Maternal lipids and leptin concentrations are associated with large-for-gestational-age births: a prospective cohort study

The change in maternal lipid, leptin and adiponectin concentrations during pregnancy and infant birth weight (BW) is still poorly characterized. Thus, the aim of the study was to evaluate the association of maternal lipids, leptin and adiponectin throughout pregnancy with large-for-gestational-age (LGA) births and BW z-score. A prospective cohort of 199 mothers was followed during pregnancy in Rio de Janeiro, Brazil. The statistical analyses comprised multiple logistic and linear regression. Women delivered 36 LGA and 11 small-for-gestational-age newborns. HDL-c rate of change throughout pregnancy was negatively associated with BW z-score (β = −1.99; p = 0.003) and the delivery of a LGA newborn (OR = 0.02; p = 0.043). Pregnancy baseline concentration of log leptin was positively associated (OR = 3.92; p = 0.025) with LGA births. LDL-c rate of change throughout pregnancy was positively associated with BW z-score (β = 0.31; p = 0.004). Log triglycerides and log adiponectin were not significantly associated with BW z-score or LGA birth. In conclusion, a higher log leptin pregnancy baseline concentration and a lower HDL-c rate of change during pregnancy were associated with higher odds of having a LGA newborn. These maternal biomarkers are important to foetal growth and could be used in prenatal care as an additional strategy to screen women at risk of inadequate BW.


Discussion
This study has two main findings. First, the HDL-c rate of change during pregnancy was negatively associated with BW z-score and the delivery of LGA newborns, whereas the pregnancy baseline log leptin concentrations, but not the rate of change, were positively associated with these outcomes. Secondly, the LDL-c rate of change over time was positively associated with BW z-score. We did not observe significant associations between gestational changes of log TG and log adiponectin and BW z-score or LGA births in the adjusted models. Moreover, BMI was not an effect modifier of the associations of lipids and leptin with BW z-score and LGA births in our sample.
One limitation of this study was the lack of an oral glucose tolerance test to diagnose gestational diabetes mellitus (GDM), which is strongly associated with BW [24][25][26] . Thirteen women reported a diagnosis of GDM during pregnancy (6.4%). We compared the analyses with and without these women and found no significant changes in the results. Furthermore, all the adjusted models were controlled for fasting glucose. Multiple regression models were additionally adjusted for other established confounders such as early pregnancy BMI, gestational weight gain (GWG) and smoking habit 3,5,7,27 . A large number of statistical comparisons were carried out, but no adjustments were made for multiple comparisons. This is also a limitation of our study, as it inflates the likelihood of Type I errors. The small number of cases of LGA can be considered another limitation of the study. However, even with a modest sample size, we were able to find statistically significant associations, indicating a stronger relationship between lipids and log leptin and BW z-score/LGA. We only measured total adiponectin in our study. Therefore, the lack of information regarding the high-molecular-weight (HMW) adiponectin, which is the form that has been reported as more strongly correlated to many outcomes such as diabetes and cardiovascular disease [28][29][30] , limits our conclusions regarding the effect of adiponectin in BW. The measurement of lipid, leptin and adiponectin concentrations during all three trimesters of pregnancy is a strength of this study. Furthermore, BW z-score and LGA were calculated using a population-based international growth curve that has evaluated 20,486 women and their newborns in eight geographically defined urban populations, including Brazil 31 . The study design enabled us to better understand the relationship between the metabolic changes in maternal biomarkers and infant BW. The use of a two-stage procedure 32 to model the association between time-dependent exposures and a non-time-varying outcome was also a strength of this study. This procedure considers that the repeated measures are correlated, accounts for different rates of change in the exposures and uses their longitudinal predictions to evaluate the association with the outcome. The HDL-c rate of change during pregnancy was inversely associated with BW z-score and the delivery of LGA newborns. In a case-control study, Kramer et al. 33 also found that HDL-c concentrations were inversely associated with BW, i.e., women who delivered SGA infants had higher concentrations of HDL-c compared to women who delivered AGA infants. Misra et al. 14 evaluated the association between HDL-c concentrations at 10-14, 16-20, 22-26 and 32-36 gestational weeks and BW in 143 American women stratified for pre-pregnancy BMI. The authors found an inverse and statistically significant association at all time points, but only in overweight/ obese women. Misra et al. 14 also tested the influence of time-dependent changes in maternal serum HDL-c on BW and concluded that the trajectory of HDL-c change over time was not significantly associated with BW in any BMI category. In our theoretical model, maternal BMI was considered a possible confounder of the relation between maternal biomarkers (lipids, leptin and adiponectin) and BW since BMI is associated with both exposures and outcome. Previous publications also tested BMI as an effect modifier of the relation between maternal lipids and BW 14 . To test if BMI had the same effect in our sample, we carried out regression models including interaction terms between lipids, leptin and adiponectin (intercept and slope) with early pregnancy BMI. In contrast to Misra et al. 14 , we did not find significant interactions in crude or adjusted models between BMI with any of the maternal biomarkers on BW. The difference observed between the two studies can be attributed to BW classifications and the statistical procedure adopted.
