Maternal diet quality and associations with plasma lipid profiles and pregnancy-related cardiometabolic health

This study of pregnant women utilises data and biological samples collected as part of the Creatine in Pregnancy Outcomes (CPO) study. A detailed outline of the CPO study protocol and methodology has been previously reported [23]. Ethical approval for the original CPO study was granted in August 2015 from Monash Health (Ref 14140B) and Monash University (Ref 7785). The current study was granted ethical approval in July 2020 by the Deakin University Human Research Ethics Committee (DUHREC) (2020-236). Study procedures followed were in accordance with ethical standards of Monash Health, Monash University and Deakin University’s committees on human experimentation and conformed to the Declaration of Helsinki.

Women attending low-risk antenatal clinics at Monash Health in Melbourne, Australia, were recruited at 10–20 weeks gestation. Monash Health is a large tertiary teaching hospital in metropolitan Melbourne, Australia, and provides antenatal care to over 9000 women per year. Women gave informed, written consent to participate and agreed to have samples collected for the CPO study and biobanked for future research approved by Monash Health. For inclusion, women needed to be aged 18–40 years and classified as having a ‘low risk’, singleton pregnancy, meaning they had no known pre-existing medical or obstetric condition. Women were excluded if they were non-English speaking, had been previously diagnosed with Type 1/ Type 2 diabetes, taking creatine supplements during pregnancy, required high-risk pregnancy care due to underlying or emergent health complications or disclosed ongoing alcohol or drug use during pregnancy. The original CPO study was conducted between 2015 and 2017 and collected data for 284 women.

The current study utilised data from three antenatal visits; the first ([mean ± SD]) (15 ± 3 weeks gestation), third (27 ± 3 weeks gestation) and fifth (35 ± 2 weeks gestation) visits. We defined these timepoints as early, mid- and late pregnancy, respectively, consistent with common terms used to describe pregnancy progression [24, 25]. Demographic data included maternal age, education, gravida, smoking status, sex of the baby and gestational weeks at delivery. Region of birth was defined using birth countries combined into the following categories: Australian/New Zealand (Australia and New Zealand); Asian/South Asian (India, Nepal, China, Afghanistan, Singapore, Sri Lanka, Philippines, Malaysia, Pakistan, Bangladesh, Thailand); European/UK/Canada (Russia, Switzerland, Ireland, UK, Canada, Greece, France, Scotland, Poland, Germany, England); Middle Eastern/African/South American (Lebanon, Israel, Kuwait, Iran, Venezuela, Mauritius, Ghana, Ethiopia, Colombia).

Dietary intake was assessed using the Cancer Council of Victoria’s (CCV) validated Dietary Questionnaire for Epidemiological Studies (DQES version 2) [26]. Design of the DQES has been described in detail elsewhere [27]. This consisted of 74 items that assessed the frequency of intake across multiple food groups. The DQES is designed to reflect usual intake over the past 12 months. It has been previously validated against 7 day food diaries among reproductive age women [26], showing comparable performance to other widely used food frequency questionnaires [28] and has been used frequently as a validated tool for assessing dietary intake among Australian women [26] and pregnant populations [29, 30]. Women completed the online DQES at early pregnancy ([mean ± SD]) (15 ± 3 weeks gestation) and again at late pregnancy (35 ± 2 weeks gestation). Completion of the DQES at early pregnancy reflected dietary intake during the first trimester of pregnancy and approximately six months leading up to pregnancy. Dietary assessment via the DQES at late pregnancy assessed women’s dietary intake across all of pregnancy specifically. Intake responses from the 74 items are grouped into categories (cereal foods, sweets and snacks, dairy products, meats and fish, fruit and vegetables) and converted to a Dietary Guideline Index (DGI) score, indicating adherence to the Australian Dietary Guidelines (2013) for adults [31].

