High-Density Lipoprotein Particle Subclass Heterogeneity and Incident Coronary Heart Disease

Abstract

Background

Raising the cholesterol of HDL particles is targeted as a cardiovascular disease prevention strategy. However, HDL particles are heterogeneous in composition and structure, which may relate to differences in antiatherogenic potential. We prospectively evaluated the association of HDL subclasses, defined by a recently proposed nomenclature, with incident coronary heart disease (CHD).

Methods and Results

Baseline HDL particle concentrations were measured by nuclear magnetic resonance spectroscopy and categorized into five subclasses (very large, large, medium, small, and very small) among 26,332 initially healthy women. During a median follow-up of 17 years, 969 cases of incident CHD (myocardial infarction, revascularization, and CHD death) were ascertained. In Cox models that adjusted for age, race/ethnicity, blood pressure, smoking, postmenopausal status, and hormone therapy, associations with incident CHD were inverse (p-trend<0.0001) for concentrations of very large (hazard ratio [HR] for top versus bottom quartile 0.49, 95% confidence interval [CI] 0.41–0.60), large (0.54, 0.45–0.64), and medium (0.69, 0.58–0.83) HDL subclasses. Conversely, HRs (95% CIs) for small and very small HDL were 1.22 (1.01–1.46, p-trend=0.08) and 1.67 (1.39–2.02, p-trend<0.0001), respectively. However, after additionally adjusting for metabolic and lipoprotein variables, associations for the spectrum of large, medium, and small HDL subclasses were inverse (p-trend<0.05 for large and small, and 0.07 for medium), while subclasses at either end of the spectrum were not associated with CHD (p-trend=0.97 for very large and 0.21 for very small HDL).

Conclusion

In this prospective study, associations with incident CHD differed by HDL particle subclass, which may be relevant for developing HDL-modulating therapies.

Robust epidemiological evidence of an inverse association between high-density-lipoprotein cholesterol (HDL-C) and coronary heart disease (CHD)¹ has failed to translate into clinical benefit.^2–4 However, HDL represents a spectrum of heterogeneous particles,⁵ which may have important implications for capturing HDL-attributable CHD risk and developing HDL modulating therapies. In particular, certain HDL particle subclasses may differ in their biological functions, including their role in reverse cholesterol transport, as well as anti-inflammatory, antioxidant, and vasodilatory functions.⁵ It is unclear whether specific subclasses of HDL are more cardioprotective or useful in assessing the benefits of HDL-related therapeutic interventions.^{6, 7} Hence, identification of specific HDL subclasses that maximize information on CHD risk may also enhance the development of more targeted therapies.

Various HDL subfractionation methods exist which also differ in their nomenclature,^{5, 8} making comparisons across these methods challenging. A recent consensus statement by Rosenson et al. proposed a new nomenclature for HDL particle subclasses based on their physiochemical properties.⁸ This nomenclature groups HDL particles based on their density and size into five distinct subclasses (very large, large, medium, small, and very small).⁸ Better characterization of HDL into these five subclasses may allow for more specific CHD risk information, although the clinical utility of this classification scheme in relation to incident CHD is uncertain.

Furthermore, the association of HDL-C with CHD risk is influenced by metabolic and lipoprotein variables, in particular triglycerides and atherogenic lipoproteins, making it important to assess HDL-related risk by accounting for such correlations.^{9, 10} Nuclear magnetic resonance (NMR) spectroscopy is a method that detects HDL subclass particle concentrations (reported in μmol/L) based on their unique subclass lipid methyl signal amplitudes.¹¹ Since NMR simultaneously quantifies the particle concentrations of the HDL subclasses, it may be particularly useful for examining correlations among these particles as well as with other lipoproteins in relation to CHD risk.

Therefore, in a prospective study of 26,332 initially healthy women followed prospectively for 17 years, we evaluated the association of baseline concentrations of HDL particle subclasses grouped by the newly proposed nomenclature described above⁸ and the risk of incident CHD events, before and after accounting for metabolic and lipoprotein variables.

METHODS

Study participants were recruited from the Women's Health Study (WHS), a completed randomized, double-blinded, placebo-controlled trial of low-dose aspirin and vitamin E in the primary prevention of cardiovascular disease (CVD) and cancer.¹² Participants were apparently healthy female health care professionals in the United States who were aged 45 years or older and free of self-reported CVD or cancer at study entry. All participants provided written informed consent, and the Institutional Review Board of the Brigham and Women's Hospital (Boston, MA) approved the study protocol. At enrollment, study participants completed questionnaires on demographics, anthropometrics, medical history, and lifestyle behaviors. A blood sample at enrollment was requested, but not required, from the 39,876 women who were randomized; 28,345 women provided one, and of these 96.7% (27,403) had complete measurements of HDL particle subclasses. The cardiovascular risk profile of women who donated blood was similar to those who did not donate blood except for a higher prevalence of baseline smoking in the latter. For the present study, we excluded women on lipid-lowering medications at baseline (n=871) and women with missing data on HDL-C or apolipoprotein A-I (ApoA-I) (n= 206), leaving 26,332 women for analysis.

