Paradoxical coronary artery disease in humans with hyperalphalipoproteinemia is associated with distinct differences in the HDL phosphosphingolipidome

Abstract

Background

Plasma HDL-C levels are inversely associated with risk of coronary artery disease (CAD) in epidemiological studies. Despite this, the directionality of this relationship and the underlying biology behind it remain to be firmly established, especially at the extremes of HDL-C levels.

Objective

We investigated differences in the HDL phosphosphingolipidome in a rare population of subjects with premature CAD despite high HDL-C levels to gain insight into the association between the HDL lipidome and CAD disease status in this unusual phenotype. We sought to assess differences in HDL composition that are associated with CAD in subjects with HDL-C > 90^th percentile. We predicted that quantitative lipidomic analysis of HDL particles would reveal novel differences between CAD patients and healthy subjects with matched HDL-C levels.

Methods

We collected plasma samples from 25 subjects with HDL-C >90^th percentile and clinically manifest CAD and healthy controls with HDL-C >90^th percentile and without self-reported CAD. Over 140 individual HDL phospho- and sphingolipid species were analyzed by LC/MS/MS.

Results

Significant reductions in HDL phosphatidylcholine (−2.41%, q-value = 0.025) and phosphatidylinositol (−10.7%, q-value = 0.047) content, as well as elevated sphingomyelin (+10.0%, q-value = 0.025) content and sphingomyelin/phosphatidylcholine ratio (+12.8%, p-value = 0.005) were associated with CAD status in subjects with high HDL-C.

Conclusions

These differences may lay the groundwork for further analysis of the relationship between the HDL lipidome and disease states, as well as for the development of biomarkers of CAD status and HDL function.

Introduction

Circulating high-density lipoprotein cholesterol (HDL-C) level, along with a variety of other “traditional” lipid risk factors such as low-density lipoprotein cholesterol (LDL-C) and triglyceride (TG) levels, have been used to stratify patients into risk categories for cardiovascular disease, yet clinicians would benefit from more sensitive biomarkers with better predictive value.

Though the epidemiological inverse correlation between HDL-C level and incident of coronary artery disease (CAD) is well established, (1–4) Mendelian randomization studies (5, 6) as well as outcome studies using HDL-C raising interventions (7–9) suggest that this relationship is not necessarily causal. Moreover, a recent study found that subjects with a rare loss of function mutation in the gene coding for the HDL receptor SR-BI have an increased risk for CAD despite profoundly elevated HDL-C levels. (10)

In an effort to explore the biological connection between HDL and CAD risk, much attention has been given to the functional properties of HDL, including its role in reverse cholesterol transport (RCT) and vasoprotection, as opposed to HDL-C level. In this context, we and others have shown that cholesterol efflux capacity constitutes a robust predictor of CAD status even after adjustment for HDL-C levels. (11–13) Of note, a reduced cholesterol efflux capacity is predictive of CAD status in patients with elevated HDL-C levels, as shown in a recent study published by our group. (14) Interestingly, we also observed significantly reduced HDL phospholipid content in patients with elevated HDL-C levels and CAD, as compared with subjects with similar HDL-C levels, but without clinically evident CAD. (14) The question remains as to whether differences in HDL phospholipid composition accompany the decrease in HDL phospholipid content of HDL in these unique patients.

Despite a growing number of publications addressing the human plasma lipidome(15) (16, 17) relatively little has been published regarding the HDL-specific lipidome.(18, 19) Recent publications by our group and others have demonstrated that lipidomic analysis of HDL has the potential to yield information about structure-function relationships, and to identify disease related biomarkers and profiles.(20) Differences in the HDL lipidome have been documented in Type 2 diabetes-associated dyslipidemia,(21, 22) rheumatoid arthritis,(23) familial apoA-I deficiency,(24) myocardial infarction,(25) and in healthy subjects with high versus low HDL-C levels.(26) HDL phospho- and sphingolipids specifically, which constitute 40–60% of the total HDL lipid, are of particular interest, given that they play an active role in many of HDL’s biological functions. (18, 20)

Furthermore, the identification of analytical biomarkers of multiple HDL functions that can serve as readily measurable surrogates for functional assays is particularly attractive, as opposed to methods that directly assess selective HDL functions. Highly quantitative mass-spectrometry (MS) based approaches including lipidomics and proteomics have the potential to reveal such biomarkers. In addition to being associated with disease status, specific changes in the HDL proteome and lipidome have been associated with HDL functional properties and thus may serve as a proxy thereof. (15, 20, 25, 27, 28)

