Relationship between serum cholesterol and indices of erythrocytes and platelets in the US population.

Whereas dyslipidemia has been associated with leukocytosis, the relationship between serum cholesterol and other hematopoietic lineages is poorly defined. Erythrocytes and platelets, anucleate cells relegated to nonspecific diffusional exchange of cholesterol with serum, have been proposed to have a distinct relationship to cholesterol from leukocytes. We examined the relationship between serum cholesterol and circulating erythrocyte/platelet indices in 4,469 adult participants of the National Health and Nutrition Examination Survey (NHANES) 2005–2006. In linear regression analyses, serum non-high density lipoprotein-cholesterol (non-HDL-C) was positively associated with mean erythrocyte number, hematocrit, hemoglobin concentration, platelet count, and platelet crit independently of age, gender, race/ethnicity, smoking, body mass index, serum folate, and C-reactive protein. The magnitude of the relationship was most marked for platelets, with lowest versus highest non-HDL-C quartile subjects having geometric mean platelet counts of 258,000/μl versus 281,000/μl, respectively (adjusted model, P < 0.001 for trend). These associations persisted in a sensitivity analysis excluding several conditions that affect erythrocyte/platelet and/or serum cholesterol levels, and were also noted in an independent analysis of 5,318 participants from NHANES 2007–2008. As non-HDL-C, erythrocytes, and platelets all impact cardiovascular disease risk, there is a need for advancing understanding of the underlying interactions that govern levels of these three blood components.

NHANES oversampled persons of low income, elderly subjects ( у 60 years), African Americans, and Mexican Americans, among others. All study participants who completed the household interview were also invited to participate in the Health Examination Component that was conducted in the mobile examination center. Detailed description of the survey design and implementation may be found online . NHANES 2005NHANES -2006 was treated as the primary study population for our analyses, and NHANES 2007-2008 as a replication study population. All participants aged у 20 years who visited the NHANES mobile examination center, and for which data were available for total cholesterol (TC), HDL-C, erythrocyte count, hemoglobin concentration, hematocrit, and platelet count were included in our analyses.

Serum cholesterol and blood cell measurements
Serum TC and HDL-C were measured using a Roche Hitachi 717 or 912 (NHANES 2005(NHANES -2006 or a Roche Modular P chemistry analyzer (NHANES 2007(NHANES -2008. For TC, coupled enzymatic reactions were used involving cholesteryl ester hydrolase, cholesterol oxidase, and peroxidase, followed by phenazone absorbance detection. HDL-C measurement was by the Roche/Boehringer-Mannheim Diagnostics direct HDL method. For blood cell analysis, a Beckman Coulter MAXM (NHANES 2005(NHANES -2006 or Beckman Coulter HMX (NHANES 2007(NHANES -2008 was used. Erythrocyte count (RBCC ) was measured directly. Hemoglobin concentration was determined by absorbance found through photocurrent transmittance. Mean corpuscular volume (MCV) was derived from the erythrocyte histogram, and used in NHANES to compute hematocrit as: RBCC × MCV/10. Platelet count and mean platelet volume (MPV) were both derived from the platelet histogram. Platelet crit (%) was calculated as follows: (platelet count × MPV)/10,000.

Statistical analysis
To account for the complex sampling design used in NHANES and to assure unbiased variance estimates, all analyses were conducted using SAS Survey statistical software (Version 9.3, SAS, Cary, NC). Descriptive statistics were generated [means or percentages and associated standard errors (SEs)]. All blood parameters were assessed for normality. Hemoglobin and platelets were not normally distributed; thus, geometric means are presented. Linear regression analyses were run, assessing the association of the blood parameters with quartiles of HDL-C and non-HDL-C; least squares means of the blood parameters, by quartiles of cholesterol, and associated 95% confi dence intervals were generated from the regression coeffi cients and variance estimates. Cholesterol quartiles derived from NHANES 2005-2006 were used to analyze both surveys: 1 ) HDL-C [low ( р 41.83 mg/dl), medium (>41.83-51.38 mg/dl), high (>51. .93 mg/dl), very high(>62.93 mg/dl)]; and 2 ) non-HDL-C [low ( р 114.42 mg/dl), medium (>114.42-140.1 mg/dl), high window of cholesterol content, above which it, along with erythrocyte and platelet lifespan, is compromised ( 4,6,17 ).