Laboratory Measurements

Blood samples obtained at enrollment were collected in EDTA tubes and stored in vapor phase liquid nitrogen (−170° C) until the time for laboratory analysis. Samples for lipoprotein particle analysis were thawed, separated into 200μL aliquots, refrozen and shipped on dry ice to LipoScience, Inc (Raleigh, NC) where NMR spectroscopy was used to analyze plasma lipoprotein particles according to the LipoProfile-3 algorithm. Concentrations of HDL particle subclasses were calculated from the measured amplitudes of the spectroscopically distinct lipid methyl group NMR signals of the HDL subclasses that constitute the spectrum of total HDL particles (HDL-P) by NMR.¹¹ Using the range of size for each HDL subclass, we reclassified HDL subclasses according to the classification scheme proposed by Rosenson et al.⁸ into five distinct HDL subclasses (very large, 10.3–13.5 nanometers [nm]; large, 8.6–10.2 nm; medium, 8.3–8.5 nm; small, 8.0–8.2 nm; and very small, 7.4–7.9 nm).. Thus, the concentration of particles for each HDL subclass was determined for every participant. Total LDL particle concentration was also measured at LipoScience, Inc by NMR spectroscopy.¹³

Additional plasma lipid measurements were analyzed in a core laboratory facility certified by the National Heart, Lung and Blood Institute/Centers for Disease Control and Prevention Lipid Standardization Program.^{13, 14} Standard lipids were measured directly with a Hitachi 917 analyzer by using reagents from Roche Diagnostics (Indianapolis, IN); specifically, HDL-C was measured with a direct homogenous polyethylene-glycol assay.¹⁵ ApoA-I and apolipoprotein B100 (ApoB) were measured with immunoturbidometric assays (DiaSorin, Stillwater, MN).

Assessment of Other Variables

Covariables of interest were self-reported on the questionnaire administered at study entry and included age, race/ethnicity, smoking, menopausal status, use of hormone replacement therapy (HRT), history of diabetes, and hypertension (defined as either physician diagnosis of hypertension, antihypertensive treatment, or self-reported systolic blood pressure ≥ 140mmHg or diastolic blood pressure ≥ 90mmHg). Body mass index (BMI) was calculated as the weight in kilograms divided by the square of the height in meters.

Ascertainment of CHD Events

We selected CHD as our pre-specified primary outcome based on a previous analysis suggesting that HDL-C and ApoA-I were not associated with ischemic stroke.¹⁶ Incident CHD was a composite of nonfatal myocardial infarction, percutaneous coronary intervention, coronary artery bypass grafting, and CHD death. Information on the occurrence of these events was ascertained via annual follow-up questionnaire, letters, and telephone calls. Following written informed consent, medical records were obtained and reviewed by a blinded Endpoints Committee for the adjudication of all reported events based on predefined criteria.¹² Myocardial infarction was confirmed if symptoms met World Health Organization criteria and the event was associated with diagnostic electrocardiogram changes and abnormal levels of cardiac enzymes. Revascularization procedures were confirmed on review of hospital records. CHD deaths were confirmed by review of autopsy reports, death certificates, information obtained from family members or postal authorities, or through the National Death Index.

Statistical Analyses

Statistical analyses were performed using SAS software version 9.3 (SAS Institute, Cary, NC). Wilcoxon rank-sum tests and chi-square tests were used to analyze the differences between women who did or did not develop CHD during follow-up. Using guidelines from the Department of Health and Human Services,¹⁷ each of the five HDL particle subclasses was divided into quartiles based on the distribution of women who were not on HRT at baseline. Variables that were not normally distributed were transformed using the natural logarithm. Person-years contributed by each participant were calculated from baseline to the date of CHD diagnosis, death, or the end of follow-up (March 14, 2012), whichever came first. Independent censoring was assumed and subjects lost to follow-up were considered censored at the time of last contact.¹⁸ Cox proportional hazard regression models were used to calculate the hazard rations (HRs) and 95% confidence intervals (CIs) for incident CHD according to quartiles and per 1-standard deviation (SD) of HDL variables using separate models for each HDL particle subclass. The proportionality hazard assumption was tested for and satisfied for all five HDL subclasses by including an interaction term between the follow-up time (log) and each HDL subclass (p>0.05 for all). Cox models were initially adjusted for age, race/ethnicity, systolic blood pressure, smoking, menopausal status, HRT, and randomization treatment assignment (Model 1). In order to account for lipoprotein correlations with each other and with other related metabolic variables, model 1 was then additionally adjusted for BMI, diabetes, LDL cholesterol and particle concentration, and triglycerides (natural logarithm transformed); this model was also mutually adjusted for the other HDL particle subclasses entered as continuous variables (Model 2). Further adjustment for high-sensitivity C-reactive protein or anti-hypertensive medication use resulted in no change in the results.