To further characterize the unique combined phenotype of elevated HDL-C levels and CAD and explore if the quantitative differences in HDL-phospholipid content observed in the study by Agarwala et al.(14) are accompanied by changes to individual phospholipid subclasses, we utilized MS based lipidomics. In the current study, we report alterations in the HDL phosphosphingolipidome observed in a subgroup of patients with the rare combined phenotype of HDL-C level > 90^th percentile and presence of clinically manifest premature CAD (heretofore referred to as “HCAD”) compared to matched healthy subjects with similar HDL-C levels and no CAD (heretofore referred to as “HHDL”).

Materials and Methods

Subjects

Plasma was collected from subjects at the University of Pennsylvania that were enrolled into one of 2 groups: a group of subjects with HDL-C >90^th percentile and clinically manifest CAD (“HCAD”, n=25) and a group of healthy subjects with HDL-C >90^th percentile, but not self-reported CAD (“HHDL”, n=25). The groups were matched by several clinical and biological parameters, including HDL-C levels, age, sex, and race. CAD was defined as a history of myocardial infarction or revascularization, documented coronary stenosis as evaluated by coronary angiogram, or presence of coronary calcium score greater than the 75^th percentile for age and sex. The age range for inclusion was 18–65 years old for male subjects and 18–85 years old for female subjects, reflecting the age of subjects with the rare HCAD phenotype at the time of recruitment. Exclusion criteria included: drugs known to affect HDL-C levels, diagnosis of cancer within the past five years, end-stage renal disease or severe renal insufficiency, any major surgery within the past 6 months, any organ transplant, and pregnancy. A third group of healthy subjects without evidence of clinically manifest CAD and HDL-C levels between the 25^th–75^th percentile was utilized as reference group (“NHDL”, n=10). The University of Pennsylvania Institutional Review Board approved the study, and all subjects gave written informed consent to participate.

Blood Samples

Blood samples were withdrawn after overnight fasting into sterile, evacuated tubes (Vacutainer) containing EDTA. Plasma was immediately separated by low-speed centrifugation at 4°C, aliquoted and frozen at −80°C until further analysis.

Standard biochemical parameters

Plasma levels of total cholesterol (TC), triglyceride (TG) and HDL-C were measured using commercially available enzymatic kits. Low-density lipoprotein cholesterol (LDL-C) was directly measured following preparatory ultracentrifugation. Plasma apolipoprotein (apo) A-I, apoB, and Lp(a) were quantitated by immunoturbidimetry.

Fractionation of lipoproteins

HDL was isolated from EDTA plasma by single step, isopycnic non-denaturing density gradient ultracentrifugation based on a modification of the method developed by Chapman et al. (29) as previously described (30). Using this procedure, VLDL and LDL (d 1.019–1.063 g/ml) were removed, followed by isolation of HDL (d 1.063–1.180 g/ml). Lipoproteins were extensively dialysed against phosphate-buffered saline (PBS; pH 7.4) at 4°C in the dark and stored at 4°C.

HDL chemical composition

The chemical composition of HDL was assessed using commercially available enzymatic kits (total protein (TP): Thermo Scientific, Villebon-sur-Yvette, France; total cholesterol, free cholesterol (FC), phospholipids (PL): DiaSys, Holzheim, Germany; TG: Biomérieux, Marcy l’Etoile, France). Cholesteryl ester (CE) was calculated using the formula CE = (TC−FC)*1.67.

