Dietary patterns related to zinc and polyunsaturated fatty acids intake are associated with serum linoleic/dihomo-γ-linolenic ratio in NHANES males and females

Identifying dietary patterns that contribute to zinc (Zn) and fatty acids intake and their biomarkers that may have an impact on health of males and females. The present study was designed to (a) extract dietary patterns with foods that explain the variation of Zn and PUFAs intake in adult men and women; and (b) evaluate the association between the extracted dietary patterns with circulating levels of serum dihomo-γ-linolenic fatty acid (DGLA) or serum linoleic/dihomo-γ-linolenic (LA/DGLA) ratio in males and females. We used reduced rank regression (RRR) to extract the dietary patterns separated by sex in the NHANES 2011–2012 data. A dietary pattern with foods rich in Zn (1st quintile = 8.67 mg/day; 5th quintile = 11.11 mg/day) and poor in PUFAs (5th quintile = 15.28 g/day; 1st quintile = 18.03 g/day) was found in females (S-FDP2) and the same pattern, with foods poor in PUFAs (5th quintile = 17.6 g/day; 1st quintile = 20.7 g/day) and rich in Zn (1st quintile = 10.4 mg/day; 5th quintile = 12.9 mg/day) (S-MDP2), was found in males. The dietary patterns with foods rich in Zn and poor in PUFAs were negatively associated with serum LA/DGLA ratio. This is the first study to associate the LA/DGLA ratio with Zn and PUFAs related dietary patterns in males and females.

Food groups. Food items (assessed in grams per day) were assigned to one of the 34 food groups predefined from the USDA's Food and Nutrient Database for Dietary Studies 2011-2012 which permitted the coding of dietary intake data according to the nutrient profiles and culinary use 28 . The dairy products group was divided into low fat and whole fat and the grain products group was also divided into good source of fiber (> 2.5 g/ per serving) and poor source (below or equal 2,5 g/per serving) because these groups may have an impact on serum fatty acids. The groups were: low fat dairy products (G1), whole fat dairy products (G2), red meat (G3), poultry (G4), organ meat (G5), fish (G6), meat, poultry, fish with nonmeat items (G7), frozen and shelf-stable plate meals, soups, and gravies with meat, poultry, fish base, gelatin and gelatin-based drinks (G8), eggs (G9), legumes (G10), nuts and seeds (G11), flour and dry mixes, yeast breads, rolls, quick bread (G12), cakes, cookies, pies, pastries, bars (G13), crackers and salty snacks from grain products (G14), pancakes, waffles, French toast, other grain products (G15), pastas, cooked cereals, rice (G16), cereals, not cooked or non-specified as to cooked (G17), grain mixtures, frozen plate meals, soups (G18), meat substitutes, mainly cereal protein (G19), cereals, grains good source of fiber (G20), cereals, grains poor source of fiber (G21), fruits (G22), fruit juices (G23), starchy vegetables (G24), dark green vegetables (G25), yellow and red vegetables (G26), other vegetables (G27), fats (G28), oils (G29), salad dressing (G30), sugars and sweets (G31), nonalcoholic beverages (G32), alcoholic beverages (G33), formulated nutrition beverages (G34). The multiple source method (MSM) was used to remove within-person variation and then to estimate the usual food group intake 26 . Serum Zn. Detailed specimen collection and processing procedures are discussed in the NHANES Laboratory Procedures Manual 29 . All NHANES analysis was conducted in CDC laboratories or by their collaborators or subcontractors. The trace metal vacutainer tubes were used only on the second or later blood draws. No fasting or special diet was required. Pre-screened polyethylene vials and pre-screened 7 mL vacutainers were used for specimen acquisition. In between 0.8 to 2.0 ml of serum was required for analysis. Inductively coupled plasma dynamic reaction cell mass spectrometry (ICP-DRC-MS) was used to measure serum Zn. Normal values for Zn: > 75mcg/dl 30 . Serum fatty acids. NHANES serum LA (18:2n-6) (µmol/L) and DGLA (20:3n-6) (µmol/L) fatty acids were measured at the CDC using gas chromatography-mass spectrometry. The detailed procedures are in 31 https:// wwwn. cdc. gov/ nchs/ data/ nhanes/ 2011-2012/ labme thods/ FAS_G_ MET. PDF. All analysis was conducted in CDC laboratories or by their collaborators or subcontractors. Briefly, a fasting 0.5 ml sample was obtained and a volume of 100 μl was required per analysis. Frozen samples were stored at − 70 °C.
Sequential treatment with mineral acid and base in the presence of heat was used to achieve esterified fatty acids. Total fatty acids were hexane-extracted with an internal standard solution containing stable isotopicallylabeled fatty acids for recovery. The extract was converted into pentafluorobenzyl esters. The reaction mixture was injected onto a capillary gas chromatograph column to separate the fatty acids of interest only, and were www.nature.com/scientificreports/ detected using electron capture negative-ion mass spectrometry. Quantitation was accomplished by comparing the peak area of the analyte in the unknown with the peak area of a known amount in a calibrator solution 31 .