Based on the prior literature, we pre-specified potential interactions by ApoB concentrations that were assessed by stratified analyses and interaction terms. A cut point of 90 mg/dL was used to discriminate between low and high concentrations of ApoB based on a previous analysis in the WHS where we found that HDL-C was not associated with cardiovascular events among women with ApoB below 90 mg/L (corresponding to the lowest tertile of ApoB in this population).¹⁶ We also examined potential interaction by HRT use.

Linear tests for trend (p-trend) were performed using the median value within each quartile as an ordinal variable. All reported p-values were two-sided, with values <0.05 considered statistically significant.

RESULTS

Baseline Characteristics and Correlations

Baseline characteristics of the 26,332 women according to whether they did or did not develop a first CHD event during follow-up are shown in Table 1. Compared with participants who remained event-free during follow-up, those who developed CHD had higher risk profiles at baseline including older age, higher blood pressure and BMI, and increased prevalence of diabetes, smoking, and dyslipidemia. Participants who developed CHD had lower HDL-C concentrations, which resulted from having lower concentrations of the very large, large, and medium HDL subclasses along with higher concentrations of the small and very small HDL subclasses.

Table 1.

Baseline characteristics according to incident CHD

	No Incident CHD N=25 363	Incident CHD N=969	P
Age, years	52.6 (48.8, 58.5)	57.9 (52.3, 64.3)	<0.0001
Systolic blood pressure, mmHg	125 (115,135)	135 (125,145)	<0.0001
Diastolic blood pressure, mmHg	80 (70, 80)	80 (70, 87)	<0.0001
Hormone replacement therapy, %	43.4	43.4	1.0
Postmenopausal, %	53.1	69.4	<0.0001
Diabetes, %	2.1	12.5	<0.0001
Race/ethnicity			0.20
Caucasian	95.3	97.0
Hispanic	1.1	0.7
African-American	1.9	1.5
Other	1.8	0.8
Smoking, %	11.2	21.2	<0.0001
Body mass index, kg/m2	24.8 (22.3, 28.3)	26.6 (23.6,30.1)	<0.0001
HDL cholesterol, mg/dl	52.3 (43.5, 62.6)	45.2 (37.9, 55.6)	<0.0001
Apolipoprtotein A-l, mg/dl	149 (133,168)	142 (126,162)	<0.0001
LDL cholesterol, mg/dl	121 (100,143)	135 (114,159)	<0.0001
LDL particle concentration, nmol/L	1201 (969, 1479)	1468 (1198, 1775)	<0.0001
Apolipoprtotein B, mg/dl	99 (83,120)	120 (99,138)	<0.0001
Triglycerides, mg/dl	117 (83,171)	161 (111,235)	<0.0001
HDL particle concentration, μmol/L
Total	36.8 (32.7,41.6)	36.5 (32.1,41.5)	0.05
Subclasses
Very large	3.1 (1.8,4.9)	2.3 (1.4, 3.8)	<0.0001
Large	5.5 (3.9, 7.4)	4.9 (3.2, 6.6)	<0.0001
Medium	9.1 (6.3, 12.4)	8.3 (5.4, 12)	<0.0001
Small	6.7 (4.3, 9.8)	6.9 (4.6, 10.4)	0.01
Very small	10.6 (7.9, 13.4)	11.9 (9.4, 14.8)	<0.0001

Values shown are medians (25^th, 75^th percentile) or percentages.

Table 2 shows Spearman correlation coefficients for the various HDL measures (including the five subclasses) and other lipid and lipoprotein measures. The correlation coefficient of HDL-C with each of the five subclasses ranged from 0.73 (for the very large HDL subclass, p <0.0001) to − 0.26 (for the very small HDL subclass, p <0.0001). In comparison, total HDL-P (the sum of the HDL subclasses) had the strongest correlation with the medium and large HDL subclasses (r= 0.65 and 0.50) and minimal correlation with the very small particle subclass (r= − 0.06), p for all <0.0001. The correlation coefficients for the five HDL particle subclasses with each other ranged from 0.38 to − 0.37, p for all <0.0001. The HDL subclasses also differed in the direction and magnitude of correlation with LDL cholesterol and particle concentration, ApoB, triglycerides, and BMI, going from negative (r= − 0.5) to positive (r= 0.4) as HDL particles became progressively smaller in size.

Table 2.

