholesterol via ATP binding cassette transporter G-1 (ABCG-1). Cholesterol is driven to the liver by the HDL subfraction 2 (HDL₂) that interacts with the scavenger receptor class B type 1 (SR-B1) or by the uptake of apoB-containing lipoproteins after receiving esterified cholesterol from HDL by the cholesteryl ester transfer protein (CETP). Cholesterol can be eliminated in the bile as free cholesterol, or after conversion into bile acids, which allows its excretion in feces [6, 8]. Besides its role along the RCT, HDL has antioxidant, anti-inflammatory, antiplatelet aggregation, and vasodilation actions that contribute to cardiovascular protection [8‐10]. HDL is a cargo lipoprotein for microRNAs, several active lipid species, and proteins that do not directly relate to atherosclerosis but can modulate HDL function and even be a marker of its functionality [9]. Proteomics evidenced around 251 proteins associated with HDL with functions related to lipid metabolism, activation of the complement system, modulation of proteases, immunogenicity, and others [11]. It is conceivable that alterations in HDL´s protein cargo may interfere with its antiatherogenic properties that cannot be seen by simply measuring HDLc or apoA-I [12]. Nonetheless, there are only a few studies dealing with the proteome of HDL in T1D, all of them dealing with total HDL fraction and mostly with subjects with DM categorized according to glycemic control [13‐15]. In the present investigation, the proteome of HDL subfractions (HDL₂ and HDL₃) was quantified by mass spectrometry using the parallel reaction monitoring (PRM) quantitative methodology. Subsequently, the proteome was related to HDL capacity in removing cell cholesterol, indicators of subclinical atherosclerosis—pulse wave velocity (PWV) and flow-mediated vasodilation (FMD) and ten-year atherosclerotic cardiovascular disease risk (ASCVDR) estimation in subjects with T1D in comparison to non-diabetes individuals.

This study aims to evaluate the proteomics of HDL subfractions in T1D subjects and controls and its association with clinical variables, subclinical atherosclerosis markers, and HDL functionality.

This was a case–control study that enrolled 50 T1D and 30 non-diabetes control individuals matched by age, gender, and body mass index (BMI). The convenience sampling method was used in the study for the recruitment of individuals with T1D. Snowball sampling was used to recruit controls. The inclusion criteria were based on age ≥ 18 years, T1D diagnosis ≥ 5 years. Individuals with a personal history of clinically evident atherosclerotic macrovascular disease (coronary disease, peripheral arterial disease, or cerebrovascular disease), active smoking, or those who stopped smoking more than ten years ago, and triglycerides (TG) concentration > 400 mg/dL were not included. Participants signed an informed written consent form previously approved by The Ethical Committee for Human Research Protocols of the Hospital das Clinicas da Faculdade de Medicina da Universidade de São Paulo (#3.796.622; 01/09/2020), in accordance with the Declaration of Helsinki. The reporting of this study conforms to STROBE guidelines [16].

Peripheral blood was drawn after overnight fasting and plasma was immediately separated in a refrigerated centrifuge (4 °C). Plasma lipids [TG, total cholesterol, (TC), and HDLc], and plasma glucose were determined by enzymatic techniques (Labtest do Brasil, Minas Gerais, Brazil). HDLc was determined after precipitation of apoB-containing lipoproteins with 0.2% dextran sulfate/3 M magnesium chloride (v/v). Low-density lipoprotein (LDL) cholesterol (LDLc) was calculated by the Friedewald formula [17] and the estimated glomerular filtration rate (eGFR), using the CKD-EPI equation [18]. Fructosamine was determined by an automated colorimetric enzymatic method (Labtest do Brasil, Minas Gerais, Brazil). HbA1c was determined by high-performance liquid chromatography, certified by the National Glyco Hemoglobin Standardization Program (NGSP-USA).

Clinical data from 50 T1D and 30 controls are presented in Table 1. Groups were similar regarding age, sex, BMI, abdominal circumference, HDLc, and eGFR. HbA1c and fructosamine levels were higher in T1D, and plasma TC, LDLc, and TG in controls. Retinopathy (48%) and CAN (30%) were the most prevalent complications in T1D. Forty percent of T1D used statins, and 15% had concomitant hypertension. Subjects with T1D and hypertension had longer duration of diabetes, and presence of albuminuria, retinopathy, and statins use. Glycemic control (HbA1c < 8.5% and ≥ 8.5%) was only statistically associated with albuminuria.

