Plasma Fatty Acid Biomarkers of Dairy Consumption Are Associated with Sex-Dependent Effects on Metabolic Syndrome Components in Mexican Adolescents

Dairy products are a major source of calcium (Ca); in developed countries, 50% or more of the requirement of Ca in adolescence is fulfilled by dairy products [1]. These products also provide other nutrients, such as high-quality protein, zinc, vitamin D, vitamins B12 and B2, but they also contribute to total and saturated fat intake [2].

The intake of saturated fatty acids (SFAs) is associated with increased risk for cardiovascular disease (CVD) during adulthood. However, the cardiometabolic risk factors of CVD often precede adulthood, suggesting the need to limit saturated fat intake at earlier stages of life [3]. Evidence relating to dairy product intake to cardiometabolic health is mixed, especially among youths. Results of two meta-analyses of cross-sectional and prospective studies point to an inverse association between dairy intake and overweight/obesity in youth [4, 5]; additionally, one cross-sectional multicenter study found an inverse association between total dairy intake and a cardiovascular risk score among females (12.5–17.5 years) [6]. In contrast, one longitudinal study in adolescents (9–14 years) found that an intake of more than 3 servings of milk per day was related to higher BMI compared to children who drank 1 or 2 servings (240 mL) [7]. Differences among studies may be attributed to measurement error since most studies rely on self-reported dairy consumption. Cross-sectional studies are also susceptible to reverse causality since having obesity could lead to dietary modifications, including altering the intake of dairy products such as whole-fat products and butter.

The use of dairy biomarkers could shed light on conflicting results relating to dairy intake to cardiometabolic risk in adults and youths. Despite contributing <1% of the total level of fatty acids (FAs), the plasma FAs pentadecanoic (C15:0), heptadecanoic (C17:0), and trans-palmitoleic (t-C16:1n-7) have been described as biomarkers of dairy intake, with varying grades of reliability between studies [8]. These FAs are found in dairy foods and non-dairy foods; however, the strongest correlations are seen for dairy consumption [9, 10].

The components of metabolic syndrome (MetS) are a cluster of cardiometabolic risk factors, including abdominal obesity, hypertension, insulin resistance, high fasting triglycerides (TG), and low high-density lipoprotein cholesterol (HDL-C) [11]. During adolescence, MetS components are considered predictors of future risk of CVD [12], the most common cause of mortality among adults worldwide [13, 14]. Given the conflicting evidence on the relationship between dairy intake and cardiometabolic risk factors in adolescents, in this study, we aimed to investigate the association between plasma FAs related to dairy intake and MetS components in a sample of Mexican adolescents. As dairy biomarkers have not been commonly used in adolescents, we also explored the correlation between FAs and dairy servings in adolescents as a complementary analysis. To the best of our knowledge, this is the first study that has employed these FAs as dairy intake biomarkers to assess the relationship between dairy and cardiometabolic risk factors during adolescence.

The analytic sample comprises of adolescents from two of three sequentially enrolled cohorts of the Early Life Exposure in Mexico City to Environmental Toxicants (ELEMENT) study [15]. Between 1997 and 2003, 1,012 mother/child dyads were recruited during pregnancy and after delivery from maternity clinics belonging to Instituto Mexicano del Seguro Social (IMSS) in Mexico City, prenatal clinics that are publicly accessible and serve low- to middle-income populations. The criteria for inclusion and exclusion of the ELEMENT study are described elsewhere [16]. In 2015, a subset of 554 participants from the original cohorts was chosen to participate in a follow-up visit when they were at peri-pubertal stages (ages 10–18 years). Participants were chosen to participate based on the availability of previously collected and stored biomarkers and health information [16]. Participants attended in-clinic evaluations where information related to health was obtained through physical examination, blood sampling, and questionnaires on health status, medical history, and lifestyle. The study protocol was approved by the Ethics, Biosafety, and Research Committees of the National Institute of Public Health in Mexico and the University of Michigan. Informed consent was obtained from parents for all participants in addition to participant assent (≤17 years), while participants ≥18 years provided their own consent. Our primary analysis included 311 participants (154 males and 157 females) with complete information on plasma FAs in addition to conventional cardiometabolic parameters (glucose, insulin, HDL-C, TG, blood pressure, and WC) and data for important covariates such as pubertal status (Fig. 1; participant flow chart for the analysis in this study) [17].