Spearman correlation coefficients

	HDL-C	ApoA-I	LDL-C	LDL-P	ApoB	TG	HDL-P	HDL-VL	HDL-L	HDL-M	HDL-S	HDL-VS
ApoA-I	0.79
LDL-C	−0.05	−0.07
LDL-P	−0.40	−0.20	0.70
ApoB	−0.32	−0.10	0.79	0.85
TG	−0.40	−0.09	0.28	0.52	0.52
HDL-P	0.47	0.74	−0.08	0.01*	0.01π	0.22
HDL-VL	0.73	0.65	−0.34	−0.53	−0.43	−0.37	0.36
HDL-L	0.48	0.50	−0.12	−0.19	−0.16	−0.07	0.50	0.25
HDL-M	0.27	0.40	−0.15	−0.10	−0.09	0.13	0.65	0.24	0.38
HDL-S	−0.05	0.11	−0.001π	0.12	0.09	0.20	0.33	−0.04	−0.11	−0.05
HDL-VS	−0.26	−0.15	0.30	0.40	0.35	0.23	−0.06	−0.37	−0.28	−0.35	−0.19
BMI	−0.26	−0.26	0.15	0.30	0.26	0.33	−0.08	0.38	−0.14	−0.08	0.06	0.22

HDL-C, HDL cholesterol; ApoA-I, apolipoprotein A-I; LDL-C, LDL cholesterol; LDL-P, total LDL particles; ApoB, apolipoprotein B100; TG, triglycerides. HDL-P, total HDL particle concentration; HDL-VL, very large HDL; HDL-L, large HDL; HDL-M, medium HDL; HDL-S, small HDL; HDL-VS, very small HDL; BMI, body mass index.

All p values <0.0001 except,

^*p = 0.03

^πp = not significant.

Associations with CHD Events

A total of 969 incident CHD events (365 myocardial infarctions, 600 revascularization procedures, and 4 CHD deaths) occurred over a median follow-up of 17 years. HDL-C was inversely associated with incident CHD: model 1 HR (95% CI) for top versus bottom quartile, 0.34 (0.28–0.41), and model 2 HR (95% CI), 0.59 (0.43–0.88), p-trend < 0.0001 for both models. Table 3 shows the association for total HDL-P (the sum of the subclasses) and each of the five HDL subclasses with incident CHD. In the minimally-adjusted model 1, there were inverse associations with CHD for increasing quartiles of total HDL-P and each of very large, large, and medium HDL subclasses; respective HRs (95% CIs) for top versus bottom quartiles were 0.77 (0.64–0.92), 0.49 (0.41–0.60), 0.54 (0.45–0.64) and 0.69 (0.58–0.83), all p-trend ≤0.008. By contrast, small and very small HDL subclasses were associated with increased risk of CHD, respective HRs (95% CIs) for top versus bottom quartiles were 1.22 (1.01–1.46), p-trend =0.08, and 1.67 (1.39–2.02), p-trend <0.0001.

Table 3.

Association of the five HDL particle subclasses with incident CHD

	Quartile 1	Quartile 2	Quartile 3	Quartile 4	P-Trend	*Per 1 SD	P
Total HDL-P(7.4–13.5nm)
Range,μmol/L	≤ 31.39	31.40–34.66	34.67–38.25	≥38.26
Model 1, HR (95% CI)	1	0.85 (0.69–1.04)	0.81 (0.66–0.98)	0.77 (0.64–0.92)	0.008	0.91 (0.86–0.97)	0.002
Model 2, HR (95% CI)	1	0.87 (0.71–1.07)	0.78 (0.63–0.95)	0.70 (0.58–0.85)	0.0003	0.88 (0.83–0.93)	<0.0001
Very large (10.3–13.5nm)
Range,μmol/L	≤ 1.53	1.54–2.69	2.70–4.32	≥4.33
Model 1, HR (95% CI)	1	0.81 (0.68–0.95)	0.60 (0.50–0.72)	0.49 (0.41–0.60)	<0.0001	0.77 (0.72–0.81)	<0.0001
Model 2, HR (95% CI)	1	0.93 (0.78–1.11)	0.85 (0.69–1.04)	1.00 (0.80–1.25)	0.97	0.95 (0.88–1.03)	0.18
Large (8.6–10.2nm)
Range,μmol/L	≤ 3.51	3.52–4.92	4.93–6.55	≥6.56
Model 1, HR (95% CI)	1	0.69 (0.57–0.82)	0.68 (0.57–0.82)	0.54 (0.45–0.64)	<0.0001	0.82 (0.77–0.86)	<0.0001
Model 2, HR (95% CI)	1	0.74 (0.61–0.89)	0.81 (0.67–0.97)	0.71 (0.58–0.86)	0.003	0.89 (0.83–0.94)	0.0002
Medium (8.3–8.5nm)
Range,μmol/L	≤ 5.52	5.53–8.02	8.03–10.80	≥10.81
Model 1, HR (95% CI)	1	0.82 (0.68–0.98)	0.68 (0.56–0.82)	0.69 (0.58–0.83)	<0.0001	0.89 (0.84–0.94)	0.005
Model 2, HR (95% CI)	1	0.96 (0.79–1.16)	0.85 (0.70–1.05)	0.84 (0.68–1.03)	0.07	0.94 (0.87–1.00)	0.06
Small (8.0–8.2nm)
Range,μmol/L	≤ 4.12	4.13–6.33	6.33–9.17	≥9.18
Model 1, HR (95% CI)	1	1.17 (0.97–1.42)	1.11 (0.91–1.34)	1.22 (1.01–1.46)	0.08	1.07 (1.01–1.13)	0.03
Model 2, HR (95% CI)	1	1.08 (0.88–1.32)	0.95 (0.78–1.16)	0.84 (0.69–1.03)	0.03	0.93 (0.87–0.99)	0.02
Very small (7.4–7.9nm)
Range,μmol/L	≤ 8.30	8.31–11	11.01–13.69	≥13.70
Model 1, HR (95% CI)	1	1.20 (0.98–1.48)	1.57 (1.30–1.91)	1.67 (1.39–2.02)	<0.0001	1.22 (1.14–1.30)	<0.0001
Model 2, HR (95% CI)	1	1.05 (0.85–1.30)	1.12 (0.91–1.39)	0.88 (0.70–1.11)	0.21	0.95 (0.88–1.02)	0.16

