What are birth cohort studies and why are they important?

Birth cohort studies are those which begin at or before the birth of its participants and continue to study the same individuals at later ages, on more than one occasion⁽⁹⁾. They are a type of observational study so there is no randomisation to exposure groups, and there is no attempt to manipulate the exposure status. They usually aim to be nationally representative but some are area-based and often such studies enrol many thousands of participants.

Birth cohorts are a powerful resource to study diet–disease associations, gene–diet interactions, epigenetic influences and the role of the microbiome because these studies involve a detailed assessment of the maternal/fetal environment before birth and include longitudinal prospective follow-up of multiple health outcomes from birth, through infancy and early childhood. This prospective measurement of maternal exposures and pregnancy characteristics is superior to the retrospective classification of exposures, as it minimises recall bias.

Many of the important birth cohort findings are consistent with advice that our mothers and grandmothers have been passing on for generations. Birth cohort studies have helped establish that ‘breast is best’: breast-fed children are at lower risk of becoming overweight during adolescence, experience better cognitive outcomes and are at lower risk of being diagnosed with attention-deficit/hyperactivity or autism spectrum disorders^{(Reference Bar, Milanaik and Adesman10,Reference Gillman, Rifas-Shiman and Camargo11)} . Through birth cohort studies, we have also come to understand the educational benefits of reading regularly to children^{(Reference Ritchie and Bates12)}; and the benefits of supine positioning for sleep⁽¹³⁾. But beyond grandmothers' advice, birth cohorts can help us to uncover the role of the in utero environment in shaping our health trajectory through the lifespan. Maternal energy deficit during pregnancy at key stages of development, especially when combined with overfeeding in the postnatal period, leads to a constellation of cardiometabolic risks, including altered glucose–insulin metabolism^{(Reference Ravelli, Stein and Susser14)}. These effects on offspring phenotype seem to be partly mediated by changes in the number and function of pancreatic β-cells, possibly via epigenetic mechanisms^{(Reference Hales and Barker15)}.

The Barker hypothesis proposed in 1990 by the British epidemiologist David Barker (1938–2013) posits that in human subjects, intrauterine growth retardation, low birth weight and premature birth have a causal relationship to the origins of hypertension, type 2 diabetes and CHD in middle age^{(Reference Barker16)}. Two observations provided the impetus for the development of Barker's hypothesis. Barker and Osmond reported a positive association between a county's neonatal death rate (a surrogate for low birthweight and its cardiovascular mortality rate)^{(Reference Barker and Osmond17)}. In 1989, Barker revisited Hertfordshire County birth records from 1911 to assess the association between birth weight and IHD, and found that low birthweight babies had three times the rate of IHD than normal-weight babies^{(Reference Barker, Winter and Osmond18,Reference Barker19)} . One long-term consequence of inadequate prenatal nutrition is impaired development of endocrine pancreas, which ‘hardwires’ the infant to be nutritionally ‘thrifty’. If the child is nutritionally deprived, the phenotype is advantageous as it matches his/her prenatal environment; but if the same infant is exposed to a high-nutritional postnatal environment, he/she will carry an increased cardiometabolic risk^{(Reference Hales and Barker15)}.

To examine whether early catch-up growth following reduced intrauterine growth modifies risk of death from CHD in adulthood, Eriksson et al. followed 3641 boys from birth through adulthood to assess the association between birth weight and death from CHD^{(Reference Eriksson, Forsen and Tuomilehto20)}. Men who died from CHD had an above average BMI at all ages from 7 to 15 years. They found that those who were born large but were small at age 11 were at no increased risk of adult CHD, whereas those who were born small and experienced large catch-up growth were at increased risk of death from CHD later in adulthood. The highest death rates from CHD occurred in boys who were thin at birth but whose weight caught up so that they had an average or above average body mass from the age of 7 years.

The Dutch Hunger Winter occurred in West Netherlands near the end of World War 2 when food rations were limited to fewer than 3347 kJ/d (800 kcal/d)⁽Reference Roseboom, de Rooij and Painter^21, Reference Lumey, Stein and Kahn²²⁾. In a historical cohort study, 300 000 19-year-old men whose mothers were exposed to the Dutch Hunger Winter pre- and post-natally were examined at military induction to test the hypothesis that prenatal and early postnatal nutrition determines subsequent obesity^{(Reference Ravelli, Stein and Susser14)}. The influence of maternal exposure to the famine was highly dependent on timing. Exposure to famine during the last trimester of pregnancy and the first months following delivery produced significantly lower obesity rates and better glucose tolerance. This is consistent with the inference that nutritional deprivation affected a critical period of development for adipose-tissue cellularity. Exposure to famine during the first half of pregnancy, however, resulted in significantly higher obesity rates and poorer glucose tolerance. These data are supported by famine studies of China, Ukraine and Austria but not from the former Soviet Union or those in Africa⁽Reference Wu, Feng and He^23–Reference Gillman²⁶⁾.