Adolescents provided a fasting blood sample, from which plasma was immediately separated and frozen at −80°C by the research staff. Plasma samples were shipped, maintaining the freezing temperature (−80°C) for storage at the University of Michigan (Ann Arbor, MI, USA). Subsequently, the determination of the FAs was carried out at the University of Michigan Metabolomics Core [18]. Fatty acid levels were measured in plasma using gas-liquid chromatography. First, the lipid layer was extracted with methanol using 100 μL of plasma, and the fatty acid methyl esters of total lipids were extracted from a TLC plate. Next, the methyl esters were resuspended in hexane and samples were analyzed with Agilent, Model 6890N. ChemStation software (Agilent) was then used to analyze the peaks, and the amounts of FAs were determined based on C19:0 methyl ester as the standard [18, 19]. Dairy biomarkers (C15:0, t-C16:1n-7, C:17) and the other 32 FAs, such as myristic acid (C14:0) and palmitic acid (C16:0), are reported as a percentage of total PFAs.

The dietary intake was assessed through a validated semiquantitative food frequency questionnaire (FFQ) [20]. Participants reported the frequency of consumption of 116 food items for the last 7 days. Ten food items were dairy products: cheese, butter, yogurt, fermented milk-based drinks, sour cream, petit cheese products, flavored chocolate milk, whole milk (3.5% or 3.8% fat milk), skim milk (0.5% or less fat). Energy, macronutrient and micronutrient intakes were obtained using a nutritional composition database of foods compiled by the National Institute of Public Health in Mexico (INSP) [21]. Total dairy servings per day were considered the sum of servings of the 10 dairy products included in the FFQ. We based the serving size on the standard servings described in the FFQ; e.g., a serving of yogurt and milk was considered 240 mL (1 cup), and a serving of cheese was considered 30 g and 5 g for butter and cream.

MetS components were as follows: HDL-C, TG, glucose, waist circumference (WC), insulin, diastolic blood pressure (DBP), and systolic blood pressure (SBP). A fasting blood sample was taken from participants by trained research staff. Using 10 mL of fasting blood, glucose and lipids were determined with a bench clinical chemistry analyzer. The samples were centrifuged and transported within the following 5 h on ice to the Nutrition Laboratory of the National Institute of Perinatology, Mexico; HDL-C and TG were measured by enzymatic techniques. Additionally, insulin levels were determined by enzyme-linked immunosorbent assay chemiluminescence method with IMMULITE^® 1000 (Erlangen, Germany) equipment [22]. Blood pressure was measured twice while the adolescents were seated, on the left arm, with a Spacelabs 90217 monitor following standard protocols.

Trained personnel performed the anthropometric measures in duplicate at the research center. The weight was taken with an InBody230 scale with a precision of 100 g. Height in centimeters was taken using a calibrated standard stadiometer (Perspective Enterprises) with a precision of 0.1 cm [23]. WC was measured by duplicate with a seca brand tape (model 201, Hamburg, Germany) with a precision of 0.1 cm [23].

The adolescents completed a questionnaire regarding lifestyle and demographic information, including physical activity and socioeconomic status (SES). Adolescents were classified into higher-medium and lower economic levels according to characteristics of the home, services, and level of education of the household head following the AMAI criteria (Asociación Mexicana de Agencias de Inteligencia de Mercado y Opinión, acronym in Spanish), an index of socioeconomic levels [24]. Sedentary behavior (self-reported mean number of hours of sedentary time activity hours per week) was assessed with a questionnaire adapted and validated for Mexican adolescents [25] and included the sedentary time spent on screens (time spent watching television, movies/DVD, or playing video games). We used the software WHO Anthro Plus to derive the BMI-for-age z-score for sex according to WHO reference growth standard. Sexual maturation, divided into pubertal and pre-pubertal statuses, was defined according to Tanner stages observed by medical personnel [26]. Possible confounders included in the present analysis were age, BMI (for-age z-scores), sedentary behavior, SES, pubertal status, and energy intake per day.