The DGI has previously been used to assess DQ and risk of cardiometabolic disease in Australian adults [20, 32, 33]. The DGI included 11 food components (Table 1): diet variety; vegetables; fruit; grains and cereals; meat and alternatives; dairy and alternatives; discretionary foods; saturated fat; unsaturated fats; sugar and alcohol. Two additional components, fluid intake and limiting foods high in salt, were not included as the DQES version 2 did not include questions appropriate to score for those items. The DGI at late pregnancy was calculated to adhere to pregnancy specific guidelines which differ for the grains and cereals food group and meat and alternative food group compared to non-pregnancy guidelines. Possible scores for each food category were between 0 and 10 for nine components and 0 and 5 for two components, with a higher score in each category indicating better DQ and higher compliance with the Dietary Guidelines. The total DGI score for each participant ranged between 0 and 110, with a higher score indicating higher adherence to the Dietary Guidelines and, therefore better DQ [20, 32].

Comprehensive lipid profiling was undertaken, and 698 species across 37 lipid classes were measured. Such extensive lipidomic analysis allows for identification of important biomarkers for assessment of disease risk [34] and enables an understanding of lipid metabolism in pregnancy which would otherwise go undetected. Plasma lipids were analysed in samples collected at early ([mean ± SD) (15 ± 3 weeks) and mid (27 ± 3 weeks)-pregnancy for the current study. Blood was collected in lithium heparin tubes, stored on ice and spun (400 g, 20 min, 4°) to isolate plasma. Lipids were isolated from plasma using lipidomics analysis [35]. In brief, 10 μL of plasma was mixed with 90 μL of butanol/methanol (1:1) and 10 µL of an internal standard mix [35]. Samples were sonicated on a sonicator bath for 1 h, maintained at 25 °C and centrifuged at 13,000g for 10 min and the supernatant transferred into glass vials with inserts for mass spectrometry analysis. Lipidomic analysis was performed using high-performance liquid chromatography in conjunction with an Agilent 6490 QQQ mass spectrometer. Liquid chromatography was done using a Zorbax Eclipse Plus C18 (Agilent Technologies, USA), 1.8 μm, 100 × 2.1 mm column with running solvents comprising of water/acetonitrile/isopropanol at ratios of 50:30:20 and 1:9:90, respectively (A and B), both with 10 mM ammonium formate. The column was maintained at 45 °C during the run and the autosampler controlled at 25 °C. Lipid extracts (1 μL) were injected and separated under a stepped gradient condition with a flow rate of 400 μL/min as previously described [36]. An additional passivation step was performed prior to the run, where 0.5% phosphoric acid in 9:1 acetonitrile/water was run through the HPLC system for an hour and subsequently washed with 1:9 acetonitrile/water overnight. Additional characterisation of phospholipid structures was done as reported previously [36]. Each lipid was integrated manually via Agilent software. Normalisation was done between batches (3) using the pooled plasma QC’s spaced 20 samples in between each sample. This was conducted by median centering, where the median concentration of the QC’s for each batch was used to align the concentrations between each batch. Data from the lipidomic analysis were integrated using MassHunter V8.00 (Agilent, Australia). Relative lipid concentrations were calculated by relating the area under the chromatogram for each lipid species to the corresponding internal standard.

Systolic and diastolic BP (mm Hg) was measured at each antenatal clinic visit by a research midwife, using Welch Allyn sphygmomanometers with manual cuff inflation and calibrated regularly as per manufacturers instruction. One reading was taken, and if abnormal, three consecutive readings were done at short intervals over 1.5 h, and the average of these readings was recorded. Medical history data (pre-existing medical conditions and GDM diagnosis) were recorded from hospital records. As part of routine care, all women without existing GDM undergo a 75 g oral glucose tolerance test at 24–28 weeks gestation [37]. A diagnosis of GDM was based on the International Association of the Diabetes and Pregnancy study groups (IADPSG) criteria; one of fasting glucose ≥ 5.1 mmol/L; 1 h level ≥ 10 mmol/L; or 2 h level ≥ 8.5 mmol/L [37]. Women’s height (cm) at the first antenatal clinic visit and weight (kg) using calibrated industrial scales at each clinic visit were measured by the same research midwife. Gestational weight gain (GWG) was calculated as weight at the last (fifth) clinic visit minus weight at the first antenatal clinic visit. Body mass index (BMI) (kg/m²) was calculated using World Health Organisation (WHO) BMI criteria for healthy weight, overweight or obese classifications [38].