We observed that the rate of maternal LDL-c change during pregnancy was positively associated with BW z-score. Although some previous studies did not report this association 6, 14, 17 , Pecks et al. 34 found that the mean LDL-c concentrations were lower in mothers of term (n = 5) and preterm (n = 10) intrauterine growth restricted newborns compared with term (n = 5) and preterm (n = 10) controls, respectively. The association reached statistical significance only between preterm groups. Merzouk et al. 35 found that obese women who gave birth to macrosomic newborns had significantly higher concentrations of LDL-c than those who delivered newborns with a healthy weight.
We performed additional analyses to understand if the results of the present study remained the same when only the AGA subsample was considered. When LGA and SGA cases were simultaneously removed from the analysis, LDL-c and HDL-c lost the significant association with BW z-score. However, when only SGA cases were removed, HDL-c remained significantly associated. When only LGA cases were excluded, only LDL-c remained significantly associated with BW z-score. These analyses revealed that the associations between lipids and BW z-score were partially driven by the extremes of the BW distribution. Cholesterol is essential for foetal development; it is part of cell membranes, necessary for activation of various signalling pathways and a precursor of steroid hormones. Although most of foetal cholesterol is endogenously obtained by de novo synthesis in the liver, there is evidence that maternal cholesterol (exogenous source) crosses the placenta and is important for foetal growth and impacts metabolic function of extraembryonic foetal tissues 10,36 . Little is known about the biological mechanism by which maternal cholesterol affects BW, but it seems to include altered sterol hormone metabolism and impaired cell cycle and signalling of growth factors (including insulin) and is indirectly by affecting placental transport of nutrients 12,13,36 . This mechanism may be involved in the positive association between LDL-c and BW; however, it does not fully explain the inverse association between HDL-c and BW and LGA. We suppose that it may also be related to its antioxidant and anti-inflammatory properties 37 .

Variables
We did not find significant associations between maternal serum concentrations of TG and BW or LGA births, in line with results from Retnakaran et al. 17 and Crume et al. 38 . TG concentrations are known to affect foetal growth in women who have gestational diabetes 16,39 ; however, in studies with non-diabetic women, it seems not to have the same impact on BW 40 .
We found a positive association between log leptin pregnancy baseline concentrations (intercept) and LGA births and no association between log leptin rate of change during pregnancy and BW z-score or LGA. Experimental studies have indicated a role of leptin in the regulation of the transfer of amino acids and lipids through the placenta 41,42 ; however, the literature remains contradictory, and there is no consensus regarding the association between leptin concentrations and infant BW in humans. Our findings are in line with a study by Shrof et al. 23 , which evaluated 1,304 American women and found that those who delivered LGA neonates had higher leptin concentrations than women who delivered term AGA neonates. Franco-Sena et al. 43 evaluated 195 women between 8 and 13 weeks of gestation and found an association between lower concentrations of leptin and a higher risk of SGA. However, other authors have identified an inverse association between leptin concentrations and BW or LGA 17 or did not find a significant association 44 . One possible explanation for these contradictory results is the difference in sample size or in the times of leptin assessment between these studies.