A potential explanation is that famine that occurs early in pregnancy is relatively more potent than famine that occurs later in pregnancy^{(Reference Tobi, Slieker and Stein27)}. This interpretation is consistent with the existence of a critical time window, one of the tenets of the developmental origins of health and disease, which posits that events during early, plastic phases of human development can have lifelong, sometimes irreversible effects^{(Reference Schulz28)}.

Example 1: Creation of harmonised dietary patterns

Methodological advances in dietary measurement in large epidemiologic studies, such as the development of valid and reproducible semi-quantitative FFQ^{(Reference Willett, Sampson and Stampfer35,Reference Rimm, Giovannucci and Stampfer36)} have facilitated the study of associations between dietary intake and health and disease outcomes, such as cancer and CVD. This is often approached with a ‘reductionist’ lens by examining associations between specific food items^{(Reference Schernhammer, Hu and Giovannucci37–Reference Giovannucci, Rimm and Stampfer40)}, single nutrients^{(Reference Giovannucci, Rimm and Stampfer40,Reference Willett, Hunter and Stampfer41)} , or sources of nutrients^{(Reference Howe, Jain and Miller42,Reference Hu, Stampfer and Manson43)} and health outcomes. This approach is reflective of public health approaches to food and nutrient recommendations, and has been quite valuable, particularly at identifying and correcting single-nutrient deficiencies such as xerophthalmia (vitamin A), beriberi (vitamin B₁), pellagra (vitamin B₃), anemia (vitamin B₁₂), scurvy (vitamin C), pernicious rickets (vitamin D) and goitre (iodine)⁽⁴⁴⁾, but has several conceptual and methodological limitations.

First, people do not eat isolated nutrients; they eat meals consisting of a variety of foods with complex combinations of nutrients that likely interact^{(Reference Hu45)}. Secondly, the high degree of intercorrelation among some nutrients in the same foods (such as potassium and magnesium) makes it challenging and unhelpful to examine their individual effects^{(Reference Lee, Reed and MacLean46)}. Thirdly, the effect of a single nutrient may be too small to detect, but the joint effects of multiple nutrients as part of a dietary pattern may be measurable^{(Reference Sacks, Obarzanek and Windhauser47)}. Fourthly, analyses based on a large number of nutrients or food items may produce chance or spurious associations^{(Reference Farchi, Mariotti and Menotti48)}. Lastly, because nutrient intakes are often associated with certain dietary patterns^{(Reference Kant, Schatzkin and Block49,Reference Randall, Marshall and Graham50)} , single-nutrient analyses may be confounded by the effect of other nutrients in foods that are often consumed together.

To overcome these limitations of single-nutrient or single-food studies, the empirical derivation of dietary patterns, defined as the quantities, proportions, variety or combinations of different foods and beverages in diets, and the frequency with which they are habitually consumed,⁽⁵¹⁾ has been proposed to more closely reflect how we consume foods and nutrients. These patterns can be assessed for their associations with health and disease^{(Reference Hu45,Reference Heidemann, Schulze and Franco52–Reference Iqbal, Anand and Ounpuu54)} . In preparation for investigations into the role of maternal nutrition on maternal and newborn outcomes, we developed an approach to derive harmonised dietary patterns in pregnant women^{(Reference de Souza, Zulyniak and Desai55)}.

In the CHILD study, maternal diet was assessed by using a semi-quantitative FFQ, adapted from the Fred Hutchinson Cancer Center tool⁽⁵⁶⁾. In the FAMILY, START and Indigenous (Aboriginal) Birth cohort studies, semi-quantitative FFQ developed for the Study of Health and Risk in Ethnic Groups Study^{(Reference Anand, Yusuf and Vuksan57)} were used to assess maternal dietary intake during pregnancy, modified to capture ethnic-specific foods. Individual FFQ items from each study were combined into thirty-six smaller groups by nutrient profile and type (e.g. poultry, leafy greens, legumes, etc.) to create common food groups across the cohorts. We identified three primary dietary patterns: plant-based, Western, and health-conscious, which collectively explained 29 % of the diet variability in a principal components analysis^{(Reference Iqbal, Anand and Ounpuu54)}. (Table 2). This study addressed a novel challenge: the merging and harmonisation of dietary data across cohorts which used different FFQ. We described a valid approach to merging both similar and distinct FFQ datasets which could be used in other studies that combine cohorts with unique diet assessment methods.