Statistical analysis and reduced rank regression (RRR).
Reduced rank regression (RRR) analysis 16 was performed with two response variables (PUFAs intake and Zn intake, in grams per day) and the 34 food groups (intake of each food groups in grams per day) in 1614 participants separated by sex (NHANES 2011(NHANES -2012. At least five individuals per food group is recommended: this study used 34 food groups and a sample of 170 people would sufficiently powered for the analyses described herein 16 ). Since nutrient intake is usually highly correlated with energy intake 32 , we calculated energy-adjusted nutrient intakes, using the regression residual method 32 .
The PROC PLS procedure with the RRR method was done with the SAS Statistical Software Version 9.3 (SAS Institute, Inc. Cary, NC, USA). We applied the RRR method to female and male populations. A detailed description of the method can be found in Hoffmann et al. 16 . Briefly, in RRR analysis, X 1 , …, X n and Y 1 , …, Y m are two sets of variables (predictors and responses). In this study, the predictors X i are intakes of food groups in grams per day and the responses Y j are intakes of Zn and PUFAs in grams per day. RRR starts from a linear function of responses called response score that will then be projected onto the space of predictors to produce a factor score, that is, a linear function of predictors. Therefore, RRR extracts successive linear combinations of the predictors, called factors or components 16 . RRR does not describe naturally occurring dietary patterns of the population under study but explains variation in important risk factors 33 and hence, this is an exploratory approach and dietary patterns are derived from the available data. This approach ignores prior knowledge of population dietary pattern completely in order to extract dietary patterns that explain the variation of Zn and PUFAs intake.
The hypothesis tested here was that intakes of Zn and PUFAs are presumed to be associated with serum PUFAs levels in males and females 10,11 . Dietary Zn and PUFAs were chosen as response variables and the outcomes of interest were serum LA/DGLA ratio and serum DGLA. The dietary pattern score "x" was calculated as the sum of z-standardized intakes (mean = 0, standard deviation = 1) of 34 food groups items multiplied by an individual weight. Further, to reduce the dimensionality of complex data, the score was simplified by including only food groups with high factor loadings > or = to 0.20 and then summing the standardized food group intake while retaining the direction of the factor loading 16 . Food groups with factor loadings equal or higher than 0.2 were considered positive contributors to the patterns, and food groups with factor loadings up to -0.2 were negative contributors to the patterns.
The simplified pattern score defined by the food groups was used to evaluate the association between the dietary pattern and the response variables (partial correlation adjusted by energy intake) but also between the pattern and NHANES 2011-2012 serum PUFAs (outcomes). The individuals simplified score "x" was distributed into quintiles. For descriptive purposes, the median (min-max) intakes of the food groups were calculated across quintiles of the simplified dietary pattern score, as well as anthropometric and serum PUFAs data and frequencies for sample characteristics. Linear trend analysis was used to compare continuous variables in their original scale throughout quintiles, adjusting for energy for food groups. Chi-square test was used to compare categorical variables throughout quintiles. We evaluated the impact of potential effect modifiers on the association between the dietary pattern and the serum PUFAs by linear regression fitting models. We applied unadjusted and adjusted multiple linear regression analysis models to the serum dihomo-γ-linolenic fatty acid and to linoleic/dihomo-γlinolenic ratio (dependent variables) for NHANES 2011-2012 population separated by sex, with the simplified scores for each dietary pattern as independent variable including the following variables as confounding variables: (model 1) unadjusted; (model 2) adjusted for age, energy intake, poverty / income ratio (PIR), race, physical activity, BMI, level of education, medication use, supplement use as confounding variables. The selection of the covariates was based on theoretical assumptions of their relationship with the outcome.
Characterizing under-and over reporters. The Vinken's equation 34 was used for predicting total energy expenditure (pTEE) in MJ/day by the following relation: where age is in years, weight is in kg, height is standing height in cm, and sex is 0 for men and 1 for women. McCrory equation described below was used to calculate cut-off points for detecting under-and over reporters 35 .
where CV wEI and CV wpTEE are the within-subject coefficient of variation obtained from energy intake and from predicted TEE equal to 32.3% and 20.2%, respectively. The number of days of energy intake (d) was equal to 2 and the within subject coefficient of variation in measured total energy expenditure (CV tmTEE ) was equal to 8.2% 35,36 .
All statistical analyses were performed using R package (version 3.6.0) and SAS software, version 9.4 (SAS Institute Inc., Cary, North Carolina, United States) and a two-sided p-value < 0.05 was considered statistically significant.
Ethics approval and consent to participate. NHANES

