Individual serum saturated fatty acids and markers of chronic subclinical inflammation: the Insulin Resistance Atherosclerosis Study[S]

Recent evidence has documented distinct effects of individual saturated FAs (SFAs) on cardiometabolic outcomes, with potential protective effects from odd- and very long-chain SFAs (VLSFAs). Cross-sectional and prospective associations of individual serum SFAs (12:0, 14:0, 15:0, 16:0, 18:0, 20:0, 22:0, and total SFA) with proinflammatory biomarkers and adiponectin were investigated in 555 adults from the IRAS. Principal component analysis (PCA) of proinflammatory markers yielded three clusters: principal component (PC) 1: fibrinogen, white cell count, C-reactive protein; PC 2: plasminogen activator inhibitor-1 (PAI-1), TNF-α, IL-18; PC 3: IL-6 and IL-8. Cross-sectional analyses on proinflammatory PCs and adiponectin, and prospective analyses on 5 year PAI-1 and fibrinogen concentrations were conducted with multiple regression. Total SFA and 16:0 were positively associated with PC 1 and PC 2, and negatively associated with adiponectin. The 14:0 was positively associated with PC 1 and negatively associated with adiponectin. In contrast, 15:0, 20:0, and 22:0 were negatively associated with PC 2, and 20:0 and 22:0 were positively associated with adiponectin. The 18:0 was negatively associated with PC 3. Prospectively, 15:0, 18:0, 20:0, and 22:0 were negatively associated with 5 year PAI-1 concentrations. The results demonstrate that individual SFAs have distinct roles in subclinical inflammation, highlighting the unique metabolic impacts of individual SFAs.

guidelines (1,4,5). A recent meta-analysis of 12 prospective observation studies found no association between SFA intake with all-cause mortality, coronary heart disease (CHD), CHD mortality, and ischemic stroke (4). Similarly, total circulating SFA was also not significantly associated with coronary outcomes in another meta-analysis of prospective cohorts (5).
Low-grade systemic inflammation is characterized by a 2-to 3-fold increase in circulating concentrations of cytokines, including TNF- and IL-6 (12), which in turn stimulates the synthesis of hepatic acute-phase proteins, such as fibrinogen, plasminogen activator inhibitor-1 (PAI-1), and C-reactive protein (CRP) (13). There is also a decreased concentration of adiponectin, a cytokine produced by adipose tissue, which has anti-inflammatory effects (12). Chronic elevation in proinflammatory markers is observed in people with established CVD and T2DM and is predictive of future cardiovascular events, as well as conversion to T2DM (12)(13)(14). Therefore, systemic inflammation has been hypothesized to be a link between obesity, atherosclerosis, and T2DM (12). Previous publications have demonstrated significant positive relationships between CRP, fibrinogen, and PAI-1 with the development of T2DM (14,15). Experimental literature has shown increased proinflammatory markers in cell lines and animal models with increased SFA exposure, with the majority of studies utilizing 16:0 (16). Similarly, studies of SFAs and inflammation in human cohorts have largely focused on even-chained SFAs and, thus, the role of other SFAs, in particular odd-chained SFAs and VLSFAs, on chronic subclinical inflammation is poorly understood (17)(18)(19)(20).
The objective of this study was, therefore, to investigate in a multi-ethnic observational study: 1) cross-sectional associations of individual serum SFAs with proinflammatory biomarkers and adiponectin; and 2) prospective associations of individual serum SFAs with PAI-1 and fibrinogen after 5 years of follow-up.

Population
A total of 1,624 participants were recruited in the IRAS. Participants without T2DM at baseline were included in the present analysis (n = 1,065). FA analyses were conducted on a subset of the population who had completed baseline and follow-up visits (n = 648). After further exclusions for participants with missing baseline inflammatory markers (n = 53), as well as CRP concentrations 10 mg/l (n = 40), the final sample size for this study was 555. There were no differences in sex between participants included in the study compared with those excluded. However, there were differences in age, ethnicity, and study center, where those included in the analysis were slightly younger and more likely to be non-Hispanic Whites (supplemental Table S1).

Variable measurements
Clinic examinations at baseline and follow-up were conducted during two clinic visits that were 1 week apart (21). Participants were asked to fast for 12 h, avoid alcohol and heavy exercise for 24 h, and avoid smoking the day of the visit. Glucose tolerance status, based on the 2006 World Health Organization criteria, was measured using a 2 h 75 g oral glucose tolerance test (Orange-dex; Custom Laboratories, Baltimore, MD) (22). Diabetes cases were defined by oral glucose tolerance test or current oral hypoglycemic or insulin use.
