Dr Lavinia Paternoster (lead), Dr Raquel Granell,
Atopic dermatitis (eczema) is a common chronic skin disease with prevalence rates of 10-20% and heritable estimates of up to 90%. 31 loci associated with AD case/control status have been identified through genome-wide association studies (GWAS) >377,000 individuals. Follow-up of these genetic findings is considered an effective strategy to identify potential drug targets. However, though identification of such variants might be informative for prevention of disease, it is unclear how useful they will be for understanding how to treat the condition. A better study design to inform design of treatments might be to find genetic variants for AD progression. This approach would identify targets that if acted on with drugs, might slow or prevent the progression of disease.
Conduct GWAS of progression traits for atopic dermatitis (such as persistence of the disease) to identify loci associated with these traits and therefore informative for drug discovery.
Follow-up these loci using multi-omic integrative data approaches to finemap the loci and identify the likely gene target.
1. Recruitment of cohorts from the EAGLE consortium with available data.
2. Generate analysis plan and scripts for dissemination to cohorts.
3. Conduct statistical association analyses on in-house datasets (e.g. ALSPAC) - genome-wide association analysis, may involve longitudinal modelling and/or correction for selection/collider bias in case-only analysis, as appropriate.
4. Meta-analyse results across cohorts.
4. Follow up variants identified using publicly available multi-omic resources (e.g. transcriptomic, epigenomic and proteomic data) and bioinformatics approaches, to identify likely causal genes.
There is also potential for engaging with industry and/or experimental biologists to validate findings.
This project will be predominantly statistical and computational in nature. Analyses will be conducted using R and some bespoke software, operated in a Linux system. Therefore some coding will be essential.
Paternoster L et al. Multi-ancestry genome-wide association study of 21,000 cases and 95,000 controls identifies new risk loci for atopic dermatitis. Nat Genet 2015, 47(12):1449-5.
Paternoster L, Tilling K, Davey Smith G. Genetic epidemiology and Mendelian randomization for informing disease therapeutics: Conceptual and methodological challenges. PLOS Genet 2017, e1006944.
Dr Emma Vincent (lead), Prof Nicholas Timpson, Prof Ann Williams
People living with type 2 diabetes have abnormal levels of circulating metabolites in their blood. This results in systemic exposure of cells and tissues to an abnormal and unregulated metabolic environment. Such exposure is likely intrinsically linked to the probability a cancer develops and the cancer cell phenotype. This is because carcinogenesis requires that tumours reprogram their metabolic pathways to adapt to their metabolic environment to satisfy demands of chronic cell proliferation.
Given the abnormal serum metabolome in people with type 2 diabetes and the potential impact on tumorigenesis it is unsurprising that type 2 diabetes is associated with an increased risk of developing cancer at several sites. However, the inherent heterogeneity of type 2 diabetes makes it challenging to characterise positive associations as causal.
This proposal is intended to investigate the impact of the dysregulated metabolic environment characteristic of people with type 2 diabetes on the incidence and progression of colorectal cancer. It will also investigate whether drugs taken to treat type 2 diabetes can reduce colorectal risk by modulating metabolism.
Aims & objectives:
We aim to understand why people with type 2 diabetes have an increased risk of colorectal cancer. To investigate this, we will focus on circulating metabolites.
3 main aims:
1) To identify the circulating metabolites that are altered in people with type 2 diabetes and identify which of these metabolites are causally associated with colorectal cancer risk and progression.
2) To investigate how certain metabolites might increase risk of cancer.
3) To investigate how type 2 diabetes drugs might reduce colorectal cancer risk by modulating cellular metabolism.
Here, we will use methods in genetic epidemiology and in particular Mendelian randomization to determine which metabolic traits characteristic of type 2 diabetes are causal for colorectal cancer (Aim 1). Laboratory based methods will be used to investigate the mechanisms underlying the associations between levels of circulating metabolites and cancer risk (Aim 2).
To investigate the role of type 2 diabetes drugs in colorectal cancer incidence and progression (Aim 3) we will start by interrogating drug prescribing data to understand how these drugs are prescribed, at what stage of type 2 diabetes people are taking them and the duration of time they are exposed to them. We will then investigate how type 2 diabetes drugs might reduce cancer risk by influencing both whole body metabolism (this will be investigated using clinical trial samples) and colorectal cell metabolism (this will be investigate using laboratory techniques in cancer cell biology).
Giovannucci E et al. Diabetes Care. 2010;33(7):1674-85.
DeNicola GM & Cantley LC. Mol Cell. 2015;60(4):514-23.
Davey Smith G & Ebrahim S. Int. J. Epidemiol. 2003;32(1):1-22.
Professor Jonathan Sterne (lead), Professor Caroline Sabing (UCL), Dr Fiona Burns (UCL) Dr Ruth Mitchell
Genome-wide association studies (GWAS) of HIV have mainly addressed correlates of acquisition and disease progression: little work has been done in the context of combination antiretroviral therapy (ART). Suppression of viral replication by ART may be insufficient to prevent premature cardiovascular events or hepato-renal pathology, or non-AIDS cancers.
The NIHR BioResource is a federated database of healthy individuals and patients, consented for recall for research. High density genotyping is performed on the Affymetrix UK Biobank array. The HIV BioResource has collected host genome sequences from ~5200 participants with ongoing recruitment, storing blood samples and derivatives together with patient information (clinical history, lifestyle factors and potential predictors of CVD). It links with the UK Collaborative HIV Cohort Study (UK CHIC) and the UK HIV Drug Resistance Database (UK HDRD): two of the world’s largest and most productive HIV cohort studies: The combined dataset is a unique resource that will enable investigation of genetic contributions to switches in antiretroviral therapy; renal disease progression and other effects of ART.
This project will focus on identification of genetic predictors of the linked outcomes CD4:CD8 ratio; Dyslipidaemia; Diabetes; and Major cardiovascular events. Genetic prediction of these outcomes could facilitate early identification of individuals who might benefit from individualised treatments, increased monitoring or intensified prevention measures. The potential mediating role of nadir CD4 count will be investigated, as will moderating effects of specific ART drugs. Analyses will address potential biases introduced through selection of patients into the HIV Bioresource.