In humans, multiple small clinical case series have identifi ed reduced serum cholesterol as a common fi nding in a variety of hemolytic anemias, and it has been proposed that this may occur through cholesterol consumption by avid erythropoiesis ( 11 ). Reports such as these suggest that, in addition to serum cholesterol impacting the population kinetics of erythrocytes and platelets, erythrocytes may reciprocally impact serum cholesterol levels. In support of this postulate, in vitro studies indicate that human erythrocytes act as a reservoir of cholesterol for serum lipoproteins, presumably because of their high nonspecifi c cholesterol loading capacity ( 18,19 ).
There have been very few studies of the relationship between serum cholesterol and indices of either erythrocytes or platelets in large human populations. While a few studies have shown positive correlations between serum cholesterol and either hematocrit or hemoglobin ( 20,21 ), others have found no such relationship ( 22 ). Findings for serum cholesterol and platelets have similarly been disparate ( 23,24 ). Given this, the nature of the relationships in humans between serum cholesterol and both erythrocytes and platelets remain undefi ned outside of disease extremes, as does the broader relevance of these relationships to public health. Given that hypercholesterolemia, erythrocytosis, and thrombocytosis, as well as the membrane cholesterol content of both erythrocytes and platelets are all risk factors for cardiovascular disease (25)(26)(27)(28); there is a need for advancing our understanding of the underlying relationships between serum cholesterol and erythrocyte and platelet lineages in humans.
The National Health and Nutrition Examination Survey (NHANES) is a biennial, cross-sectional population-based survey of the US population that includes measurements of erythrocyte and platelet indices, and serum cholesterol. We hypothesized that, in humans, as observed in rodent models, HDL-C would have an inverse association with abundance indices of erythrocytes (erythrocyte number, hematocrit, hemoglobin concentration) and platelets (platelet crit, platelet count). Given that non-HDL-C may possibly promote both production and destruction of both cell types, we had no clear a priori hypothesis regarding the relationship between non-HDL-C and erythrocyte/platelet indices.

Study population
Data were obtained from the NHANES 2005-2006 and NHANES 2007-2008, which used a complex multistage design to assess the health and nutritional status of the civilian noninstitutionalized US population. NHANES uses a randomization scheme to select US counties and, within them, households for survey each year, and thus by design minimizes the likelihood of resampling individuals across 2 year survey installments. To ensure adequate sample sizes of certain subgroups of the population, NHANES sample sizes and to account for the multiple blood parameters examined.

RESULTS
The characteristics of the NHANES 2005-2006 and  2007-2008 study populations are shown in Table 1 . The  NHANES 2005-2006 study population was approximately equally divided between genders, with a mean ± SE age of 46.8 ± 0.7 years, and was predominantly (72.4%) non-Hispanic White, with the remainder represented by non-Hispanic Black, Mexican American, and Other categories. A little under half of the subjects had fasted (i.e., у 9 h) at the time of laboratory analysis, and 13.3 ± 0.8% reported using a statin drug within the past 30 days. TC and HDL-C were measured in both fasting and nonfasting NHANES (>140.1-168.28 mg/dl), very high (>168.28 mg/dl)]. Cholesterol quartiles for NHANES 2007-2008 were similar, as follows: 1 ) HDL-C [low ( р 40.04 mg/dl), medium (>40.04-49.14 mg/dl), high (>49.14-60.61 mg/dl), very high (>60.61 mg/dl)]; and 2 ) non-HDL-C [low ( р 115.06 mg/dl), medium (>115.06-140.67 mg/dl), high (>140.67-170.08 mg/dl), very high (>170.08 mg/dl)]. Five sets of models were run: 1 ) unadjusted; 2 ) adjusted for age, race/ethnicity, gender, smoking, and BMI; 3 ) adjusted for age, race/ethnicity, gender, smoking, BMI, and fasting time; 4 ) adjusted for age, race/ethnicity, gender, smoking, BMI, and CRP; and 5 ) adjusted for age, race/ethnicity, gender, smoking, BMI, CRP, and fasting time. Because adjustment for fasting time and CRP did not affect the observed associations, only the crude models (Model 1) and the models adjusting for age, race/ethnicity, gender, smoking, and BMI (Model 2) are presented in the results section. A test for trend was used to statistically evaluate variations in the blood parameters across quartiles of cholesterol. A P value of р 0.001 was set as a cutoff for statistical signifi cance. This value was chosen given the large and BMI, this inverse association did not persist. By contrast, mean hematocrit increased across ascending quartiles of non-HDL-C, and this association persisted after controlling for age, race/ethnicity, gender, smoking status, and BMI. Additional adjustment for fasting time, erythrocyte folate, and CRP did not appreciably impact results (data not shown). No relationship was found between ascending categories of non-HDL-C and transferrin saturation or serum folate (data not shown). As shown in Table 2 , we repeated these analyses using data from NHANES 2007-2008, and while absolute mean values varied, the associations were replicated.
