Eating Disorders (ED) are serious mental health disorders affecting approximately 5-10% of adults (Hudson et al., 2007; Swanson et al., 2011) and have a peak of onset in adolescence between the ages of 15-19 (Micali et al., 2013; Field et al., 2012). The etiology of ED remains poorly understood, though and interplay of genetic and environmental factors is likely to be at play. Research into the clarification of risk factors has been hampered by uncertainties about clear phenotypic distinctions across ED categories, and by a lack of integrative studies using a longitudinal approach to clarify risk for ED. This study aims to focus on how cognitive, emotional and social processes cause ED behaviors in adolescence and young adulthood in interaction with environmental factors from infancy onwards.

Cross-sectional studies have identified cognitive, emotional and social difficulties that are associated with ED.

There is cross-sectional evidence that anorexia nervosa (AN) is associated with emotional and social communication difficulties and that specific cognitive profiles characterize bulimic type disorders.

AN shares common features with anxiety (Silberg & Bulik, 2005; Micali et al, 2011); and a range of social communication deficits, including interpersonal problems and poor emotion recognition, are present in individuals with AN (Tchanturia et al., 2012; Treasure et al., 2012).

Bulimic-type ED are cross-sectionally associated with specific cognitive profiles characterized by poor attention (Dobson & Dozois, 2004; Faunce, 2002) and low inhibition (Galimberti, et al., 2012; Rosval et al., 2006).

Most of the studies cited above are cross-sectional and have often focused on one specific area of behavior or cognition.

In relation to genetic risk for ED poor replication of early candidate genes studies has led to newer approaches heralding possible successful identification of risk markers. Large genome-wide studies of psychiatric disorders have highlighted genetic similarities across disorders, indicating either poor specificity of genetic markers or poor specificity of categorical classification systems. It also remains to be determined what the role of genetic mechanisms that affect gene expression, such as epigenetic processes, is in relation to ED.

The purpose of this study is to investigate the contribution of specific Research Domain Criteria (RDoC) constructs (across behavioral/self report and biological units of analyses (genetic, epigenetic and biomarkers data)) and their interaction with environmental factors to the risk for ED behaviors (restrictive eating, excessive exercise, bingeing, purging at ages 13,14, 16, 18, 24) in a longitudinal developmental fashion. We aim to develop a risk prediction algorithm for ED behaviors that incorporates these predictors.

We propose to use data collected prospectively (and carry out a new wave of data collection) from a unique ongoing cohort study: the Avon Longitudinal Study of Parents and Children (ALSPAC), based in the United Kingdom, to investigate the prospective association of anxiety and social communication and AN-type behaviors (restrictive eating and excessive exercise); and cognitive control and attention and BN-type behaviors (bingeing, purging).

The availability of biological (blood samples and cord blood for DNA at birth, age 7/9 and 15/17 years), neuropsychological (between ages 1 and 5, 8 and 13 years), behavioral (every two years from 3 1/2 until 12 years and more detailed at 7,10 and 13 years), social (ages 1 and 5, ages 8-10, 12 years) data at repeated time points in 7,000-9,000 adolescents/young adults over a period of 20 years makes this cohort a unique resource.

We will investigate the following specific aims:

1. To determine whether anxiety and social communication difficulties:

a) are longitudinally predictive of developing AN-type behaviors (restrictive eating, excessive exercise) in adolescence/young adulthood (ages 13-24);

b) interact with life events to predict AN-type behaviors (restrictive eating, excessive exercise)

We hypothesize that high childhood anxiety and poor social communication will be predictive of restrictive eating and excessive exercise in adolescence/young adulthood; and that they will interact with childhood life events.

2. To determine whether low cognitive control and poor attention:

a) are longitudinally predictive of developing BN-type behaviors (bingeing, purging) in adolescence/young adulthood (ages 13-24);

b) interact with life events/bullying to predict bingeing and purging from ages 13 to 24.

We hypothesize that low cognitive control and poor attention will be predictive of bingeing and purging in adolescence/young adulthood; and will interact with life events/bullying.

3. To build a risk model across domains and derive a risk prediction algorithm that can be used to identify high-risk individuals.

Exploratory aim: Exploratory analyses on the effect of poor emotion recognition on restrictive eating and excessive exercise in 500 young males, using functional and structural neuroimaging will complement the above aims.

We will look at whether these patterns and associations vary by gender. We propose to use one of the largest population-based longitudinal study in the world with a long enough follow-up and data available across most units of analyses, to investigate longitudinal associations between specific constructs and ED behaviors in adolescence/young adulthood. The size of our sample and the repeat and comprehensive assessments will allow an investigation of risk for ED across specific cognitive, social and behavioral constructs and their interaction with environmental factors in a unique and cost-effective way.

Methodology

Outcomes:

ED behaviours collected at ages 13,14,16,18. A questionnaire will be included in the 24/25 clinic to assess ED. We will use the same questionnaire used at age 14.

Predictors:

This grant focuses on using dimensional behavioural/cognitive predictors recently identified in the Research Domain Criteria (RDoC) proposal (one of the strategic aims of NIMH). RDoC focuses on understanding specific neurobiological dimensional phenotypes by studying each across a series of indicators-called units of analyses (including available biomarkers, self-report measures and behavioural observation).

Data collected throughout childhood on anxiety/emotional disorders, social communication and neuropsychological tasks (constructs) will be used as predictors. Additionally we will use data on evironmental risk factors, such as life events throughout childhood and bullying/teasing as predictors.

Genetic data already collected on ALSPAC participants will be used to investigate the role of genome and epigenome across the constructs under study. In particular a polygenic risk score will be generated using results (top SNPs) from current MEGA-analyses from two consortia. Genome-wide methylation data available from ARIES will also be used.

Biomarkers, i.e. salivary cortisol at age 11/12 will also be used as predictors.

Confounders:

data collected throughout childhood on socio-demographic data. Basic demographic characteristics and an overall environmental adversities index will be constructed from mothers' questionnaire data as collected from pregnancy through to child age 11. Information on gender, ethnicity, SES, employment, education, income, housing, family type and size, mother age at birth, gestational age and birth weight will be recorded.

Child IQ collected at age 8 will also be used as a confounder.

Data analyses:

After initial dscriptive analyses across each variable a stepped approach will be used.

Firstly each construct under study will be investigated across its units of analyses using factor analysis or by deriving latent variables.

Secondly the association between hypothesised constructs and outcomes will be investiagated using univariable analyses.

Fourthly multivariable models will be applied.

In relation to aim 3. we will use a multivariable Cox proportional hazard regression backward elimination model for each outcome under study, by including variables identified in objective 1 and 2 to build a riks prediction model.