Inflammatory markers. Baseline and 5 year follow-up fasting plasma samples were centrifuged and stored at 70°C within 90 min of collection to measure a range of laboratory parameters, including biomarkers for this study (23). At follow-up, only PAI-1 and fibrinogen concentrations were measured. Analysis for the inflammatory markers, PAI-1, TNF-, CRP, and white cell count (WCC), were conducted at the Laboratory for Clinical Biochemistry Research at the University of Vermont (15). Fibrinogen was measured in citrated plasma with a modified clot-rate assay using the Diagnostica Stago ST4 instrument (American Bioproducts, Parsippany, NJ), with a coefficient of variation (CV) of 3.0% (24). PAI-1 in citrated plasma was measured using a two-site immunoassay, which was sensitive to free PAI-1, but not to PAI-1 complexed with tissue plasminogen activator (CV of 14%) (25). TNF- in citrated plasma was measured using a Quantikine HS Human TNF- immunoassay (R&D Systems, Minneapolis, MN) according to the manufacturer's instructions, with a CV of 8.4-11.8% (26). CRP was measured by in-house competitive immunoassay (antibodies and antigens from Calbiochem, La Jolla, CA) with an interassay CV of 8.9% (27). WCCs were measured using standard techniques (28). Adiponectin was measured using a sandwich enzyme-linked immunosorbent assay (CV 12.5%) (29). IL-6, IL-8, and IL-18 were quantified using molecular counting on the ZeptX™ system (Singulex) in a 10 l volume using a 384-microwell immunoassay format (30,31). Attached to each microwell were capture antibodies specific for an individual biomarker. After serum samples were added to the microwell, AlexaFluor 647-labeled secondary antibodies were added and bound. Fluorescence-labeled antibody complexes were chemically released from each well and pumped through a capillary flow system to detect laserinduced fluorescence. Each photon signal above the background level represents a labeled antibody molecule (30). A more detailed account of biomarker assays and detection technology has been published previously (30).
FA measurement. FAs were quantified from baseline serum samples (stored at 70°C) using methods previously described (32). Total lipids from serum were extracted in the presence of internal standards using Folch's method (33). The total lipid extract was trans-esterified in 1% sulfuric acid in methanol in a sealed vial under a nitrogen atmosphere at 100°C for 45 min. The samples were then neutralized with 6% potassium carbonate and the FA methyl esters were extracted with hexane and prepared for gas chromatography. Individual FAs were separated using capillary gas chromatography (Agilent Technologies model 6890) equipped with a 30 m DB-88MS capillary column (Agilent Technologies) and quantified using a flame-ionization detector. Each FA peak and area under the peak was compared with the internal standard control to quantify FA concentrations. A total of 35 serum FAs were analyzed and quantified and all results passed internal quality assurance and quality control processes. Seven SFAs were identified in this cohort and were analyzed in the current work: even-chained SFAs (12:0, 14:0, 16:0, and 18:0), odd-chained SFA (15:0), and VLSFAs (20:0 and 22:0). FAs in this work are expressed as a mole percentage (mol%) of total FAs, which normalize any increases or decreases in total serum lipid levels.
Other measurements. Weight, height, and waist circumference were measured in duplicate to the nearest 0.1 kg (weight) and 0.5 cm (height and waist circumference). A validated physical activity recall assessed total estimated energy expenditure (kilocalories per kilogram per week) as the sum of energy expenditure activities and energy expenditure from sleep during the past year (21). Age, ethnicity, and smoking status were self-reported and medical history was assessed using structured interviews. Usual dietary intake over the past year prior to baseline was assessed using a validated semi-quantitative 114-item food frequency questionnaire. The food frequency questionnaire was modified from the National Cancer Institute Health Habits and History Questionnaire to include ethnic and regional foods, and was validated in a subsample of 186 women (34). Nutrient and energy intakes were estimated using an expanded nutrient database (HHHQ-DIETSYS analysis software, version 3.0; National Cancer Institute, Bethesda, MD). Participants were asked to recall the usual frequency and portion sizes of foods and beverages over the past year. Servings per day were standardized to the medium serving size by multiplying the intake frequency with the portion size after applying a weighting factor (small = 0.5, medium = 1.0, large = 1.5).