As shown in Table 3 , patterns of association of cholesterol with mean hemoglobin concentration, an alternate clinically used metric of erythrocyte mass, were similar to those observed for hematocrit. In adjusted analyses, there was no signifi cant association between HDL-C and hemoglobin, while there was a signifi cant increase in hemoglobin across increasing quartiles of non-HDL-C. As for hematocrit, the associations of hemoglobin to HDL-C and non- Consistent with the fi ndings for hematocrit and hemoglobin, a signifi cant increase was also observed in unadjusted and adjusted erythrocyte number across ascending categories of non-HDL-C in both surveys ( Table 4 ). By contrast, while the inverse relationships of HDL-C to hematocrit participants, whereas low density lipoprotein-cholesterol (LDL-C) was only measured in subjects who had been instructed to fast. Non-HDL-C (i.e., TC minus HDL-C), a composite measure of atherogenic LDL-C and very low density lipoprotein-cholesterol (VLDL-C), has comparable or better predictive value than LDL-C for cardiovascular disease ( 29,30 ), and both fasting and nonfasting non-HDL-C are predictive of cardiovascular disease ( 31 ). Thus, all primary analyses were based upon non-HDL-C (derived as TC minus HDL-C) and HDL-C measured in a combined fasting and nonfasting study population, as previously reported ( 32 ). The mean ± SE serum non-HDL-C in the 2005-2006 study population was 144.4 ± 0.9 mg/dl, and the mean serum HDL-C was 54.6 ± 0.3 mg/dl. Mean ± SE values for hematocrit, hemoglobin, platelet count, and platelet crit were 42.7 ± 0.2%, 14.4 ± 0.1 g/dl, 272.1 ± 2.0 × 10 3 / l, and 0.225 ± 0.002%, respectively. For the 2007-2008 study population, the mean ± SE serum non-HDL-C was 145.3 ± 0.6 mg/dl, and the mean serum HDL-C was 52.0 ± 0.5 mg/dl. Mean ± SE values for hematocrit, hemoglobin, platelet count, and platelet crit were 41.6 ± 0.2%, 14.3 ± 0.1 g/dl, 258.2 ± 1.3 × 10 3 / l, and 0.204 ± 0.002%, respectively. Table 2 presents mean hematocrit by quartiles of HDL-C and non-HDL-C. In the unadjusted model, mean hematocrit decreased as HDL-C increased. However, upon adjustment for age, race/ethnicity, gender, smoking status, 2005-2006 study population in which we excluded subjects (N = 1,375) with one or more of the following conditions known to impact serum cholesterol levels and/or blood cell counts: 1 ) history of liver disease; 2 ) history of cancer; 3 ) statin use within past 30 days; 4 ) current pregnancy; and 5 ) treatment for anemia within the past 3 months. Neither the magnitude nor the statistical significance of any of the relationships of HDL-C and non-HDL-C to hematocrit, hemoglobin, erythrocyte count, MCV, platelet crit, and platelet count was changed after these exclusions (supplementary Tables I-VI). Analysis of the lipoprotein strata in both surveys revealed a signifi cant decline in males across ascending HDL-C quartiles. Conversely, an increase in males across ascending non-HDL-C quartiles was observed in NHANES 2005-2006 (supplementary Table VII). Given this, in order to address possible persisting effects of gender upon our analysis, we also repeated the analyses within gender strata. The signifi cant increases in hematocrit, hemoglobin, erythrocyte count, and platelet count seen in both surveys across increasing quartiles of non-HDL-C were also observed in gender-stratifi ed analyses (i.e., within both males and females separately) (supplementary Ta- Increasing MCV was observed across ascending quartiles of HDL-C in both adjusted and unadjusted models; this was observed in both NHANES surveys ( Table 5 ). This relationship persisted after controlling for transferrin saturation and erythrocyte folate (data not shown). No relationship was observed between non-HDL-C and MCV.