We did not find significant associations between total adiponectin and BW z-score or LGA births. Ong et al. 28 also did not find significant associations between total adiponectin and BW in a sample of 58 women of Caucasian descent with singleton pregnancies. However, they found a borderline significant association between HMW adiponectin and BW and a significant inverse association between the ratio of HMW to total adiponectin and BW (β = −19.2; p = 0.018). In contrast, Retnakaran et al. 17 found a significant inverse association between total adiponectin in the third trimester and BW in a sample of 422 women without GDM. Although we did not find statistically significant associations between total adiponectin and BW, we observed the same trend reported by Retnakaran et al. 17 . We hypothesize that the lack of association observed in our study and in the one conducted by Ong et al. 28 could be attributed to the sample sizes.
This prospective study of low-income women found that maternal HDL-c and LDL-c rates of change during pregnancy were associated with BW z-score, even after adjusting for important confounders such as maternal early pregnancy BMI, GWG and fasting glucose. We also observed that leptin concentrations were positively associated with LGA births. The association between HDL-c and log leptin with LGA births persists even when we enter both variables in the fully adjusted model. There are no established gestational cut-off points for the assessment of lipid concentrations, so any alteration is considered a physiological adaptation of pregnancy. However, our results indicate that lipids and leptin are important to foetal growth and that in the future, the evaluation of lipid changes and leptin concentrations during pregnancy may be used as an additional strategy to screen women at risk of delivering LGA newborns. Although we found relevant associations, additional studies exploring these relationships in different populations and with lager sample sizes are needed to propose specific cut-off points.

Methods
Setting and participants. We conducted a prospective cohort study in pregnant women at a municipal health centre in Rio de Janeiro, Brazil from November 2009 to June 2012. Eligibility criteria were: age between 20 to 40 years and pregnancy between 5 and 13 completed weeks of gestation, with no known chronic non-communicable diseases (except obesity).
Women were studied at three time points during pregnancy:  pre-pregnancy diagnosis of chronic non-communicable diseases, including women with fasting glucose values ≥126 mg/dL at the 1 st trimester (n = 12); the presence of infectious or parasitic diseases (n = 9); twin pregnancy (n = 4); and miscarriage (n = 25). We further excluded women with missing values for BW or gestational age at birth (n = 24), with baseline underweight (BMI < 18.5 kg/m²; n = 4) and women with no lipid measurements at the first trimester (n = 22). The baseline sample comprised 199 pregnant women. Thirteen women reported a diagnosis of GDM during pregnancy (6.5%), and two developed hypertension (>140 and/or >90 mmHg systolic and diastolic respectively) during pregnancy.

Measurements. BW (g) was obtained from the child vaccination booklet at the post-partum interview.
We also evaluated BW z-score for gestational age and sex according to the international foetal and newborn growth consortium for the 21 st Century (Intergrowth-21 st ) curves 31 . We classified newborns as LGA when the BW, according to the gestational age and sex-specific Intergrowth-21 st curves, was above the 90 th percentile and as SGA when it was below the 10 th percentile.
The gestational age was calculated based on data from the first ultrasonography examination if it was performed prior to 24 weeks of gestation (n = 189; 95.0%). In cases where this measure was unavailable, the date of the last menstrual period was used (n = 10; 5.0%). The gestational age at delivery was calculated based on the date of birth reported at the post-partum visit.
During each trimester of pregnancy, a nurse technician collected two fasting blood samples (2.5 mL) from each woman in vacutainer tubes containing EDTA or separator gel. The samples were centrifuged (5 minutes, 5031 g), and the serum and plasma were immediately stored at −80 °C. Plasma leptin (ng/mL) and total adiponectin (µg/mL) concentrations were measured during the three pregnancy trimesters using commercial ELISA kits (Millipore, St. Charles, Missouri, USA), with sensitivities of 0.50 ng/dL and 0.78 µg/mL, respectively.