This project involves cleaning and analysing a novel dataset to study genetic predictors of outcomes of treatment with antiretroviral therapy: these are questions of global health importance with important implications for early identification of individuals who might benefit from individualised treatments, increased monitoring or intensified prevention measures. The data are derived from linking the NIHR HIV BioResource links (~6000 individuals) with the UK Collaborative HIV Cohort Study (UK CHIC) and the UK HIV Drug Resistance Database (UK HDRD), two of the world’s largest and most productive HIV cohort studies. The student will learn state of-the-art techniques for analysis of genome-wide association studies (eg LD regression), supported by colleagues in the world-leading MRC Integrative Epidemiology Unit (IEU). They will have full access, free of charge, to courses in research methods run by the MRC IEU and the University of Bristol Department of Population Health Sciences, as well as to specialist expertise and guidance in all areas required for successful completion of the project, including genetic epidemiology, Mendelian randomisation, epidemiology, statistical methodology and bioinformatics. The student will receive training in HIV clinical epidemiology at the UCL HIV Epidemiology and Biostatistics Group (co-led by Sabin), and in HIV medicine at the UCL Centre for Sexual Health & HIV Research, which shares accommodation with the largest clinic for sexual health and HIV disease in Europe.
Professor Nicholas Timpson (lead), Dr Kaitlin Wade,
Whilst there has been an exponential growth in the literature implicating the human gut microbiome as having a role in health[1-3], there is a concerning lack of robust evidence able to discern correlation from causation. Despite this lack of evidence, there is a market for commercial initiatives targeting the microbiome as a consumer-driven intervention (e.g., SmartGut, Viome and Atlas BioMed) and increased clinical probiotic recommendation for disease treatment or following antibiotic use[5-7]. These activities may mark an untapped opportunity for an important population-based health intervention but there is little evidence able to establish the impact of these developments in the absence of large-scale, population-based trials.
This PhD will focus on understanding factors that shape the human gut microbiome (measured by proxy from faecal samples) and the causal impact it has on important health outcomes, by combining multiple data sources. This is a PhD which will be coordinated with a Wellcome Trust funded programme of research that explicitly seeks to use the most contemporary and powerful study designs, metabolic data capture and analytical techniques to explore BMI as a risk factor. This programme is affiliated with the MRC Integrative Epidemiology Unit (University of Bristol).
With established collaborations and combining epidemiological, genetic and causal inference methods (including genome-wide association studies (GWAS), Mendelian randomization and recall-by-genotype analyses), this PhD will address the need for causal assessment of the role played by the human gut microbiome on health. For example, both the Flemish Gut Flora Project, which includes ~3,000 individuals with questionnaire, clinic, genetic and microbiome data, and the meta-analysis to which this will contribute (N>18,000) will act as the initial resources for the human gut microbiome to which other studies can be added. Information on chosen exposures and health outcomes will be obtained from existing summary-level GWAS results. The product of this work will be to identify targets causally related to disease, which can then be dissected further by the precise measurement and functional annotation of the gut microbiome and advanced to functional studies. Importantly, there is foreseeable scope to contribute to the manufacture of pharmaceutical or food products as a direct consequence of translating findings from this proposed research to industry partners and policymakers.
1.Falony G, et al. Science. 2016;352:560. 2.Subbarao P, et al. Thorax. 2015;70:998. 3.Tigchelaar EF, et al. BMJ Open. 2015;5. 4.Harley ITW, et al. Mol Metab. 2012;1:21. 5.Boyle RJ, et al. AJCN. 2006;83:1256. 6.Islam SU. Med. 2016;95:e2658. 7.Verna EC, et al. Therap Adv
Gastroent. 2010;3:307. 8. MacArthur J, et al. Nucl Acid Res. 2017;45:D896. 9.Haycock PC, et al. AJCN. 2016;103:965. 10.Corbin LJ, et al. Nat Comms. 2018;9:711.11.Wang J, et al. Microbiome. 2018;6:101.
Professor Nicholas Timpson (lead), Dr Laura Corbin,
There is strong evidence that BMI causally influences a wide range of health outcomes, but there is little understanding of the mechanisms driving BMI effects and at the level of the population we are limited in our ability to alter BMI. Targeted lifestyle and pharmaceutical interventions have failed to deliver large reductions in BMI and the only effective intervention is surgery.
The main theme of this work is to dissect the causal effects which have been shown to lie between BMI and health outcomes; i.e. to better understand how body mass index (BMI) exerts an effect on human health. The aim is to do this using metabolomics and (in this case) studies of the human microbiome in complementary study designs and through applied genetic epidemiology. This is a unique opportunity to study in a research team dedicated to this and to get access to a series of mature data sets able to address this question. This is a PhD which will be coordinated with a Wellcome Trust funded programme of research that explicitly seeks to use the most contemporary and powerful study designs, metabolic data capture and
analytical techniques to explore BMI as a risk factor. This programme is affiliated with the MRC Integrative Epidemiology Unit (University of Bristol).
The proposal here is that metabolomic measurements can be used to understand the mechanisms by which BMI contributes to disease. Circulating metabolites are the product of
genetic and non-genetic factors and are a useful read-out of physiological function. Despite the similarity of metabolites to complex health outcomes, it remains possible to map metabolites to genotypes and translatable biological pathways. Consequently, identifying metabolites important in the link between BMI and disease is a promising approach to further understanding BMI as a risk factor. To better understand how body mass index (BMI) exerts an effect on human health and disease using metabolomics in complementary study designs and
through applied genetic epidemiology. This particular PhD will look to integrate data from microbiome and metabolomic data collections in efforts to assess these as mediating routes between BMI variation and disease. This will include a causal mapping of the human faecal microbiome onto the blood metabolome using data from the Flemish Gut Flora Project, which includes ~3,000 individuals with questionnaire, clinic, genetic and microbiome data, and the meta-analysis to which this will contribute (N>18,000). These are new and exciting data within a broader programme of research focused on the breakdown of BMI impact.
1.Wang J, et al. Microbiome. 2018;6:101.
Professor Nicholas Timpson (lead), Dr Laura Corbin,
There is strong evidence that body mass index (BMI) causally influences a wide range of health outcomes but little understanding of the mechanisms driving these effects. Investigating the role of BMI in causing disease using a traditional epidemiology framework is challenging for several reasons. BMI is typically associated with a wide range of confounding factors making isolating its effect difficult in observational settings. Furthermore, there are issues with the acceptability and feasibility of weight change interventions for research with such studies often
restricted to low numbers. So far, approaches such as Mendelian randomization (MR) have provided a useful insight into the causal effect of BMI on a range of health outcomes. However, such methods also have limitations, for example, concerns about the impact of genetic structure on such analyses.