Serum cholesterol-platelet relationships were next examined. As for hematocrit, a signifi cant increase in platelet crit was observed across ascending categories of non-HDL-C in unadjusted and adjusted models ( Table 6 ). Similar to the fi ndings for erythrocyte number, a signifi cant increase in mean platelet count was also observed across ascending categories of non-HDL-C in unadjusted and adjusted models ( Table 7 ). This relationship was seen in both NHANES surveys, and also persisted after adjustment for fasting time, CRP, and erythrocyte folate (data not shown). However, no consistent relationship was seen between either HDL-C or non-HDL-C and MPV ( Table 8 ).
In order to evaluate the robustness of these associations, we performed a sensitivity analysis of the NHANES HDL-C or non-HDL-C and MPV. Consistent fi ndings were observed in an independent study population from NHANES 2007-2008. Erythrocytes and platelets, anucleate blood cells with no significant capacity for cholesterol storage but with high capacity for diffusional exchange of cholesterol with plasma, have long been proposed to have a relationship to extracellular cholesterol that differs sub stantially from that of leukocytes ( 7 ). Coordinate ab normalities of erythrocytes and platelets have been observed in gene-targeted rodent models of dyslipidemia ( 4, 6, 13 ). Conversely, low serum cholesterol has been reported in several types of anemia and found to reverse upon treatment of anemia ( 11 ). However, the broader relevance of these relationships to human health has remained undefined.
Studies extending back over 30 years have elegantly shown that in vitro incubation of erythrocytes and platelets with cholesterol-enriched lipid dispersions or LDL leads to cholesterol incorporation into the cell membrane, and that cholesterol incorporation may regulate cell populations through impacting membrane stability ( 16,33 ). Membrane fl uidity is reduced and order increased in parallel with increases in membrane cholesterol/phospholipid ratio ( 16 ). This may explain clinical reports, generally consistent with the present one, that LDL-C, erythrocyte membrane stability, and hematocrit within females in the adjusted model in NHANES 2007-2008 (supplementary Tables XVI-XVII).
Differences in the percentage of fasting subjects were also noted across HDL-C and non-HDL-C quartiles, although most differences fell short of statistical significance (supplementary Table XVIII). In order to more confi dently exclude confounding by fasting, we evaluated the relationship of HDL-C and non-HDL-C to blood cell parameters among the subset of the study population that had fasted (i.e., у 9 h). As shown in supplementary Tables XIX-XXIV, we obtained very similar results to those obtained in the mixed fasting-nonfasting study population.

DISCUSSION
Examining US national data from NHANES 2005-2006, we report that serum non-HDL-C is positively related to abundance measures of both erythrocytes (erythrocyte number, hematocrit, hemoglobin concentration) and platelets (platelet crit, platelet count). This is independent of age, race/ethnicity, gender, smoking status, and BMI. Conversely, an inverse relationship was found between HDL-C and erythrocyte number. We also report that HDL-C is directly related to erythrocyte MCV, whereas no relationships were found between either including megaloblastic anemia, hereditary spherocytosis, sickle cell disease, aplastic anemia, glucose-6-phosphate defi ciency, and anemia associated with liver disease ( 11 ). These studies suggest that, in addition to effects of serum cholesterol on erythrocyte populations, erythrocyte kinetics may reciprocally affect cholesterol status. Remarkably, following treatment of several disparate anemias, ranging from B12/folate repletion for megaloblastic anemia, to splenectomy for hereditary spherocytosis, to red blood cell transfusion for sickle cell disease or aplastic anemia, an increase in serum cholesterol has been noted that parallels the correction in hematocrit ( 11,37 ). While this has led some investigators to hypothesize that serum cholesterol may be reduced during anemia by hemodilution, low serum cholesterol during hemolytic anemias has been attributed by others to consumption by avid erythropoiesis in the bone marrow ( 38,39 ). Additional hypotheses for anemia-associated hypocholesterolemia have included reduced cholesterol biosynthesis by the liver and increased cholesterol clearance by the reticuloendothelial system. There have been no studies, to our knowledge, that have investigated serum cholesterol levels during primary platelet disorders.
