Red Blood Cell Aggregation-Associated Dietary Pattern Predicts Hyperlipidemia and Metabolic Syndrome

Red blood cell (RBC) aggregation and iron status are interrelated and strongly influenced by dietary factors, and their alterations pose a great risk of dyslipidemia and metabolic syndrome (MetS). Currently, RBC aggregation-related dietary patterns remain unclear. This study investigated the dietary patterns that were associated with RBC aggregation and their predictive effects on hyperlipidemia and MetS. Anthropometric and blood biochemical data and food frequency questionnaires were collected from 212 adults. Dietary patterns were derived using reduced rank regression from 32 food groups. Adjusted linear regression showed that hepcidin, soluble CD163, and serum transferrin saturation (%TS) independently predicted RBC aggregation (all p < 0.01). Age-, sex-, and log-transformed body mass index (BMI)-adjusted prevalence rate ratio (PRR) showed a significant positive correlation between RBC aggregation and hyperlipidemia (p-trend < 0.05). RBC aggregation and iron-related dietary pattern scores (high consumption of noodles and deep-fried foods and low intake of steamed, boiled, and raw food, dairy products, orange, red, and purple vegetables, white and light-green vegetables, seafood, and rice) were also significantly associated with hyperlipidemia (p-trend < 0.05) and MetS (p-trend = 0.01) after adjusting for age, sex, and log-transformed BMI. Our results may help dieticians develop dietary strategies for preventing dyslipidemia and MetS.

We hypothesized that obesity-related inflammation may cause the upregulation of hepcidin and sCD 163, and the altered serum hepcidin may affect iron levels, which, in turn, may influence RBC aggregation. The aims of this study were to: (1) investigate the relationship between serum iron biomarkers and RBC aggregation and (2) identify dietary patterns associated with RBC aggregation and their predictive effects on hyperlipidemia and MetS in 212 Taiwanese adults.

Participants
In total, 212 Taiwanese adults aged 20-64 years were recruited at the Division of Gastroenterology and Hepatobiliary Diseases, Department of Internal Medicine, Taipei Medical University Hospital, from 29 April 2015 to 28 April 2016. A non-probability volunteering sampling was used as the sampling method. All participants were Han Chinese and were excluded from the study if they had at least one of the following: (1) were pregnant or breast feeding; (2) were taking hormone drugs; (3) had been diagnosed with hepatitis virus B or C or liver carcinoma; (4) failed to give a blood sample. This procedure was approved by the Taipei Medical University Institutional Review Board (TMU-JIRB 201502018), and written informed consent was signed by all participants.

Questionnaires
A simple questionnaire was used to record the basic information of participants, including age, sex, anthropometry, alcohol consumption, and history of diseases and medication. A food frequency questionnaire (FFQ) was used to investigate the dietary patterns of the participants. This was modified from a Chinese FFQ, which originally consisted of 64 items [29], and included three major categories: (1) food intake frequency; (2) cooking method used; (3) frequency of eating outside the home. The modified FFQ contained 66 food items that were categorized into 32 food groups, including five commonly used cooking methods for protein-rich foods and the frequencies of eating outside and homemade food. The food frequency was divided into eight levels: (1) 0-1 time/week; (2) 2-3 times/week; (3) 4-5 times/week; (4) 6-7 times/week; (5) 8-10 times/week; (6) 11-13 times/week; (7) 14-16 times/week; (8) ≥17 times/week.

Anthropometric Measurements
Body weight, height, waist and hip circumference of each participants were recorded, and the body mass index (BMI) was then calculated in kg/m 2 . The waist circumference was measured around the midpoint between the lower margin of the last rib and the top of the iliac crest.

