Abstract
Prevalence of cardiovascular disease (CVD) is increased in patients with obstructive sleep apnea (OSA), possibly related to dyslipidemia in these individuals. Insulin resistance is also common in OSA, but its contribution to dyslipidemia of OSA is unclear. The study's aim was to define the relationships among abnormalities of lipoprotein metabolism, clinical measures of OSA, and insulin resistance.
Cross-sectional study. OSA severity was defined by the apnea-hypopnea index (AHI) during polysomnography. Hypoxia measures were expressed as minimum and mean oxygen saturation, and the oxygen desaturation index. Insulin resistance was quantified by determining steady-state plasma glucose (SSPG) concentrations during the insulin suppression test. Fasting plasma lipid/lipoprotein evaluation was performed by vertical auto profile methodology.
Academic medical center.
107 nondiabetic, overweight/obese adults.
Lipoprotein particles did not correlate with AHI or any hypoxia measures, nor were there differences noted by categories of OSA severity. By contrast, even after adjustment for age, sex, and BMI, SSPG was positively correlated with triglycerides (r = 0.30, P < 0.01), very low density lipoprotein (VLDL) and its subclasses (VLDL1+2) (r = 0.21–0.23, P < 0.05), and low density lipoprotein subclass 4 (LDL4) (r = 0.30, P < 0.01). SSPG was negatively correlated with high density lipoprotein (HDL) (r = −0.38, P < 0.001) and its subclasses (HDL2 and HDL3) (r = −0.32, −0.43, P < 0.01), and apolipoprotein A1 (r = −0.33, P < 0.01). Linear trends of these lipoprotein concentrations across SSPG tertiles were also significant.
Pro-atherogenic lipoprotein abnormalities in obstructive sleep apnea (OSA) are related to insulin resistance, but not to OSA severity or degree of hypoxia. Insulin resistance may represent the link between OSA-related dyslipidemia and increased cardiovascular disease risk.
INTRODUCTION
Atherosclerotic cardiovascular diseases (CVD) such as coronary heart disease, peripheral vascular disease, and stroke are highly prevalent in obstructive sleep apnea (OSA).1 Abnormalities in lipoprotein concentrations may be implicated in the pathogenesis of OSA-related atherosclerosis.2 In addition to elevations in total cholesterol and triglycerides (TG), decreases in high density lipoprotein cholesterol (HDL-C) levels have been described in patients with OSA,2 and dyslipidemia may be exacerbated by increasing OSA severity.3 It has been postulated that recurrent hypoxia may induce lipid alterations via mechanisms related to oxidative stress4,5 or activation of lipid biosynthesis pathways in the liver.6 However, high TG and low HDL-C are also characteristic of insulin resistance,7 and insulin resistance is independently associated with OSA8 and excess CVD risk.9 Therefore, it is conceivable that the lipoprotein abnormalities present in OSA may be mediated by mechanisms related to insulin resistance rather than OSA per se. Prior investigations of the dyslipidemic profile in OSA either did not contain measures of insulin resistance, had mixed populations of patients with and without diabetes, or included patients who already had underlying CVD.3,5,10–13 Thus, the extant literature does not permit distinction between the contributions of OSA or insulin resistance to the lipid phenotype, prior to clinical CVD event onset.
In an attempt to address some of the above issues, we initiated the present cross-sectional study with the primary goal to elucidate the relationships among lipoprotein concentrations, OSA, and insulin resistance. Patients with OSA but without diabetes or clinically evident CVD underwent direct measurement of insulin action by the insulin suppression test. Because the role of lipoprotein subclasses, especially pro-atherogenic lipoprotein particles, has also not been well-characterized within the context of OSA, comprehensive lipid/lipoprotein evaluation by vertical auto profile (VAP) methodology was performed. We hypothesized that lipid abnormalities would be more closely associated with insulin resistance than severity of OSA.