Forty-five proteins were selected and quantified in association with HDL₂ and HDL₃ of T1D and controls. From those, 18 were primarily associated with lipid metabolism [apo(a), apoA-I, apoA-II, apoA-IV, apoA-V, apoB, apoC-I, apoC-II, apoC-III, apoC-IV, apoD, apoE, apoF, apoM, CETP, LCAT, PCSK9, and PLTP], five related to acute inflammatory response (SAA1, SAA4, HP, HPHPR, and Orm1), one with complement system (C3), one antithrombotic (apoH), four antioxidants (APMAP, clusterin, PON1, and PON3), six transport proteins (ALB, GC, IGFALS, RBP4, TF, and transthyretin), five protease inhibitors (AHSG, AMBP, CST3, A1AT, and vitronectin), two enzymes not related with lipid metabolism (GPLD1 and PCYOX) and three with not well-known function (apoL-I, A1BG, and HBB) (Table 2). The differences between the abundance of proteins detected in HDL₂ between T1D and controls are shown in the Volcano Plot (Fig. 1A). Ten proteins (APMAP, apoB, apoC-I, apoC-II, apoE, apoF, apoM, C3, GPLD1 and SAA4) were more abundant in HDL₂ from T1D, and three (A1BG, apoC-III and HBB) in controls. The T1D was more associated with greater abundance of seven proteins (C3, apoE, APMAP, apoC-II, apoB, GPLD-1, and PLTP) (Fig. 1B). In the controls, the association was with four proteins (A1BG, transthyretin, HBB, and apoC-III) (Fig. 1C). Six proteins (APMAP, apoC-II, apoC-III, C3, HBB, and PLTP) were found to discriminate the HDL₂ proteomics between T1D and controls. Thirty-three proteins were differentially expressed in HDL₃ from T1D and controls (Fig. 1D), being 23 (APMAP, apoA-I, apoA-II, apoA-V, apoB, apoC-I, apoC-II, apoC-III, apoC-IV, apoD, apoE, apoL-I, apoM, CETP, GPLD1, HPHPR, PCSK9, PLTP, PON1, PON3, SAA1, SAA4, and TF) more abundant in T1D, and ten (A1BG, AHSG, ALB, apoF, apoH, CST3, GC, HBB, RBP4, and transthyretin) in controls. T1D was more associated with the abundance of 17 proteins (apoM, apoC-I, apoC-III, SAA4, apoC-II, HPHPR, PLTP, PCSK9, GPLD1, SAA4, apoB, TF, CETP, apoA-I, apoC-IV, LCAT, and APMAP) (Fig. 1E), and the controls with seven proteins (HBB, ALB, apoF, GC, A1BG, apoH, and CST3) (Fig. 1F). Nonetheless, only apoA-I, HBB, APMAP, LCAT, and PCSK9 had discriminatory capacity in HDL₃ proteomics between groups.

In HDL₂ of T1D, the amount of apoB and Lp(a) was negatively correlated, while PLTP, PON1, and A1AT were positively correlated with plasma HDLc. In controls, apoF presented a negative correlation, and apoA-I, GPLD1, PLTP, and PON1 had a positive association with HDLc. Only two proteins presented a similar correlation in both groups (PLTP and PON1). For HDL₃, the abundance of three proteins (C3, IGFALS, and PON1) was positively correlated with HDLc of T1D. In controls, three proteins in HDL₃ (AMBP, apoF and apoH) were negatively correlated with HDLc; while five proteins (C3, clusterin, IGFALS, LCAT, and PON1) were positively correlated. Only two proteins presented a similar correlation in both groups (C3 and IGFALS). PON1 correlated well in controls, but not in T1D (data not shown).

APMAP, apoB, apoC-I, C3, and SAA4 in HDL₂ was greater in T1D with HbA1c ≥ 8.5%, while IGFALS was less abundant in those subjects. In HDL₃, 12 proteins (APMAP, apoA-I, apoA-II, apoA-V, apoC-I, apoC-II, apoC-III, apoD, apoM, HPHPR, PCSK9, and SAA4) were more abundant in individuals with HbA1c ≥ 8.5%, while the content of eight proteins (A1AT, A1BG, Alb, apoF, GC, HBB, RBP4, and transthyretin) was greater in T1D with HbA1c < 8.5% (Additional file 1: Table S1). Regarding the categorization by statin use, a higher abundance of C3, HBB and apo(a), and a lower abundance of apoM, IGFALS, and PON3 in HDL₃ was observed (Additional file 1: Table S1).