HDL Phosphoshingolipidome analysis by mass spectrometry

Quantification of the HDL phosphospingolipidome was accomplished using an original method published earlier by Camont et al. (20) In brief, HDL (30μg total phospholipid mass determined using the commercially available assay as above) was added to 4ml of cold CHCl₃/acidified CH₃OH (5:2 v/v) containing seven internal lipid standards (Avanti Polar Lipids, Alabaster, AL, USA). A blank and quality controls (one every 5 samples) were extracted in parallel with each batch to ensure for quality control. K₄EDTA (200mM) solution was added (1:5 v/v) and the mixture was vortexed and centrifuged. The organic phase was dried under nitrogen and lipids were reconstituted into isopropanol/hexane/water (10:5:2 v/v), transferred into LC/MS vials, dried under nitrogen and resuspended in isopropanol/hexane/water (10:5:2 v/v). Eight principal phospholipid (PL) subclasses (Phosphatidylcholine (PC), lysophosphatidylcholine (LPC), phosphatidylethanolamine (PE), lysophosphatidylethanolamine (LPE), phosphatidylinositol (PI), phosphatidylglycerol (PG), phosphatidylserine (PS), and phosphatidic acid (PA)) and two principal sphingolipid (SL) subclasses (sphingomyelin (SM) and ceramide (Cer)), which together comprise 162 individual molecular lipid species and account for >95% of total plasma PL and SM (31, 32) were assayed by LC/MS/MS. Lipids were quantified using a QTrap 4000 mass spectrometer (AB Sciex, Framingham, MA, USA), an LC20AD HPLC system, and the Analyst 1.5 data acquisition system (AB Sciex, Framingham, MA, USA). Quantification of PLs and SLs was performed in positive-ion mode, except for PI species that were detected in negative-ion mode (20). Sample (4μl) was injected onto a Symmetry Shield RP8 3.5μm 2.1×50mm reverse phase column (Waters Corporation, Milford, MA, USA) using a gradient from 85:15 to 91:9 (v/v) methanol/water containing 5mM ammonium formate and 0.1% formic acid at a flow rate of 0.1ml/min for 30 mins. Lipid species were detected using multiple reaction monitoring reflecting the headgroup fragmentation of each lipid class and quantified using calibration curves specific for the ten individual lipid classes with up to 12 component fatty acid moieties; 17 calibration curves were generated in non-diluted and 10-fold diluted matrices to correct for matrix-induced ion suppression effects. More abundant lipid species which displayed a non-linear response in non-diluted extracts were quantified from a 10- or 100-fold diluted sample. An in-house developed script was used to compile data from the three successive injections. Abundance data for individual species and classes is presented as percentage of total PL and SL, rather than mg/dl. This normalization step is necessary to correct for variations in extraction efficiency between samples.

Statistical analyses

Statistical analysis was performed using Statistica 6 software (StatSoft, USA) and R, a freely available language (33). By design, primary analysis focused on the comparison between the HCAD and HHDL groups, with NHDL group provided as a reference population only. Therefore, pairwise comparisons were conducted between the HCAD and HHDL group. Discontinuous variables were analysed using Fisher’s exact test. Data distributions were assessed for normality within subject groups using a Shapiro-Wilk test. Statistical outliers were identified using the Grubb’s test. Variables that were normally distributed were compared using a Student’s t-test, while variables that deviated significantly from a normal distribution, as well as all phosphosphingolipid species, were compared using a non-parametric Mann-Whitney test. For lipidomic data, Benjamini-Hochberg methodology was employed to control the false discovery rate (FDR) to adjust for multiple testing, since it has been argued that FDR analysis is more appropriate and robust in -omics analyses than the more stringent Bonferroni correction.(15) (34, 35) Q-values were determined using the R package “qvalue,” and a significance threshold of 0.05 was used.

Results

Clinical and biological parameters

Clinical characteristics are shown in Table 1. The HCAD and HHDL groups did not differ significantly in age, sex or racial distribution. Median age of cardiovascular event in the HCAD group was 57 (43–69) years for men and 60 (25–82) years for women. HDL-C, apoA-I, and TG levels were not significantly different between the HCAD and HHDL group. Plasma concentrations of TC, LDL-C, VLDL-C and apoB were significantly reduced in the HCAD group, likely due to the higher use of statins (76%) and other lipid lowering drugs (44%) than in the HHDL group (8% and 0% respectively). Lp(a) levels were not different between the HCAD and HHDL groups. As expected, both apoA-I and HDL-C levels were markedly and significantly higher in the HCAD (1.5-fold for apoA-I, 2.1-fold for HDL-C) and HHDL (1.5-fold for apoA-I, 2.4-fold for HDL-C) than the NHDL group, and TG levels were significantly lower.

Table 1.

Clinical parameters, plasma lipid levels, and treatment. Continuous variables are expressed as median (range).