METHODS
Subjects
Nondiabetic, overweight to obese (body mass index, BMI 26–38 kg/m2) adult volunteers aged 30 to 70 years old were recruited between July 2010 and October 2013 (n = 107). Participants recruited from the Stanford Sleep Medicine Center were diagnosed with OSA, defined as apnea-hypopnea index (AHI) ≥ 5 events/h in addition to characteristic symptoms. Additional volunteers were sought through local print, online, or radio advertisements soliciting individuals with symptoms of but no known prior diagnosis of OSA. Main exclusion criteria were having a history of major medical illnesses including type 2 diabetes, CVD, kidney or liver diseases, use of medications that may affect glucose metabolism or weight loss, and having previously received treatment for OSA. History regarding use of prescription or non-prescription lipid-lowering therapy was obtained. All subjects gave written informed consent, and the study protocol was approved by the Stanford Administrative Panels for the Protection of Human Subjects.
Procedures
After an overnight fast, subjects were admitted to the Stanford Clinical and Translational Research Unit. BMI (kg/m2) was calculated based on height and weight measured while subjects wore light clothing and no shoes. Fasting plasma samples were frozen at −80°C until lipoprotein analysis was performed by VAP methodology (Atherotech, Inc).14 All 5 major lipoprotein classes were measured by VAP: (1) HDL; (2) very low density lipoprotein (VLDL); (3) intermediate density lipo-protein (IDL); (4) lipoprotein (a) [Lp(a)]; and (5) low density lipoprotein (LDL). LDL was further reported as total LDL or LDL-Real, i.e., LDL cholesterol without Lp(a) and IDL. In addition, subclasses HDL2 (large and buoyant subclass), HDL3 (small and dense subclass), LDL subclasses ranging from least dense (LDL1) to most dense (LDL4), and VLDL subclasses (VLDL1+2, VLDL3) were measured. Determination of LDL pattern A (large and buoyant LDL subclass) and LDL pattern B (smaller, dense LDL) was also made based on LDL max time value, which decreases as LDL peak density increases. Measurements of apolipoprotein B (apoB) and apolipoprotein A1 (apoA1) allowed for calculation of apoB/apoA1 ratio.
Insulin Suppression Test
Insulin-mediated glucose uptake was quantified by the modified octreotide15 insulin suppression test as originally described and validated by our research group.16 After a 12-h overnight fast, volunteers were infused over 180 min with octreotide (0.27 mg/m2/min), insulin (32 mU/m2/min), and glucose (267 mg/m2/min). Insulin and glucose concentrations were measured every 10 min during the final 30 min of the infusion and averaged to obtain steady-state plasma insulin (SSPI) and steady-state plasma glucose (SSPG) concentrations. Because SSPI is comparable across all individuals under these experimental conditions, SSPG provides a direct measure of the ability of insulin to mediate disposal of the infused glucose load. Thus, the higher the SSPG value, the more insulin-resistant the individual. Results from the insulin suppression test have been shown to be highly correlated with measures of insulin sensitivity derived from the hyperinsulinemic euglycemic clamp (r ≥ −0.87).17,18 Subjects were further subdivided by tertiles of SSPG.
Polysomnography
Full-night in-laboratory nocturnal polysomnograms (PSG) were performed at the Stanford Sleep Medicine Center according to standard procedures. Each PSG was scored manually by a single experienced certified technologist in accordance to the American Academy of Sleep Medicine Manual (2007).19 Severity of OSA was defined as follows: (1) mild: AHI 5–14.9 events/h; (2) moderate: AHI 15–30 events/h; and (3) severe: AHI > 30 events/h. Other measurements included minimum oxygen saturation (minO2), mean oxygen saturation (meanO2), and oxygen desaturation index (ODI, defined as the number of times per hour that O2 sat drops ≥ 3% from baseline). AHI was further subdivided into events occurring during REM (REM-AHI) sleep and NREM (NREM-AHI) sleep.
Statistical Analysis
Statistical analyses were performed using IBM SPSS Statistics 21.0 (Armonk, NY). Lipoprotein concentrations that were not normally distributed were log-transformed prior to analysis. Centering and scaling was employed to maximize the numerical stability of regression fitting algorithms and provide some mitigation of collinearity. Pearson simple and partial (adjusted for age, sex, and BMI) correlation coefficients were calculated between lipid/lipoprotein concentrations and fasting plasma glucose (FPG), SSPG, AHI, minO2, meanO2, and ODI. Where variables were not normally distributed Spearman correlations were performed. A generalized linear model (ANCOVA) to adjust for age, sex, and BMI was used to assess for significant associations across tertiles of SSPG or categories of OSA. P values were adjusted for multiple pairwise comparisons using sequential Bonferroni adjustment. Trends of lipo-protein concentrations across the groups were estimated by ANOVA. Chi-squared tests were performed to compare LDL density patterns across tertiles of SSPG concentration and OSA categories. P < 0.05 denoted statistical significance and all hypothesis testing was two-tailed.