The focus of this project is to bring together data and methodologies from two areas of epidemiology, applied genetic epidemiology and experimental epidemiology, to investigate the metabolic profile of BMI reduction (achieved through contrasting interventions).
We believe that randomised controlled trials (RCT) of interventions to change (typically reduce) BMI could provide a valuable source of complementary data for exploring the causal relationship between BMI and disease. More specifically, data will be available from two RCTs, the first involving a medical intervention (caloric restriction) and the second a surgical intervention. Metabolomic profiling have or will be conducted for samples collected from patients at baseline (pre-intervention) and study end, using data generated from two platforms (nuclear magnetic resonance (NMR)/mass spectrometry). Circulating metabolites are the product of genetic and nongenetic factors and are a useful read-out of physiological function. Identifying metabolites important in the link between BMI and disease is thus a promising approach to further understanding BMI as a risk factor. The re-use of the RCT data in an alternative causal framework will necessarily involve the development of appropriate statistical approaches to deal with the complexities of the data. One approach is likely to be to consider the clinical trials within an instrumental variable framework where randomization is the instrumental variable. Additional statistical methods that may be used include mixed models and latent class analysis, and potentially nonlinear modelling to assess the importance of absolute and relative levels of weight change on the metabolome. This is a PhD which will be coordinated with a Wellcome Trust funded programme of research that explicitly seeks to use the most contemporary and powerful study designs, metabolic data capture and analytical techniques to explore BMI as a risk factor. This programme is affiliated with the MRC Integrative Epidemiology Unit (University of Bristol).
1. Leslie, W, et al. BMC Family Practice 17:20(2016).
Professor Nicholas Timpson (lead), Dr Kaitlin Wade,
Cardiovascular diseases are major causes of mortality in the growing world, despite major advances of medical research. Such diseases are thought to originate in early life, where preclinical events and vascular dysfunction have been documented in children and adolescents. However, there is inconsistency in the literature. Whilst various risk factors of adverse cardiovascular health have been implicated in playing a causal role, such as adiposity[1, 2] and cardiometabolic factors[3, 4], most studies in children and adolescence rely on an observational design that is unable to distinguish causation from correlation. Mendelian randomization (MR) is a technique that uses genetic variants as instrumental variables in observational epidemiological studies. MR has provided evidence to support a causal role of higher body mass index (BMI) on increasing the risk of various cardiometabolic diseases, predominantly in large populations of adults. However, due to the technique’s requirement for large sample sizes to provide adequate statistical power, MR studies use routinely collected clinical measures or data generated from high-throughput technologies. Recall by genotype (RbG) studies are an innovative extension of MR, designed to improve study efficiency, enable deep-phenotyping and improve causal inferences.
This PhD will aim to use recall by genotype (RbG) and compare this to other causal inference methodologies within the ALSPAC cohort and independent studies to further understand the causal role of various exposures on cardiovascular health. This is a PhD which will be coordinated with a Wellcome Trust funded programme of research that explicitly seeks to use the most contemporary and powerful study designs, metabolic data capture and analytical techniques to explore BMI as a risk factor. This programme is affiliated with the MRC Integrative Epidemiology Unit (University of Bristol).
In collaboration with Professor Alun Hughes (UCL), this PhD will make best use of data from a subsample of the Avon Longitudinal Study of Parents and Children (ALSPAC) with precise measures of cardiovascular health using magnetic resonance imaging (MRI) and very high resolution echocardiography. Of those with genetic data (N=8350), individuals were invited to a RbG study based on the lower/upper 30% of a genome-wide genetic risk score distribution associated with BMI. Of those invited (N=2071), 419 individuals were recalled at 21 years and underwent precise phenotyping including MRI-derived left ventricular mass index, stroke volume and cardiac output.
1. Cote, et al. J Am Coll Card. 2013;62:1309. 2. Tounian, et al. Lancet. 2001;358:1400. 3. Lewington, et al. Lancet. 2007;370:1829. 4. Prospective Studies Collaboration. Lancet. 1995;346:1647. 5. Davey Smith & Hemani. Hum Mol Gen. 2014;23:R89. 6. Corbin, et al. Nat Comms. 2018;9:711. 7. Boyd, et al. Int J Epi. 2013;42:111. 8. Wade, et al. Circulation. 2018;138.
Laura Howe (lead), Neil Davies,
Relationships between socioeconomic factors and health are likely to be bidirectional, with socioeconomic processes influencing both physical and mental health (social causation), but poor health also limiting educational potential and the ability to work (social drift).
The role of socioeconomic factors in shaping health is well studied, but the reverse process - poor health affecting socioeconomic outcomes - has been less well explored.
This project aims to generate high quality evidence that improving population levels of health will benefit educational attainment, employment and other socioeconomic outcomes, thus strengthening the economic argument for cross-governmental investment in health.
Observational evidence suggests that many aspects of poor health are associated with lower educational attainment and economic outcomes. However, these observational studies are likely to be plagued with bias due to confounding by background socioeconomic factors.
In this project, we will use cutting edge causal inference methods to assess the causal effects of health on educational attainment and economic outcomes. We will also use data from across the life course to understand how these relationships shift with age, and intergenerational data to assess effects of parental health on a child's socioeconomic outcomes.
You will carry out analysis of existing data from population-based cohort studies including UK Biobank and the Avon Longitudinal Study of Parents and Children (ALSPAC). This analysis will address questions such as:
1. Are associations between health and social and economic outcomes causal?
2. Are there periods of the life course during which health problems are particularly detrimental for economic outcomes?
3. Does parental health causally affect social and economic outcomes in children?
The project offers the opportunity to become skilled in techniques such as Mendelian Randomization, life course analyses and intergenerational analyses. There is also potential to use evidence synthesis approaches such as Multi-Parameter Evidence Synthesis to combine results from our analyses with external sources of evidence in order to gain better estimates of the potential economic return on improvements to population health, and the uncertainty around these estimates.
Lê F, Diez Roux A, Morgenstern H. Soc Sci Med. 2013 Jan;76(1):57-66
Tyrell et al. BMJ. 2016 Mar 8;352:i582
Davey Smith G, Hemani G. Hum Mol Genet. 2014 Sep 15;23(R1):R89-98.