Table 2. Principal component analysis food group loading scores

Food items with a loading score ≥ |0·30| are presented and characterise each of the three dietary patterns within the NutriGen Birth Cohort Alliance cohort (n 4880). Originally published in reference^{(Reference Sacks, Obarzanek and Windhauser47)}

* Proportion of the total dietary variation in the dataset that is explained by considering 1, 2, or 3 underlying dietary patterns.

Example 2: Understanding the causes and consequences of gestational diabetes mellitus

The reasons for the increased risk of gestational diabetes among South Asian women are not well-understood. Using data from the START study, one of the participating cohorts in the NutriGen Birth Cohort Alliance, we sought to identify the determinants of gestational diabetes and its impact on newborn health^{(Reference Anand, Gupta and Teo58)}. We collected health information and physical measurements from 1012 women and administered an oral glucose tolerance test. We obtained birth weight and skinfold thickness measurements from newborns, as well as cord blood glucose and insulin levels. We reported an incidence of gestational diabetes of 36·3 %; the age-standardised rate was 40·7 %. Newborns of dysglycemic mothers had increased birth weight and body fat, and reduced insulin sensitivity, which influences future risk of excess adiposity and type 2 diabetes.

Factors associated with gestational diabetes included maternal age, family history of diabetes, pre-pregnancy weight and low diet quality, which had a combined population attributable risk of 65 %. Maternal height was protective against gestational diabetes. The population attributable risk due to the modifiable risk factors, pre-pregnancy BMI and low diet quality was 37·3 %. This suggests that, if South Asian women could achieve an optimal pre-pregnancy weight (i.e. BMI <23) and improve their diet quality, about one-third of cases of gestational diabetes in this population could be prevented.

Our findings highlight the importance of public health messaging to South Asian women who are contemplating pregnancy to aim for an optimal weight before pregnancy as a potential prevention strategy against gestational diabetes^{(Reference Poston, Caleyachetty and Cnattingius59)}. To our knowledge, primary care physicians or public health specialists do not provide this message routinely, and this will require an integrated approach involving primary health care and policy initiatives^{(Reference Hanson, Barker and Dodd60)}.

Example 3: Ethnicity and diet-related differences in the healthy infant microbiome

The developing gastrointestinal microbiome in the first years of life is important for immune function, nutrient metabolism and protection from pathogens^{(Reference Falk, Hooper and Midtvedt61–Reference Newburg and Walker63)}. Microbial colonisation of the infant gut proceeds through infancy and establishment of an adult-like microbiome is estimated to occur within the first 3 years^{(Reference Yatsunenko, Rey and Manary64)}. Identifying factors that shape the gut microbiome is currently an active area of research and early evidence suggests that host genetics^{(Reference Li, Oosting and Deelen65)} and early life exposures, including delivery method, antibiotics^{(Reference Bokulich, Chung and Battaglia66,Reference Azad, Konya and Maughan67)} and diet, influence the infant gut microbiome^{(Reference Goodrich, Waters and Poole68,Reference Backhed, Roswall and Peng69)} . Although a stable microbiome may not be established until 1–3 years after birth, the infant gut microbiota appears to be an important predictor of health outcomes in later life, possibly influencing the progression of chronic diseases and has been associated with adverse health outcomes^{(Reference Aron-Wisnewsky and Clement70)}.

We analysed stool at age 1 year from a sub-sample of 173 white Caucasian and 182 South Asian infants from the CHILD and START birth cohorts to gain insight into how ethnicity, along with maternal and early infancy exposures influence the gut microbiota, using established methods of microbiome assessment^{(Reference Stearns, Zulyniak and de Souza71–Reference Whelan and Surette75)}. More species richness was observed in South Asian babies than white Caucasian infants after considering breastfeeding at the time of collection^{(Reference Stearns, Zulyniak and de Souza71)}. The effect of ethnicity was larger than the effect of geographic location, which is typically an important source of variation^{(Reference Dugas, Fuller and Gilbert76)}. Numerous studies have found the infant gut microbiome to vary between infants born by Caesarean section and those born vaginally, although the effect diminishes with age^{(Reference Yasmin, Tun and Konya77–Reference Rutayisire, Huang and Liu79)}. In our study, delivery method was not a significant predictor of the structure of the gut microbiome, but this is not surprising as the effect of delivery mode on the gut microbiome could have diminished by age 1 year. Larger analyses from the CHILD cohort have found delivery method to exert a strong influence on gut microbiome composition at 3–4 months^{(Reference Azad, Konya and Maughan67,Reference Yasmin, Tun and Konya77,Reference Tun, Bridgman and Chari80)} .

