RESUMO
Observational studies are rarely representative of their target population because there are known and unknown factors that affect an individual's choice to participate (the selection mechanism). Selection can cause bias in a given analysis if the outcome is related to selection (conditional on the other variables in the model). Detecting and adjusting for selection bias in practice typically requires access to data on nonselected individuals. Here, we propose methods to detect selection bias in genetic studies by comparing correlations among genetic variants in the selected sample to those expected under no selection. We examine the use of four hypothesis tests to identify induced associations between genetic variants in the selected sample. We evaluate these approaches in Monte Carlo simulations. Finally, we use these approaches in an applied example using data from the UK Biobank (UKBB). The proposed tests suggested an association between alcohol consumption and selection into UKBB. Hence, UKBB analyses with alcohol consumption as the exposure or outcome may be biased by this selection.
RESUMO
Common genetic variation has been associated with multiple phenotypic features in Autism Spectrum Disorder (ASD). However, our knowledge of shared genetic factor structures contributing to this highly heterogeneous phenotypic spectrum is limited. Here, we developed and implemented a structural equation modelling framework to directly model genomic covariance across core and non-core ASD phenotypes, studying autistic individuals of European descent with a case-only design. We identified three independent genetic factors most strongly linked to language performance, behaviour and developmental motor delay, respectively, studying an autism community sample (N = 5331). The three-factorial structure was largely confirmed in independent ASD-simplex families (N = 1946), although we uncovered, in addition, simplex-specific genetic overlap between behaviour and language phenotypes. Multivariate models across cohorts revealed novel associations, including links between language and early mastering of self-feeding. Thus, the common genetic architecture in ASD is multi-dimensional with overarching genetic factors contributing, in combination with ascertainment-specific patterns, to phenotypic heterogeneity.
Assuntos
Transtorno do Espectro Autista , Transtorno Autístico , Humanos , Transtorno Autístico/genética , Transtorno do Espectro Autista/genética , Fenótipo , Idioma , Modelos EstruturaisRESUMO
BACKGROUND: The number of words children produce (expressive vocabulary) and understand (receptive vocabulary) changes rapidly during early development, partially due to genetic factors. Here, we performed a meta-genome-wide association study of vocabulary acquisition and investigated polygenic overlap with literacy, cognition, developmental phenotypes, and neurodevelopmental conditions, including attention-deficit/hyperactivity disorder (ADHD). METHODS: We studied 37,913 parent-reported vocabulary size measures (English, Dutch, Danish) for 17,298 children of European descent. Meta-analyses were performed for early-phase expressive (infancy, 15-18 months), late-phase expressive (toddlerhood, 24-38 months), and late-phase receptive (toddlerhood, 24-38 months) vocabulary. Subsequently, we estimated single nucleotide polymorphism-based heritability (SNP-h2) and genetic correlations (rg) and modeled underlying factor structures with multivariate models. RESULTS: Early-life vocabulary size was modestly heritable (SNP-h2 = 0.08-0.24). Genetic overlap between infant expressive and toddler receptive vocabulary was negligible (rg = 0.07), although each measure was moderately related to toddler expressive vocabulary (rg = 0.69 and rg = 0.67, respectively), suggesting a multifactorial genetic architecture. Both infant and toddler expressive vocabulary were genetically linked to literacy (e.g., spelling: rg = 0.58 and rg = 0.79, respectively), underlining genetic similarity. However, a genetic association of early-life vocabulary with educational attainment and intelligence emerged only during toddlerhood (e.g., receptive vocabulary and intelligence: rg = 0.36). Increased ADHD risk was genetically associated with larger infant expressive vocabulary (rg = 0.23). Multivariate genetic models in the ALSPAC (Avon Longitudinal Study of Parents and Children) cohort confirmed this finding for ADHD symptoms (e.g., at age 13; rg = 0.54) but showed that the association effect reversed for toddler receptive vocabulary (rg = -0.74), highlighting developmental heterogeneity. CONCLUSIONS: The genetic architecture of early-life vocabulary changes during development, shaping polygenic association patterns with later-life ADHD, literacy, and cognition-related traits.