Ades T. Stat Med. 2003 Oct 15;22(19):2995-3016
Dr Luisa Zuccolo (lead), Dr Gemma Sharp, Dr Cheryl McQuire, Dr Matt Suderman
Fetal alcohol spectrum disorders (FASDs) are lifelong disabilities caused by prenatal alcohol exposure, thought to be the leading preventable cause of developmental disability in the world. Understanding which individuals are most at risk to develop FASD is key. Early identification of FASD risk through robust maternal and infant biomarkers will offer an early window for intervention, will focus specialist diagnostic efforts in a resource-limited healthcare system, and will contribute towards mitigating FASD-related disabilities in later life. It may also prevent or reduce feetal exposure to alcohol in subsequent pregnancies.
This project will focus on DNA methylation (DNAm). A DNA methylation score was recently discovered that can robustly distinguish between current heavy drinkers and non-drinkers in a general population. In addition, strong evidence from animal models indicates that alcohol use during pregnancy can alter offspring DNAm.
We have developed an algorithm to screen for FASD in the Avon Longitudinal Study of Parents and Children (ALSPAC). This allows to identify individuals at high risk of FASD regardless of a formal (and rare!) diagnosis. Using the algorithm as the best available proxy for an FASD diagnosis in the ALSPAC study, we propose to run analyses integrating molecular data (DNAm-based markers) to find out whether a very early (prenatal or early perinatal) risk assessment would be possible.
To identify optimal maternal and infant biomarkers of FASD, by integrating DNA methylation data and rich phenotypic data on FASD risk in the ALSPAC study and replication cohorts.
1. To review current evidence on epigenetic (specifically methylation)-based biomarkers in prediction of alcohol use and/or health consequences of prenatal alcohol use, including FASD
2. To identify novel maternal DNAm-based predictors of FASD and compare their diagnostic performance to previously known predictors
3. To identify novel infant DNAm-based predictors of FASD and compare their diagnostic performance to previously known predictors (maternal and offspring)
4. To replicate the findings of Objectives 2 and 3 in independent cohorts from populations with different rates of FASD (eg Western Europe Vs South Africa)
5. To extend the predictor by including genetic and phenotypic risk factors for FASD
Univariate and multivariate regressions
Epigenome-wide association studies (EWAS)
Liu, C. et al. A DNA methylation biomarker of alcohol consumption. Mol. Psychiatry 23(2):422-433 (2018)
Mandal, C., Halder, D., Jung, K. H. & Chai, Y. G. In Utero Alcohol Exposure and the Alteration of Histone Marks in the Developing Fetus: An Epigenetic Phenomenon of Maternal Drinking. Int. J. Biol. Sci. 13, 1100–1108 (2017).
McQuire C, Mukherjee R, Hurt L, Higgins A, Greene G, Farewell D, Kemp A, Paranjothy, S. Screening prevalence of fetal alcohol spectrum disorders in a region of the United Kingdom: a population-based birth-cohort study. Preventive Medicine (submitted).
Chudley AE, Conry J, Cook JL, Loock C, Rosales T, LeBlanc N. Fetal alcohol spectrum disorder: Canadian guidelines for diagnosis. Canadian Medical Association Journal. 2005;172(5 suppl):S1-S21.
Dr Gemma Sharp (lead), Dr Luisa Zuccolo, Prof Anita Thapar
There is growing concern that paternal exposures before conception have been greatly neglected. Studying how these impact on future generations’ health could open new avenues for prevention-prospective fathers are not generally advised to change their behaviour. In animal studies, one of the paternal exposures showing the largest and most consistently reported effects relative to the prenatal period is alcohol, however convincing human evidence is lacking to date. In this project, we propose to investigate the relationship between pre-conception paternal alcohol use and offspring mental health, and in particular attention deficit hyperactivity disorder (ADHD).
1 Is paternal alcohol use before conception associated with offspring ADHD, and/or its brain morphology correlates?
2 Is this due to shared genetic influences?
3 Is this association robust to different causal inference methods of analyses such as Mendelian randomization, and the use of negative controls (eg non-biological fathers)?
4 Are offspring cord blood DNA methylation and ADHD structural brain correlates on the causal pathway from paternal alcohol exposure childhood ADHD?
To improve the chance of disentangling correlation from causation, we propose to use a paradigm similar to that of (lab studies of) rodent models of paternal effects: a pseudo-experimental study design (employing Mendelian randomization and other analytical approaches to improve causal inference), restricting to families with no intrauterine exposure to alcohol (like in the animal studies), focusing on early manifestation of the outcome occurring before the onset of own drinking to remove confounding by own drinking, and even earlier potential biological mediators of the effects (eg cord-blood DNA methylation, eliminating confounding by paternal drinking in the postnatal period, both set to ‘zero’ in animal studies).
Finegersh A, Alcohol 2015 49:461-70
Rogers JC, JAMA Psychiatry 2016 73:64-72
Chen YC, Mol Psychiatry 2017 Mar 21 -Epub ahead of print
Dr Tom Richardson (lead), Professor George Davey Smith, Professor Tim Frayling
Leveraging high dimensional biomedical and molecular datasets provides an unprecedented opportunity to develop our understanding of both the environmental and genetic components of complex disease. This project will focus on developing expertise in applying state-of-the-art techniques in causal inference and bioinformatics to large-scale and phenotypically rich datasets. In doing so, the successful student will undertake leading research in the field of epidemiology by uncovering risk factors, genes and biological pathways which contribute to disease susceptibility.
Our understanding of how our genes confer a predisposition to disease from birth has made substantial advances in the last decade. Polygenic risk scores (PRS), commonly defined as the weighted sum of risk alleles for a given disease, are now beginning to demonstrate their full potential in terms clinical utility and disease prediction. Furthermore, these profiles provide an unparalleled opportunity to evaluate the causal relationship between modifiable risk factors and outcomes by applying the principles of Mendelian randomization (MR).
- Build polygenic risk scores for various risk factors and complex traits
- Evaluate causal relationships between modifiable risk factors and disease outcomes using these scores
- Dissect scores for identified associations to develop understanding into how lifestyle factors confer risk of disease
- Evaluate the individual genes driving observed associations to improve biological interpretation of results
This PhD will involve constructing PRS for hundreds of complex traits and using them in MR analyses to identify and characterise novel epidemiological relationships. There will be an emphasis on innovative development and application of these scores, as well as dissecting identified associations, for example to discern which pathway sets of genes appear to be driving the identified effect. Extensive follow-up of these results will involve using high-dimensional molecular datasets to develop our understanding of the underlying biological mechanisms in disease. For instance, integrating molecular datasets can help us unravel causal genes and mechanisms in disease, as well as detect biomarkers which may prove valuable for early disease prognosis and prevention.