HDL-P, HDL particle concentration; nm, nanometer; HR, hazard ratio; CI, confidence interval.

^*Natural log transformation was used to compute values for very large and large HDL subclasses.

Model 1 included age, race/ethnicity, blood pressure, smoking, postmenopausal status, hormone replacement therapy, and treatment assignment.

Model 2 included model 1 covariables plus body mass index, diabetes, LDL cholesterol and particle concentration, triglycerides, and the other 4 HDL subclasses.

Total HDL particle concentration was not adjusted for the HDL subclasses as it is the sum of the subclasses. Due to participants with missing information on adjusted covariates, Model 1 included a total of 25 706 participants (947 events) and Model 2 included a total of 25 232 participants (911 events)

After additionally adjusting for the metabolic and lipoprotein variables, including mutual adjustment for the other HDL particle subclasses (model 2, Table 3), higher concentrations of the large and small HDL subclasses showed a statistically significant trend for lower CHD, respective HRs (95% CIs) for top versus bottom quartiles were 0.71 (0.58–0.86) (p-trend =0.003) and 0.84 (0.68–1.03) (p-trend =0.03), with a borderline significant trend for medium HDL: 0.84 (0.68–1.03), p-trend =0.07. Total HDL-P also was associated with a reduced risk of incident CHD (p-trend =0.0003). Similar results were obtained when the HDL subclasses were examined per 1-SD increments.

Stratified Analyses

Event rates differed in participants with ApoB<90 versus ≥90 mg/dL (1.4% and 4.9%, respectively). The associations of HDL subclasses with incident CHD were significant only among participants with ApoB ≥90mg/dL (Tables 4 and 5), with statistically significant interactions by ApoB for the association of total HDL-P and the large HDL subclass with incident CHD (p for interaction =0.01 and 0.003, respectively). CHD events rates were similar in baseline users and non-users of HRT (3.7%). Somewhat attenuated associations were seen among HRT users, with only the large HDL subclass having statistically significant interaction by HRT use (p for interaction =0.02, data not shown).

Table 4.

Association of HDL particle subclasses with incident CHD in participants with apolipoprotein B ≥ 90mg/dL (N=17 227, CHD events =838)