Statistical analysis
Baseline characteristics overall and across tertiles (T1-T3) of increasing total SFA are presented. Normally distributed variables are presented as mean ± SD, nonnormally distributed variables are presented as medians with interquartile ranges, and categorical variables are presented as number and percent of participants in each tertile, with differences across tertiles tested using ANOVA, Kruskal-Wallis tests, and Chi-square tests, respectively. Principal component analysis (PCA) was used to identify proinflammatory variable clusters for the purpose of reducing the number of dependent variables in the analyses and optimizing the inflammatory marker signal. PCA is a variable reduction procedure that creates a smaller number of new independent variables, called principal components (PCs), which explain most of the variation in the original variables (35). All proinflammatory variables, except for fibrinogen, were log transformed prior to conducting PCA to achieve normality. PCA was conducted in SAS using PROC FACTOR with eight proinflammatory variable inputs: fibrinogen, PAI-1, TNF-, CRP, WCC, IL-6, IL-8, and IL-18. The number of PCs to be retained was determined based on eigenvalues >1 and the evaluation of scree plots. Generated PCs were interpreted using variable factor loadings of 0.40. A score for each pattern was calculated for each participant and these scores used as were the dependent variables in linear regression analysis. Adiponectin was chosen, a priori, to be analyzed as a separate dependent variable due to its documented anti-inflammatory properties (36).
Multiple linear regression analyses were conducted to evaluate the cross-sectional and prospective associations of total and individual SFAs with inflammatory markers. The regression models were sequentially adjusted for potential demographic, lifestyle, and dietary confounders: age, sex, ethnicity, center, physical activity, smoking status, alcohol intake, education, total energy intake, fruit and vegetable intake, red meat intake, soft drink intake, and fiber intake. In prospective analyses of 5 year PAI-1 and fibrinogen, baseline PAI-1 or fibrinogen was also added as a covariate. Effect modification analyses were conducted to assess potential interactions between the SFAs and variables of a priori interest, namely sex and ethnicity, in light of observations within this cohort and results of previous literature (18,37). Subgroup regression analyses were conducted when interaction terms were significant (P < 0.05). All analyses were conducted using SAS version 9.4 (SAS Institute Inc., Cary, NC). Statistical significance was set at P < 0.05.

RESULTS
Baseline characteristics of the participants overall and across tertiles of total SFA (mol% of total FAs) are presented in Table 1. Sex differed across tertiles, with a higher proportion of women in the highest tertile, and the highest proportion of males in the lowest tertile. Ethnicity differed significantly, with African Americans having the lowest total SFA compared with non-Hispanic Whites and Hispanics across all tertiles. BMI, waist circumference, and impaired glucose tolerance status prevalence significantly increased across tertiles. Smoking status, alcohol intake, and physical activity did not differ significantly; however, total energy intake increased significantly. All proinflammatory markers, except fibrinogen, increased while adiponectin decreased across tertiles.
PCA yielded three clusters derived from baseline proinflammatory marker concentrations. Factor loadings are presented in Table 2. Major inflammatory marker contributors to PC 1 were fibrinogen, WCC, and CRP. PC 2 was characterized by high positive loadings of PAI-1, TNF-, and IL-18, while PC 3 was characterized by IL-6 and IL-8.  In prospective analyses of baseline SFAs with 5 year PAI-1 and fibrinogen concentrations, 15:0, 18:0, 20:0, and 22:0 were inversely associated with PAI-1 at 5 years (Fig. 2). Total SFA, 12:0, 14:0, and 16:0 were not significantly associated with PAI-1. There was no association between total or individual SFAs with fibrinogen at follow-up.
Sex significantly modified the associations of 12:0 and 14:0 on PC 1 in cross-sectional analyses. In subgroup analyses by sex, 12:0 was inversely associated with PC 1 in females, while 14:0 was positively associated with PC 1 in males ( Table 3). In addition, sex significantly modified the associations of 15:0 with adiponectin and total SFA with fibrinogen prospectively; however, there were no marked differences in the associations in subgroup analyses by sex (Tables 4, 5). Moreover, ethnicity was a significant effect modifier for 20:0 and 22:0 on PC 2. Ethnicity significantly modified associations of VLSFA with PC2. In Hispanics, 20:0 and 22:0 were inversely associated with PC 2, while 22:0 was inversely associated with PC 2 in non-Hispanic Whites ( Table 6). In addition, sensitivity analyses excluding participants with morbid obesity (n = 20) or CHD (n = 23) at baseline did not change results appreciably (data not shown).

