Soluble CD14 and CD14 Variants, Other Inflammatory Markers, and Glucose Dysregulation in Older Adults: The Cardiovascular Health Study

The role of chronic low-grade inflammation in insulin resistance and subsequent development of diabetes has been well documented experimentally (1). Consistent with such experimental findings, epidemiologic studies have linked various inflammatory biomarkers, including CRP and interleukin-6 (IL-6), to risk of incident diabetes in women (2), middle-aged European cohorts (3), and middle-aged to older multiethnic cohorts (4). This also includes previous work from the Cardiovascular Health Study (CHS), which found a significant association in older adults between CRP and development of diabetes 3 to 4 years later (5). Additionally, a meta-analysis in 2010 combined data from 20 studies to conclude that an increased total white blood cell (WBC) count is associated with a higher risk of diabetes (6). The pathways implicated in such chronic inflammation and consequent dysregulation of glucose metabolism remain incompletely understood.

CD14, a pattern-recognition receptor that has specificity for lipopolysaccharides (LPSs) and other bacterial wall–derived components, is expressed primarily by myeloid cells but also by a range of nonmyeloid tissues (7). The receptor is anchored to the plasma membrane by a glycosylphosphatidylinositol moiety but is also detectable in plasma as a soluble form lacking this membrane-anchoring component (7). Such soluble (s)CD14 mainly arises from monocytes/macrophages following cleavage of the membrane receptor by proteases or from direct secretion of the anchor-free form by hepatocytes and adipocytes (8).

Experimental data suggest that the composition of the gut microbiome and translocation of bacterial components, namely LPSs, contribute importantly to obesity-related inflammation, so-called metabolic endotoxemia (9,10). As a key factor in signal transduction of LPS-related proinflammatory cascades by Toll-like receptors, CD14 has been shown to play a pivotal role in this process (11,12). sCD14 is an acute-phase reactant, but it can have differing effects on LPS-related inflammatory signaling (7). On one hand, sCD14 can act to potentiate LPS-related inflammation in cells both harboring and lacking membrane-bound CD14 (13). On the other hand, high levels of sCD14 can buffer LPS by promoting its transfer to lipoprotein particles, preventing binding of surface CD14 on monocytes/macrophages (14). Previous work by our group in CHS has shown that elevated sCD14 is associated with incident cardiovascular disease and mortality in older adults (15). The link between sCD14 level and insulin resistance has been examined in small clinical studies, which have suggested an inverse association (16) or one that differs by obesity status (17), but findings have been inconclusive. The relationship between sCD14 level and incident diabetes has not been examined to date. In this study, we sought to determine the associations of sCD14 with abnormal glucose metabolism in a large cohort of older adults, both alone and in the context of other inflammatory biomarkers not previously evaluated in relation to long-term incidence of type 2 diabetes in this age- group. We also undertook an instrumental variable analysis using Mendelian randomization to investigate whether observed associations of sCD14 with insulin resistance may have a causal basis.

The design and rationale of CHS have been described before (18). Briefly, CHS is a prospective population-based cohort study of older adult men and women recruited from four U.S. field centers: Forsyth County, NC; Sacramento County, CA; Washington County, MD; and Pittsburgh, PA. Candidates were randomly sampled from Medicare eligibility lists, and those who were institutionalized, wheelchair-bound, in hospice, receiving treatment for cancer, not expected to remain in the area for the next 3 years, or unable to give informed consent were excluded. A total of 5,201 participants (original cohort) were recruited in 1988–1989 from random samples of Medicare eligibility lists. An additional 687 primarily black participants were recruited in 1992–1993 (supplemental cohort) for a total cohort of 5,888 individuals. During in-person visits, participants underwent evaluation for demographic and lifestyle factors, physical examination, and fasting blood collection for laboratory measurement and storage using standardized protocols, as previously described (18,19). After excluding individuals with treated diabetes (type 2, in this age range) at baseline, we had a data set of 5,380 participants. Of those, 5,296 individuals had measures of both fasting glucose and fasting insulin at baseline, who were included in analyses of insulin resistance measures. For these analyses, we had baseline measures of sCD14 in 4,999 participants and of IL-6, CRP, and WBC count in 4,927, 5,302, and 5,305 participants, respectively. After further exclusion of 515 individuals with prevalent type 2 diabetes at baseline by fasting glucose (≥126 mg/dL), random glucose (≥200 mg/dL), or antihyperglycemic medication, our study sample comprised 4,865 participants for assessment of the association with incident type 2 diabetes (hereafter referred to as diabetes). For this set of analyses, we had baseline measures of sCD14 in 4,558 participants and of IL-6, CRP, and WBC count in 4,502, 4,836, and 4,850 participants, respectively. For instrumental variable analyses of sCD14 in relation to insulin resistance measures using Mendelian randomization, which were stratified by race, genotypic data were available in 3,387 white and 422 African American participants.