	HCAD (N=25)	HHDL (N=25)	NHDL (N=10)
Age (years)	61 (28 – 82)	63 (29 – 75)	64 (47 – 74)
Sex (M/F)	8/17	4/21	4/6
Race (W/B)	24/1	24/1	9/1
BMI (kg/m2)	23.8 (19.3 – 35.1)	23.0 (16.9 – 38.5)*	25.9 (20.1 – 29.9)
TC (mg/dl)	183 (113 – 297) ###	224 (164 – 360)*	194 (127 – 291)
HDL-C (mg/dl)	96 (63 – 146)***	110 (60 – 178)***	46 (39 – 65)
LDL-C (mg/dl)	73 (37 – 159) ###**	109 (72 – 226)	120 (72 – 178)
VLDL - C (mg/dl)	12 (1 – 33)*	15 (6 – 41)	21 (10 – 48)
TG (mg/dl)	58 (37 – 128)**	67 (12 – 128)*	97 (54 – 162)
ApoA-I (mg/dl)	197 (151 – 324)***	200 (179 – 295)***	136 (121 – 168)
ApoB (mg/dl)	63 (43 – 104) ##**	81 (57 – 113)	95 (57 – 123)
Lp(a) (mg/dl)	24.4 (0 – 224)	13.9 (0 – 146)	9.3 (0 – 67.8)
Statin treatment	19/25###*	2/25	3/10
Other lipid treatment	11/25###*	0/25	0/10
Hypertension	18/25###*	5/25	3/10
treatment
Diabetes treatment	3/25	1/25	1/10
Aspirin	20/25###***	5/25	1/10

^##p<0.01;

^###p<0.001 compared to HHDL subjects

^*p<0.05;

^**p<0.01;

^***p<0.001 compared to NHDL subjects

W=White; B=Black; TC = Total Cholesterol; TG = Triglyceride; ApoA-I, apolipoprotein A-I; ApoB, apolipoprotein B

HDL chemical composition

No significant differences in the chemical composition of the HDL fraction as a whole were observed between the HCAD and HHDL groups when expressed as percentage of total HDL mass (Table 2). As expected, differences were observed between the HCAD and HHDL groups and the NHDL reference group. Compared with the NHDL group, total HDL mass was increased 1.5-fold and 1.7-fold in the HCAD and HHDL groups respectively; TP content was decreased by approximately 15% in both groups); CE and PL content were increased by approximately 20% in both groups; TG content was markedly reduced (by 26% in the HCAD and by 33% in the HHDL groups, respectively. The ratio of HDL TC isolated by ultracentrifugation to HDL TC measured directly in apoB-precipitated serum, a metric of HDL recovery, was 1.086 ± 0.078, indicative of an adequate recovery. Chemical composition data from three subjects from the HCAD and one subject from the HHDL group were identified as statistical outliers by Grubbs’ test, likely reflecting error during measurement, and were removed from this analysis.

Table 2.

Chemical composition of HDL. Values expressed as mean ± SD.

	HCAD (n=24)	HHDL (n=22)	NHDL (n=10)
Total mass (mg/dl)	519.7 ± 103.3***	562.2 ± 103.0***	337.3 ± 48.5
TP (wt %)	43.3 ± 3.6 ***	42.6 ± 2.39***	50.7 ± 3.4
CE (wt %)	21.2 ± 3.6 **	21.6 ± 1.79 ***	17.6 ± 2.34
FC (wt %)	3.5 ± 0.7	3.7 ± 0.60	3.6 ± 2.14
PL (wt %)	29.4 ± 2.7***	29.8 ± 1.84***	24.7 ± 3.34
TG (wt %)	2.6 ± 0.9	2.3 ± 1.0**	3.5 ± 1.24

^*p<0.05;

^**p<0.01;

^***p<0.001 compared to NHDL subjects. Chemical composition data from three subjects from the HCAD and one subject from the HHDL group were identified as statistical outliers by Grubbs’ test and removed from this analysis. TP, Total protein; CE, Cholesterol ester; FC, Free cholesterol; PL, Phospholipid; TG, Triglyceride

HDL phosphosphingolipidome

We assessed ten phospho- and sphingolipid classes across groups. Significant differences were observed between the HCAD and HHDL groups (Figure 1, Online Table 1). In the HCAD group, a reduction in HDL-PC content (−2.41%, q-value = 0.025) and an increase in HDL-SM content (+10.0%, q-value = 0.025) were observed (Figure 1A), leading to an increased SM/PC ratio (+12.5%, p-value = 0.0048, Figure 1E). HDL-PI, a minor, negatively charged phospholipid, was also decreased in the HCAD group (−10.7%, q-value = 0.047). (Figure 1B, Online Table 1). Similar differences were noted when the HCAD group was compared with the reference NHDL group.