Laboratory Measurements
Samples consisting of 15 mL of blood were collected after 8 h of fasting. Blood analyses included a complete blood cell count, inflammation biomarkers analysis, lipid profile, and serum iron biomarkers analysis. Serum iron biomarker analysis included serum ferritin (SF), serum iron, total iron-binding capacity (TIBC), serum-free Hb, serum hepcidin, and sCD163. Serum transferrin saturation (%TS) was calculated using the formula: (serum iron ÷ TIBC) × 100%. Serum iron was measured by a colorimetric method (Le-Zen Clinical Laboratory, Taipei, Taiwan). Serum-free Hb (Immunology Consultants Laboratory, Portland, OR, USA), serum hepcidin (DRG International Inc, Springfield, NJ, USA), and sCD163 (R&D Systems, Minneapolis, MN, USA) were analyzed by an enzyme-linked immunosorbent assay (ELISA) according to the manufacturer's procedures. RBC aggregation was analyzed by pipetting 500 µL of whole blood into the left well of an RSD-K01 chip (MicroStar Instruments, Seoul, Korea) that was inserted into RheoScan-AnD300 (MicroStar Instruments, Seoul, Korea), a microfluidic ektacytometer, for measurement. The measurements provided the critical shear stress (CSS), which stands for the minimum amount of shear stress exerted by blood stream currents to disperse RBCs [30]. A greater value of CSS indicates higher RBC aggregability.

Statistical Analysis
Analyses were carried out using IBM ® SPSS ® 21 (IBM Corp., Armonk, NY, USA), SAS version 9.4 (SAS Institute Inc., Cary, NC, USA) and GraphPad Prism 5 (GraphPad Software, La Jolla, CA, USA). Categorical data were presented as the number (percentage (%)), and continuous data were presented as the mean ± standard deviation (SD). RBC aggregation data were divided into quartiles (Q) using SPSS by assigning Q1 to the smallest value. A general linear model was used to analyze the p-trend between variables for continuous data, and Chi-squared was used for categorical data. A normality test was carried out to test for the distribution of each variable. Variables that were not normally distributed were log-transformed. An age-, sex-, log-transformed BMI-adjusted Poisson regression model was performed to estimate the prevalence rate ratio (PRR) and 95% confidence interval (CI) of hyperlipidemia and MetS [31]. A multivariate linear regression analysis was implemented to examine the relationships between RBC aggregation and potential variables. A reduced rank regression (RRR) was carried out to derive RBC aggregation-associated dietary patterns with the 32 food groups from the FFQ as predictors, and biomarkers determined from the multivariate linear regression analysis as responses [32]. Food groups with factor loadings of ≥0.20 or ≤−0.20 were used to describe RBC aggregation-associated dietary patterns. The dietary pattern score was derived from each participant and represented the sum of food intake variables weighted by a factor loading. These scores indicated the conformity of food consumption to the RBC aggregation-associated dietary pattern. The directed acyclic graph below explains the conceptual framework of the RRR (Figure 1). If the p-values were ≤0.05, the differences were considered significant.

RBC Aggregation Shows a Positive Correlation with Dysregulated Iron and Is Positively Associated with Hyperlipidemia
We next stratified individuals according to the RBC aggregation concentrations. Table 1 shows that RBC aggregation was positively correlated with BMI (p-trend = 0.001), hyperlipidemia (p-trend < 0.001), MetS (p-trend = 0.001), blood lipids, and hepcidin (both p-trend < 0.01). On the other hand, %TS was a negatively correlated with RBC aggregation (p-trend = 0.001). Table 1. Baseline characteristics of the study population according to quartiles of RBC aggregation levels (N = 196). Age-, sex-, and log-transformed BMI-adjusted Poisson regression model showed that quartile levels of RBC aggregation CSS had a significant positive correlation with hyperlipidemia (p-trend < 0.05 (Figure 2)).

Figure 2.
Prevalence rate ratio (PRR) and 95% confidence intervals of red blood cell (RBC) aggregation critical shear stress (CSS) quartile levels for hyperlipidemia adjusted by age, sex, and log-transformed body mass index (BMI); * p ≤ 0.05.

RBC Aggregation Is Positively Correlated with Hepcidin and sCD163, but Negatively Correlated with %TS
We next investigated potential confounding variables associated with RBC aggregation. A multivariate linear regression analysis was used to explore variables that could independently predict RBC aggregation. A univariate regression analysis indicated that log-transformed BMI (β = 0.316 (0.120-0.512), p < 0.