RESULTS
Approximately two-thirds of the study population was comprised of men (Table 1). On average they were middle-aged (50 years), obese (BMI 30.6 kg/m2), and had impaired fasting glucose (101 mg/dL). Insulin sensitivity as quantified by SSPG varied 6-fold among study participants (range 51 to 309 mg/dL). One-quarter had mild OSA, 31% had moderate OSA, and 45% had severe OSA. Twenty-nine (27.1%) individuals reported taking prescription and/or non-prescription lipid-lowering therapy. Among these individuals, 18 were on statin therapy. The proportion of individuals on lipid-lowering therapy or taking a statin did not differ by tertiles of SSPG or categories of OSA severity (P > 0.05).
Table 2 presents a correlation matrix between lipoprotein measurements and variables pertaining to glucose metabolism (FPG, SSPG), and OSA (AHI, minO2, meanO2, ODI). Signifi-cant relationships with lipoprotein concentrations were demonstrated most frequently with SSPG. SSPG was positively correlated with TG, total VLDL, VLDL1+2, LDL4, apoB/apoA1 ratio, and inversely correlated with total HDL and its subclasses (HDL2 and HDL3), LDL2, apoA1, and Lp(a). Notably, the correlations between SSPG and LDL subclasses were detected despite the lack of an association between SSPG and total LDL or LDL-Real. Statistical significance was preserved for SSPG and these lipoprotein measurements even after adjustment for age, sex, and BMI. Several significant relationships were also observed between FPG and lipoprotein concentrations, although the magnitude of the associations was of lesser degree. After multivariate adjustment, FPG was positively correlated with TG, VLDL1+2, LDL3, LDL4, apoB, and apoB/apoA1 ratio.
Simple and partial correlation coefficients (r) between lipoprotein measurements and metabolic or sleep variables.
Simple and partial correlation coefficients (r) between lipoprotein measurements and metabolic or sleep variables.
In contrast to the significant relationships between SSPG and FPG and lipid/lipoprotein concentrations, none of the OSA-related variables were correlated with lipoprotein concentrations after adjustment for age, sex, and BMI. Additional analysis of correlations between lipoprotein concentrations and REM-AHI or NREM-AHI yielded no significant differences (data not shown). Importantly, there were also no statistically significant correlations between SSPG and AHI (r = 0.13, P = 0.18), SSPG and REM-AHI (r = 0.15, P = 0.12), or SSPG and NREM-AHI (r = 0.13, P = 0.20).
Evaluation of lipoprotein measurements across tertiles of SSPG concentrations is shown in Table 3. The trends for TG, VLDL, VLDL1+2, LDL4, and apoB/apoA1 ratio were such that concentrations increased across SSPG tertiles. Conversely, HDL and its subclasses (HDL2 and HDL3), apoA1, LDL2, and Lp(a) decreased progressively with increasing SSPG tertiles. The associations between the aforementioned lipoprotein measurements and SSPG tertiles remained statistically significant when adjusted for age, sex, and BMI. Pairwise comparisons of the significant associations revealed that the most insulin-resistant group (ter-tile 3) differed from the most insulin-sensitive (tertile 1) in all cases. Additionally, differences between tertile 3 and tertile 2 were significant for total HDL and its subclasses, TG, VLDL1+2, LDL2, LDL4, and Lp(a).
The results in Table 4 indicate that the results were quite different when considering categories of OSA. Comparison of lipoprotein measurements across OSA categories yielded no significant associations or trends. Thus, there is no evidence that measures of increasing severity of OSA had any relationship with lipoprotein metabolism.
Finally, LDL density patterns were compared by SSPG ter-tiles and OSA categories (Table 5). Whereas the most insulin-sensitive group (SSPG tertile 1) was predominantly (80%) characterized by LDL pattern A, the majority (64%) of the most insulin-resistant group (SSPG tertile 3) was characterized by LDL pattern B. However, there was no appreciable association between LDL pattern and OSA category.