The presence of CAN, assessed in T1D, was associated with a higher abundance of AMBP, ApoB, Lp(a), Orm1, and transthyretin in HDL₂, and lower content of clusterin, HBB, and PON3 in HDL₃ (data not shown). A positive correlation was observed between A1BG content in HDL₂ with ten-year ASCVDR assessed by the QRISK3; AMBP was positively, and apoE in HDL₂ negatively correlated with both QRISK3 and Steno T1 Engine (Additional file 1: Table S1). Also, the ten-year ASCVDR by Steno T1 Engine showed a positive correlation with AHSG, AMBP, and apoH in HDL₃, and a negative correlation with apoC-I, apoE, apoL-I, and HPHPR. The ten-year ASCVDR calculated by QRISK3 presented a positive correlation with eight proteins in HDL₃ (A1BG, AHSG, ALB, AMBP, apoA-IV, apoH, RBP4, and transthyretin) and negative with apoC-I, apoE, apoL-I, and apoM (data not shown).

Thirty individuals in T1D and 30 in controls underwent PWV and post-reactive hyperemia FMD tests. T1D presented higher PWV than controls. In the assessment of endothelial function, the basal diameter of the brachial artery was similar, but FMD after reactive hyperemia was lower in the T1D than controls (Table 1). In both groups, PWV, but not FMD, correlated with ten-year ASCVDR by Steno T1 Engine Risk and QRISK3. CAN and HDLc were not associated with changes in vascular function.

In T1D, six proteins in HDL₂ (apoL-I, A1BG, apoA-II, apoB, apo(a),and SAA4) positively correlated with PWV while PCSK9 and clusterin presented an inverse association. Post-RH FMD was positively correlated with apoA-I and PCYOX. In the controls, HBB and Orm1 showed a negative correlation with PWV, and post-RH FMD was not related to any protein associated with HDL₂. After nitrate, A1AT, apoA-IV, apoF, clusterin, PCSK9, and vitronectin negatively correlated with FMD, while IGFALS directly related to FMD in T1D. In the controls, a negative association was observed between HBB and Orm1 with PWV, with no association of any protein with FMD (Table 3).

In HDL₃ of T1D, nine proteins (clusterin, A1AT, apoA-I, apoE, apoL-I, complement C3, HBB, IGFALS, LCAT, and PON1) were inversely related to PWV. FMD after RH and nitrate were inversely correlated with GPLD1. In the controls, apoC-IV was negative, while CETP and RBP4 positively correlated with PWV. Regarding post-RH FMD, a negative association was observed with CETP (Table 3). CAN was associated with the higher abundance of AMBP, apoB, Lp(a), Orm, and transthyretin in HDL₂, and lower content of clusterin, HBB, and PON3 in HDL₃. In HDL₂, A1BG presented a positive correlation with QRISK3; while and AMBP and apoE, respectively, had a positive and a negative correlation with QRISK3 and Steno 1 Engine. Regarding HDL₃, AHSG, AMBP, and apoH positively correlated, while apoC-I, apoE, apoL-I e HPHPR, negatively correlated with Steno T1 Engine. For QRISK3, the positive association was observed with A1BG, AHSG, ALB, AMBP, apoA-IV, apoH, RBP4, and transthyretin) and a negative correlation with apoC-I, apoE, apoL-I, and apoM.

The percentage of 6-h cholesterol efflux from macrophages was similar between T1D [HDL₂: 24.1% (18.0–30.2); HDL₃: 18.8% (14.8–24.4)] and controls [HDL₂: 22.1% (18.5–25.7); HDL₃: 18.5% (16.1–22.6)]. It was not observed association between clinical variables with cholesterol efflux capacity, except for an inverse correlation between the percentage of HDL₂-mediated efflux with albuminuria in T1D (r = − 0.339; p = 0.02). Three proteins in HDL₃ of T1D negatively correlated with cholesterol efflux: A1AT (r = − 0.33; p = 0.03), GC (r = − 0.32; p = 0.03), and TF (r = − 0.31; p = 0.04). In HDL₃ of controls, eight were negatively correlated with cell cholesterol efflux: A1BG (r = − 0.41; p = 0.03), AHSG (r = − 0.48; p = 0.01), apoF (r = − 0.46; p = 0.01), apoH (r = − 0.46; p = 0.01), CST3 (r = − 0.38; p = 0.04), GC (r = − 0.46; p = 0.01), RBP4 (r = − 0.434; p = 0.02), and transthyretin (r = − 0.60; p = 0.01). Only RBP4 in HDL₂ of controls positively correlated with cell cholesterol removal (r = 0.46; p = 0.02).