We will also harness large-scale sequencing data (e.g. exome and whole genome sequencing data from the UK Biobank study) and build gene risk scores (GRS) to elucidate genes which influence disease risk. Furthermore, these analyses will likely yield insight into proposed therapeutic intervention, for instance evaluating potential adverse effects and possible drug repurposing.
Professor Kate Tilling (lead), Dr Jonathan Bartlett, Dr Rachael Hughes Rosie Cornish
Missing data can cause bias, and common solutions (MI and IPW) are not always applicable – particularly where data are missing not at random (MNAR). We will develop methods for analysing incomplete MNAR data – including Bayesian methods and Instrumental Variables – and apply them to cohort studies, particularly those with linkage to routine data (e.g. GP records).
The central aim of this PhD is to develop sensitivity analyses which allow for data to be MNAR, and which make optimal use of external data.
Typically such MNAR sensitivity analyses have involved analysts specifying priors for sensitivity parameters which are difficult to interpret and hence also difficult to elicit prior beliefs or knowledge about (1). We will build on recent work (2) to develop approaches which instead rely on estimates of simple population quantities (e.g. from the linked data), such as the average BMI or proportion with depression. This will involve investigating how sensitive results are to these priors/estimates and to the particular non-response model used. Methods explored will include Bayesian and Instrumental Variables models.
The methods developed will be applied to important epidemiological questions using data from cohorts, including the world-leading Avon Longitudinal Study of Parents and Children (ALSPAC) cohort, and UK Biobank.
1. Stat Med. 2018 Jul 10;37(15):2338-2353. On the use of the not-at-random fully conditional specification (NARFCS) procedure in practice. Tompsett et al.
2. Population-calibrated multiple imputation for a binary/categorical covariate in categorical regression models. Statistics in medicine. ISSN 0277-6715 DOI: https://doi.org/10.1002/sim.8004. Pham et al.
Dr Eleanor Sanderson (lead), Dr Rhian Daniel (Cardiff), Professor George Davey Smith Dr Laura Howe
Many of an individual’s traits are observationally associated with their health outcomes. Understanding the exact relationships between these factors is critical to effective public health intervention. When multiple traits are potentially associated with a disease or health outcome, it is often not clear how much of the observed effect of each single trait is due to the effect of that trait or behaviour on other traits and behaviours, which then affect the outcome, and how much is “direct” in the sense that it is not mediated by the other traits being considered.
Causal mediation analysis is one approach that can be used to determine the proportion of the effect of a trait on an outcome that is via a mediating variable. However, this method relies on many strong assumptions. An alternative approach, relying on different assumptions, is Mendelian Randomisation (MR). MR is a method of instrumental variable analysis which utilises genetic variation between individuals to help understand causal effects.
The aim of this project is to conduct research on the strengths and limitations of MR when trying to understand the causal effects of multiple exposures on a health outcome. This will include investigation of how novel methods of MR analysis such as Multivariable MR relate to mediation analysis and developing and extend existing MR methods to deal with multiple mediators. This project will also explore possibilities of combining the two approaches.
In this PhD you will have the opportunity to work with leading researchers in the fields of population health, statistics, and Mendelian randomisation to further develop statistical methods for causal analysis based around MR mediation analysis.
This project will involve mathematical derivation of the properties of the extended methods and verification using simulation studies. This project will also involve analysis of both individual-level and summary-level data to apply the methods developed.
Although this project will be methodological in focus, the student will have the opportunity to develop a relevant application of these methods based on their personal research interests. Prospective applicants should have a strong quantitative background and an interest in developing methods for causal analysis within a population health setting; however, no particular background knowledge is required.
Professor Jon Tobias (lead), Dr Louise Millard, Dr Celia Gregson
Osteoarthritis (OA) is a major cause of morbidity in older people representing a significant economic burden for the NHS through costs of surgery. The AUGMENT study (funded by a Wellcome collaborative grant) aims to reduce this impact, by underpinning new preventative strategies for OA based on improved understanding of causal pathways, and development of predictive tools for progression to joint replacement. Involving 100,000 individuals from UK Biobank, AUGMENT will focus on the role of joint shape, derived from DXA scans by principal components analysis. However, OA progression is related to several other features that are detectable by DXA scans but independent of joint shape, including thickening of the bone immediately adjacent to the joint (sclerosis), bone mineral density (BMD) (1), and bone texture (2); these features also need evaluation if the potential of using high-resolution DXA to identify those at risk of OA progression is to be fully realised.
I. To develop automated methods for evaluating sclerosis, BMD and texture using knee DXA scans.
II. To use machine learning to predict risk of OA progression from knee DXA scans by combining modalities in (I), with joint shape including osteophytes.
III. To evaluate how well our classifier predicts subsequent total knee replacement.
IV. To identify novel genetic pathways contributing to knee OA progression, based on relationships between genome wide genetic data and our classifier output.
I. Sclerosis will be derived from pixel density profiles adjacent to the knee joint. Positioning of ROIs previously used to measure medial/lateral BMD will be automated (1). Methods for evaluating lumbar spine bone texture (3) will be applied to the tibial subchondral region.
II. Modalities from (I) and knee shape evaluated in AUGMENT including osteophytes, will be obtained in the OAI bone study (n=629), an early knee OA cohort at high risk of subsequent progression. These data will be used to 1) train a classifier to predict radiographic knee OA progression and 2) evaluate classifier performance using cross-validation. Following validation in a manually labelled subsample of UK Biobank, the classifier will be applied to all 100,000 UK Biobank DXA images.
III. We will test whether our OA progression classifier predicts total knee replacement based on HES-linked incidence data, in UK Biobank.