First, we conducted a descriptive analysis of the main characteristics of the study sample. The categorical variables are expressed as frequencies and proportions, and the continuous variables are presented as mean with standard deviation (SD) or median with percentiles 25th and 75th, according to their distribution. We estimated statistical differences between males and females using χ² tests for the categorical variables and Student’s t test or Wilcoxon rank-sum test for the continuous variables. In exploratory analyses, we found that sex was a modifier in the association between FAs and MetS components, and for this reason, we conducted all the analyses stratified by sex.

Our data did not satisfy the required assumptions for all linear regression models (homoscedasticity and normality), even after we transformed the outcome variables. Therefore, we decided to assess our associations using quantile regressions. In a quantile regression model, the result is the relationship between a set of predictor variables (independent variables) and a quantile of the target variable [27]. It is important to mention that this approach does not involve “breaking” a continuous target variable into quantiles; what it does is model a particular quantile, such as the median in this case. Using the median instead of the mean to explain the relationship between the variables has advantages over conventional least squares regression, including that the models are more robust or less sensitive to outliers [27, 28]. Furthermore, quantile regression makes no assumption of the normal distribution of the dependent variable [28]. Also, it requires no assumptions about the distribution of the model residuals or the assumption of homoscedasticity [27]. Regarding the exposure variables, the FAs C15:0, t-C16:1n-7, and C17:0 were divided into four categories defined by quantiles (Q1, Q2, Q3, Q4). We categorized the independent variables as quartiles to avoid assuming a linear relation (or any other fixed functional form) with the outcome variables, thus giving flexibility to the models, alongside ease of interpretation. Quantile regression models stratified by sex were fit to explore the association between plasma FAs (Q1, Q2, Q3, Q4) and each MetS component as a continuous variable, adjusted by potential confounders. Covariates included in all the quantile regression models were age, BMI Z-score, sedentary behavior (total amount of hours per week spent in sedentary behaviors), and energy intake (total amount of kcal consumed per day), all of which were included as continuous variables. We also considered the SES, in which adolescents were classified into three categories (low, medium, and high) according to income and household characteristics. Finally, we included pubertal status, divided into pre-pubertal and pubertal according to the Tanner staging assessment carried out by medical personnel. We also explored the association between FAs C15:0, t-C16:1n-7, and C17:0 and each MetS component using quantile regression models, such that exposure and outcome variables were both introduced in the models as continuous variables (in contrast to the exposures categorized into quartiles), adjusting by the same potential confounders. Thus, we evaluated the association between continuous FAs and the median of each outcome variable. For this analysis, we multiplied the three FAs (C15:0, t-C16:1n-7, and C17:0) by 10 to simplify the interpretation of the β coefficient in these regressions. The results in this analysis represent the effects for 0.10 unit increase (0.1%) in the median fatty acid exposure.

For our complementary analysis to examine the association between FAs and dairy intake assessed through the FFQ, we estimated Spearman’s correlations between the plasma FAs (C15:0, t-C16:1n-7, and C17:0) and total dairy servings consumed per day. Spearman’s correlations were stratified by sex and adjusted by age and BMI z-scores. We also estimated Spearman’s correlations between FAs (C15:0, t-C16:1n-7, and C17:0) and each type of dairy from our FFQ (cheese, butter, yogurt, fermented milk-based drinks, sour cream, petit cheese products, and all types of milk) to explore if these FAs explained the intake of some specific type of dairy. These were also adjusted by age and BMI z-scores.