Maternal characteristics recorded at baseline included age (years), monthly per capita family income (US$), education (years of schooling), current smoking habits (no or yes), alcohol consumption (no or yes), parity (0 or ≥1 parturitions), and pre-pregnancy LTPA (no or yes). The sex of the newborn (male or female) was reported in the post-partum questionnaire.
Maternal body weight (kg) was obtained using a digital scale (Filizzola PL 150, FilizzolaLtda, Brazil). Height was measured twice, using a portable stadiometer at baseline (Seca Ltd., Hamburgo, Germany). Early pregnancy BMI [weight (kg)/height (m) 2 ] was calculated based on first trimester weight and height. The cutoff point proposed by the Institute of Medicine 46 was used to classify the women during early pregnancy as normal weight (18.5 to 24.9 kg/m 2 ) or as overweight/obese (≥25.0 kg/m 2 ). Anthropometric measures were collected according to standardized procedures and performed by trained interviewers 47 . GWG (kg) was calculated as the difference between the last weight measured before delivery (mean gestational age = 37.9 weeks; SD = 2.3) and the first trimester weight (mean gestational age = 9.6 weeks; SD = 2.2).
Fasting glucose (mg/dL) was measured in all pregnancy trimesters using the glucose oxidase-peroxidase enzymatic colorimetric method and a Wiener Lab kit (Rosario, Argentina).
Total energy intake was assessed using a semi-quantitative food frequency questionnaire (FFQ) validated for the adult population of Rio de Janeiro 48 . The FFQ was composed of 81 food items, eight frequency options and household measure portion options. The FFQ was administered at the first gestational trimester (5-13 weeks of gestation) and referred to food intake 6 months prior to pregnancy. For statistical analysis, frequency options were transformed into daily frequencies and household measures into grams (g) or milliliters (ml) 49 . The daily amount consumed (g or ml/day) of each food item was obtained by multiplying the daily frequency (3x/day; 2 to 3x/day; 1x/day; 5 to 6x/week; 2 to 4x/week; 1x/week; 1 a 3x/month and never or almost never) per portion size. The Brazilian Food Composition Table (TACO) 50 was used as the main database to determine food nutritional composition and The National Nutrient Database for Standard Reference provided by the United States Department of Agriculture 51 was considered as a secondary option when a food item was not available in the TACO database.  signed a consent agreement, which was obtained freely and spontaneously, after all necessary clarifications had been provided. All ethical procedures of this study involving human beings followed the Brazilian Resolution 196/96.

Statistical analysis.
General characteristics of the sample were described as the means and standard deviations (SD) for continuous variables and proportions (%) for categorical variables. Student's unpaired t test was used to compare means, and the chi-square test was used for proportions. The mean variation (SD) between the first and the third trimester biomarkers values was calculated.
To evaluate the association between biomarker changes during pregnancy and BW z-scores and LGA, we used a two-stage method. (1) We constructed a linear mixed-effect model (LME) for each exposure (maternal lipids, leptin and adiponectin) including gestational age at sampling as fixed and random effects and estimated the best linear unbiased prediction (BLUP) of random coefficients. The predicted intercept refers to the mean lipid, leptin or adiponectin exposure level, i.e., the biomarker concentrations when the gestational age was zero, and was labelled as the pregnancy baseline concentration. The predicted slope refers to the rate in concentration changes per gestational age during pregnancy. (2) The BLUP predicted intercept and slope were simultaneously included as continuous predictors in linear and logistic regression models having BW z-score and LGA as outcomes, respectively. This approach considers that repeated measures are correlated and estimates time trends of exposure even for women with missing values across pregnancy, increasing the power of the analysis 32 . Since the LME model assumes that the dependent variable is normally distributed, we have log transformed variables with skewed distribution (TG, leptin and adiponectin).