Assuntos
Transtorno do Deficit de Atenção com Hiperatividade , Alfabetização , Adolescente , Humanos , Lactente , Transtorno do Deficit de Atenção com Hiperatividade/genética , Transtorno do Deficit de Atenção com Hiperatividade/diagnóstico , Cognição , Estudo de Associação Genômica Ampla , Estudos Longitudinais , Fenótipo , VocabulárioRESUMO
The use of spoken and written language is a fundamental human capacity. Individual differences in reading- and language-related skills are influenced by genetic variation, with twin-based heritability estimates of 30 to 80% depending on the trait. The genetic architecture is complex, heterogeneous, and multifactorial, but investigations of contributions of single-nucleotide polymorphisms (SNPs) were thus far underpowered. We present a multicohort genome-wide association study (GWAS) of five traits assessed individually using psychometric measures (word reading, nonword reading, spelling, phoneme awareness, and nonword repetition) in samples of 13,633 to 33,959 participants aged 5 to 26 y. We identified genome-wide significant association with word reading (rs11208009, P = 1.098 × 10-8) at a locus that has not been associated with intelligence or educational attainment. All five reading-/language-related traits showed robust SNP heritability, accounting for 13 to 26% of trait variability. Genomic structural equation modeling revealed a shared genetic factor explaining most of the variation in word/nonword reading, spelling, and phoneme awareness, which only partially overlapped with genetic variation contributing to nonword repetition, intelligence, and educational attainment. A multivariate GWAS of word/nonword reading, spelling, and phoneme awareness maximized power for follow-up investigation. Genetic correlation analysis with neuroimaging traits identified an association with the surface area of the banks of the left superior temporal sulcus, a brain region linked to the processing of spoken and written language. Heritability was enriched for genomic elements regulating gene expression in the fetal brain and in chromosomal regions that are depleted of Neanderthal variants. Together, these results provide avenues for deciphering the biological underpinnings of uniquely human traits.
Assuntos
Estudo de Associação Genômica Ampla , Individualidade , Leitura , Fala , Adolescente , Adulto , Criança , Pré-Escolar , Loci Gênicos , Humanos , Idioma , Polimorfismo de Nucleotídeo Único , Adulto JovemRESUMO
Marked physiological changes in pregnancy are essential to support foetal growth; however, evidence on the role of specific maternal metabolic traits from human studies is limited. We integrated Mendelian randomisation (MR) and metabolomics data to probe the effect of 46 maternal metabolic traits on offspring birthweight (N = 210,267). We implemented univariable two-sample MR (UVMR) to identify candidate metabolic traits affecting offspring birthweight. We then applied two-sample multivariable MR (MVMR) to jointly estimate the potential direct causal effect for each candidate maternal metabolic trait. In the main analyses, UVMR indicated that higher maternal glucose was related to higher offspring birthweight (0.328 SD difference in mean birthweight per 1 SD difference in glucose (95% CI: 0.104, 0.414)), as were maternal glutamine (0.089 (95% CI: 0.033, 0.144)) and alanine (0.137 (95% CI: 0.036, 0.239)). In additional analyses, UVMR estimates were broadly consistent when selecting instruments from an independent data source, albeit imprecise for glutamine and alanine, and were attenuated for alanine when using other UVMR methods. MVMR results supported independent effects of these metabolites, with effect estimates consistent with those seen with the UVMR results. Among the remaining 43 metabolic traits, UVMR estimates indicated a null effect for most lipid-related traits and a high degree of uncertainty for other amino acids and ketone bodies. Our findings suggest that maternal gestational glucose and glutamine are causally related to offspring birthweight.
RESUMO
Two-sample summary data Mendelian randomization is a popular method for assessing causality in epidemiology, by using genetic variants as instrumental variables. If genes exert pleiotropic effects on the outcome not entirely through the exposure of interest, this can lead to heterogeneous and (potentially) biased estimates of causal effect. We investigate the use of Bayesian model averaging to preferentially search the space of models with the highest posterior likelihood. We develop a Metropolis-Hasting algorithm to perform the search using the recently developed MR-RAPS as the basis for defining a posterior distribution that efficiently accounts for pleiotropic and weak instrument bias. We demonstrate how our general modeling approach can be extended from a standard one-component causal model to a two-component model, which allows a large proportion of SNPs to violate the InSIDE assumption. We use Monte Carlo simulations to illustrate our methods and compare it to several related approaches. We finish by applying our approach to investigate the causal role of cholesterol on the development age-related macular degeneration.