South Asians had higher abundances of several genera within the Actinobacteria including Bifidobacterium, Collinsella, Actinomyces and Atopobium compared to white Caucasians^{(Reference Stearns, Zulyniak and de Souza71)}. Genera within the phylum Firmicutes within two distinct taxonomic groups were associated with ethnicity. Genera such as Streptococcus, Enterococcus and Lactobacillus (class Bacilli, order Lactobacillales) were more abundant within South Asians whereas genera such as Blautia, Pseudobutyrivibrio, Ruminococcus and Oscillospira (order Clostridiales) were more abundant in white Caucasians. The most differentially abundant bacteria were members of the Lachnospiraceae which were higher in white Caucasians. Ethnic differences in the gut microbiome may reflect maternal and/or infant dietary differences. Whether these differences are associated with future cardiometabolic outcomes will only be determined through prospective follow-up.

Example 4: The association of maternal dietary patterns with birth weight differs by ethnicity

Birth weight is an indicator of newborn health and a strong predictor of health outcomes in later life^{(Reference Law81)}. Significant variation in diet during pregnancy between ethnic groups in high-income countries provides an ideal opportunity to investigate the influence of maternal diet on birth weight. We investigated the association between maternal diet and birth weight in our multiethnic cohort using a previously developed dietary pattern analysis approach^{(Reference de Souza, Zulyniak and Desai55,Reference Zulyniak, de Souza and Shaikh82)} . A total of 3997 full-term mother–infant pairs with principal component analysis-derived diet pattern scores for the plant-based, Western and health-conscious diets, along with birthweight recorded, were included. No associations were found between the Western and health-conscious diet patterns and birth weight; however, the plant-based dietary pattern was inversely associated with birth weight, and an interaction with non-white ethnicity and birth weight was observed. Among white Europeans, maternal consumption of a plant-based diet was associated with lower birth weight, increased risk of small-for-gestational and reduced risk of large-for-gestational-age. Among South Asians, maternal consumption of a plant-based diet was associated with a higher birth weight.

In post-hoc analyses conducted separately in white Europeans and South Asians, we identified fifteen food groups for which the distribution differed between the individuals in the first and fourth quartiles of the plant-based diet. When we included terms for these food groups in the multivariable models, only the addition of cooked vegetables reduced the magnitude of the plant-based dietary pattern association in South Asians by 6 %. No other foods, nor multi-vitamin use influenced the association of plant-based diet and birth weight in either ethnic group. Differences in food preparation methods can significantly alter the chemical and nutritional composition of dishes^{(Reference Lesser, Gasevic and Lear83)}, notably the addition of fat for frying. Higher fat intake has been associated with newborn length and adiposity in other studies of South Asian pregnant women^{(Reference Rao, Yajnik and Kanade84)}. These results require replication to elucidate potential mechanisms that underlie these ethnic-specific associations.

Metabolomics

The existing literature for both maternal and infant diet reveals inconsistent associations between foods, nutrients and various health outcomes, which arise from the methodological challenges and errors associated with estimating dietary intake and assessing the contribution of nutrition independent of other lifestyle and biological confounders^{(Reference Parsons, Power and Logan85,Reference Jacques and Tucker86)} . Recognising these limitations we look to use a more direct measurement of nutrient status, function and effect. Metabolomics-based approaches offer an unprecedented opportunity to derive a more precise measurement of dietary intake, reflecting many steps including gut absorption and liver metabolism. We will use previously identified candidate metabolites found to be associated with (1) adolescent body weight^{(Reference Wurtz, Wang and Kangas87)}; (2) gestational diabetes mellitus or hyperglycemia^{(Reference Lehmann, Friedrich and Krebiehl88,Reference Scholtens, Muehlbauer and Daya89)} ; (3) low birthweight^{(Reference Ivorra, Garcia-Vicent and Chaves90)} or (4) childhood risk of obesity^{(Reference Isganaitis, Rifas-Shiman and Oken91)}, together with untargeted metabolites which pass a pre-specified statistical threshold. We will conduct analyses separately for each cohort (CHILD, FAMILY and START) and pool results using appropriate meta-analytic techniques^{(Reference Kelley and Kelley92)}.