sCD14 was measured in baseline plasma specimens using a commercial ELISA (R&D Systems, Minneapolis, MN) with an interassay coefficient of variation of 5.3–12.4% (20). CRP was measured using an ultra-sensitive ELISA developed at the CHS Central Laboratory at the University of Vermont with an interassay coefficient of variation of 5.5% (20). IL-6 was measured using commercial ELISA kits (Quantikine IL-6; R&D Systems, Minneapolis, MN) with an interassay coefficient of variation of 6.3%. Complete blood counts (including WBC counts) were measured at each of the participating clinical centers using the following instruments: Coulter Stack S cell counter (Coulter, Inc., Hialeah, FL) (University of Pittsburgh, Pittsburgh, PA, and University of California, Davis, Davis, CA) and the Sysmex NE8000 counter (Toa Electronics, Inc., Chicago, IL) (Wake Forest University, Winston-Salem, NC, and Johns Hopkins University, Baltimore, MD).

Genotyping done in this cohort has been described in detail (15). Fine-mapping of the CD14 locus was performed in 3,950 European American (3,660 with sCD14 measured) and 792 African American (683 with sCD14 measured) participants who were genotyped using the custom gene-centric IBCv2 genotyping array that contains high single nucleotide polymorphism (SNP) marker density and linkage disequilibrium coverage for various cardiovascular, metabolic, and inflammation-related genes, including CD14 (21). IBCv2 array used a “cosmopolitan” tagging approach to capture the genetic diversity across candidate genes in the multiple ethnic populations represented in the HapMap, including both Europeans and West Africans. Tag SNPs were selected to capture known variation with minor allele frequency >0.02 and an r² of at least 0.8 in HapMap populations. The top SNP in whites was identified as rs5744455, while that in African Americans was identified as rs5744451.

Fasting glucose and insulin values were analyzed using standard enzymatic methods at the University of Vermont Central Laboratory. Measurements were performed in serum at baseline, as well as in all but one of the follow-up examinations, which used EDTA plasma. Specimens were frozen at −70°C <1 h after collection (22). Values were harmonized across examinations, as described previously (23). Two measures of insulin resistance were selected: 1) quantitative insulin-sensitivity check index (QUICKI), which is derived from a formula using fasting insulin and fasting glucose and correlates well with glucose clamp studies (24,25), and 2) HOMA 2 index of insulin resistance (HOMA2IR), which improves upon the original HOMA index by using a computer program (26,27).

Participants were followed up through year 2011 for incident diabetes, defined as a fasting glucose ≥126 mg/dL, random glucose ≥200 mg/dL, or use of antihyperglycemic medication. As previously described, after the 1989–1990 examination, glucose measurements were obtained in 1992–1993, 1994–1995, 1996–1997, 1998–1999, and, in a subset, 2005–2006 (23). All were on fasting samples, with the exception of 1994–1995, which were random.