	Quartile 1	Quartile 2	Quartile 3	Quartile 4	P-Trend	*Per 1 SD	P
Total HDL-P (7.4–13.5 nm)
Range, μmol/L	≤ 31.39	31.40–34.66	34.67–38.25	≥38.26
Model 1, HR (95% CI)	1	0.84 (0.68–1.04)	0.76 (0.61–0.94)	0.73 (0.60–0.89)	0.003	0.90 (0.85–0.96)	0.0007
Model 2, HR (95% CI)	1	0.85 (0.68–1.05)	0.72 (0.58–0.89)	0.65 (0.53–0.80)	<0.0001	0.86 (0.81–0.92)	<0.0001
Very large (10.3–13.5 nm)
Range, μmol/L	≤ 1.53	1.54–2.69	2.70–4.32	≥4.33
Model 1, HR (95% CI)	1	0.84 (0.71–1.01)	0.67 (0.55–0.81)	0.62 (0.50–0.77)	<0.0001	0.82 (0.77–0.88)	<0.0001
Model 2, HR (95% CI)	1	0.93 (0.78–1.12)	0.84 (0.68–1.04)	1.08 (0.85–1.38)	0.70	0.96 (0.88–1.04)	0.26
Large (8.6–10.2 nm)
Range, μmol/L	≤ 3.51	3.52–4.92	4.93–6.55	≥6.56
Model 1, HR (95% CI)	1	0.71 (0.59–0.86)	0.71 (0.59–0.86)	0.54 (0.44–0.65)	<0.0001	0.82 (0.77–0.87)	<0.0001
Model 2, HR (95% CI)	1	0.75 (0.62–0.91)	0.80 (0.66–0.98)	0.65 (0.52–0.80)	0.0003	0.87 (0.82–0.93)	<0.0001
Medium (8.3–8.5 nm)
Range, μmol/L	≤ 5.52	5.53–8.02	8.03–10.80	≥10.81
Model 1, HR (95% CI)	1	0.82 (0.68–1.00)	0.69 (0.56–0.84)	0.69 (0.58–0.84)	0.0001	0.89 (0.83–0.95)	0.0002
Model 2, HR (95% CI)	1	0.93 (0.76–1.14)	0.82 (0.66–1.02)	0.80 (0.64–1.00)	0.04	0.93 (0.86–1.00)	0.04
Small (8.0–8.2 nm)
Range, μmol/L	≤ 4.12	4.13–6.33	6.33–9.17	≥9.18
Model 1, HR (95% CI)	1	1.16 (0.94–1.43)	1.09 (0.88–1.35)	1.14 (0.94–1.39)	0.35	1.04 (0.98–1.11)	0.20
Model 2, HR (95% CI)	1	1.08 (0.87–1.33)	0.94 (0.76–1.17)	0.81 (0.65–1.01)	0.01	0.91 (0.85–0.98)	0.01
Very small (7.4–7.9 nm)
Range, μmol/L	≤ 8.30	8.31–11	11.01–13.69	≥13.70
Model 1, HR (95% CI)	1	1.24 (0.98–1.57)	1.45 (1.16–1.82)	1.52 (1.23–1.89)	<0.0001	1.16 (1.08–1.24)	<0.0001
Model 2, HR (95% CI)	1	1.17 (0.91–1.49)	1.14 (0.90–1.46)	0.93 (0.72–1.20)	0.22	0.94 (0.87–1.02)	0.16

HDL-P, HDL particle concentration; nm, nanometer; HR, hazard ratio; CI, confidence interval.

^*Natural log transformation was used to compute values for very large and large HDL subclasses.

Model 1 included age, race/ethnicity, blood pressure, smoking, postmenopausal status, hormone replacement therapy, and treatment assignment.

Model 2 included model 1 covariables plus body mass index, diabetes, LDL cholesterol and particle concentration, triglycerides, and the other 4 HDL subclasses Due to participants with missing information on adjusted covariates, Model 1 included a total of 16 800 participants (817 events) and Model 2 included a total of 16 459 participants (786 events)

Table 5.

Association of HDL particle subclasses with incident CHD in participants with apolipoprotein B < 90 mg/dL (N=9 100, CHD events =131)

	Quartile 1	Quartile 2	Quartile 3	Quartile 4	P-Trend	*Per 1 SD	P
Total HDL-P (7.4–13.5 nm)
Range, μmol/L	≤ 31.39	31.40–34.66	34.67–38.25	≥38.26
Model 1, HR (95% CI)	1	1.26 (0.65–2.45)	1.47 (0.78–2.75)	1.37 (0.76–2.47)	0.40	1.04 (0.90–1.20)	0.61
Model 2, HR (95% CI)	1	1.23 (0.63–2.40)	1.41 (0.75–2.65)	1.32 (0.72–2.43)	0.46	1.02 (0.87–1.19)	0.85
Very large (10.3–13.5 nm)
Range, μmol/L	≤ 1.53	1.54–2.69	2.70–4.32	≥4.33
Model 1, HR (95% CI)	1	0.96 (0.48–1.89)	0.80 (0.42–1.51)	0.70 (0.38–1.29)	0.14	0.85 (0.69–1.06)	0.14
Model 2, HR (95% CI)	1	0.94 (0.46–1.90)	0.89 (0.45–1.75)	0.82 (0.41–1.65)	0.54	0.93 (0.72–1.20)	0.58
Large (8.6–10.2 nm)
Range, μmol/L	≤ 3.51	3.52–4.92	4.93–6.55	≥6.56
Model 1, HR (95% CI)	1	0.78 (0.41–1.48)	0.93 (0.52–1.69)	1.06 (0.62–1.81)	0.43	1.01 (0.98–1.04)	0.57
Model 2, HR (95% CI)	1	0.83 (0.43–1.61)	1.10 (0.60–2.05)	1.36 (0.76–2.43)	0.11	1.08 (0.88–1.32)	0.48
Medium (8.3–8.5 nm)
Range, μmol/L	≤ 5.52	5.53–8.02	8.03–10.80	≥10.81
Model 1, HR (95% CI)	1	1.30 (0.69–2.45)	1.22 (0.66–2.24)	1.29 (0.72–2.30)	0.53	1.03 (0.88–1.20)	0.72
Model 2, HR (95% CI)	1	1.31 (0.69–2.48)	1.17 (0.62–2.19)	1.13 (0.60–2.14)	0.96	0.96 (0.80–1.17)	0.71
Small (8.0–8.2 nm)
Range, μmol/L	≤ 4.12	4.13–6.33	6.33–9.17	≥9.18
Model 1, HR (95% CI)	1	1.16 (0.69–1.96)	1.16 (0.69–1.94)	1.36 (0.83–2.24)	0.23	1.10 (0.94–1.28)	0.22
Model 2, HR (95% CI)	1	1.10 (0.64–1.87)	1.03 (0.61–1.75)	1.10 (0.65–1.86)	0.78	1.03 (0.86–1.22)	0.77
Very small (7.4–7.9 nm)
Range, μmol/L	≤ 8.30	8.31–11	11.01–13.69	≥13.70
Model 1, HR (95% CI)	1	0.78 (0.50–1.21)	1.21 (0.78–1.88)	0.72 (0.38–1.35)	0.69	0.98 (0.81–1.18)	0.84
Model 2, HR (95% CI)	1	0.71 (0.44–1.15)	1.21 (0.74–1.97)	0.59 (0.29–1.20)	0.43	0.93 (0.74–1.16)	0.50