IV. GWAS will be performed in UK Biobank to identify genetic influences on knee OA progression.
1 Lo GH Periarticular bone predicts knee OA progression. Semin Arthritis Rheum 2018
2 Lespessailles E Bone texture analysis on radiographic images. Calcif Tiss Int 2007;80:97-102
3 Schousboe JT Association of Trabecular Bone Score With Vertebral Fractures J Bone Miner Res 2017;32:1554-8
Prof Tom Gaunt (lead), Prof Jules Hancox,
The orderly sequence of electrical excitation of the heart measured at the body surface as the electrocardiogram (ECG) arises due to the combined activity of multiple ion channel proteins and electrogenic transporters. Mutations to the underlying genes lead to arrhythmia disorders, whilst more common single nucleotide polymorphisms (SNPs) produce more subtle effects that underlie some of the observed population differences in ECG parameters. Genome wide association studies (GWAS) have identified SNPs in a number of ion-channel and non-ion channel genes that associate with ECG traits [1, 2], providing the opportunity to explore the causal relationship between ECG traits and other phenotypes. An approach called Mendelian randomization (MR)  pioneered in Bristol offers potential to exploit SNPs as causal “anchors” to determine whether ECG traits cause other phenotypes, and also whether other phenotypes causally affect ECG traits. MR has been widely applied, eg in demonstrating the likely inefficacy of HDL raising treatment without requiring drug development .This methodology can exploit a wealth of publicly available datasets integrated with cutting-edge statistical methods in the MR-Base platform for systematic MR (http://www.mrbase.org/) developed in Bristol. An additional analytical platform (LD Hub, http://ldsc.broadinstitute.org/) exploits the same data to enable evaluation of genetic correlation between phenotypes.
The aims of this project are (1) to identify causal factors influencing ECG traits and (2) to identify phenotypes causally influenced by ECG traits. The specific objectives include:
1. Collating published data on genetic associations with ECG traits
2. LD score regression (using LD Hub) to analyse genetic correlation between ECG traits and other traits.
3. Mendelian randomization (MR) analyses to identify risk factors, drug targets and lifestyle exposures that alter specific ECG traits.
4. MR analyses to identify phenotypes and health outcomes that are causally influenced by ECG traits
5. Performing in vitro analyses of pharmacologically tractable molecular pathways
Collating published data: data from GWAS of ECG traits (such as QT-interval, QRS-duration, PR-interval) will be collated, processed and uploaded to the MR-Base/LD Hub database.
LD score regression: These analyses will use LD Hub to analyse genetic correlation between ECG traits and other traits in the database. The genetic correlation indicates the extent to which two traits share a common genetic basis, and enables common molecular pathways to be identified. These will be subject to pathway analysis.
Mendelian randomization: Two different categories of analysis will be performed. In the first, genetic variants related to specific druggable genes, lifestyle exposures (eg smoking) and other traits (eg blood pressure) will be tested for their influence on ECG traits to identify novel risk factors and potential intervention targets for abnormal ECG traits. In the second category genetic variants related to ECG traits will be tested for their influence on other phenotypes and health outcomes to determine whether ECG traits causally influence other aspects of health.
Performing in vitro analyses: For genetic variants and pathways that are pharmacologically tractable, causality will be tested using electrophysiological recording from appropriate isolated cardiac cell preparations and pharmacological agents to disrupt or enhance gene product function.
1. Arking DE, Pulit SL, Crotti L, van der Harst P, Munroe PB, Koopmann TT, Sotoodehnia N, Rossin EJ, Morley M, Wang X et al: Genetic association study of QT interval highlights role for calcium signaling pathways in myocardial repolarization. Nat Genet 2014, 46(8):826-836.
2. Gaunt TR, Shah S, Nelson CP, Drenos F, Braund PS, Adeniran I, Folkersen L, Lawlor DA, Casas J-P, Amuzu A et al: Integration of Genetics into a Systems Model of Electrocardiographic Traits Using HumanCVD BeadChip. Circulation-Cardiovascular Genetics 2012, 5(6):630-638.
3. Davey Smith G, Ebrahim S: 'Mendelian randomization': can genetic epidemiology contribute to understanding environmental determinants of disease? Int J Epidemiol 2003, 32(1):1-22.
4. Burgess S, Harshfield E: Mendelian randomization to assess causal effects of blood lipids on coronary heart disease: lessons from the past and applications to the future. Curr Opin Endocrinol Diabetes Obes 2016, 23(2):124-130.
Prof Tom Gaunt (lead), Prof Peter Flach, Dr Ben Elsworth
Population health research is being transformed by the increasing wealth of complex data. New high-dimensional epidemiological datasets provide novel opportunities for systematic approaches to understanding the relationships between risk factors and disease outcomes. Moving beyond individual hypothesis testing to fully exploit the available data requires new approaches to data mining, including the development of machine learning approaches, natural language processing and application of ontologies/knowledge representation.
This studentship will work on data mining within a novel platform (EpiGraphDB) being developed within the MRC IEU. EpiGraphDB integrates genomic and population health data with information mined from the scientific literature and from a range of bioinformatic databases.
The aim of this studentship is to develop and apply data mining methods within a complex graph database representing epidemiological relationships between traits and other relevant biological information.
1. Develop efficient methods for identification of informative sub-graphs utilising a range of methods
2. Develop approaches to knowledge representation within the graph database to support more effective data mining
3. Develop approaches to the triangulation of different types of evidence to prioritise findings
4. Identify novel risk factors for disease
5. Identify potentially spurious established risk factors
The student will be encouraged and supported in developing their own research ideas.
A wide range of research methods may be used, including machine learning, network analysis, natural language processing, causal inference. The exact methods will depend on the background and interests of the successful candidate.
Dr Sarah Lewis (lead), Dr Evie Stergiakouli, Dr Gemma Sharp
Orofacial clefts are the most common types of birth defects worldwide and occur in around 650 live births in the UK. Although we know that Aaround 30% of clefts arise as a result of genetic syndromes and follow a monogenic model of inheritance, but the majority of clefts follow a multifactorial model with both genetic and prenatal environmental risk factors that are still largely unknown. A better understanding of the causes of orofacial clefts will be essential to inform better prediction and prevention strategies, and to improve outcomes amongst those born with a cleft. Previous genome wide association studies (GWAS) of non-syndromic cleft lip and/or palate have identified several new genetic loci (Leslie et al, 2016). However, as has been shown in other diseases, further GWAS in different populations and with larger sample sizes are likely to uncover new variants. In addition, existing cleft GWAS have compared genes in people with a cleft to people without. However, maternal genes may also determine whether a child develops a cleft or not since these influence the prenatal environment. This project will use a combination of GWAS analyses and Mendelian randomization to identify genetic and modifiable causes of cleft.
Aim: The overall aim of this project is to identify causal genetic and lifestyle factors for cleft
1) To carry-out a GWAS using genetic data from the Cleft Collective to identify genetic variants associated with cleft
2) To carry-out a GWAS of mothers whose children were born with cleft compared to control mothers to identify maternal risk variants
3) To perform Mendelian randomization analyses to determine whether Mother’s BMI and folic acid intake and other nutrients are risk factors for cleft
Generate analysis plan and conduct GWAS of child’s genetic variants and mother’s genetic variants with child’s cleft phenotype as the outcome.