Hypertension was defined by systolic and diastolic cutoffs of 140 and 90 mmHg, respectively, or by self-report and antihypertensive therapy. Trained personnel performed anthropometric measurements in standardized fashion. Measurement of physical activity in kilocalories has been previously reported (23). Heavy alcohol use was defined as consumption of >14 drinks/week in men or >7 drinks/week in women. Prevalent atherosclerotic cardiovascular disease (ASCVD) consisted of coronary heart disease, stroke or transient ischemic attack, and peripheral arterial disease. Along with heart failure (HF), prevalent ASCVD components were ascertained at the 1989–1990 and 1992–1993 examinations and in the intervening period through a combination of CHS questionnaires, medical record review, and physician confirmation (19,28). Additionally, fasting baseline samples were used to measure creatinine and lipids (22). Estimated glomerular filtration rate (eGFR) was calculated using the Chronic Kidney Disease Epidemiology equation (29).

Baseline characteristics are presented as median (interquartile range) for continuous variables and count (percent) for categorical variables, both in the cohort overall and by quartiles of sCD14 level. We tested for trend across increasing quartiles of sCD14 by using contrasts in generalized linear models for continuous variables and the Cochran-Armitage test for categorical variables. The four biomarkers showed positive skew and were converted to base-2 logarithm to normalize their distributions and to evaluate their associations with outcomes uniformly per doubling of their levels. Pearson correlation coefficients were calculated between log-transformed inflammatory biomarkers. Using minimally and fully adjusted linear regression models, we tested the associations between the biomarkers and QUICKI and HOMA2IR. To avoid disproportionate influence of extreme upper values for HOMA2IR, this measure was winsorized at the 99th percentile. The initial model adjusted for age, sex, and race, while the full model additionally adjusted for BMI, systolic blood pressure, use of antihypertensive medication, smoking status, heavy alcohol use, physical activity, estrogen use, prevalent ASCVD, prevalent HF, and eGFR. A subsequent model examined whether additional adjustment for LDL cholesterol influenced the findings; as triglycerides and HDL cholesterol are influenced by insulin resistance, they were considered part of the outcome and therefore not included for adjustment. Additional sensitivity analyses evaluated the impact of further adjustment for use of statins, oral corticosteroids, and nonsteroidal anti-inflammatory drugs. In the second part of the analyses, participants with prevalent diabetes at baseline based on fasting glucose, random glucose, and medication use were excluded. Using Cox proportional hazards models with the same covariates for the initial and full models, we tested for association between the four biomarkers and incident diabetes. In both linear regression and Cox regression analyses, we assessed the functional form of the relationship between inflammatory markers and outcome measures using generalized additive model (GAM) plots. For analyses of insulin resistance and incident diabetes, we tested for interaction by age, sex, race, BMI, prevalent ASCVD, and prevalent HF by including appropriate cross-product terms in the full models. Last, we also tested for associations of the biomarkers with QUICKI, HOMA2IR, and incident diabetes independent of each other by including the four biomarkers simultaneously in the full model.

For the Mendelian randomization analysis, we used the two-stage least-squares method using sCD14 as the endogenous regressor; SNPs rs5744455, rs778584, and rs4914 in whites and rs5744451 in African Americans as instrumental variables; and QUICKI and HOMA2IR as the continuous outcomes.

We used SAS 9.4 (SAS Institute, Cary, NC) and Stata 15 (StataCorp, College Station, TX) for all analyses. A two-tailed P value <0.05 was considered statistically significant.

Table 1 describes the baseline characteristics of the cohort both overall and divided by quartiles of sCD14 level. There were decreasing proportions of men, blacks, estrogen users (among women), and former smokers, along with declining levels of BMI, physical activity, and eGFR, with increasing quartiles of sCD14. By contrast, age, LDL cholesterol, and triglycerides tended to be higher at increasing quartiles of sCD14, as were proportions of heavy alcohol users, participants using oral corticosteroids, and those with hypertension, prevalent ASCVD, and prevalent HF. Levels of QUICKI and HOMA2IR did not differ across sCD14 quartiles, but those for IL-6, CRP, and WBC count tended to increase with increasing sCD14 quartile categories.

sCD14 was positively correlated with IL-6, CRP, and WBC count, with Pearson ρ values of 0.17, 0.24, and 0.11, respectively (all P < 0.001). IL-6 was also positively correlated with CRP and WBC count, with Pearson ρ values being 0.50 and 0.30, respectively (all P < 0.001). Meanwhile, CRP was positively correlated with WBC count, with a Pearson ρ of 0.27 (P < 0.001).