Gene–environment interactions

Complex interactions between genetic variants and selected environmental factors likely exist with cardiometabolic traits, allergic disorders and asthma. Investigation of gene–environment interactions has been challenging because optimal studies require large sample sizes, careful measurement of environmental factors and robust clinical outcomes^{(Reference Kauffmann and Demenais5)}. Most investigations which have identified gene–environment interactions have selected candidate SNP identified through genome-wide association studies and tested their association with the outcome together with an environmental exposure which is also related to the outcome. Gene–environment interactions have been demonstrated and replicated in myocardial infarction, type 2 diabetes, obesity and asthma^{(Reference Morales, Bustamante and Vilahur93,Reference Joseph, Pare and Anand94)} . However gene–environment interaction studies among infants/children which necessitate large sample sizes and finely phenotyped cohorts are only recently underway. The NutriGen Birth Cohort Alliance will facilitate the study of gene–diet interactions on maternal and infant health outcomes. Our analysis of gene–diet interactions will be prioritised in three steps: Step (1) SNP demonstrated to be significant in genome wide studies will be prioritised for testing for gene–diet interactions. Step (2) a genome wide analysis of mother genotype against maternal traits and baby genotype against infant/child traits will be performed. Significant SNP will be tested with dietary patterns and highly prioritised dietary factors of mother and infant to determine if a gene–diet interaction exists. Step (3) any main effect of SNP, diet, or interaction will be tested for replication in partner cohorts.

Epigenetics

There is increasing evidence that maternal diet modifies maternal/fetal DNA (DNA methylation, histone structure and small non-coding RNA) which affects gene expression in the offspring^{(Reference Waterland and Jirtle95)}. This represents the most likely biological explanation for the persistent effects of environmental exposures during pregnancy through successive generations^{(Reference Leon and Moser96)}. We aim to investigate maternal nutrition in conjunction with newborn epigenetic marks to determine their interactions and potential influences on an array of health outcomes in the offspring. Rather than performing genome-wide methylation in all samples, which is expensive and unfocused, we will use an ’omics approach to target specific methylation sites guided by gene expression information^{(Reference Relton, Groom and St Pourcain97,Reference Yang, Adelstein and Kassis98)} .

Genome wide chip-based methylation analysis has a high degree of reproducibility. However, the challenge of multiple testing and generation of false-positive results with poor replication remains^{(Reference Hayes99,Reference Michels and Binder100)} . Target regions will be identified by (1) performing whole genome expression analysis using RNA in 500 cord blood samples from the START and CHILD birth cohorts, (2) investigating gene expression differences in newborn samples based on maternal dietary extremes i.e. Western diet v. Prudent diet, (3) performing genome wide methylation analysis using the Illumina 450 K Infinium methylation assay to interrogate >450 000 methylation sites per sample at single-nucleotide resolution in the same cord blood samples, and (4) investigating if genes differentially expressed in newborns have corresponding variation in methylation patterns specifically in the promoter region. These targeted methylation sites will then be tested in the remaining cord blood samples (n 4000) from infants across four cohorts and analysed in relation to maternal dietary patterns and selected infant outcomes including adiposity, cardiometabolic traits, allergy and asthma.

Bridging ‘omics data

Underlying associations between the genome (an organism's complete set of DNA, including its genes), epigenome (compounds that attach to and ‘mark’ the genome, altering its function but not sequence) and metabolome (the totality of metabolites present within an organism) can provide additional insights regarding the transmission of the metabolome from mother to infant^{(Reference West and Caudill101,Reference West, Weir and Smith102)} . Recently, independent studies identified lipids and amino acids as being highly heritable^{(Reference Kettunen, Tukiainen and Sarin103)}. Understanding the role of epigenetic modification as a method of controlling heritability of certain traits will help advance our understanding of childhood adiposity and other metabolic syndrome traits.

Family health behaviours

Among young children, poor diet quality along with physical inactivity contribute to the risk of excess body weight and metabolic syndrome^{(Reference Eloranta, Schwab and Venalainen104–Reference Pan and Pratt106)}. Metabolic syndrome is a cluster of conditions: increased blood pressure, high blood sugar, abdominal fat and abnormal cholesterol or TAG levels that occur together, increasing the risk of heart disease, stroke and diabetes^{(Reference Tarrade, Panchenko and Junien107–Reference Johns, Hartmann-Boyce and Jebb109)}. There is increasing evidence that a child's energy balance is mediated by their family environment because the home environment is where a child's food habits are established^{(Reference Adair and Popkin110–Reference St-Onge, Keller and Heymsfield112)}. With the long-term goal of developing an intervention to prevent the metabolic syndrome in children aged 10 years, we will use quantitative information collected from questionnaires from children in the NutriGen Birth Cohort Alliance (i.e. self-reported dietary intake and physical activity) along with semi-structured interviews to guide inquiries among families to understand their health behaviour experiences.