Comparison of LDL density patterns by tertiles of SSPG concentrations and OSA categories.
Comparison of LDL density patterns by tertiles of SSPG concentrations and OSA categories.
DISCUSSION
Our findings highlight the heterogeneity of lipid/lipoprotein metabolism in individuals with OSA, in the absence of diabetes or clinically evident CVD. Consistent with our hypothesis, lipoprotein abnormalities were strongly correlated with insulin resistance, including positive associations with TG, VLDL, lipoprotein subclasses VLDL1+2 and LDL4, and negative associations with HDL-C and its subclasses (HDL2 and HDL3), LDL2, and apoA1. These relationships were also borne out in comparison of the most insulin-resistant with the insulin-sensitive tertile. Thus, among patients with OSA, the most insulin-resistant individuals have the most clinically adverse lipid/lipoprotein profile, and thereby are at greatest risk for increased CVD events. In contrast, neither increasing severity of OSA nor measures of overnight hypoxia had any appreciable impact on lipoprotein particle sizes or subclasses. Taken together, these results suggest that the mechanism by which dyslipidemia may increase CVD risk among individuals with OSA is via insulin resistance, rather than as a consequence of OSA per se.
The present findings are consistent with published literature regarding the atherogenic dyslipidemia of insulin resistance in apparently healthy individuals,20–22 and extend their applicability to individuals with OSA. Thus, the relationship between insulin resistance and lipoprotein abnormalities appears to be similar among nondiabetic individuals with and without OSA. In addition to the characteristic pattern of plasma TG and HDL-C, TG-rich VLDL and its subclasses VLDL1+2 varied with insulin action even after adjustment for covariates. There was also a borderline significant trend of increasing VLDL3 (the smaller and denser cholesterol-rich VLDL subfraction) across SSPG tertiles. Small, dense LDL was also increased with insulin resistance, as reflected by the relative enrichment of LDL4 particles accompanied by a reciprocal reduction in LDL2 in the insulin-resistant third. Accordingly, there was a preponderance of LDL density pattern B in this group, whereas LDL density pattern A was more frequent among the insulin-sensitive. These details underscore the potential value of lipoprotein measurements beyond standard lipid panels, given that these variations were not reflected in total LDL, the primary lipoprotein CVD risk factor. Not surprisingly, both HDL subclasses—especially HDL2, which is felt to be the most beneficial HDL particle in terms of CVD protection—demonstrated robust inverse relationships with insulin resistance. These variations in HDL levels were accompanied by corresponding changes in apoA1. Interestingly, although Lp(a) is an independent risk factor for CVD events,23 Lp(a) was decreased in the most insulin-resistant group in the present study. Similar results were previously observed among South Asians,24 but not among whites.25 Our cohort was predominantly white, and distribution of race across SSPG tertiles did not differ. While Lp(a) has a strong genetic basis, it is possible that other epigenetic factors may underlie the association. It is important to re-emphasize that the lipoprotein abnormalities described were detected in patients who—while diagnosed with OSA—did not have type 2 diabetes or clinically evident CVD. Identification of at-risk individuals prior to disease onset would allow initiation of treatment to reduce or prevent CVD events.