Table 2 describes the associations of sCD14 and the other inflammatory biomarkers in minimally and fully adjusted linear regression models in relation to QUICKI and HOMA2IR, as well as Cox proportional hazards models in relation to incident diabetes. All four biomarkers were significantly inversely related with QUICKI (decreasing QUICKI values reflect higher insulin resistance) after adjustment for demographic variables. These inverse associations were attenuated but remained significant after full adjustment. The magnitude of these associations was comparable for sCD14, IL-6, and CRP, but not for WBC count, which showed a numerically stronger relationship with QUICKI. In the case of HOMA2IR, significant positive associations were again observed for all biomarkers. These associations were attenuated after full adjustment for all biomarkers except for sCD14; all of the relationships remained significant. Once again, the numerically strongest association was observed for WBC count, followed by numerically similar associations for sCD14 and IL-6, and a numerically weaker association for CRP. Additional adjustment for LDL cholesterol did not meaningfully alter the observed associations. GAM plots showed that associations of inflammatory biomarkers with QUICKI and HOMA2IR were consistent with linear relationships (data not shown).

To test for the association of the four biomarkers with QUICKI and HOMA2IR independent of each other, we next tested the full model with the four biomarkers included together. As shown in Fig. 1A and B, after such adjustment, sCD14 ceased to have a statistically significant association with QUICKI and HOMA2IR, while the remaining three biomarkers continued to have a significant association. CRP showed attenuated associations, while WBC count seemed to retain the strongest numerical association with either outcome.

Over a median follow-up of 11.6 years, we documented 466 cases of incident diabetes. As shown in Table 2, sCD14 was not significantly associated with new-onset diabetes in minimally or fully adjusted models. By contrast, IL-6, CRP, and WBC count were significantly positively associated with incident diabetes after adjustment for demographic factors. These associations were attenuated after full adjustment for potential confounding variables but remained statistically significant. GAM plots were consistent with linear relationships. Once again, the strongest numerical association noted was for WBC count. Further adjustment for LDL cholesterol did not materially alter the findings. When all biomarkers were additionally adjusted for each other, the relationship of WBC count with diabetes did not change substantively, whereas the associations for IL-6 and CRP became nonsignificant (Fig. 1C).

In the sensitivity analysis undertaking adjustment for use of statins, oral steroids, or nonsteroidal anti-inflammatory drugs in addition to covariates in the main model (model 2), there were no meaningful changes in risk estimates for insulin resistance or incident diabetes observed for the four inflammatory markers (Supplementary Tables 1 and 2). Assessment for interaction by age, sex, race, BMI, prevalent ASCVD, and prevalent HF did not reveal consistent evidence of effect modification across outcomes for any of the four inflammatory biomarkers (data not shown).

Table 3 shows the results of instrumental variable analysis using Mendelian randomization of sCD14 in relation to QUICKI and HOMA2IR, stratified by race. In the two-stage least-squares approach, F values for all gene variants in whites and African Americans were >10, attesting to the SNPs serving as good instruments. Second-stage P values for sCD14 for whites with SNPs rs5744455, rs778584, and rs4914 were nonsignificant for both QUICKI and HOMA2IR. Similarly, for African Americans, SNP rs5744451 exhibited second-stage P values that were nonsignificant for QUICKI and HOMA2IR.

In this population-based cohort of older adults, we evaluated the association of plasma sCD14, alongside the three inflammatory markers, IL-6, CRP, and WBC count, with insulin resistance and incident diabetes. Like IL-6, CRP, and WBC count, sCD14 was cross-sectionally associated with insulin resistance as determined by QUICKI and HOMA2IR after adjustment for potential confounders. However, Mendelian randomization analysis did not support a causal basis for the observed association between sCD14 and QUICKI or HOMA2IR. Upon adjustment for one another, IL-6, CRP, and WBC count remained associated with these two insulin resistance measures, but sCD14 did not. sCD14 was also not significantly associated with incident diabetes, as opposed to IL-6, CRP, and WBC count, all of which showed significant associations with this outcome. When all four biomarkers were entered simultaneously in the model, only WBC count remained significantly associated with incident diabetes.