DISCUSSION
In the IRAS cohort, we found that among the evenchained SFAs, shorter chain lengths (14:0 and 16:0) were PC loading is a measure of the contribution of each (baseline) proinflammatory marker to each PC pattern. The number of PCs retained were based on eigenvalues >1 and scree plots. a Loadings >0.40 indicate the proinflammatory markers that define the PC. Covariates: age, sex, ethnicity, center, physical activity, smoking status, alcohol intake, education, total energy intake, fruit and vegetable intake, red meat intake, soft drink intake, and fiber intake. a P < 0.05, b P < 0.01, c P < 0.001, d P < 0.0001. positively related to proinflammatory markers and negatively related to adiponectin, while longer even-chained SFAs (20:0 and 22:0) and odd-chained SFA (15:0) had inverse associations. Our findings are consistent with the emerging literature demonstrating different cardiometabolic effects of individual SFAs, specifically, with potential protective effects observed for odd-chained SFAs and VLSFAs, and detrimental effects of even-chained SFAs (5,11). Our results extend this literature in a number of important ways, including measuring a variety of circulating SFAs and inflammatory markers in this cohort and, for the first time, showing heterogeneity in their associations both cross-sectionally and prospectively in a multi-ethnic cohort.
Our results are most likely explained either by a structure-function relationship between FA chain length and inflammatory processes or an increase in de novo lipogenesis in concert with the increased subclinical inflammation. A structure-function relationship based on chain-length would indicate an active role for SFAs in the triggering of subclinical inflammation, perhaps by activating toll-like receptor 4 (38) or through other lipotoxic mechanisms (39). Alternatively, an increase in de novo FA synthesis as a result of inflammation and metabolic stress would lead to production of shorter chain FAs (e.g., 14:0 and 16:0) by FA synthase and, thus, a relative enrichment in their concentrations. The FA 15:0 is unique among the FAs presented here in that: 1) it is not of mammalian origin and likely derives from the intake of ruminant animal products; and 2) despite being a shorter chain SFA, it is inversely associated with subclinical inflammation. This indicates a possible role for 15:0 or the foods containing 15:0 in the suppression of inflammatory processes. In this study, 15:0 was the only odd-chained SFA available, as 17:0 was used as an internal standard in the FA analysis.
In the current study, we used PCA to identify three proinflammatory clusters, not only to reduce the number of outcome variables in order to avoid multiple testing, but also to better understand the covariance structure or interactions between proinflammatory markers in this cohort, especially because diagnostic criteria for chronic systemic inflammation have yet to be established (40). Although the precise functions of individual inflammatory markers likely vary, this method better captures the overall inflammatory state, as it accounts for the combined effects of multiple markers, which might be underestimated in analyses evaluating single markers (41). The use of statistical methods to optimize multiple inflammatory markers is increasingly being used in the literature (41)(42)(43)(44).
To our knowledge, no other studies have determined the association between multiple individual SFAs with a variety of inflammatory markers. Previous epidemiological studies examining the association between circulating SFAs with inflammatory markers are limited in number and conflicting. In 264 elderly Swedish men and women, 18:0 was inversely correlated with CRP, while there was no significant correlation of 14:0 and 16:0 with markers of inflammation and endothelial function cross-sectionally (17). Among males in the EPIPorto study, 12:0 and 14:0 were positively associated with hs-CRP (18), although there was no association with adiponectin or leptin after multivariate adjustment. The lack of association in females may be due to their higher accumulation of total central fat compared with men, which the authors speculate to be the most important source of inflammation, or from the effects of sex hormones on circulating levels of inflammatory markers.