Assuntos
Variação Genética , Análise da Randomização Mendeliana , Teorema de Bayes , Causalidade , Pleiotropia Genética , Humanos , Análise da Randomização Mendeliana/métodos , Polimorfismo de Nucleotídeo ÚnicoRESUMO
Body mass index (BMI) is a complex disease risk factor known to be influenced by genes acting via both metabolic pathways and appetite regulation. In this study, we aimed to gain insight into the phenotypic consequences of BMI-associated genetic variants, which may be mediated by their expression in different tissues. First, we harnessed meta-analyzed gene expression datasets derived from subcutaneous adipose (n = 1257) and brain (n = 1194) tissue to identify 86 and 140 loci, respectively, which provided evidence of genetic colocalization with BMI. These two sets of tissue-partitioned loci had differential effects with respect to waist-to-hip ratio, suggesting that the way they influence fat distribution might vary despite their having very similar average magnitudes of effect on BMI itself (adipose = 0.0148 and brain = 0.0149 standard deviation change in BMI per effect allele). For instance, BMI-associated variants colocalized with TBX15 expression in adipose tissue (posterior probability [PPA] = 0.97), but not when we used TBX15 expression data derived from brain tissue (PPA = 0.04) This gene putatively influences BMI via its role in skeletal development. Conversely, there were loci where BMI-associated variants provided evidence of colocalization with gene expression in brain tissue (e.g., NEGR1, PPA = 0.93), but not when we used data derived from adipose tissue, suggesting that these genes might be more likely to influence BMI via energy balance. Leveraging these tissue-partitioned variant sets through a multivariable Mendelian randomization framework provided strong evidence that the brain-tissue-derived variants are predominantly responsible for driving the genetically predicted effects of BMI on cardiovascular-disease endpoints (e.g., coronary artery disease: odds ratio = 1.05, 95% confidence interval = 1.04-1.07, p = 4.67 × 10-14). In contrast, our analyses suggested that the adipose tissue variants might predominantly be responsible for the underlying relationship between BMI and measures of cardiac function, such as left ventricular stroke volume (beta = 0.21, 95% confidence interval = 0.09-0.32, p = 6.43 × 10-4).
Assuntos
Índice de Massa Corporal , Moléculas de Adesão Celular Neuronais/genética , Doença da Artéria Coronariana/genética , Diabetes Mellitus Tipo 2/genética , Obesidade/genética , Proteínas com Domínio T/genética , Tecido Adiposo/metabolismo , Tecido Adiposo/patologia , Encéfalo/metabolismo , Encéfalo/patologia , Moléculas de Adesão Celular Neuronais/metabolismo , Doença da Artéria Coronariana/metabolismo , Doença da Artéria Coronariana/patologia , Diabetes Mellitus Tipo 2/metabolismo , Diabetes Mellitus Tipo 2/patologia , Proteínas Ligadas por GPI/genética , Proteínas Ligadas por GPI/metabolismo , Perfilação da Expressão Gênica , Regulação da Expressão Gênica , Loci Gênicos , Variação Genética , Genoma Humano , Estudo de Associação Genômica Ampla , Humanos , Análise da Randomização Mendeliana , Redes e Vias Metabólicas/genética , Obesidade/metabolismo , Obesidade/patologia , Volume Sistólico/fisiologia , Proteínas com Domínio T/metabolismo , Relação Cintura-QuadrilRESUMO
Autism Spectrum Disorder (ASD) and Attention-Deficit/Hyperactivity Disorder (ADHD) are complex co-occurring neurodevelopmental conditions. Their genetic architectures reveal striking similarities but also differences, including strong, discordant polygenic associations with educational attainment (EA). To study genetic mechanisms that present as ASD-related positive and ADHD-related negative genetic correlations with EA, we carry out multivariable regression analyses using genome-wide summary statistics (N = 10,610-766,345). Our results show that EA-related genetic variation is shared across ASD and ADHD architectures, involving identical marker alleles. However, the polygenic association profile with EA, across shared marker alleles, is discordant for ASD versus ADHD risk, indicating independent effects. At the single-variant level, our results suggest either biological pleiotropy or co-localisation of different risk variants, implicating MIR19A/19B microRNA mechanisms. At the polygenic level, they point to a polygenic form of pleiotropy that contributes to the detectable genome-wide correlation between ASD and ADHD and is consistent with effect cancellation across EA-related regions.