In sensitivity analyses, we adjusted our main analysis (FAs and MetS components) for fish intake (g/day) using data from the FFQ since C15:0 and C17:0 have been identified as possible biomarkers of fish consumption [29]. Additionally, we adjusted C17:0 models for fiber intake (g/day) since it has also been pointed out as a possible fiber biomarker [30]. Finally, we conducted adjusted analyses that included all previously described confounders but excluded BMI z-scores. We also examined Spearman’s correlations among the FAs we employed (C15:0, t-C16:1n-7, and C17:0) with all measured FAs in plasma in the adolescents of our sample. We used a threshold of p < 0.05 to denote statistical significance in all our analyses. All data were analyzed by using the statistical package STATA 14.0.

Descriptive data are provided in Table 1. Among the 311 adolescents in the analytic sample (49% were male), the mean age was 13 ± 2 years and mean BMI z-score was 0.6 ± 1.3. There were no statistically significant differences between the sexes for mean age and BMI z-score. Males had significantly higher levels of glucose and SBP and lower levels of HDL-C, while females had significantly higher TG levels and insulin. Male participants also had significantly more hours of sedentary behavior and were more likely to be in the pre-pubertal stage compared to female participants. No significant differences were observed between males and females in the percentage of FAs C15:0, t-C16:1n-7, and C17:0.

Differences between adolescents included (n = 311) and not included in our principal analysis (n = 243) are presented in online supplementary Table 1 (for all online suppl. material, see https://doi.org/10.1159/000531972). Adolescents of both sexes included in the main analysis were significantly younger, had fewer hours of sedentary behavior per week, and presented significantly lower diastolic and systolic blood pressures. Regarding differences found in just one sex, female participants had significantly lower insulin levels; males had significantly lower WC, BMI z-score; and a greater percentage were classified as pre-pubertal compared to those not included in the analysis.

Intake of macronutrients and sources of FAs collected from the FFQ is provided in Table 2. There were no differences between males and females in the intake of specific types of dairy, except for milk. Male participants consumed significantly more servings per day of milk in contrast to female participants. In addition, male participants consumed significantly more total energy than females as well as specific nutrients and foods, including total grams per day of fat, saturated fat, fiber, and total dairy servings (Table 2). Spearman correlation coefficients between the FAs C15:0, t-C16:1n-7, and C17:0 and dairy servings are provided in Table 3. We found weak correlations between these FAs and dairy servings derived from the FFQ. The strongest non-statistically significant correlation was found with C15:0 in females (unadjusted r = 0.122, p value: 0.08), although correlations were essentially null after accounting for age and BMI z-score.

Spearman’s correlations between FAs (C15:0, t-C16:1n-7, and C17:0) and each type of dairy from FFQ are provided in online supplementary Table 2. We found weak significant associations; in females, C15:0 was significantly correlated with sour cream intake before (r = 0.160, p value: 0.02) and after (r = 0.002, p value: <0.01*) the adjustment for age and BMI z-score. In males, t-C16:1n-7 was significantly correlated with fermented milk-based intake, but only after the adjustment (r = 0.001, p value: 0.05), and C15:0 was significantly associated with petit cheese product intake in females only after the adjustment (r = 0.001, p value: 0.01). None of the other specific types of dairy (butter, cheese, yogurt, and milk) showed a significant correlation with the concentrations in plasma of C15:0, t-C16:1n-7, and C17:0 (the correlation coefficient for all cases was r < 0.1 and r < 0.01 after the adjustment). The strongest non-statistically significant correlation was found with C15:0 and fermented milk-based drinks in males (unadjusted r = 0.129, p value: 0.07).