In terms of odd-chained SFAs, in a nested case-control study, the sum of 15:0 and 17:0 in serum phospholipids was inversely correlated with PAI-1 and leptin (45). On the other hand, in the Multi-Ethnic Study of Atherosclerosis (MESA) (n = 2,837), both 14:0 and 15:0 were not crosssectionally associated with CRP after adjustment for sociodemographic, lifestyle, and dietary factors (7). However, subgroup associations by sex or ethnicity were not analyzed in the MESA cohort. Further investigations on potential differences on the associations of individual SFAs with inflammatory markers by sex, ethnicity, and other subgroups are warranted. Heterogeneity in the current literature may Covariates: age, sex, ethnicity, center, physical activity, smoking status, alcohol intake, education, total energy intake, fruit and vegetable intake, red meat intake, soft drink intake, fiber intake, and baseline PAI-1. a P < 0.05, b P < 0.01. PC 1: fibrinogen, log WCC, and log CRP (all measured at baseline); Interaction term p values: 12:0 (P = 0.005); 14:0 (P = 0.047). Covariates: age, ethnicity, center, physical activity, smoking status, alcohol intake, education, total energy intake, fruit and vegetable intake, red meat intake, soft drink intake, and fiber intake. SFAs are expressed as mole fractions of total FA concentration. Standardized  coefficients are presented. Interaction term P value = 0.014; Covariates: age, ethnicity, center, physical activity, smoking status, alcohol intake, education, total energy intake, fruit and vegetable intake, red meat intake, soft drink intake, and fiber intake. 15:0 is expressed as mole fraction of total FA concentration. Standardized  coefficients are presented. Interaction term P = 0.040. Covariates: age, ethnicity, center, physical activity, smoking status, alcohol intake, education, total energy intake, fruit and vegetable intake, red meat intake, soft drink intake, and fiber intake. Total SFA is expressed as mole fraction of total FA concentration. Standardized  coefficients are presented. be due to differences in population characteristics, the FA pool in which SFAs were measured, covariates included in statistical analyses, and the SFAs and inflammatory markers evaluated.
Studies examining the mechanisms in which SFA affects concentrations of inflammatory markers have largely focused on even-chained SFAs, most notably 16:0. In vitro studies have demonstrated that high 16:0 exposure in adipocytes and myotubules increased TNF-, IL-10, and IL-6 mRNA and protein expression (16). Moreover, SFAs can stimulate the inflammatory signaling cascade by binding to and activating toll-like receptor 4 in adipocytes and macrophages, resulting in NFB and JNK activation and cytokine production (16,38). On the other hand, the mechanism underlying the observed protective effects of odd-chained SFAs and VLSFAs is not known, as these SFAs are largely understudied. Studies have yet to determine whether these SFAs are bioactive or are rather a marker of other potential protective components found in dairy, such as calcium, vitamin D, proteins, or other dairy specific FAs, such as c9,t11-CLA (37,46). VLSFAs are major components of sphingomyelins and ceramides (47). These FAs have been shown to be involved in myelin formation and hepatic homeostasis in mice (48) and to have opposing biological activities from ceramides containing 16:0 (49). Furthermore, VLSFAs have distinct metabolism through peroxisomal oxidation (50). However, the biological pathways that may link VLSFAs with cardiometabolic health are largely unknown, as there is a lack of mechanistic studies on the metabolism of these SFAs.
The main strength of this study is that the IRAS is a large well-characterized cohort with detailed and precise measurements of outcomes, exposures, and covariates. Multiple SFAs, adiponectin, and inflammatory markers were measured in serum, allowing a more comprehensive evaluation of associations between individual SFAs and the inflammatory phenotype. In addition, we optimized proinflammatory marker clustering using PCA. However, one limitation is that SFAs and most inflammatory markers, except PAI-1 and fibrinogen, were only measured at baseline. Therefore, assessing the temporality of many of the associations reported herein was not possible. Second, because circulating even-chained SFAs reflect both dietary and endogenous production, we were unable to distinguish any differences in effects based on the SFA source; thus, one must be cautious in extrapolating our results to dietary recommendations on SFA intake. Moreover, there is the potential for residual confounding in our analyses beyond the demographic, lifestyle, dietary, and metabolic covariates included in our models.
In conclusion, this study found that individual SFAs in serum affect subclinical inflammation differently, highlighting the unique metabolic impacts of individual SFAs. Lower circulating 15:0 and VLSFAs and higher even-chained SFAs were related to worsened subclinical inflammation status. Further studies on the sources (de-novo lipogenesis or dietary), metabolism, and mechanisms of individual SFAs as they relate to inflammatory and cardiometabolic pathways are warranted, especially for the understudied odd-chained SFAs and VLSFAs. While circulating levels of odd-chained SFAs may be altered through diet, the implications of these findings for intake recommendations for endogenously produced even-chained SFAs are less clear. Further studies on the biology of individual SFAs will help to elucidate the heterogeneity within SFA species and help to inform future public health recommendations on the role of SFA intake on cardiometabolic health. PC 2: log PAI-1, log TNF-, and log IL18 (all measured at baseline). Interaction term P values: 20:0 (P = 0.028); 22:0 (P = 0.021). Covariates: age, sex, center, physical activity, smoking status, alcohol intake, education, total energy intake, fruit and vegetable intake, red meat intake, soft drink intake, and fiber intake. SFAs are expressed as mole fractions of total FA concentration. Standardized  coefficients are presented.