Results
Of the 9756 individuals in the NHANES 2011-2012 database, 1614 participants (771 females and 843 males) met the inclusion criteria ( Fig. 1). Demographics characteristics of studied females and males in NHANES 2011-2012 are in Table 1. Females and males were similar regarding age, body mass index, poverty/ income ratio and percentage of individuals using lipid lowering drugs but have small differences in ethnic background and education level. Dietary supplement use was more prevalent in females, and males practice more vigorous activity. No statistically significant associations were found between Zn intake alone and DGLA or LA/DGLA ratio without (spearman correlation females r = 0.013, p = 0.71, r = − 0.05, p = 0.14, respectively; and spearman correlation males r = 0.06, p = 0.07, r = − 0.06, p = 0.06, respectively) or with adjustment for energy intake (female r = 0.04, p = 0.19, r = − 0.05, p = 0.12, respectively; male r = − 0.04, p = 0.25, r = 0.03, p = 0.32, respectively).
Since dietary Zn alone was not associated with these circulating fatty acids in men or women, either with or without adjustments for energy intakes, dietary patterns with intake of different ratios of Zn and PUFAs may better explain DGLA or LA/DGLA ratios. Reduced Rank Regression analysis was used to assess whether these two combined response variables (Zinc intake and PUFAs intake) were associated with levels of these specific fatty acids.
Reduced rank regression. RRR applied to females' diet intakes extracted a female dietary pattern 1 (FDP1) which had more foods higher in Zn and PUFAs and a female dietary pattern 2 (FDP2) which had more foods higher in Zn and lower in PUFAs (Supplemental Table S1). RRR applied to male population extracted a male dietary pattern 1 (MDP1), which had more foods higher in Zn and PUFAs, and a male dietary pattern 2 (MDP2), which had more foods higher in Zn and lower in PUFAs (Supplemental Table S1).  Dietary patterns. The dietary pattern scores were simplified as described in Methods in female and male populations by using food items with absolute factor loadings equal to or above 0.20 as being significant contributors to a pattern (  (Table 2).