In Spearman correlations among FAs, we observed significant strong correlations between C15:0 and other FAs such as C14:0 and C16:0 (r = 0.80 and 0.70, respectively). C17:0 and t-C16:1n-7 also correlated significantly with C14:0 and C16:0, but these correlations were weak (r < 0.30). Therefore, we also explored Spearman correlations between FAs (C14:0 and C16:0) and dairy servings and each type of dairy from our FFQ (results are presented in online suppl. Table 2). We found significant weak correlations between FAs C14:0 and C16:0 and fermented milk-based drinks in males before the adjustment (r = 0.217, p value: <0.01 and r = 0.245, p value: <0.01, respectively). In both cases, the associations lost significance and became essentially null after the adjustment. C14:0 and C16:0 in females were significantly correlated with petit cheese product intake, C14:0 after the adjustment (r = 0.005, p value: 0.01) and C16:0 before the adjustment (0.136, p value: 0.05).

We found significant associations between FA categories defined by quantiles (Q1, Q2, Q3, Q4) and the median of MetS components (Table 4). In females, we found that t-C16:1n-7 (Q4 vs. Q1: β = −2.97; 95% CI: −5.79, −0.16) was inversely associated with WC. In males, C15:0 (Q4 vs. Q1: β = 5.84; 95% CI: 1.82, 9.85) was associated with increased SBP. Results were opposite by sex for HDL-C. In males, we found that C15:0 (Q3 vs. Q1: β = −4.23; 95% CI: −7.98, −0.48) was associated with decreased HDL-C, while in females, C15:0 (Q3 vs. Q1: β = 4.75; 95% CI: 0.68, 8.82) and t-C16:1n-7 (Q4 vs. Q1: β = 6.54; β = 2.01, 11.07) were associated with increased HDL-C. In both sexes, C15:0 was associated with increased TG (in males, Q4 vs. Q1: β = 25.82; 95% CI: 1.62, 50.3, and in females, Q4 vs. Q1: β= 22.18; 95% CI: 1.04, 43.31). Additionally, C17:0 (Q2 vs. Q1: β = 27.19; 95% CI: 4.10, 50.29), (Q3 vs. Q1: β = 23.64; 95% CI: 1.03, 46.24), and t-C16:1n-7 (Q4 vs. Q1: β = 24.47; 95% CI: 5.66, 43.28) were associated with increased TG only in males. We also explored the associations between FAs and WC without adjusting for BMI z-score but including all the other covariates. Overall, results were similar except that C15:0 (Q3 vs. Q1: β = 9.32; 95% CI: 1.11, 17.53) was associated with increased WC in males. For females, we obtained results in the same direction as the BMI z-score-adjusted estimates (t-C16:1n-7 Q2 vs. Q1: β = −7.07; 95% CI: −13.82, −0.31, was associated with WC; data not shown). In sensitivity analysis, after adjusting C15:0 and C17:0 models for fish intake (g/day), there were no significant differences for most of the β coefficients obtained in quantile regression models between FAs categories and MetS components; there was just one significant change, such that the association that related C15:0 with a decreased HDL-C in males lost significance after the adjustment (Q3 vs. Q1: β = −3.54; 95% CI: −7.24, 0.16, data not shown). There were no significant changes in the β coefficients after we adjusted the associations between C17:0 and MetS components for fiber intake (g/day), and the association found between C17:0 and the increase in TG in males remained significant (data not shown).

We also obtained the associations between continuous C15:0, t-C16:1n-7, and C17:0 and each MetS component using quantile regression models (results are presented in online suppl. Table 3). In general, all the associations found were similar to those observed with FAs into quartiles. t-C16:1n-7 in females was negatively associated with WC (β = −1.28; 95% CI: −2.34, −0.22) and positively associated with HDL-C (β = 1.80; 95% CI: 0.04, 3.57). C15:0 and t-C16:1n-7 in males were positively associated with TG (β = 21.99; 95% CI: 3.84, 40.14; and β = 13.39; 95% CI: 5.05, 21.72, respectively).