Stratify by cleft sub-types and repeat the above GWAS analyses.
To identify genetic instruments for nutritional factors hypothesized to cause cleft lip and or cleft palate.
To perform two sample Mendelian randomization to identify causal risk factors for cleft.
Dr Evie Stergiakouli (lead), Dr Emma Anderson, Dr Laura Howe
Patients with childhood neurodevelopmental disorders, such as attention-deficit/hyperactivity
disorder (ADHD) and autism spectrum disorder (ASD), could be at increased risk of developing
Alzheimer’s disease but this has not been investigated adequately due to the lack of long-term
follow-up data. Utilizing the existing information generated by genetic studies and large population
cohort studies we will test if the genetic risk factors causing childhood neurodevelopmental
disorders are also implicated in Alzheimer’s disease. We will also use rapid high-throughput
analysis methods developed in the MRC IEU to take advantage of publicly available information on
the genetics of complex diseases to test if childhood neurodevelopmental disorders can cause
Alzheimer’s disease. This project could have implications for identifying patients at increased risk of
dementia and supporting a particularly vulnerable group of patients.
We have the expertise to train a student in applying polygenic risk score analysis and Mendelian randomization with the aim to investigate if there is genetic overlap between childhood neurodevelopmental disorders and Alzheimer’s disease and if childhood neurodevelopmental disorders can cause Alzheimer’s disease, using genetic data. These methods have not been applied to study the genetic overlap of Alzheimer’s disease with childhood measures and neither have the causal effect of childhood ADHD and ASD on Alzheimer’s disease been investigated in a framework that takes horizontal pleiotropy into account. Research on Alzheimer’s disease in children is important because it avoids issues of selection bias which can occur when studying
The specific hypotheses to be tested are:
1. There is shared genetic susceptibility between childhood neurodevelopmental disorders (including ADHD and ASD) and Alzheimer’s disease.
2. ADHD and/or ASD are causally associated with Alzheimer’s disease.
To test the first hypothesis we will calculate polygenic risk scores for ADHD and ASD in adults from the general population and test whether they are associated with trajectories of cognitive decline from mid- to late-life and brain imaging measures. Polygenic risk scores for Alzheimer’s disease will also be calculated to test whether they are associated with trajectories of ADHD and ASD symptoms in children from the general population.
To test the second hypothesis we will conduct 2-sample MR and perform sensitivity analyses to test and adjust for pleiotropic effects.
Callahan BL, Bierstone D, Stuss DT, Black SE. Adult ADHD: Risk Factor for Dementia or Phenotypic Mimic? Frontiers in Aging Neuroscience. 2017;3(9):260
Golimstok A, Rojas JI, Romano M, Zurru MC, Doctorovich D, Cristiano E. Previous adult attention-deficit and hyperactivity disorder symptoms and risk of dementia with Lewy bodies: a case-control study. European Journal of Neurology. 2011;18(1):78
Lambert JC, Ibrahim-Verbaas CA, Harold D, Naj AC, Sims R, Bellenguez C, et al. Metaanalysis of 74,046 individuals identifies 11 new susceptibility loci for Alzheimer's disease. Nature Genetics. 2013;45(12):1452-8.
Demontis D, Walters RK, Martin J, Mattheisen M, Als TD, Agerbo E et al. Discovery of the first genome-wide significant risk loci for ADHD. BioRxiv. Jun. 3, 2017; doi: http://dx.doi.org/10.1101/145581
Grove J, Ripke S, Als TD, Mattheisen M, Walters R, Won H, et al. Common risk variants identified in autism spectrum disorder . BioRxiv. Nov. 25, 2017; doi: http://dx.doi.org/10.1101/224774
Dr Evie Stergiakouli (lead), Dr Yvonne Wren, Dr Sarah Lewis
Approximately 1000 children are born in the UK each year with a cleft of the palate. Surgical repair takes place within the first year of life, but the impact of the cleft and the surgical repair can have long lasting consequences on speech and language development as well as other outcomes. While some children have cleft lip/palate as part of a syndrome, most are non-syndromic and multifactorial in nature. The Cleft Collective Cohort Studies was set up to provide a data resource for researchers and clinical academics to address questions related to risk factors and outcomes for individuals affected by cleft. The dataset includes DNA and recent funding has facilitated genotyping of the entire sample. Combined with information from parent report and clinical records, there is a rich dataset of genetic, speech, language and environment information.
Speech and language problems are heritable as shown by family and twin studies (Newbury and Monaco 2010). The genetic mechanisms underlying susceptibility to speech and language disorders are multifactorial in nature, involving both genetic factors and environmental influences. Recently genome-wide association studies have started to identify specific genetic variants contributing to speech and language impairments (Reader et al. 2014). This project provides a unique opportunity to contribute to our understanding of why some children born with cleft palate are more likely to have problems with speech and language than others. This information will be of use to clinicians in terms of knowing who to prioritise for treatment and/or provide preventative intervention. Moreover, it will enable parents to be fully informed about their child’s prognosis and options for management.Recent work by the Cleft Collective team using polygenic risk scores and bidirectional Mendelian randomisation found strong evidence of genetic overlap between nonsyndromic cleft lip and palate and philtrum width (Howe et al, 2018). The associations of cleft with education have also been studied using similar methods.
The aim of this project is to use quantitative genetics methods and causally informative designs to investigate genetic influences on speech and language among those born with a cleft palate and to compare this with genetic risk factors for speech and language impairment in the general population. The project will also apply causally informative designs to investigate whether having a cleft palate causes adverse outcomes in speech as opposed to being correlated with them . The candidate will have access to detailed speech and language data from the Avon Longitudinal Study of Parents and Children and the Cleft Collective which are both based in Bristol.
This a highly interdisciplinary project which involves applying cutting-edge genetic epidemiological methods to clinical phenotypes that have not been at the forefront of research on genetics. It also highly original, since it will apply causally informative designs based on genetics findings to investigate if a phenotype (in this case cleft lip/palate) is causing adverse outcomes as opposed to being correlated with them. The supervisors involved come from completely different backgrounds (clinical SLT, cleft, genetics, epidemiology, linguistics) and the candidate would have the opportunity of developing into one of the few experts with in-depth understanding of these fields. The data required for this project have already been collected or are publicly available making it less likely for setbacks to appear. One of the unique features of this project is that the candidate will be able to gain a good understanding of running a cohort study by attending weekly Cleft Collective meetings on questionnaire planning, public and patient involvement, research governance and funding applications.