Subclinical atherosclerosis is more frequently observed in T1D subjects including increased intima-media thickness, artery calcification, and abnormal vascular function, namely PWV and FMD [32‐35]. Nonetheless, the mechanisms involved in the evolution to clinical established CVD and possible markers are not well known. HDL protein signature may provide new insights on the modulation of HDL functionality that may not be evident by the classical metrics of this lipoprotein in plasma. In the present investigation, it was demonstrated a remodeling in HDL proteomics in T1D as compared to controls which related to markers of subclinical atherosclerosis.

T1D presented similar concentrations of plasma HDLc and lower levels of TG, LDLc, and TC as compared to controls. On the other hand, subclinical atherosclerosis markers, such as greater PWV and reduced FMD, were greater in T1D. Most studies that evaluated arterial stiffness in DM were conducted in T2D and those with T1D were carried out mainly in children, with different results and assessment methods [14, 34, 36]. Increased arterial stiffness observed in T1D corroborates other studies carried out with individuals of similar ages that also showed an association of higher PWV with inflammatory biomarkers in T1D [37]. The concomitance of other cardiovascular risk factors, such as hypertension and dyslipidemia, was associated with greater arterial stiffness (PWV), but not with endothelial dysfunction (FMD). Interestingly, glycemic control was not associated with abnormal vascular function, although it is noted to mention that the current investigation is a cross-sectional study, and HbA1c levels represent the mean glycemic rate for the last three months. Both the increased values of HbA1c and serum fructosamine in individuals with T1D suggest that there were no acute changes in glycemic control. CAN that may affect up to 30% T1D individuals [38, 39] was not associated with vascular test results, although Liatis et al. [40] demonstrated that cardiovascular autonomic function, especially parasympathetic activity, is related to arterial stiffness in individuals with T1D. The two cardiovascular risk assessment tools validated in T1D—Steno T1 Engine and QRISK3—indicated an association between increased PWV and reduced FMD with higher ten- year ASCVDR [37].

HDLc were not associated with vascular dysfunction in T1D reinforcing the hypothesis that HDLc may not represent the best prediction for CVD in DM. The proteome analysis revealed differences in abundance of 13 peptides in HDL₂, and 33 in HDL₃ between the T1D and controls. A study demonstrated a distinct profile of proteins carried by total HDL fraction in T1D as compared to controls, including apoA-IV, apoE, and apoD, fibrinogen, albumin, and others, together with irreversible post-translational modifications such as oxidation, deamidation, and glycation. Those changes were related to a reduced ability of serum from subjects with T1D in removing cholesterol from macrophages as well as a diminished antioxidant capacity of HDL from subjects with T1D, independently of the glycemic control and HDLc [13]. In another case–control study, a labeled-free SWATH peptide quantification revealed 78 proteins bound to HDL that were differentially expressed in youth people with T1D in comparison to controls although no association with HDL functionality (³H-cholesterol efflux analysis) was made [14]. In a subgroup of subjects enrolled in Diabetes Control and Complications Trial/Epidemiology of Diabetes Intervention and Complications Study (DCCT/EDIC), 46 proteins were quantified in HDL by isotope dilution tandem-mass spectrometry, and after stringent analysis, eight proteins were associated with albuminuria. Particularly, PON1 in HDL was inversely correlated with coronary calcium score, and positively with albumin excretion rate [41]. Nine proteins in HDL₂, and 22 in HDL₃ whose difference in abundance were observed in the present investigation were not reported in the two case–control studies described above. The AMBP, albumin, and apoA-II were in agreement with at least one other study. ApoA-IV, apoH, A1BG, and C3 were also found in at least one of the two studies and had a significant difference between the groups, but the result was not in agreement regarding the group with greater abundance [14, 15].