Within this sample of Mexican adolescents, we found sex-specific associations between FA categories defined by quantiles and MetS components. In females, we found an inverse association between t-C16:1n-7 and WC. In males, higher plasma C15:0 was associated with higher SBP (comparing Q4 vs. Q1). Interestingly, we observed opposite directions in the association between C15:0 and HDL-C between sexes. In males, C15:0 was associated with a decrease in HDL-C, while in females, it was associated with an increase in HDL-C (comparing in both sexes Q3 vs. Q1). We found positive associations between plasma FA biomarkers and TG. In both sexes, C15:0 was associated with higher TG (comparing Q4 vs. Q1). Also in males, compared with the lowest quantile, the second and third quantiles of C17:0, and the highest quantile of t-C16:1n-7, were associated with increased TG. However, the FAs we employed as possible dairy biomarkers were not associated with total self-reported dairy intake, but we found weak significant associations between FAs and specific dairy types. In females, C15:0 was associated with sour cream intake before and after the adjustment for age and BMI z-score. In males, t-C16:1n-7 was associated with fermented milk based after the adjustment, and in females, C15:0 was associated with petit cheese product intake after the adjustment. In sensitivity analyses, we noticed the FAs had significant correlations with C14:0 and C16:0. C14:0 is mostly found in milk fat, and other natural sources are palm oil, coconut oil, and butter fat. C16:0 is found naturally in palm oil and palm kernel oil, as well as in other dairy products like butter, cheese, milk, and meat [31, 32]. For this reason, we cannot dismiss the possibility that the results we obtained are also reflecting the intake of other foods.

The association found between t-C16:1n-7 and the reduction in WC in females is consistent with the results of other studies. One cross-sectional multicenter study with adolescents from 10 European cities found that dairy was the food group that best identified in adolescents (12.5–17.5 years) with low CVD risk in terms of WC and sum of skinfolds [6]. In both genders, WC and sum of skinfolds were inversely associated with yogurt, milk, and yogurt-based beverages intake [6]. Dairy products like cheese, yogurt, and milk also provide protein to the diet, which helps maintain satiety to a greater extent than carbohydrates and fats and may prevent excessive energy consumption. In contrast, protein intake can stimulate an increase in thermogenesis and muscle protein anabolism, promoting fat mass retention in some individuals [31]. However, we cannot dismiss the possibility of reverse causality in our results due to the design of our study. Participants who were overweight could modify their consumption of certain types of dairy foods. Ultimately, longitudinal analyses that employ biomarkers are needed to clarify these results.

In both sexes, we observed positive associations between plasma biomarkers and TG, which is considered a biomarker of cardiovascular risk [32]. In adults, the evidence suggests that dietary fat could influence the composition and size of triacylglycerol-rich lipoproteins (TRL). In a randomized controlled trial, a diet rich in monosaturated FAs reduced the number of total TRL postprandial particles compared with the other meals, including one rich in butter [33]. This suggests that SFA in dairy products could have an impact on these biological pathways. In contrast, one cross-sectional study of 1,088 Spanish children (8–11 years) found that children with normal levels of TG and HDL-C consumed more whole-fat milk and less reduced-fat milk than children with dyslipidemic patterns [34]. But these results need to be interpreted cautiously due to the design of this study.

The associations between dairy biomarkers and HDL-C were different between the sexes, such that in males, C15:0 was related with a decrease in HDL-C, while in female participants, C15:0 and t-C16:1n-7 were associated with an increase in HDL-C. According to one meta-analysis of 60 randomized controlled trials, the specific SFA in dairy products increases HDL-C [35]. Further, the positive association of dairy biomarkers with HDL-C in female participants is consistent with a longitudinal study which evaluated the effect of usual eating patterns (9–17 years) on lipid levels at 18–20 years of age. They found that diets characterized by higher intakes of dairy products and whole grains had benefits on total cholesterol and LDL-C [36]. However, to the best of our knowledge, no other studies have reported differences in the influence of dairy intake on HDL-C depending on sex in adolescents. The sex differences we report could be attributed to sex-specific behaviors during adolescence or the hormonal differences involved in sexual maturation in males and females [37, 38]. In adults, similar results were reported in a cross-sectional analysis in the EPIC-Potsdam study. In women, C15:0 in erythrocyte membranes was positively associated with plasma concentrations of HDL-cholesterol, while in men, C17:0 in erythrocyte membranes was inversely associated with plasma HDL-cholesterol [39]. As another example, in males, SFAs have been associated consistently with CVD [40], while in the Estrogen Replacement and Atherosclerosis (ERA) trial, higher saturated fatty acid and lower polyunsaturated fatty acid intakes were associated with less progression of coronary atherosclerosis in females [41]. This could indicate that women have different associations between dietary factors and CVD risk progression than men [42], although the exact mechanism is unclear.