Figure 1. HDL phospho- and sphingolipid composition.

Lipid classes are expressed as the percentage of total HDL phosphosphingolipid content. Error bars represent standard deviation. * denotes q-value < 0.05, ** denotes q-value < 0.01. PC, Phosphatidylcholine; LPC, lysophosphatidylcholine; PE, phosphatidylethanolamine; LPE, lysophosphatidylethanolamine; PI, phosphatidylinositol; PG, phosphatidylglycerol; PS, phosphatidylserine; PA, phosphatidic acid; SM, sphingomyelin; Cer, ceramide.

Despite efforts to match the HCAD and HHDL groups as closely as possible, some difference remained. We therefore performed additional analyses to exclude the possibility that some of the observed differences were due to these differences. The number of males was small and imbalanced between the HCAD and HHDL groups. To assess whether any of the differences in phospholipid classes were related to this imbalance, we conducted a subgroup analysis excluding male subjects. The differences in PC, SM, and PI remained significant when only female subjects in the HCAD (n=17) and HHDL(n=21) were analyzed (PC: −2.56%, p-value = 0.012; SM: +12.4%, p-value = 0.0073; PI: −11.7%, p-value = 0.0326). Additionally, to exclude the possibility that the changes observed were due to the presence in the HCAD group to subjects older than 75 years old, we repeated the analysis excluding these subjects. The differences in PC, SM, and PI between the HCAD (n=21) and HHDL (n=25) groups remained significant (PC: −2.42%, p-value = 0.011; SM: +9.07%, p-value = 0.031; PI: −11.3%, p-value = 0.032). Finally, as there was an imbalance in the prevalence of statins and other lipid lowering treatments in the HCAD group compared with the HHDL group, we repeated the analysis excluding subjects on statin treatment. Importantly, the trends in PC, SM, and PI content persisted in this subgroup analysis, however the number of subjects in this analysis was too small for these trends to reach statistical significance (Online Table 2).

In order to more deeply interrogate potential differences within these classes, we assessed 55 individual molecular species of PC, SM, and PI (Online Table 3). Of these, 6 species of PC, 11 species of SM, and 5 species of PI were found to be significantly different between the HCAD and HHDL groups after adjustment for multiple testing (Figure 2). In general, PI species tended to be decreased and SM species tended to be elevated in the HCAD group, while the distribution of PC species was more variable. PC species with short to mid chain length and a moderate degree of unsaturation (PC 34:2, PC 36:2, PC 36:3) tended to be reduced in the HCAD subjects, while for the most part PC species with longer fatty acid chain length and higher degree of unsaturation (PC 36:4, PC 38:4) tended to be increased in the HCAD subjects (Figure 3). Interestingly, some of these species showed differences also when the HCAD group was compared with the NHDL group. An unbiased approach analyzing differences across all PL and SL species was also conducted but did not reveal any additional differences.

Figure 2. Changes in phospho- and sphingolipid classes and species in subjects with HCAD compared to healthy HHDL subjects.

Percent difference for lipid classes (PI, PC, SM) are shown as bars. Individual lipid species within each class are shown as circles whose areas are proportional to the percent abundance in the overall class. Grey shading denotes a significant q-value. A list of all species included in this figure is provided in Online Table 3. PI, phosphatidylinositol; PC, phosphatidylcholine; SM, sphingomyelin

Figure 3. Percent differences between HCAD and HHDL groups in PI, PC, and SM.

Lipid species are arranged by fatty acid tail length and degree of unsaturation. The color of dots represents the percent difference in a given phospholipid species between the HCAD group relative to the HHDL group. * denotes q-value < 0.05, ** denotes q-value < 0.01, *** denotes q-value < 0.001. PI, phosphatidylinositol; PC, phosphatidylcholine; SM, sphingomyelin

Both the HCAD and the HHDL showed significant decrease in HDL-PS and PG content and a trend toward an increase in HDL-LPC compared to the reference, NHDL group (Figure 1B–D). In order to allow a direct comparison of our data to the previously published work by Yetukuri et al.,(26) we compared the HDL-LPC concentration normalized by plasma apoA-I concentration in all subjects with high HDL-C to the NHDL reference group (Online Table 4). Our data confirmed the published findings that HDL LPC is increased in high HDL-C subjects compared to subjects with lower HDL-C levels (+ 39.5%, p-value = 0.012). Interestingly, after normalization SM was elevated and PS was reduced in all high HDL-C subjects vs. the NHDL group (Online Table 4).

