This project has two aims:

1. To examine the causal effect of maternal smoking and drinking during pregnancy, and maternal BMI before pregnancy, on child development. More specifically, we are interested in outcomes relating to the child's birth, such as birth weight, as well as measures later in life such as academic outcomes, behavioural problems and child health (e.g. blood pressure).

2. To examine the effect of smoking and drinking on BMI or fat mass, both at the mother and child level.

As the relationships between (maternal) smoking, drinking, BMI and (child) development are likely to be subject to confounding in observational studies, we will identify the causal effect by using specific genetic markers that have been robustly identified to affect smoking, drinking or BMI as instrumental variables in a Mendelian randomization set-up.

We are aware that other researchers are also currently working on similar research questions using the ALSPAC data. For example, Sarah Lewis is involved in a project on alcohol led by Ron Gray, and we are aware that Rachel Freathy is looking at smoking-related variants. Hence, our aim is to explore this research question from a slightly different angle than that used in these existing projects. First, rather than examining the effect of these risky behaviours at the mean, we are specifically interested in their effects at different points in the outcome distribution. We will explore this distributional aspect with an instrumental variable quantile regression (IVQR) approach (Chernozhukov and Hansen, 2005, 2006). This will allow us to examine whether smoking and drinking has differential effects at different points in the (e.g.) birth weight or academic attainment distribution. Second, we are interested in the effect of different timings of smoking and drinking. For example, Freathy et al. (2009) show that the common polymorphismrs1051730 in the nicotinic acetylcholine receptor gene cluster (CHRNA5-CHRNA3-CHRNB4) is associated with heavier smoking in the first, but not third trimester. We will explore the effects of these different timings on our outcomes of interest.

We will use the variants CHRNA3 (rs1051730) and ADH1B (rs1229984) as instruments for smoking and alcohol consumption respectively. Finally, FTO (rs9939609) and MC4R (rs17782313) will be used as instruments for maternal BMI. Within the ALSPAC cohort, data have already been collected on a series of SNPs, with robust evidence that these SNPs affect these risky behaviours, which has been replicated in different independent samples (see e.g. Freathy et al., 2009; Zuccolo et al, 2009; Frayling et al., 2007; Loos et al., 2008). As such, we wish to use these in IVQR analyses, allowing us to specifically examine the distributional effects on these outcomes of interest.

We will also discuss the potential limitations of power within this setting. We are aware that (a) the minor allele rs1229984 is relatively rare in the ALSPAC population (MAF 2.4%), and that (b) the genetic effects on the intermediate trait are relatively small. Hence, depending on the variation required to shift the outcome of interest, this may lead to low power. The implications will differ depending on the outcome of interest, which is something we will discuss in the paper. For example, it may be easier to find effects on child birth weight compared to effects on their educational outcomes.

As the principle applicant already has access to the ALSPAC data on child and family background characteristics, we only wish to obtain (a) the more detailed measures of maternal smoking and drinking and (b) the genetic markers related to mother's BMI, smoking and drinking.