In males, C15:0 was associated with an increase of 5.67 mm/hg in SBP. In general, previous studies suggest that dairy product intake may have antihypertensive effects on blood pressure among youths. Yet other studies also hint at possible sex differences. One longitudinal analysis in adolescents (12–17 years) reported a decrease in mean systolic and arterial BP of 1.04 (p = 0.03) and 1.10 mm Hg (p = 0.02), for each dairy serving in females, but they did not observe significant associations in males [43]. Another study in children and adolescents showed that high dairy intake, defined as ≥2 servings of dairy per day, was associated with 1.74 mm Hg lower SBP and with 0.87 mm Hg lower DBP [44]. Nevertheless, they found no significant association of Ca or potassium intake on children’s blood pressure, suggesting the role of other antihypertensive components in dairy products. For example, proteins existing in milk have inhibitory effects in angiotensin-I-converting enzyme, which could result in BP decrease [45]. However, the effect of dairy on SBP could also be different by other factors like ethnicity [46].

Overall, the fact that our results were sometimes in opposite directions than the current literature could be related to the wide and complex variety of SFA in dairy foods, including short-, medium-, long-, odd-, and branched-chain FAs. These SFA may have different effects on cardiometabolic outcomes depending on the length and particular structure of the fatty acid. Another factor could be the tri-layered globule membrane found in dairy foods, which is rich in phospholipids and proteins capable of encapsulating milk fat [47]. The content and physical structure of this membrane is different between dairy products, and animal studies showed that this membrane could partially lower plasma cholesterol [47]. We also cannot dismiss the possibility that our associations could be due to the intake of some other foods or group foods correlated with dairy product intake.

Diet is difficult to measure accurately due to self-reported consumption and can be affected by some errors and reporting bias. Therefore, biomarkers for dietary food intake assessment can provide an alternative or complementary approach to assessing dietary intake as objective methods. To our knowledge, this is the first cross-sectional study that has employed these FAs as an objective measure of dairy intake to explore the association between dairy and cardiovascular risk factors in adolescents. In addition to milk and other usual sources, dairy products may be consumed in small amounts in various traditional preparations like fried corn foods and also in industrialized foods, such as baked goods, cream sauces, pizzas, coffee drinks, and others [48]. Therefore, small amounts of these foods are not accounted for when estimating dairy consumption from semiquantitative FFQs. Further, there can be many different sources of dairy in the diet that are hard to capture through the questionnaire alone. As one example, in our sample, 42% of the adolescents consumed more than one type of milk, from skim milk plus whole milk to flavored. Yet, the use of these FAs as dairy intake biomarkers has been little explored in adolescents [49]. Due to the multiple tests that we conducted since we had multiple exposures and outcomes, we cannot dismiss the possibility that some of our results could be spurious.