RBC Aggregation-Associated Dietary Patterns Independently Predict Hyperlipidemia and MetS
RBC aggregation-associated dietary pattern scores were derived by the RRR. The response variables were selected on the basis of strong correlations between the independent variables, which were hepcidin (p < 0.001), log-transformed sCD163 (p < 0.01) and %TS (p < 0.001), and RBC aggregation (Table 2; model 2). Table 3 shows the percentage of food variation corresponding to the first dietary pattern scores and factor loading of the food groups. Noodles and deep-fried foods were positively correlated with the first dietary pattern scores (factor loadings ≥0.20). On the other hand, steamed, boiled, and raw foods, dairy products, orange, red, and purple vegetables, white and light-green vegetables, seafood, and rice were negatively correlated with the dietary pattern scores (factor loadings ≤−0.20).  Table 4 shows a positive correlation between RBC aggregation-related dietary pattern scores and RBC aggregation after adjusting for age, sex, log BMI, and hyperlipidemia (p-trend = 0.01, model 5). Age-, sex-, and log-transformed BMI-adjusted Poisson regression model showed that the quartile levels of dietary pattern scores had a significant positive correlation with hyperlipidemia (p-trend < 0.05 (Figure 4)) and MetS (p-trend = 0.01 (Figure 4)).

Discussion
Our study found that RBC aggregation is closely linked to obesity and dysregulated iron metabolism (as indicated by decreased %TS and increased hepcidin and sCD163). Increased RBC aggregation also increased the rate of hyperlipidemia (p-trend < 0.05). The RBC aggregation and iron-associated dietary pattern, which was characterized by high-frequency consumption of noodles and deep-fried foods and low-frequency intake of steamed, boiled, and raw food, dairy products, orange, red, and purple vegetables, white and light-green vegetables, seafood, and rice, was also significantly associated with both hyperlipidemia (p-trend < 0.05) and MetS (p-trend = 0.01).
One of the interesting findings of our results was the relationship between RBC aggregation and the consumption frequency of staple foods. Our study showed that noodles, but not rice, had the highest impact on the RBC aggregation and iron-associated dietary patterns (highest percentage of explained variation, 12.66%, and highest factor loading, 0.38). Both noodles and rice are refined carbohydrates which also yield a high glycemic index [33]. A recent study from the Chinese Nutrition and Health survey showed that, compared to rice, increased consumption of noodles produced a higher risk of developing type 2 diabetes [34]. Another study also showed negative effects of noodles on serum lipid levels and glucose metabolism [35]. Our dietary patterns showed that an increased intake of noodles was associated with deep-fried food consumption but decreased consumption of rice, steamed, boiled, and raw food, dairy products, orange, red, and purple vegetables, white and light-green vegetables, and seafood. This suggests that individuals who prefer noodles as a staple food are also less likely to eat steamed, boiled, and raw foods, vegetables, and seafood. Several prospective studies have suggested that a low intake of dairy products and vegetables increases the risks of dyslipidemia and MetS [35,36]. Our results also agree with previous studies indicating that deep-fried foods increase RBC aggregability, dyslipidemia, and MetS [36].
In this study, RBC aggregation and the three independent RBC aggregation factors, i.e., the iron biomarkers hepcidin, %TS, and sCD163, were selected as responses. The selected iron biomarkers had different relationships with RBC aggregation. Using Pearson correlation coefficients (data not shown), the r values of RBC aggregation and hepcidin, %TS, and sCD163 were 0.262 (p < 0.001), −0.254 (p < 0.001), and 0.285 (p < 0.001), respectively. Also, hepcidin was significantly correlated with %TS (r = 0.258, p < 0.001). However, correlations of sCD163 with hepcidin and %TS were not statistically significant. We further analyzed the relationship between the selected responses (RBC aggregation, hepcidin, %TS, and sCD163) and the selected food groups to clarify which kind of foods affected each response (data not shown). Using a linear regression, only steamed, boiled, and raw foods were negatively correlated with RBC aggregation (β: −0.030 (−0.056-0.003), p < 0.05). Hepcidin was positively correlated with deep-fried foods (β: 23.508 (3.717-43.298), p < 0.05), and sCD163 was negatively correlated with white and light-green vegetables (β: −0.064 (−0.125-0.003), p < 0.05). Noodles and steamed, boiled, and raw foods may also affect sCD163 as their p-value were close to significance (both p = 0.51). The %TS was not statistically significantly correlated with any of the selected food groups. Taken together, our data suggests that RBC aggregation-associated dietary pattern is largely influenced by RBC aggregation or the iron biomarkers hepcidin and sCD163.