HDL-P, HDL particle concentration; nm, nanometer; HR, hazard ratio; CI, confidence interval.

^*Natural log transformation was used to compute values for very large and large HDL subclasses.

Model 1 included age, race/ethnicity, blood pressure, smoking, postmenopausal status, hormone replacement therapy, and treatment assignment.

Model 2 included model 1 covariables plus body mass index, diabetes, LDL cholesterol and particle concentration, triglycerides, and the other 4 HDL subclasses Due to participants with missing information on adjusted covariates, Model 1 included a total of 8 901 participants (130 events) and Model 2 included a total of 8 768 participants (125 events)

DISCUSSION

In this prospective study of 26,332 initially healthy women followed for a median duration of 17 years, differential associations with incident CHD events were found for baseline concentrations of five HDL subclasses measured by NMR spectroscopy and grouped according to a newly proposed classification scheme. Before accounting for the correlations of the HDL subclasses with each other and with metabolic and lipoprotein variables, the very large, large, and medium HDL subclasses had inverse association with CHD, while small and very small HDL subclasses had positive association. Once the correlations were accounted for, associations for the spectrum of large, medium, and small HDL subclasses showed a tendency towards a reduced risk of CHD (p-trend<0.05 for large and small, 0.07 for medium), while the subclasses at either end of the spectrum were not associated with CHD (p-trend =0.97 and 0.21 for very large and very small HDL, respectively). These findings underscore the heterogeneity of HDL particle subclasses in conveying clinical CHD risk information.

This is the first study to examine incident CHD associations in relation to NMR-measured HDL particle subclasses grouped according to the five subclasses that were recently recommended.⁸ Related studies that have assessed the association between HDL subclasses and CHD risk by NMR spectroscopy have previously grouped HDL particles into three subclasses (large, medium, and small).^{9, 13, 19–22} The previously designated NMR derived “large” HDL subclass corresponds to the “very large” HDL subclass assessed in the present study, while the previously designated “small” HDL particle subclass is a combination of both the “very small” and “small” HDL subclasses assessed in the present study.⁸ Using the new classification scheme additionally identified a very small HDL subclass, which was not associated with CHD in our study, and refined the range of medium to large HDL subclasses. Hence, this new HDL subclass distribution may provide better assessment of CHD risk attributable to specific HDL particle subclasses.

In a previous case-control study of high-risk men with established CHD and low HDL-C, all three subclasses tended towards a reduced risk of CHD, although only small and medium HDL were statistically significant after accounting for lipoprotein correlations.²⁰ Similar results were obtained in a secondary prevention study, with only small and medium HDL subclasses having significant inverse association with CHD after adjustment for lipoprotein correlations, while the association of large HDL was attenuated after adjustment.²³ Taken together, our results in a low-risk primary prevention population of women are in agreement with these prior studies.