Assuntos
Transtorno do Deficit de Atenção com Hiperatividade/patologia , Transtorno do Espectro Autista/patologia , Transtorno do Deficit de Atenção com Hiperatividade/genética , Transtorno do Espectro Autista/genética , Feminino , Predisposição Genética para Doença/genética , Estudo de Associação Genômica Ampla , Humanos , Masculino , Polimorfismo de Nucleotídeo Único/genética , Análise de RegressãoRESUMO
Several abilities outside literacy proper are associated with reading and spelling, both phenotypically and genetically, though our knowledge of multivariate genomic covariance structures is incomplete. Here, we introduce structural models describing genetic and residual influences between traits to study multivariate links across measures of literacy, phonological awareness, oral language, and phonological working memory (PWM) in unrelated UK youth (8-13 years, N = 6453). We find that all phenotypes share a large proportion of underlying genetic variation, although especially oral language and PWM reveal substantial differences in their genetic variance composition with substantial trait-specific genetic influences. Multivariate genetic and residual trait covariance showed concordant patterns, except for marked differences between oral language and literacy/phonological awareness, where strong genetic links contrasted near-zero residual overlap. These findings suggest differences in etiological mechanisms, acting beyond a pleiotropic set of genetic variants, and implicate variation in trait modifiability even among phenotypes that have high genetic correlations.
RESUMO
Individual differences in early-life vocabulary measures are heritable and associated with subsequent reading and cognitive abilities, although the underlying mechanisms are little understood. Here, we (i) investigate the developmental genetic architecture of expressive and receptive vocabulary in early-life and (ii) assess timing of emerging genetic associations with mid-childhood verbal and non-verbal skills. We studied longitudinally assessed early-life vocabulary measures (15-38 months) and later-life verbal and non-verbal skills (7-8 years) in up to 6,524 unrelated children from the population-based Avon Longitudinal Study of Parents and Children (ALSPAC) cohort. We dissected the phenotypic variance of rank-transformed scores into genetic and residual components by fitting multivariate structural equation models to genome-wide genetic-relationship matrices. Our findings show that the genetic architecture of early-life vocabulary involves multiple distinct genetic factors. Two of these genetic factors are developmentally stable and also contribute to genetic variation in mid-childhood skills: One genetic factor emerging with expressive vocabulary at 24 months (path coefficient: 0.32(SE = 0.06)) was also related to later-life reading (path coefficient: 0.25(SE = 0.12)) and verbal intelligence (path coefficient: 0.42(SE = 0.13)), explaining up to 17.9% of the phenotypic variation. A second, independent genetic factor emerging with receptive vocabulary at 38 months (path coefficient: 0.15(SE = 0.07)), was more generally linked to verbal and non-verbal cognitive abilities in mid-childhood (reading path coefficient: 0.57(SE = 0.07); verbal intelligence path coefficient: 0.60(0.10); performance intelligence path coefficient: 0.50(SE = 0.08)), accounting for up to 36.1% of the phenotypic variation and the majority of genetic variance in these later-life traits (≥66.4%). Thus, the genetic foundations of mid-childhood reading and cognitive abilities are diverse. They involve at least two independent genetic factors that emerge at different developmental stages during early language development and may implicate differences in cognitive processes that are already detectable during toddlerhood.
Assuntos
Cognição , Inteligência/genética , Vocabulário , Criança , Pré-Escolar , Estudos de Coortes , Correlação de Dados , Feminino , Estudo de Associação Genômica Ampla , Genótipo , Humanos , Lactente , Desenvolvimento da Linguagem , Estudos Longitudinais , Masculino , Análise Multivariada , Fenótipo , Polimorfismo de Nucleotídeo Único , Leitura , SoftwareRESUMO
BACKGROUND: The heritability of language and literacy skills increases from early-childhood to adolescence. The underlying mechanisms are little understood and may involve (a) the amplification of genetic influences contributing to early language abilities, and/or (b) the emergence of novel genetic factors (innovation). Here, we investigate the developmental origins of genetic factors influencing mid-childhood/early-adolescent language and literacy. We evaluate evidence for the amplification of early-childhood genetic factors for vocabulary, in addition to genetic innovation processes. METHODS: Expressive and receptive vocabulary scores at 38 months, thirteen language- and literacy-related abilities and nonverbal cognition (7-13 years) were assessed in unrelated children from the Avon Longitudinal Study of Parents and Children (ALSPAC, Nindividuals ≤ 6,092). We investigated the multivariate genetic architecture underlying early-childhood expressive and receptive vocabulary, and each of 14 mid-childhood/early-adolescent language, literacy or cognitive skills with trivariate structural equation (Cholesky) models as captured by genome-wide genetic relationship matrices. The individual path coefficients of the resulting structural models were finally meta-analysed to evaluate evidence for overarching patterns. RESULTS: We observed little support for the emergence of novel genetic sources for language, literacy or cognitive abilities during mid-childhood or early adolescence. Instead, genetic factors of early-childhood vocabulary, especially those unique to receptive skills, were amplified and represented the majority of genetic variance underlying many of these later complex skills (≤99%). The most predictive early genetic factor accounted for 29.4%(SE = 12.9%) to 45.1%(SE = 7.6%) of the phenotypic variation in verbal intelligence and literacy skills, but also for 25.7%(SE = 6.4%) in performance intelligence, while explaining only a fraction of the phenotypic variation in receptive vocabulary (3.9%(SE = 1.8%)). CONCLUSIONS: Genetic factors contributing to many complex skills during mid-childhood and early adolescence, including literacy, verbal cognition and nonverbal cognition, originate developmentally in early-childhood and are captured by receptive vocabulary. This suggests developmental genetic stability and overarching aetiological mechanisms.