The concept that insulin resistance and diabetes have an inflammatory basis is supported by laboratory and, increasingly, clinical studies. Expansion of the adipose compartment from caloric excess subjects adipocytes to metabolic and ischemic stress, resulting in macrophage infiltration, inflammation, and adipocyte insulin resistance (30). The resultant lipid excess promotes ectopic fat deposition, inflammation, and insulin resistance in liver and skeletal muscle, further impairing glucose disposal (31). Compensatory hypersecretion of insulin by the pancreas maintains glucose homeostasis until β-cell failure ultimately supervenes, sometimes fostered by pancreatic islet inflammation of its own (32), eventuating in hyperglycemia.

However, obesity has also been linked to perturbations in the gut microbiome (dysbiosis) and intestinal barrier function (33). Attendant low-grade increases in circulating LPSs have been shown in animal models of obesity to promote adipose tissue inflammation and glucose dysregulation (9,10). Such so-called metabolic endotoxemia has been supported by some, but not all, clinical studies (34).

In this context, the multifunctional pattern-recognition receptor CD14, which mediates LPS’s potent proinflammatory actions, has drawn interest, demonstrating a role in adipose tissue inflammation and insulin resistance in experimental studies (35). Clinical studies have in turn reported an inverse association between sCD14 and insulin resistance in men only after adjustment for triglycerides (16) or disparate associations of sCD14 with insulin resistance in nonobese (positive) and morbidly obese (inverse) individuals (17), but these studies have been modest in size. Hence, whether CD14 is a central player in the intestinal–adipose tissue–dysmetabolic axis in humans has remained unclear.

In this study, we found sCD14 to be associated with insulin resistance independent of multiple covariates, including postmenopausal estrogen replacement, with which it was inversely associated, but we did not detect a similar association with incident diabetes. Moreover, the observed relationship with insulin resistance was lost upon concurrent adjustment for other inflammatory markers, and there was no support for a causal association from instrumental variable analysis using Mendelian randomization. Our findings therefore would not appear to support a preeminent role for CD14 as a lynchpin of dysbiosis, inflammation, and glucose dysregulation. This may owe to the fact that sCD14 is an acute-phase reactant (36), reflecting varied inflammatory stimuli that may not set it apart from other inflammatory markers, or not itself acting as a proximate driver of key pathogenic pathways, as relates to insulin resistance. The lack of a predominant role seen for sCD14 could also be attributable to sCD14’s duality of effects. sCD14 may serve to buffer circulating LPS’s proinflammatory consequences by promoting its sequestration to triglyceride-rich lipoproteins (14), but can also act to induce proinflammatory responses in cells lacking (37), and even expressing (13), membrane-bound CD14. The extent to which sCD14’s protective and harmful functions are determined by tissue provenance, cellular and molecular milieu, stoichiometry of sCD14 and LPS levels, and other factors is not well defined, but such offsetting influences complicate assessment of the molecule’s pathophysiologic contributions using epidemiologic approaches. Thus, the present findings cannot exclude an important pathophysiologic role for LPS-CD14 in metabolic dysregulation not discernible from the evaluation of circulating sCD14 levels.

The other notable finding of the current study is the emergence of WBC count as the most robust inflammatory measure associated with glucose dysregulation, showing the strongest association with insulin resistance, and remaining the single marker related to long-term incidence of diabetes, after accounting for all others. Previous studies have shown prospective associations between various inflammatory markers and diabetes, including CRP, IL-6, and WBC count (2–4). Of note, in a previous analysis of CHS assessing new-onset diabetes in the middle term, CRP but not WBC count showed a significant relationship with this outcome, but the number of events was small (15). To our knowledge, ours is the first study to document the stronger and independent association of WBC count with incidence of diabetes after simultaneous adjustment for other commonly studied markers. That WBC count superseded CRP, and IL-6, with which it is related, supports its role as an integrative measure of systemic inflammation in ubiquitious use in clinical practice. Apart from confirming the association between inflammation and glucose dysregulation among older adults, the segment of the population at highest risk of diabetes (38), this finding suggests that WBC count could potentially have a role in stratification of diabetes risk. As anti-inflammatory therapies are tested for prevention of diabetes or its complications (39), the question of whether WBC count may be useful in identifying individuals for such interventions will merit scrutiny. Likewise, while the current investigation is not focused on prediction, whether addition of WBC count could enhance prediction of diabetes risk in elders will warrant separate study.