Dr Evie Stergiakouli (lead), Dr Gemm Sharp, Dr Sarah Lewis
Approximately 1000 children are born in the UK each year with a cleft lip and/or palate. Surgical repair takes place within the first year of life, but the impact of the cleft and the surgical repair can have long lasting consequences. While some children have cleft lip/palate as part of a syndrome, most are non-syndromic and multifactorial in nature. The Cleft Collective Cohort Studies (http://www.bristol.ac.uk/dental/cleft-collective/) was set up to provide a data resource for researchers and clinical academics to address questions related to risk factors and outcomes for individuals affected by cleft. The dataset includes DNA and recent funding has facilitated genotyping of the entire sample. Combined with information from parent report and clinical records, there is a rich dataset of genetic and behavioural development outcomes.
Recent studies have suggested that children with nonsyndromic clefts have a higher risk of neurodevelopmental disorders, including ADHD, autism and psychotic disorders (Tillman et al. 2018). In addition, nonsyndromic clefts have been found to be associated with poor academic achievement (Richman et al. 2012). Recent work by the Cleft Collective team using polygenic risk scores and bidirectional Mendelian randomisation did not identify evidence of genetic overlap between nonsyndromic cleft lip and/or palate and educational attainment (Dardani et al. BioRxiv 2018). This indicates that the risk is unlikely to be explained by familial influences such as inherited genetic factors.
This project provides a unique opportunity to contribute to our understanding of why some children born with cleft lip and/or palate are more likely to have neurodevelopmental problems than others. This information will be of use to clinicians in terms of knowing who to prioritise for treatment and/or provide preventative intervention. Moreover, it will enable parents to be fully informed about their child’s prognosis and options for management.
The aims of the project are:
To describe neurodevelopmental outcomes in children born with cleft lip and/or palate in the UK and compare them to children of the same age from the general population.
To investigate genetic influences on neurodevelopmental disorders among those born with a cleft lip and/or palate and to compare this with genetic risk factors for neurodevelopmental disorders in the general population.
To investigate whether having a cleft causes adverse neurodevelopmental outcomes as opposed to being correlated with them.
For the first aim longitudinal data on neurodevelopmental disorders from the Cleft Collective will be utilized and epidemiological analysis will be employed.
For the second aim we will use quantitative genetics methods and causally informative designs to investigate genetic influences on neurodevelopmental disorders among those born with a cleft lip and/or palate.
For the third aim will also apply causally informative designs to investigate whether having a cleft causes adverse neurodevelopmental outcomes as opposed to being correlated with them.
Tillman et al. J Am Acad Child Adolesc Psychiatry 2018;57(11):876–883
Dardani et al. BioRxiv 2018 doi: https://doi.org/10.1101/434126
Dr Evie Stergiakouli (lead), Dr Dheeraj Rai,
Neurodevelopmental disorders start in childhood and include Attention Deficit Hyperactivity Disorder (ADHD), autism spectrum disorder (ASD) and learning/intellectual disabilities. These conditions are common, collectively affecting around 5-10% of children and often associated with long term disabilities. Although previously considered as childhood-limited, the difficulties associated with these conditions can be life-long and impair many aspects of adult life, including physical health and wellbeing. Although neurodevelopmental disorders are considered as categorical diagnoses, there is evidence to suggest that they behave as continuous traits in the general population.
Neurodevelopmental disorders and traits are highly heritable with a contribution from common genetic variants. These variants can be utilized to investigate associations and causal links between neurodevelopmental disorders and physical health outcomes in adults with a neurodevelopmental disorder or with a high genetic risk for a neurodevelopmental disorder.
This projects aims to investigate the link between neurodevelopmental disorders and traits and poor physical health outcomes and to test for a causal relationship by using epidemiological methods.
The student will have the opportunity to receive training in:
· epidemiological analysis within large population cohort studies including ALSPAC and the Stockholm Youth Cohort
· genetic epidemiological analysis including polygenic risk score analysis and LD score regression in large datasets including ALSPAC and the UK Biobank
· 2-sample mendelian randomization analysis and other causal methods developed in the MRC Integrative Epidemiology Unit to assess and take into account pleiotropic effects.
Dr Sarah Lewis (lead), Dr Dheeraj Rai,
Background: There is a body of evidence that heavy metal exposure in the prenatal period and in the first two years of life may lead to impaired neurodevelopment. Previous studies have suggested adverse effects of arsenic and lead exposure (even at low levels) on children's cognitive function, including lower IQ scores, impaired attention and memory, and behavioural problems. In addition, copper and methylmercury have been shown to have negative effects on the developing brain. The causal nature of these associations is still unclear, particularly for a wider range of neurodevelopmental disorders such as autism and attention deficit/hyperactivity disorder (ADHD). It is also unclear whether any effects of heavy metals on the brain are long lasting and whether they could be related to negative behaviours or mental health issues later in adolescence or adulthood.
Aim: The overall aim of this project is to use Mendelian randomization to determine whether early life heavy metal exposure is causally related to neurodevelopmental disorders and with mental health problems and problem behaviours in adolescents and young adults.
1) To identify genetic instruments for lead, arsenic, copper and methylmercury
2) To carry-out association analyses of prenatal (and early life) heavy metal exposure with symptoms and diagnoses of behavioural (conduct disorder), neurodevelopmental (autism, ADHD, intellectual disability); and mental health (depression, anxiety and psychotic experiences) problems in a population based birth cohort (ALSPAC)
3) To perform Mendelian randomization analyses to determine whether these associations are likely to be causal.
Generate analysis plan and draw possible causal diagrams.
Carry-out standard multivariate logistic regression analyses based on observational data.
To perform two sample Mendelian randomization to determine whether heavy metals are causal risk factors for conduct disorder, depression and anxiety in adolescents and young adults.
Smith GD and Ebrahim S. Int J Epidemiol. 2003;32(1):1-22.
Lanphear, B. P. et al. 2005. Environmental Health Perspectives 2005; 113, 894-899.
Daniels, J. L., Longnecker, M. P., Rowland, A. S., and Golding, J. Epidemiology 2005; 15, 394-402.