To our knowledge, there is just one randomized controlled trial study that has evaluated the use of these FAs as biomarkers of dairy intake during adolescence, and they did not find significant associations between plasma circulating FAs C15:0, t-C16:1n-7, and C17:0 and dairy intake in the short or long term [49]. However, dairy intake was assessed using just one 24-h recall, and it is necessary to consider that one single day could be insufficient to generate reliable estimates of habitual intake. We also found weak correlations between these FAs and dairy servings per day from the FFQ. The FFQ we employed for dietary assessment was not necessarily the best instrument for assessing the association between the employed FA biomarkers and dairy intake. Further, plasma concentrations of FAs C15:0, t-C16:1n-7, and C17:0 are considered short-term biomarkers of dairy intake, while erythrocyte concentrations are considered long-term biomarkers [50, 51] which could also partially explain why these FAs did not correlate with total dairy servings per day. Twenty-four-hour dietary recalls can provide a more accurate estimation of absolute intake of foods in contrast to FFQs [52].

Another consideration is that C15:0 and C17:0 could also be biomarkers from fish intake [29]. Therefore, it has been pointed out that its use may not be appropriate in populations with a high consumption of fish. However, fish consumption is modest in the Mexican population, and the mean intake in our study was 17.4 g per day, according to the FFQ. In one study in adults, C17:0 was associated with fish fat intake, while C15:0 was associated with dairy fat intake, independent of fish fat intake [29]. Further, it is worth mentioning that circulating fatty acid measures may reflect both metabolism and dietary intake. More studies are required to explore the use of these FAs as biomarkers of dairy consumption since their use must be characterized for different ages and populations. Finally, our results should not be interpreted as causal effects due to the cross-sectional design of the study.

Our results suggest FAs C15:0, t-C16:1n-7, and C17:0, biomarkers of dairy intake, may be associated in different directions with cardiovascular risk factors and, for some parameters, in a sex-dependent manner. Our findings thus point out the complexity of the relationship between dairy intake and MetS, which could be related to nutrients found within dairy products and saturated fats of varying chain lengths that may have different directions of effects on MetS components. Longitudinal epidemiological studies that employ objective measurements are needed to explore these associations in adolescents due to their plausible contribution to CVD risk in adulthood.

We are grateful to Dr. Arun Das for his expertise on the laboratory fatty acid assessments. We are also grateful to the American British Cowdray Medical Center (ABC) in Mexico for providing research facilities and the study team: María Guadalupe Rodríguez, Beatriz Escobedo, Ana Benito, María de Jesús Rodríguez, Jorge Zúñiga-Ramírez, Tomás Villa, Rubén Valencia, and Nicolás Reynoso.

All the subjects included in this study participated voluntarily and gave consent. Adolescents provided written informed consent (age ≥18 years), and for adolescents under 18 years of age, the mother or legal guardian provided written informed consent. Additionally, adolescents under 18 years of age provided written assent. This study protocol was reviewed and approved by the Research Ethics Committee, National Institute of Public Health, approval number (1583).

The authors have no conflicts of interest to declare.

Financial resources for monitoring the cohort were provided by the National Institutes of Health (NIH) (Grant No. 1P01ES022844-01). Dr. Jansen was supported by the NHLBI (Grant No. K01HL151673). Funding for fatty acid analysis was provided by the University of Michigan Momentum Center Pilot and Feasibility Grant and conducted by the Metabolomics Core Services, which reports support from the grant U24 DK097153 of the NIH Common Fund Project to the University of Michigan. The funding sources had no role in the design and conduct of the study or in the preparation of the manuscript.

Erica C Jansen, Alejandra Cantoral, and Rebeca Trejo-Reyes conceived of the research question and designed the study; Martha María Téllez-Rojo and Karen E Peterson conducted the research and provided essential materials; Rebeca Trejo-Reyes performed statistical analysis and wrote the first draft of the paper; Erica C Jansen, Alejandra Cantoral, Héctor Lamadrid-Figueroa, Ana Baylin, and Larissa Betanzos-Robledo interpreted the data; Alejandra Cantoral, Erica C Jansen, Ana Baylin, and Larissa Betanzos-Robledo contributed to literature reviews and drafting of this paper; and all authors contributed to the interpretation of the data, critically revised the manuscript, read, and approved the final manuscript.

All data generated or analyzed during this study are included in this article and its online supplementary material files. Further inquiries can be directed to the corresponding authors.