Specific foods consumed by females.
Females with the highest adherence (Table 3, 5th quintile) to the S-FDP1 (foods rich in Zn and in PUFA) were eating more (1) low fat dairy products (that may have contributed to Zn intake), (2) eggs, (3) nuts and seeds, (4) cakes, cookies, pies, pastries, bars, (5) crackers and salty snacks from grain products, (6) cereals not cooked or non-specified as to cooked, (7) grain mixtures, frozen plate meals, soups, (8) cereals and grain products good source of fibers, (9) starchy vegetables, and (10) salad dressing; while females with the highest adherence (5th quintile) to the S-FDP2 (foods rich in Zn and poor in PUFA) were eating more (1) dairy products, (2) red meat, (3) cereals not cooked or non-specified as to cooked, and less (4) cakes, cookies, pies, pastries, bars and (5) salad dressing (Table 3). Serum Zn levels did not change throughout the quintiles ( Table 3). The prevalence of females below the lower cutoff for serum Zn concentration was only 3.4% (26 individuals).
One of the objectives of this study was to test associations of dietary patterns with the levels (i.e., outcomes) of serum DGLA and LA/DGLA ratio. In trend analysis, males with the best adherence to a dietary pattern with foods rich in Zn and poor in PUFAs (S-MDP2) had a trend to lowest values for LA/DGLA ratio (p = 0.06). The highest values for serum DGLA and the lowest values for LA/DGLA ratio were observed in females with the best adherence to a dietary pattern with foods rich in Zn and poor in PUFAs (S-FDP2) ( Table 5).

Multiple linear regression analysis. Dietary patterns in females and biomarkers. Since a relationship
was found between dietary patterns and serum DGLA and LA/DGLA through linear trend analysis, we applied unadjusted and adjusted multiple linear regression analysis models. Regression analyses of both the unadjusted (model 1) and adjusted for all confounding variables (model 2) models confirmed that there was a positive association between females with a dietary pattern with foods rich in Zn and poor in PUFA (S-FDP2) and serum DGLA (unadjusted model: R 2 adjusted 0.016; p < 0.01; adjusted model: R 2 adjusted 0.168; p < 0.01). There was also a negative association between S-FDP2 and serum LA/DGLA ratio in females (unadjusted model: R 2 adjusted 0.018; p < 0.01; adjusted model: R 2 adjusted 0.123; p < 0.01) (Supplemental Table S2).
Dietary patterns males and biomarkers. The simplified dietary pattern scores did not show any association with serum DGLA in males with linear regression analyses using either the unadjusted (model 1) or adjusted for all confounding variables (model 2) models. However, a negative association was found between S-MDP2 Table 3. Food groups, response variables intakes and serum Zn across quintiles of the simplified dietary pattern score. Female population NHANES 2011-2012. 1 Differences in quintiles for food and nutrients intakes were tested using linear trend, adjusting for energy. The top number in each cell is the median value and the data in parenthesis is the range.  Table S3).

Simplified dietary pattern 1 (foods rich in Zn and PUFAs) Simplified dietary pattern 2 (foods rich in Zn and poor in PUFAs)
Serum Zn and biomarkers. Serum Zn was not correlated with the LA/DGLA ratio in females and there was a negative trend in males (Spearman: r = − 0.05, p = 0.39; r = − 0.11, p = 0.07, respectively). Serum Zn was statistically correlated with serum DGLA in both females and males (Spearman: female r = 0.13, p = 0.04; male r = 0.14, p = 0.02). Figure 3 graphically summarizes the analysis pipeline and main results obtained with the reduced rank regression approach.

Discussion
RRR was used to extract dietary patterns. This revealed dietary patterns that are negatively associated with LA/ DGLA ratio. The patterns are similar across males and females, containing foods rich in Zn and poor in PUFAs.
No correlations were found between Zn intake alone and serum DGLA or serum LA/DGLA ratio supporting previous observations that associations between intake of a single nutrient and serum metabolites may potentially be confounded by the effect of dietary patterns 8,9,37 . The amount of Zn and PUFAs intakes found in the present study are in accordance with EAR and RDA 38 and with other published studies 39,40 . Table 4. Food groups, response variables intakes and serum Zn across quintiles of the simplified dietary pattern score. Male population NHANES 2011-2012. 1 Differences throughout quintiles for food and nutrients intakes were tested using linear trend, adjusting for energy. The top number in each cell is the median value and the data in parenthesis is the range.     www.nature.com/scientificreports/ Food groups for RRR analysis were adapted from groups previously defined in the Food and Nutrient Database for dietary studies (FNDDS 2011(FNDDS -2012 [45][46][47] , which include many foods, mixtures and brand name items. No specific criteria in the literature exists to aggregate food items into food groups although others have used culinary customs, nutrient content 16,48 , or expert opinion 49,50 to create groups.