While it is plausible that disease extremes such as hemolytic anemia may serve to reveal some of the mechanisms that govern blood cell-cholesterol relationships during health, it is likely that additional mechanisms may be at work in large human populations. For example, malnutrition are all positively correlated ( 34 ). It has also been proposed that the stability of erythrocytes may be maximal within an optimal range of membrane fl uidity (and thus membrane cholesterol) ( 17 ). Thus, under cholesterolloading conditions that exceed the critical range, erythrocyte membrane stability is impaired, likely accounting for the increased erythrocyte osmotic fragility and spur cell and hemolytic anemias seen in experimental animals fed a high-cholesterol diet ( 10,12 ), and the erythrocyte membrane damage observed in hypercholesterolemic humans ( 9 ). Conversely, LDL-C reduction in hypercholesterolemic multiple sclerosis patients with statin therapy has been shown to increase erythrocyte stability ( 17 ).
As has been proposed for erythrocytes, it is possible that non-HDL-C may impact platelet number in part through effects on membrane stability. Interestingly, however, recent studies have suggested the potential for additional mechanisms. Thus, it has been reported that high LDL-C induces thrombocytosis in mice in part through delocalization of megakaryocytes in the bone marrow due to an altered gradient of stromal cell-derived factor-1 ( 35 ). It is also reported that cholesterol loading of megakaryocyte progenitors induces thrombocytosis through enhancing cell surface expression and activation of the thrombopoietin receptor, c-MPL ( 36 ).
Several case series have documented hypocholesterolemia as a common fi nding in a wide variety of anemias, cholesterol in patients with primary hypercholesterolemia, but did not report on erythrocyte abundance indices ( 44 ). The few studies of platelet number during cholesterol disorders have yielded disparate fi ndings. Pathansali, Smith, and Bath ( 24 ) reported no alteration in platelet count in eight patients with primary hypercholesterolemia. By contrast, in a Japanese study of 387 men and 550 women, platelet counts in women correlated negatively with HDL-C but were unrelated to non-HDL-C, whereas in men they were unrelated to HDL-C but were positively correlated to non-HDL-C ( 23 ). The latter result is consistent with our fi nding in NHANES. Last, of interest but of uncertain signifi cance, thrombocytopenia has been described in Tangier disease, a condition of low HDL-C due to mutation of the ATP binding cassette transporter A1 ( 45 ). Interestingly, we found that HDL-C was positively related to MCV, although apparently unrelated to MPV, suggesting important differences in the relationship of HDL to erythrocytes and platelets. HDL-C was also inversely related to erythrocyte count. These fi ndings are somewhat reminiscent of the SR-BI-null mouse, which, along with marked increases in HDL-C, is reported to have macrocytic anemia ( 6,13 ). In that setting, it is thought that increased erythrocyte membrane cholesterol deriving from HDL may both impair erythrocyte maturation ( 13 ) and promote hemolysis through effects on osmotic fragility has the potential to reduce both serum cholesterol and circulating numbers of erythrocytes (e.g., through iron defi ciency) and platelets (e.g., through folate or vitamin B12 defi ciency) ( 40 ). Similarly, infl ammation has complex effects, as it can be associated with anemia ( 41 ), altered serum cholesterol transport ( 42 ), and thrombocytosis ( 43 ). Arguing against an important role for nutritional factors in our fi ndings, the relationships persisted after adjustment for red blood cell folate and BMI, and no relationship was detected between non-HDL-C and either transferrin saturation (a marker of iron status) or serum folate. In addition, our fi nding that the relationships persisted after adjustment for CRP argues against confounding by infl ammation.