Many proteins directly involved in lipid metabolism presented a greater content in HDL subfractions in T1D and may represent interplay of metabolic pathways that modulate lipid metabolism in T1D. Noteworthy that the abundance of PLTP, apoA-I, LCAT and PCSK9 had a discriminatory power between T1D and controls. PLTP is responsible for HDL remodeling and elevated PLTP activity is reported in T1D being considered as a contributor for enhancing apoA-I in fused HDL particles. ApoA-I mediates many atheroprotective actions of HDL, including RCT, antioxidative and anti-inflammatory activities, although increased modification of apoA-I by glycation, oxidation and others impairs its functionality contributing to HDL loss of function in diabetes [42]. The LCAT activity in T1D is reported as increased according to hyperglycemia [43] or unchanged [44] and its role in atherogenesis is controversial. Shao et al. [41] demonstrated that LCAT was inversely related to coronary calcium score in T1D individuals which agrees with the negative correlation observed between this enzyme with PWV observed in the present investigation. Interestingly, PCSK9 was more abundant in HDL₃ of T1D especially in those with HbA1c values ≥ 8.5% and its amount in HDL₂ was associated with endothelial dysfunction. Plasma levels of PCSK9 were found increased in T1D subjects, which correlated with glycemic control [45]. Moreover, PCSK9 was associated with smaller HDL particles in T1D individuals with a poor glycemic control, although its role in HDL is not well established [46]. Other proteins related to lipid metabolism abundant in HDL subfractions of T1D included apoA-II, apoC-III, apoA-V, and apoD apoC-I, and apoC-II, that correlated with glycemic control, and apoA-IV, inversely correlated with PWV. Interestingly, apoC-III was greatly expressed in HDL₂ of controls and HDL₃ of T1D subjects. This apolipoprotein reflects insulin sensitivity and negatively modulates LPL activity driving plasma levels of triglycerides and being a target of drugs for hypertriglyceridemia control. Moreover, it drives VLDL secretion and relates to inflammatory processes in the vasculature and in the pancreas. In the present investigation, the differential expression of apoC-III in HDL was not correlated to HDL functionality (¹⁴C-cholesterol efflux analysis) or vascular tests, which deserves further exploration.

Acute inflammatory proteins such as SAA1, SAA4 that presented a discriminatory power between both groups were related by others to the impairment in RCT and anti-inflammatory activity of HDL [6, 47, 48]. Complement C3 in HDL₃ that in the present study was inversely related to PWV, was described as enhanced in the HDL proteome of subjects with established CVD [49] and elevated in the HDL proteome in T1D (11). PON1 and PON3 were higher in HDL₃ being PON1 inversely related to PWV. In the DCCT/ EDIC those enzymes were inversely related to calcium coronary score [43].

Curiously, the hemoglobin subunit beta (HBB) in both HDL₂ and HDL₃ had a discriminatory capacity for T1D and controls, being HBB reduced in T1D and with a negative correlation with PWV. APMAP was also discriminatory but more abundant in T1D with worse glycemic control. Finally, the clusterin content in HDL₂ and HDL₃ was positive and negative correlated with, respectively, PWV and FMD. Moreover, clusterin was negatively correlated with CAN, agreeing with its role in preventing familial amyloidotic polyneuropathy [50].

The PRM analysis is a very sensitive methodology making detectable the presence of very small amounts of proteins such as apoB and apo(a) in HDL, as the hydrated density of large HDL₂ is similar to the densities of lipoprotein (a) and LDL [27].

Reduced cholesterol efflux was shown in T1D [13, 14], although Vaisar et al. [49] did not find a difference in cholesterol removal comparing subjects with T1D with and without vascular complication. One study showed that improved glycemic control is associated with increased cholesterol efflux in T1D [51]. In this investigation, both HDL₂ and HDL₃-mediated cholesterol efflux was similar between T1D and controls and not related to vascular function tests.

This study detected differences in HDL proteomics between individuals with T1D and controls that had not been found in previous studies. Nine proteins in HDL₂, and 22 in HDL₃ were detected, whose observed abundance differences were not reported in other case–control studies. Vascular tests corroborate previous studies in individuals with T1D that indicated worse markers of subclinical atherosclerosis and higher estimated cardiovascular risk than individuals without diabetes, even with an apparently better serum lipid profile in individuals with T1D. The association of clinical and laboratory variables, HDL proteomics data and results of vascular function tests and HDL functionality made it possible to explore the various atheroprotective functions of HDL. In addition to proteins involved in lipid metabolism, HDL carries acute-phase inflammatory, complement system, antithrombotic, antioxidant, protease inhibitor and transporter proteins, with significant differences between T1D and controls.