Discussion

We observed marked differences in two HDL-phospholipid classes, PC and PI, as well as in SM, the major sphingolipid class in HDL, between high HDL subjects with and without clinically manifest CAD. Within each of these classes, a distinct profile of individual molecular species was observed between the HCAD and HHDL groups. These different profiles may provide biomarkers to distinguish subjects with residual risk of CAD despite having favorable HDL-C levels.

Using a highly quantitative LCMS-based lipidomic analysis, we found that the percentage of PC present in the total HDL fraction was significantly reduced in the HCAD subjects compared to HHDL subjects. These results are in line with a recently published report in a larger, overlapping cohort,(14) which showed a reduction in total HDL-PL in the HCAD subjects compared to HHDL subjects. The higher resolution provided by our lipidomics results suggest that this finding is likely due primarily to a reduction in PC, and may actually mask an increase in SM. Using a colorimetric enzymatic test, which measured PC, SM, and LPC, similar to that used by Agarwala et al.,(14) we observed a trend toward decreased total HDL-PL, though it did not reach statistical significance, likely as a result of the smaller sample size.

A previous study assessing the lipidome of HDL from patients with extremely low HDL-C (mean 21 mg/dl) due to heterozygous ApoA-I deficiency also revealed diminished PC in HDL 2b, 3b, and 3c subfractions.(24) Additionally, patients with low HDL-c (mean 36 mg/dl) after ST-elevation myocardial infarction (STEMI) showed a 9% reduction in PC in the HDL 3c subfraction compared to healthy, normolipidemic controls.(25) Intriguingly, the HDL-PC content in the HCAD group is lower than the levels observed in both the HHDL and NHDL groups. Thus our finding of reduced HDL-PC percentage in the context of high HDL-C levels and CAD is consistent with that of low HDL-C in the context of ApoA-I deficiency or STEMI and may represent a general finding of defective HDL. Furthermore, it is of interest to note that the HDL-SM content observed in the HCAD group is higher not only than that of the NHDL subjects, but also that of HHDL. Thus perhaps the most striking difference between the HCAD and HHDL groups is that of an increased SM/PC ratio.

Levels of PC, PI, and SM, as well as the SM/PC ratio, have been associated with functional properties of HDL,(20, 36–38) including cholesterol efflux.(37, 39) In our analysis, PC species with short to mid chain length and a moderate degree of unsaturation (2–3 double bonds) tended to be reduced in the HCAD subjects, while for the most part PC species with longer fatty acid chain length and higher degree of unsaturation (4+ double bonds) tended to be increased (Figure 3). Differences in PC species are known to affect cholesterol efflux.(40) According to Davidson and colleagues, the capacity of reconstituted HDL (rHDL) particles to accept FC correlates with surface fluidity, which varies based on phosphatidylcholine fatty acid tail length and degree of unsaturation. The surface of discoidal rHDL particles containing saturated PC species with longer tail lengths resides in the less-fluid gel state at 37°C, while that of rHDL particles with mono or poly-unsaturated PC species with shorter tail lengths resides in the liquid crystal state at 37°C.(40) Since Davidson and colleagues only utilized PC species with either zero, one, or two degrees of unsaturation, it is impossible to make direct statements about the effects of PC species with three or four degrees of unsaturation.