Simplified dietary pattern 1 (foods rich in Zn and PUFAs) Simplified dietary pattern 2 (foods rich in Zn and poor in PUFAs)
The simplified pattern approach was used to construct less data-dependent pattern variables even though it results in a minor loss of information 51 . Absolute factor loadings equal or above 0.20 were considered as significant contributors to a dietary pattern. This value has frequently been used in the context of determining those food items that are significantly associated with factor-analysis-based dietary patterns [49][50][51] . Less than 10%, of information was lost by using the simplified instead of the original pattern variables. Correlations between response variables and simplified scores were modest to moderate, similar to those found in other studies using RRR analyses of dietary patterns 43,44,50,51 . Our study suggests that the simplified dietary pattern analysis reduced the number of food variables necessary to create a dietary pattern.
The foods in the dietary patterns with more Zn included meat, cereals not cooked, or non-specified as cooked (Zn-fortified foods) which are known to have absorbable Zn 7 . Ready-to-eat breakfast cereals are fortified with Zn, making them a primary Zn source in United States 52 . Lean red meat, whole-grain cereals, provide concentrations of Zn that may vary between 25-50 mg/kg raw weight. The food groups found to contribute to Zn intake in the present study were also found in another study that associated Zn intake and food groups 53 .
Foods rich in PUFAs in the dietary pattern identified in our study were eggs, most vegetable oils, breads, baked goods, margarine, salad dressing and nuts 54 . Individuals who eat salad in USA may also consume more total fat and unsaturated fatty acids and oils in salad dressing 39 .
Our results showed that serum Zn did not correlate with dietary patterns with foods rich in Zn intake which confirms that blood level of Zn is not a good biomarker of its dietary intake 7,53,55-60 . Serum zinc status. The present study followed CDC protocols and considered 75 mcg/dL as lower cutoff of fasting Zn concentration 30 . Others have suggested 74 mcg/dL for males and 70 mcg/dL for females for assessing the risk of Zn deficiency 61,62 . The prevalence of individuals with serum Zn deficiency in this study was very low. However, serum Zn (SZCs) has been reported to be higher compared to Zn levels measured in plasma (PZCs) 63 . Nevertheless, PZCs and SZCs are both considered valid estimates of Zn status 64 .