Another interesting finding was that, although there was a statistically significant p-trend for the relationship between dietary pattern scores and hyperlipidemia (p-trend < 0.05), the PRRs between dietary pattern scores and hyperlipidemia did not consistently increase. Individuals with dietary pattern scores Q2 (PRR = 1.904 (0.921-3.934)) and Q4 (PRR = 1.889 (0.882-4.043)) had a higher risk of developing hyperlipidemia compared to Q1 (Figure 4). From the analysis between dietary pattern scores and variables (data not shown), only total C exhibited a significant difference between dietary pattern scores Q2 and Q3 (p < 0.05). Since total C was significantly higher in dietary pattern scores Q2 (208.38 ± 40.86) than in Q3 (193.57 ± 37.66), and one of the criteria for hyperlipidemia is elevated total C, high total C could be a possible reason for the Q2 dietary pattern scores having a higher rate of developing hyperlipidemia than Q3. Another possible explanation could be due to the gender element. Studies already showed that there are gender differences in eating behaviors [37][38][39]. Therefore, the phenomenon of Q2 having higher a risk of developing hyperlipidemia than Q3 may be due to the bias caused by different food preferences between sexes.
In the current study, hepcidin, %TS, and sCD163 were identified as independent factors which predicted RBC aggregation, with hepcidin and log-transformed sCD163 being positively correlated, and %TS being negatively correlated. However, the pathways through which iron biomarkers regulate RBC aggregation are still unknown. Hepcidin is regulated through a complex interplay of signals, mainly inflammation, the iron status, and RBC production. During obesity-induced inflammation, anemia of inflammation (AI) may occur as hepcidin increases. Inflammatory cytokines, predominately interleukin (IL)-6, promote the secretion of hepcidin through the Janus kinase (JAK)/signal transducer and activator of transcription 3 (STAT3) pathway. Therefore, the characteristics of AI include a decreased availability of circulating iron for the production of RBCs, despite adequate iron stores [40].
CD163 is the macrophage scavenger receptor which takes up Hp-Hb complexes, but sCD163 levels increase with obesity and metabolic disorders. Studies showed that the shedding of CD163 from macrophages during chronic inflammation is more robust than during acute inflammation [41]. Increases in sCD163 under inflammatory conditions are due to activation of TACE/ADAM17 [42]. TACE/ADAM17 is located in cell membranes and quickly responds to various physiological stimuli. Activation of the enzyme can be stimulated by Toll-like receptor (TLR) [43], which is activated by lipopolysaccharides [44], cross-linking oxidative stress [45], and thrombin [46]. Several studies indicated that sCD163 is a predictor of obesity-related diseases. sCD163 has strong correlations with MetS and inflammatory serum markers [22]. CD163 + macrophages can be found in vessel walls, and increased expressions of heme oxygenase (HO)-1 and CD163 were positively correlated with tissue iron content and symptomatic plaques [47]. However, how sCD163 affects RBCs and RBC aggregability is still unclear.
This study has several limitations. First, the sample size was relatively small. Second, the FFQ only represents the frequency of participants' food intake, while the actual amount of food consumed by participants remains unknown. Thus, the precise amounts of nutrients could not be determined. Third, the RRR is a method that requires in-depth knowledge of the relationships between diet and disease in order to select the response variables. Using different variables for the RRR analysis will yield different results. Since different variables have different relationships with each other and with the food groups, choosing different biomarkers as response variables will affect the outcome of the RRR analysis.

Conclusions
The study results suggest that individuals with the highest consumption frequencies of noodles and deep-fried foods and the lowest intake of steamed, boiled, and raw food, dairy products, orange, red, and purple vegetables, white and light-green vegetables, seafood, and rice were more likely to develop RBC aggregation, dyslipidemia, and MetS. Our findings may help clinicians and dieticians develop dietary strategies for preventing dyslipidemia and MetS particularly in those individuals with RBC and iron dysfunctions.