Sphingomyelin content is also known to affect HDL function. Increased sphinglomyelin content can inhibit lecithin:cholesterol acyltransferase (LCAT) activity (41), which may in turn impair LCAT mediated nascent HDL maturation (42). Sphingomyelin content affects SR-BI-mediated cholesterol efflux capacity of HDL in a nuanced, bidirectional fashion. Though an increased proportion of SM decreases the fluidity of the HDL PL surface, it has been shown to increase the cholesterol efflux capacity of rHDL containing unsaturated fatty acids (oleic and linoleic acids). (39) Direct interactions between SM and FC may counteract the effect of diminished surface fluidity and the effect of the enrichment of HDL SM on cholesterol efflux capacity may be beneficial in some contexts. (39, 43). However, sphingomyelin has also been shown to decrease uptake of cholesterol from HDL into SR-BI expressing Cos7 cells. (37) This suggests the intriguing possibility that the high HDL-C in subjects with the HCAD phenotype may result from a bottleneck in reverse cholesterol transport due to inhibited HDL cholesterol uptake to the liver. Indeed, this model is consistent with our group’s previous findings that a loss-of-function mutation in SCARB1, the gene encoding SR-BI, reduces HDL-C uptake and profoundly raises HDL-C levels, and is associated with increased risk of coronary heart disease in humans. (10)

Lastly, we observed a decrease in HDL-PI in subjects in the HCAD group compared to the healthy HHDL group. Phosphatidylinositol, an anionic class of phospholipid, which comprises 1–2% of total HDL phospho- and sphingolipid, may also affect the cholesterol efflux capacity of HDL. Indeed, enrichment of rHDL with PI increased efflux capacity by >2-fold from J774 macrophages stimulated with cAMP. (36)

The reproducibility of these potential biomarkers should be assessed in a larger population. Further studies should also be conducted to directly compare the HDL lipidome of subjects with the combined phenotype of elevated HDL-C and CAD with that of subjects that have CAD but not elevated HDL-C. This comparison would help to determine whether the differences in the HDL lipidome of subjects with high HDL-C and CAD are unique to that phenotype or are rather generalizable to subjects with CAD and normal or low HDL-C levels.

This study has several limitations, including the cross-sectional nature of the analysis, the relatively small number of subjects, and the highly disproportionate number of subjects taking statins or aspirin in the HCAD group. Some of these limitations are difficult to overcome given the rare combined phenotype involved. In the HCAD group, 19 subjects received statin treatment, with 5 subjects on atorvastatin and 9 subjects on rosuvastatin. The effects of rosuvastatin and atorvastatin on the plasma lipidome have been previously investigated. Rosuvastatin treatment was found to be associated with an increase in plasma PC and a decrease in SM, while atorvastatin treatment was associated with a decrease in both PC and SM. (44) At least part of these changes may reflect changes in non-HDL lipoproteins, which are more strongly affected than HDL-C in response to statin treatment.(45) While it is not possible to draw direct comparisons between the HDL lipidome and the plasma lipidome, these results suggest that the directionality of the potential effects of statin treatment on the HDL phosphosphingolipidome in the HCAD group may be opposite to our observed differences. This is supported by a subgroup analysis performed on subjects who were neither on statins or any other lipid-lowering treatment (Online Table 2), which still shows trends toward lower PC and PI and higher SM in the HCAD subjects. While aspirin has been shown to have profound effects on the lipidome of platelets, its primary therapeutic target, (46) we are not aware of any studies assessing the effects of aspirin on the plasma or HDL lipidome. Further studies are needed to appreciate the full effect of aspirin on the lipidome of plasma or lipoprotein compartments.

In summary, our data expands on the findings of Agarwala and colleagues (14) and suggest that variations in the HDL phosphosphingolipidome may have implications for CVD risk. Despite the relatively modest difference observed, it is possible that these differences are able to affect HDL function. Indeed, a significant reduction in cholesterol efflux capacity was observed by Agarwala et al. in a larger, partially overlapping cohort with a similar phenotype of high HDL-C and CAD.(14) Effects of altered HDL phosphosphingolipidome on other HDL functions, including antioxidative and anti-inflammatory activity, have also been demonstrated.(20)

We believe that the high sensitivity of MS-based HDL lipidomics has the potential to identify differences in HDL lipid composition that are relevant to CAD status and HDL function, even though these differences may be too subtle to be captured by current assays of HDL function. In this study we utilized lipidomic methodology to identify marked alterations in HDL particle composition in subjects who present with clinically manifest CAD, even though they would be classified as low-risk based on their HDL-C and LDL-C levels.

Highlights.

• Lipidomic anaylsis was performed on HDL from patients with CAD despite high HDL-C.

• Compared to controls, phosphatidylcholine (PC) and phosphatidylinositol (PI) are decreased.

• Sphingomyelin (SM) is increased, leading to an increased SM/PC ratio.