It is striking that neither OSA severity, nor any of our hypoxia measures had any appreciable effect on lipoprotein concentrations. Moreover, given the lack of association between SSPG and AHI, any indirect effect of OSA via insulin resistance on lipoprotein abnormalities seems unlikely. These results are in contrast to prior investigations,3,10–13,26 most of which have focused on the association of OSA with conventional lipid measurements, of which evidence has been strongest for a relationship with TG3,11 or HDL-C.3,10,11,13,26 Few studies have examined lipoprotein concentrations beyond conventional lipid panels.12,27 Luyster et al. reported that moderate-to-severe OSA was associated with LDL pattern B, but only among individuals with normal waist circumference.12 On the other hand, Sopkova et al. did not detect a relationship between OSA and LDL size or subclass.27 In order to interpret our findings in context of these studies, it is necessary to discuss 3 key differences in study design and population: (1) Individuals without OSA were frequently combined in the analysis of patients who had OSA.3,10,11,13,26 For example, in a subset of individuals from the Sleep Heart Health Study, HDL-C was negatively and TG positively associated with increasing OSA severity.3 However, more than half the cohort did not have OSA, and the distribution of RDI was skewed towards lower values (median respiratory disturbance index was 4.0). (2) Evaluation of insulin resistance was either not performed or accounted for3,11,13 and may have been a confounding variable in many studies, particularly since patients with diabetes were typically not excluded.3,10–13,26 As demonstrated in the present study, individuals can also be quite insulin-resistant even in the absence of having clinical diabetes. To that end, only one other study tried to separate the effects of OSA and insulin resistance (using metabolic syndrome as a surrogate marker) on plasma lipoprotein concentrations, specifically LDL size and subclasses.27 In comparison of patients with and without metabolic syndrome who had OSA, only the presence of metabolic syndrome (but not severity of OSA) remained an independent predictor for LDL size. These results are consistent with our findings. (3) Several studies used portable unattended monitors to obtain sleep measurements,3,12,13 which although correlate with in-laboratory PSG, may not be as reliable especially among patients with mild or moderate OSA.28 In sum, the studies referenced here have not been equipped to address the question we posed, that is, to what degree does insulin resistance or OSA severity contribute to lipoprotein abnormalities in OSA? It seems apparent from our results that insulin resistance is the culprit.
One can approach this question another way, by asking whether conventional treatment for OSA with positive airway pressure (PAP) therapy influences lipid profiles. Several10,29–32 but not all33–35 studies reported beneficial effects of PAP therapy on lipids, including increases in HDL-C10 or lowering of total cholesterol,29,31,32 postprandial TG,30 and apoB.29,32 However, only one of the positive studies included an estimate of insulin resistance (homeostasis model of insulin resistance, HOMAIR),29 which declined along with total cholesterol, TG, and apoB after 8 weeks of PAP therapy. Therefore, in this instance it could be hypothesized that the improved lipid profile was related to a PAP-induced improvement in insulin sensitivity. Finally, there is only one published randomized controlled trial that evaluated the effect of PAP therapy on CVD events; there was not a statistically significant reduction in events, although it may have lacked adequate statistical power.36 Thus, to date PAP therapy has yet to demonstrate convincingly a clinically meaningful effect on lipoprotein concentrations and/or CVD outcomes.
There are several potential limitations of our study. The lipid analysis, while extensive, did not include qualitative assessment of lipid function. For example, impaired ability of HDL to inhibit LDL oxidation5 as well as increases in oxidized LDL particles4 have been reported in patients with OSA. Nonetheless, we are unaware of prior publications that have performed as comprehensive of a lipoprotein and metabolic analysis in the same OSA cohort, as in the present study. It is also possible that facets of sleep disordered breathing that we did not measure may have contributed to lipid abnormalities. In addition to AHI and hypoxia measures reported here, we also examined and did not find a relationship between lipoprotein concentrations and proportion of slow wave sleep (data not shown), which is felt to be the most restorative stage of sleep and of which disruptions may lead to metabolic derangements.37 Finally, failure to detect an association with OSA may be attributed to measurement error; inherent night-to-night variability of AHI has been reported, and it has been suggested that 2 consecutive nights of PSGs would improve measurement accuracy.38 We conducted additional analyses to address the possibility of measurement noise (supplemental material), of which the results demonstrated no significant relationships between AHI and lipid markers. This suggests that measurement error did not contribute to the lack of an association between AHI and plasma lipids.
In conclusion, the present study demonstrates that individuals with OSA, who are otherwise healthy without diabetes or CVD, represent a heterogeneous population with variable insulin sensitivity and plasma lipid/lipoprotein profiles. Insulin resistance was a strong independent predictor of lipid/lipoprotein abnormalities, including higher levels of pro-atherogenic TG-rich lipoprotein classes and small, dense LDL particles, and lower levels of protective HDL-C. By contrast, neither severity of OSA, nor degree of hypoxia had any appreciable impact on lipoprotein concentrations. These findings do not prove that the increased prevalence of CVD in subjects with OSA is secondary to the presence of insulin resistance and its associated atherogenic lipoprotein profile. On the other hand, if abnormalities of lipid/lipoprotein contribute to the enhanced atherogenesis in these patients with OSA, it is necessary to acknowledge the central role played by insulin resistance.