The current study has several strengths, including its large and well-characterized population-based sample of older adults with long-term follow-up and regular measures of glycemia, its concurrent assessment of multiple inflammatory markers, and availability of CD14 sequencing for Mendelian randomization. Various weaknesses must also be acknowledged. Neither hemoglobin A_1c nor postload glucose was regularly available in CHS, which may have led to misclassification of baseline or follow-up diabetes status. Glucose was mostly measured in serum and not in sodium fluoride or citrated tubes. Specimens were frozen soon after collection, however, which should have minimized glucose consumption by blood-cell glycolysis and attendant nondifferential bias. The hyperinsulinemic-euglycemic clamp method, the gold standard to measure insulin sensitivity, was not obtained in CHS, yet QUICKI and HOMA2IR have been shown to yield good estimates of insulin resistance for use in large-scale epidemiologic studies (24–27). Concurrent measurement of LPS was not available, which would have permitted more direct assessment of the contribution of microbial translocation to circulating sCD14 and other inflammatory markers, although accurate measurement of LPS poses challenges (34). WBC differential was not performed in CHS, and we could not evaluate the specific relationships of WBC subtypes with study outcomes, including neutrophils, for which the role in the pathogenesis of dysglycemia is being increasingly recognized (40). Information on recent infection or antibiotic use was not obtained at the baseline examination, which may have biased the associations under study toward the null hypothesis. Last, our findings come from a biracial, but predominantly European ancestry, population and are not necessarily generalizable to other ethnic groups.

In conclusion, in this sample of community-dwelling older adults, sCD14 was significantly associated with insulin resistance, but not with incident diabetes. The association of sCD14 with insulin resistance was not independent of other inflammatory markers, nor did it appear to have a causal basis on instrumental variable analysis using Mendelian randomization. By contrast, CRP, IL-6, and WBC count were significantly associated with incident diabetes, but this association persisted only for WBC count upon mutual adjustment, such that WBC count emerged as the strongest inflammatory marker of future diabetes risk. These findings do not support a key role for sCD14 in diabetes risk in elders, although complex actions of sCD14 preclude direct assessment as to the role of metabolic endotoxemia in this population. The foremost association for WBC count provides impetus for further study of this marker’s potential role in diabetes risk stratification and prediction.

Acknowledgments. A full list of principal CHS investigators and institutions can be found at CHS-NHLBI.org.

Funding. This research was supported by National Heart, Lung, and Blood Institute contracts HHSN268201200036C, HHSN268200800007C, HHSN268201800001C, N01HC55222, N01HC85079, N01HC85080, N01HC85081, N01HC85082, N01HC85083, and N01HC85086 and grants U01-HL-080295, U01-HL-130114, and R01-HL-071862, with additional contribution from the National Institute of Neurological Disorders and Stroke. Additional support was provided by National Institute on Aging grant R01-AG-023629. J.R.K. is supported by National Heart, Lung, and Blood Institute grant K24-HL-135413.

Duality of Interest. J.R.K. reports stock ownership in Bristol-Myers Squibb, Johnson & Johnson, Merck, and Pfizer. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. S.G.S. contributed in analyzing the data. S.G.S. and J.R.K. contributed in writing the manuscript. S.G.S., M.L.B., A.P.R., K.J.M., L.D., J.H.I., J.I.B., R.P.T., D.S., and J.R.K. contributed in reviewing and editing and suggesting intellectual contributions to the manuscript. J.R.K. contributed in formulating the hypothesis. S.G.S. and J.R.K. are the guarantors of this work and, as such, had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. The research abstract for this study was presented in poster form at the American Heart Association Scientific Sessions, New Orleans, LA, 12–16 November 2016.