Data were analysed using STATA/SE statistical software version 16.0. Descriptive analyses reported maternal demographics, systolic and diastolic BP, GDM diagnosis and maternal anthropometry as mean ± SD or n (%). Dietary Guideline Index scores were reported as mean ± SD and the score range. Change in DGI scores from early to late pregnancy was assessed via paired t-tests for women who had complete FFQ data at both early and late pregnancy (n = 203). Paired-samples t-tests were conducted to assess mean changes in all lipids between early and mid-pregnancy. Linear and logistic regression models were used to assess associations between total DGI score at early pregnancy with continuous (plasma lipids, BP and GWG) and categorical (GDM diagnosis) outcomes, respectively. Predetermined factors known widely to influence lipids including age [39], BMI [40, 41] and gestational week [42, 43] were included as covariates in the regression models. The model examining the association between early pregnancy diet and mid-pregnancy lipid concentrations also included early pregnancy lipid concentrations as a covariate. Additional covariates included well established factors known to impact (i) BP in pregnancy (smoking [44], age [19, 45], early pregnancy BMI [19, 45] and gestational week at BP measurement [46]); (ii) GDM diagnosis (age [19, 47] early pregnancy BMI [48], country of birth [47]) and (iii) excess GWG ((age [49], early pregnancy BMI [50] and education [49]). These were further included in models assessing association of DGI with pregnancy complications. Finally, the Benjamini–Hochberg method [51] was applied to regression models involving lipid outcomes at both early and mid-pregnancy. This method is widely used for controlling the false discovery rate (FDR) using sequential modified Bonferroni correction for multiple comparisons [51, 52] and has previously been applied to regression models involving the plasma lipidome [36, 39]. A cut-off of FDR of 0.05 and a less conservative level for comparison of 0.2 were used for significance. In all other models (unadjusted and adjusted), statistical significance was set as p < 0.05.

Of the 286 women enrolled, four women were excluded due to complete missing data, 52 women were excluded due to missing FFQ and anthropometric data and 15 women were excluded as their GDM status was unknown and not available to be retrieved from the electronic medical record system. This left data for 215 included the analyses (Supplemental material). Women were 31.5 ± 3.9 years ([mean ± SD]), and 64% of women were tertiary educated (Table 2). Just over half (54.9%) of all women were born in Australia/New Zealand, over one third (35.8%) were born in Asia/South Asia/Middle East/Africa/South America and 9.3% were born in Europe/UK/Canada. Over half of all women (56.3%) were pregnant with their first baby, and mean gestational age at delivery was 39.2 ± 1.4 weeks. Mean early pregnancy BMI was 25.0 ± 4.6 kg/m². Women gained 10.9 ± 4.6 kg during pregnancy, and 9.7% of women were diagnosed with GDM. Mean blood pressure measures across pregnancy (systolic/diastolic) were 110/67 ± 12/9.3 mm Hg, 111/67.6 ± 11.0/7.8 mm Hg and 114./70.2 ± 12.1/9.8 mm Hg at early, mid- and late pregnancy, respectively, with a highest overall recorded measure of 122/77.1 ± 1.8/8.7 mm Hg.

Compared to early pregnancy total DGI score ([mean ± SD]) (70.7 ± 11.4), DGI score at late pregnancy was statistically significantly lower (66.5 ± 12.6) (p < 0.0005) (Table 3). Scores for total cereal decreased significantly from early to late pregnancy (3.3 ± 1.2 and 2.5 ± 1.1) (p < 0.0005) and scores for total meat and alternatives also decreased significantly from early to late pregnancy (4.5 ± 1.1 and 3.9 ± 1.4) (p < 0.0005). Scores for total dairy and alternatives increased significantly from early to late pregnancy (6.3 ± 2.5 and 7.1 ± 2.5) (p < 0.0005). Intakes were poor for multiple key individual food components. Scores for limiting discretionary food items were well below the maximum score of 10 at both early and late pregnancy (4.0 ± 4.9 and 4.0 ± 4.9, respectively). DGI scores for vegetables and fruit (scored out of 10) were suboptimal at early and late pregnancy (3.3 ± 1.9 and 3.1 ± 1.8) (p = 0.078) and (7.2 ± 2.8 and 7.5 ± 2.8) (p = 0.128), respectively, where overall scores decreased for vegetables, although this difference was not significant. Further, DGI scores for limiting saturated fats and sugars were suboptimal decreased significantly from early to late pregnancy (3.6 ± 4.5 to 3.0 ± 4.4) (p = 0.018) and (5.1 ± 5.0 to 3.9 ± 4.9) (p ≤ 0.003), respectively.