A possible biomarker of dietary patterns related to Zn intake and PUFAs intakes. The present
study is the first one to show that LA/DGLA ratio may be a biomarker of dietary patterns related to Zn and PUFAs intakes in males and females. We did not find statistically significant correlation between Zn intake alone and LA/DGLA ratio or with DGLA alone. Others found inverse associations between Zn intake alone and LA/ DGLA ratio and positive associations with DGLA in humans 12,53,56 and in birds 65,66 . Serum DGLA and the LA/ DGLA ratio in the present study were related to short-term Zn and PUFAs intake 67 in unadjusted and adjusted regression models. Further studies are needed to validate these metabolites as biomarkers and should consider the following factors: plausibility, dose-response, time-response, robustness, reliability, stability and reproducibility. New studies should determine if DGLA or the LA/DGLA ratio are recovery/predictive biomarkers or concentration biomarkers 68 . The present study strengthens the association between LA/DGLA ratio and dietary patterns defined by Zn and PUFAs intakes. Serum Zn, DGLA and LA/DGLA ratio. An inverse correlation trend was found between the LA/DGLA ratio and serum Zn (p = 0.07) and a positive correlation between serum Zn and DGLA alone. Others also found a negative correlation trend between LA/DGLA ratio and plasma Zn, but not an association with plasma DGLA alone 53 . On the other hand, another study showed positive correlations between plasma Zn and DGLA and an inverse association between plasma Zn and LA/DGLA ratio 56 . Limitations and strengths of the study. The limitations and strengths of the present study are discussed in Box 1. Although a large number of influential studies on US trends in aspects of diet 69 rely on dietary data collected by the National Health and Nutrition Examination survey, only two 24-h-dietary-recalls are used for data collection. The lack of the usual third recall or the use of other methods decreases accuracy and might affect variation in the estimates of Zn and PUFAs intakes with subsequent alteration in dietary patterns. These effects may be mitigated by the use of the United States Department of Agriculture's 5-step Automated Multiple-Pass Method which has been shown to reduce bias in dietary intake data 70 .
The present study excluded under/over-reporters that could alter outcomes of the analysis. Several approaches have been suggested to adjust for under/over-reporting 36 . However, excluding large amounts of data that fall above or below the cutoff and a large residual unexplained variation in basal metabolic rate (BMR) determination may decrease the accuracy of the results 71 . Despite the above disadvantages, studies still indicate that implausible reporters be excluded from the analysis 72 .
Correlating diet intake with biomarkers is considered a solution to the limitations of dietary recall methods 68 . However, dietary intake biomarkers also have limitations 73,74 . Indices of nutrient exposure may not necessarily reflect nutrient status, which, in turn, may not necessarily reflect the impact or function of that nutrient 75 . Hence, the analysis pipeline was used to determine whether an association between the dietary patterns and the LA/ DGLA ratio exists. In addition, using the LA/DGLA ratio in nutritional epidemiology studies as a biomarker should be preceded by an assessment of its validity, replicability, ability to detect changes, and suitability for the population under study.
RRR shares a number of limitations with other data-driven approaches, one of which is that any identified food intake pattern may be specific to the population under study 17  www.nature.com/scientificreports/ validation in a different population 8 and across different ethnic groups, separated by sex or regardless of sex. Finally, although the models were adjusted by relevant confounders, the selection of the covariates was based on theoretical assumptions, and we cannot rule out other uncontrolled potential confounding factors. The strengths of the study were that (i) dietary patterns were analyzed instead of individual nutrients which is generally more directly related to dietary recommendations. (ii) Zn and PUFAs intakes were adjusted based on energy intakes which reduces the 24-HDR measurement errors [76][77][78][79] ; (iii) the study had a reasonable sample size despite the exclusion of under/over-reports-a sample size of at least five individuals per food group is recommended (our study used 34 food groups and a sample of 170 people would have been sufficient 80 ); (iv) associations on simplified dietary patterns reduce dimensionality of complex data. (3) Several interindividual factors can operate and generate variation in LA/DGLA ratio levels, which does not reflect solely differences in dietary intake (4) RRR shares a number of limitations with the data-driven approaches, including that the identified food intake patterns are specific to the population under study (5) The regression models were adjusted by relevant confounders and the selection of the covariates was based on theoretical assumptions, and we cannot rule out other uncontrolled potential confounding factors

Strengths
The followings are the strengths of the study: (1) United States Department of Agriculture's 5-step Automated Multiple-Pass Method has been shown to reduce bias in dietary intake data (2) Reasonable sample size even after under/over-reporters exclusion. Excluding implausible reports resulted in a dataset of much higher quality according to literature (3) Energy adjustments were made for Zn and PUFAs intakes which substantially weakened the impact of the 24-HDR measurement error on total nutrient intakes (4) We reduced the dimensionality of data by constructing simplified dietary patterns (5) Epidemiologic analyses based on foods, as opposed to nutrients, are generally most directly related to dietary recommendations

Conclusions and future considerations
This is the first study to associate LA/DGLA ratio with dietary patterns related to Zn and PUFAs intake. Two dietary patterns were negatively associated with serum LA/DGLA ratio. The patterns are similar across males and females, containing foods rich in Zn and poor in PUFAs. Further intervention or observational studies should be done to validate results. These future studies should also consider the integration of omics data to provide insights into the metabolism mechanisms and interference of genetic profile 81-83 .

Data availability
Data described in the manuscript, code book, and analytic code are available from the corresponding author on request.