To date, there have been very few systematic evaluations of red cell abundance across varying levels of serum cholesterol in large human populations. In 1972, Böttiger and Carlson reported a positive correlation between serum cholesterol and hemoglobin in 2,458 nonanemic subjects ( 20 ); a similar correlation between serum cholesterol and hematocrit was noted by another group 20 years later ( 21 ). More recently, no relationship was found between serum cholesterol and erythrocyte number, hematocrit, hemoglobin concentration, or MCV ( 22 ). However, this study involved just 463 subjects, all of whom were elderly and from South Korea. Another small study documented increased erythrocyte membrane analyzed HDL-C and non-HDL-C, and not just TC (the sum of HDL-C and non-HDL-C), allowing us to identify distinct relationships for these different lipoprotein categories.
The effect size of the adjusted relationships of non-HDL-C to erythrocyte parameters is modest when considered in isolation. A somewhat more impressive relationship was found for platelets, where a nearly 10% difference in mean platelet count was observed between the lowest and highest non-HDL-C quartiles in the adjusted model. While these relationships are of uncertain clinical signifi cance, erythrocytosis, thrombocytosis, and hypercholesterolemia are all risk factors for thrombosis ( 50,51 ). If, as our data suggest, these variables track together in human subjects, it is possible that they may synergize in promoting cardiovascular disease. Indeed, emerging data suggest that erythrocytes, and erythrocyte membrane cholesterol, in particular, are independently associated with clinical instability in coronary artery disease patients ( 26,52 ). Interestingly, red blood cell distribution width, a strong prognostic marker in cardiovascular disease, is positively associated with erythrocyte membrane cholesterol ( 53 ). Similarly, platelet cholesterol overload correlates with platelet activation and coronary artery disease ( 28 ).
Atorvastatin is reported to decrease erythrocyte membrane cholesterol in human subjects ( 54 ). The lack of a and deformability ( 6 ). The analogy to our study may be imperfect, however, given that SR-BI-null mice have abnormally large HDL particles as well as increased free cholesterol, and their elevated MCV may in part derive from reticulocytosis ( 6,13 ).
Our study has limitations. Importantly, the cross-sectional design of the NHANES precludes inferences of causality between blood cell abundance and serum cholesterol. It is possible that serum cholesterol and erythrocyte/platelet levels, rather than causally impacting each other, may track together as biomarkers of a separate underlying condition. Our sensitivity analysis, nonetheless, indicates that the relationships do persist after several common conditions affecting blood cell levels and cholesterol are excluded. Although the ratio of free cholesterol:TC has been linked to platelet abnormalities in mouse and man ( 4,46 ), we were unable to analyze free cholesterol as it was not measured in the NHANES. Also, the ABO blood group type, recently shown in a genome-wide association study to be associated with LDL-C ( 47 ), was not determined in NHANES. Two additional variables that we did not analyze, but that have important effects upon erythrocyte membrane stability, are albumin concentration ( 48 ) and glucose concentration ( 49 ). Strengths of our analysis include the large size of our study population and our replication of results in a separate cohort. In addition, we change in our results after excluding statin-treated subjects may suggest that pharmacologic reduction of cholesterol is associated with similar effects on blood cell indices as other environmental or genetic infl uences on cholesterol present in the general population. Nonetheless, whether interventions upon serum cholesterol such as statins impact population kinetics of erythrocytes and/or platelets is an important question that may require investigation using a prospective study design. Finally, our fi ndings at least raise the possibility that non-HDL-C may act as a disease modifi er of primary disorders of erythrocytes and platelets. For example, elevated non-HDL-C could conceivably attenuate anemia from iron defi ciency and/or aggravate erythrocytosis in polycythemia vera. Conversely, our fi ndings raise the possibility that circulating erythrocyte and/or platelet counts could modify the expression of primary hypercholesterolemia. In closing, we report for the fi rst time that non-HDL-C is directly related to abundance measures of circulating erythrocytes and platelets in the US population, whereas HDL-C is directly related to MCV. Given that elevated erythrocytes, elevated platelets, and hypercholesterolemia are all established risk factors for coronary disease and that hypercholesterolemia impairs erythrocyte deformability ( 55 ) and activates platelets ( 56,57 ), our fi ndings suggest an important need for characterizing possible mechanisms by which serum cholesterol and the population kinetics of erythrocytes and platelets may impact one another.