Changes in lipid level were observed for the majority of the lipids, with most increasing significantly from early to mid-pregnancy, while a smaller proportion decreased significantly from early to mid-pregnancy (Supplemental material). There were no statistically significant associations found between DGI score at early pregnancy and lipid outcomes at either early or mid-pregnancy when the model was adjusted for a FDR of 0.05 using the Benjamini–Hochberg procedure (Supplemental material). When we applied a FDR of 0.2, there were multiple statistically significant associations between early pregnancy DGI score and plasma lipids at mid-pregnancy, but not early pregnancy. We run an unadjusted model for all lipids and found that only one lipid (LPCO221) reached significance at a FDR of 0.2, despite not being significant in the adjusted model (data not presented). There was no significant association found in unadjusted models at a FDR of 0.05. Figure 1 presents associations between DGI score at early pregnancy with 698 lipid species at mid-pregnancy. Under the FDR of 0.2, significant associations were found for 13 of the 698 lipids. These significant associations are further summarised in Table 4. Total DGI score at early pregnancy was significantly associated with seven plasma TGs (out of 113 total TG species) at mid-pregnancy, six of which were negative associations. Three TGs contained the 18:0 saturated fatty acid stearic acid ((TG54:0)[NL18:0]; (TG48:0)[NL18:0]); (TG50:0)[NL18:0]). Total DGI score was significantly positively associated with one unsaturated fatty acid ((TG52:4)[NL16:1]), palmitoleic acid.

A statistically significant inverse association between DGI score and diastolic BP at late pregnancy was found (−0.14 mm Hg, 95% CI −0.26 −0.02, p = 0.028) (Table 5). No statistically significant associations were found between DGI score at early pregnancy with BP at any time point, GDM diagnosis (26–28 weeks gestation) or total GWG.

A major strength of our study was the utilisation of comprehensive human lipidomic analysis. This robust method has enabled insights into the potential impact of diet on lipid metabolism at the molecular level in pregnant women. This was also the first study to our knowledge, to have assessed the relationship between DQ and maternal plasma lipidomics during pregnancy, including measures across two trimesters. An additional strength of this study was using the Australian DGI to calculate DQ at both early and late pregnancy. Dietary intake assessment has shifted from evaluating individual nutrient intake to whole foods and food patterns [73] to account for important interactions of nutrients and non-nutrient components [73]. Therefore, a major strength of the DGI is that it is a validated, food-based index and is translatable to public health messages related to the whole diet [74]. Our study also has some limitations. Recruitment of a high proportion (64%) of tertiary educated women with low-risk pregnancies means that findings are not generalizable to all pregnant women, especially those at ‘high risk’ of maternal complications. Assessment of DQ in less educated populations would enable important identification of vulnerable pregnant women, as poorer DQ may be expected in women of lower socioeconomic status and lower education level. Future research might also recruit women with more complicated pregnancies or women who are at high risk for adverse pregnancy outcomes, specifically women with a pre-pregnancy BMI > 30 kg/m². A further limitation was the use of self-reported dietary data which may be prone to recall bias or social desirability (e.g. overreported fruit and vegetable intake). However, as mean DGI scores of healthy foods including fruit and vegetables fell short of meeting dietary guidelines, this would suggest that women did not overreport intake of healthy foods. The DQES is commonly used to assess usual intake over the past 12 months and despite having been previously validated against weighed food records, it may not entirely reflect usual dietary intake due to recall of intake over a relatively long timeframe. Further, we did not account for women’s physical activity levels in our study, which is important when considering the effect on plasma TGs. We recognise that there is a need to consider the effect of lifestyle factors including physical activity in interpreting lipidomic profiles in pregnancy, as physical activity has shown positive effects on lowering fasting and postprandial blood glucose levels and blood triglyceride concentrations [75]. Finally, as this was a low-risk sample of pregnant women without serious complications, we could not assess associations of DQ with preeclampsia. However, including BP outcomes across pregnancy was an important aspect of our study as BP monitoring of levels below clinical thresholds of gestational hypertension or preeclampsia is still important in the assessment of overall CVD-related health.