The modelling process was performed in three steps. We constructed three linear (outcome: BW z-score) and three logistic (outcome: LGA) regression models to test the association between maternal biomarkers (lipids, leptin and adiponectin) and BW z-score or LGA births, reporting the regression coefficient (β) and odds ratio (OR), respectively, and their 95% confidence intervals (95% CI). In the first models, lipids, leptin and adiponectin intercept and slope variables were included together in the same model. The second models were additionally adjusted for age (years), education (years of schooling), pre-pregnancy LTPA (no/yes), smoking habit (no/yes), parity (number of parturitions), total pre-pregnancy energy intake (kcal/day), fasting glucose (mg/dL) and GWG (kg). In the third models, the early pregnancy BMI variable was added. We also tested if there was an interaction between maternal BMI and lipids, leptin or adiponectin on BW z-score and LGA births. The adjustment variables were chosen based on the biological plausibility of the association.
Since the models were adjusted for variables that could be correlated with each other, we tested the correlation between them. Variables with strong correlations (Pearson or Spearman coefficient ≥0.7) were candidates to be excluded. The TC was not included in the fully adjusted models due to its strong correlation with HDL-c and LDL-c.  We further investigated the occurrence of multicollinearity in the full models using the variance inflation factor. We predicted residuals and fitted values of the outcomes for our final models. We checked normality of the residuals and constructed two-way scatter plots between the residuals and predicted values of BW to detect outlying observations and to check the assumption of constant variability of outcomes across values of exposure (homoscedasticity) and scatter patterns of the residuals 52 .   Table 4. Logistic regression between maternal lipids, log leptin and log adiponectin time trends during pregnancy and large-for-gestational-age (LGA) 1 in a sample of women and their newborns followed at a public health centre in Rio de Janeiro city, Brazil, 2009-2012. 1 LGA was classified according to the international foetal and newborn growth consortium for the 21 st Century (Intergrowth-21 st ) curves. 2 Odds ratio. 3 p-value refers to the logistic regression. Notes: Intercept variables represent the prediction of the mean exposure level, i.e. biomarkers concentrations when the gestational age was zero and the slope the trend of change in concentrations during pregnancy. Model 1 included lipids, leptin and adiponectin intercepts and slopes variables; Model 2 was additionally adjusted for women's age, education, parity, pre-pregnancy practice of leisure time physical activity, pre-pregnancy energy intake, glycaemia and gestational weight gain. Model 3 was additionally adjusted for early pregnancy BMI. The models were not adjusted by smoking habit since none of the mothers of LGA smoked during pregnancy. Abbreviations: CI = confidence interval; HDL-c = high-density lipoprotein; LDL-c = low-density lipoprotein; LGA = large-for-gestational-age; LTPA = leisure time physical activity.   Table 5. Linear regression between maternal lipids, log leptin and log adiponectin time trends during pregnancy and birth weight z-score 1 in a sample of women and their newborns followed at a public health centre in Rio de Janeiro city, Brazil, 2009-2012. 1 Birth weight was classified according to the international foetal and newborn growth consortium for the 21 st Century (Intergrowth-21 st ) curves. 2 Linear regression coefficient. 3 p-value refers to the linear regression. Notes: Intercept variables represent the prediction of the mean exposure level, i.e. biomarkers concentrations when the gestational age was zero and the slope variables represent the trend of change in concentrations during pregnancy. Model 1 included lipid, leptin and adiponectin intercepts and slopes variables; Model 2 was additionally adjusted for women's age, education, parity, smoking habit, pre-pregnancy practice of leisure time physical activity, pre-pregnancy energy intake, glycaemia and gestational weight gain. Model 3 was additionally adjusted for early pregnancy body mass index. Abbreviations: CI = confidence interval; HDL-c = high-density lipoprotein; LDL-c = low-density lipoprotein; LTPA = leisure time physical activity.