Besides NMR spectroscopy, HDL particle subclass concentrations can also be measured with ion mobility, but the two methods have not been directly compared.⁸ Ion mobility currently subfractionates HDL particles into two subclasses (large and small); both subclasses were inversely associated with CHD in a prior study.²⁴ Other commonly used HDL subfractionation methods do not specifically quantify the particle concentrations of HDL subclasses but instead classify HDL by other physical properties.^{2, 10, 25} Prior studies that evaluated HDL size using these other subfractionation methods have mostly found inverse association for large HDL size with CHD but equivocal results for smaller HDL size.^26–34

HDL metabolism is a dynamic process during which HDL particles are constantly being remodeled in a lipidation-delipidation cycle, and effective metabolism depends on a system of recycling between larger and smaller HDL particles.^{5, 35} Individual HDL subclasses differ in their ability to promote cholesterol efflux.^{35, 36} This, along with relative differences in other non-transport related functions of HDL,⁵ may explain the lack of uniform CHD risk across the spectrum of HDL subclasses in our study. The pattern of association that we observed suggests that metabolic processes involved in the maturation and delipidation of HDL may be coupled to CHD risk. The lack of an association between either very large HDL or very small HDL after adjusting for metabolic and lipoprotein variables may thus be unrelated to the cholesterol content of these particles and may be relevant for understanding HDL-attributable CHD risk. By contrast, in the middle of the spectrum, increasing concentrations of progressively larger HDL particles (small, medium, and large HDL) had inverse associations with CHD.

The metabolic fate of very large HDL is predominantly the regeneration of small HDL either through transporting cholesterol to the liver for excretion or through the concerted action of lipid transfer proteins and lipases.³⁵ In particular, very large HDL particles may be limited in their cholesterol efflux capacity as noted by their inability to interact with ATP binding cassette transporter A1, a potent pathway for cholesterol efflux.^{5, 35} Very large HDL may also have other diminished atheroprotective functions.^{37, 38} For instance, accumulation of very large HDL at the expense of small HDL particles after intervention with cholesteryl ester transfer protein (CETP) inhibitors is debated as a possible reason for their inability to reduce cardiovascular events.^{37, 39} Events involved in the maturation and recycling of HDL particles may thus underlie the observations in our study.

On the other hand, HDL metabolism may also be altered in dyslipidemic conditions such as diabetes and insulin resistance states, with a redistribution of HDL particles towards smaller particles as a result of enhanced CETP – lipase activity.¹⁸ Hence, the preponderance of small HDL particles usually encountered in association with atherogenic dyslipidemia may reflect enhanced enzymatic remodeling of large HDL into small HDL particles. It is also possible that very small HDL may accumulate as a consequence of impaired HDL maturation. The result of this study and others^{20, 23, 24} do not support the notion that small HDL particles are atherogenic per se, but instead suggest that the apparent increased CHD risk for small HDL particles may be due to their co-occurrence with other metabolic and lipoprotein derangements. Analysis and interpretation of data relating to HDL and CHD risk should therefore account for the correlation between different HDL particle subclasses and atherogenic lipids/lipoproteins. From a therapeutic standpoint, conversion of large lipid rich HDL particles into small lipid poor HDL particles by autologous delipidation of selective HDL may diminish atheroma volume,³⁸ but whether this improves clinical outcomes has not been determined. Besides biological activities that directly influence HDL particle size, HDL subclasses differ in other functions, including anti-inflammatory, antioxidant, and immunomodulatory functions.^{5, 8} Specifically, anti-inflammatory and antioxidative capacities of HDL have been found to preferentially localize to smaller protein-rich HDL particles.⁵ While we found that both HDL-C and total HDL-P were significantly inversely associated with CHD, our results differ from other studies that found total HDL-P was a better predictor than HDL-C.^20,10,25 This may be in part due to the different study populations, in that the WHS is a lower risk population of women. Notably, we found evidence of interaction by atherogenic lipoprotein burden (as quantified by ApoB ≥90 mg/dL), with a stronger inverse association of total HDL-P and CHD among WHS participants with high ApoB, consistent with our prior finding for HDL-C.¹⁶ Other differences in the studies include the method of HDL-C measurement, sample collection and storage, and timing of the NMR measurements.

Strengths of this study include the long prospective follow-up (median 17 years), large sample size, and the well characterized information on cardiovascular health and outcomes. Limitations of our study include that HDL particle concentrations were only measured at baseline. It is possible that processing and storage duration of the blood samples may introduce measurement error, however this would not be expected to differentially affect the HDL subclasses. HDL particle subclasses were obtained by NMR spectroscopy, hence we were not able to make comparison based on subclasses obtained from other techniques or in relation to HDL function.⁸ The study population of low-risk female health professionals who were middle-aged and older at baseline may have different health behaviors from men or women in the general population which may limit the generalizability of these results to other populations. Residual confounding or chance cannot be ruled out due to the observational nature of this study. Finally, as the current study did not set out to assess risk stratification, we note that the data presented does not suggest a recommendation for the measurement of HDL subclasses as a clinical tool to assess CHD risk.

In conclusion, we provide evidence that concentrations of HDL particle subclasses are differentially related to incident CHD. This heterogeneity in HDL particle subclasses and their attributable CHD risk may have important implications for the development of HDL modulating therapies.