Assuntos
Idioma , Vocabulário , Adolescente , Criança , Humanos , Desenvolvimento da Linguagem , Alfabetização , Estudos LongitudinaisRESUMO
Cranial growth and development is a complex process which affects the closely related traits of head circumference (HC) and intracranial volume (ICV). The underlying genetic influences shaping these traits during the transition from childhood to adulthood are little understood, but might include both age-specific genetic factors and low-frequency genetic variation. Here, we model the developmental genetic architecture of HC, showing this is genetically stable and correlated with genetic determinants of ICV. Investigating up to 46,000 children and adults of European descent, we identify association with final HC and/or final ICV + HC at 9 novel common and low-frequency loci, illustrating that genetic variation from a wide allele frequency spectrum contributes to cranial growth. The largest effects are reported for low-frequency variants within TP53, with 0.5 cm wider heads in increaser-allele carriers versus non-carriers during mid-childhood, suggesting a previously unrecognized role of TP53 transcripts in human cranial development.
Assuntos
Alelos , Loci Gênicos , Variação Genética , RNA Mensageiro/genética , Crânio/metabolismo , Proteína Supressora de Tumor p53/genética , Adolescente , Adulto , Idoso , Idoso de 80 Anos ou mais , Cefalometria , Criança , Feminino , Regulação da Expressão Gênica no Desenvolvimento , Frequência do Gene , Genoma Humano , Humanos , Masculino , Pessoa de Meia-Idade , Crânio/anatomia & histologia , População BrancaRESUMO
Interpreting polygenic overlap between ADHD and both literacy-related and language-related impairments is challenging as genetic associations might be influenced by indirectly shared genetic factors. Here, we investigate genetic overlap between polygenic ADHD risk and multiple literacy-related and/or language-related abilities (LRAs), as assessed in UK children (N ≤ 5919), accounting for genetically predictable educational attainment (EA). Genome-wide summary statistics on clinical ADHD and years of schooling were obtained from large consortia (N ≤ 326,041). Our findings show that ADHD-polygenic scores (ADHD-PGS) were inversely associated with LRAs in ALSPAC, most consistently with reading-related abilities, and explained ≤1.6% phenotypic variation. These polygenic links were then dissected into both ADHD effects shared with and independent of EA, using multivariable regressions (MVR). Conditional on EA, polygenic ADHD risk remained associated with multiple reading and/or spelling abilities, phonemic awareness and verbal intelligence, but not listening comprehension and non-word repetition. Using conservative ADHD-instruments (P-threshold < 5 × 10-8), this corresponded, for example, to a 0.35 SD decrease in pooled reading performance per log-odds in ADHD-liability (P = 9.2 × 10-5). Using subthreshold ADHD-instruments (P-threshold < 0.0015), these effects became smaller, with a 0.03 SD decrease per log-odds in ADHD risk (P = 1.4 × 10-6), although the predictive accuracy increased. However, polygenic ADHD-effects shared with EA were of equal strength and at least equal magnitude compared to those independent of EA, for all LRAs studied, and detectable using subthreshold instruments. Thus, ADHD-related polygenic links with LRAs are to a large extent due to shared genetic effects with EA, although there is evidence for an ADHD-specific association profile, independent of EA, that primarily involves literacy-related impairments.
Assuntos
Transtorno do Deficit de Atenção com Hiperatividade/genética , Escolaridade , Idioma , Alfabetização , Herança Multifatorial , Adolescente , Criança , Feminino , Estudo de Associação Genômica Ampla , Humanos , Inteligência , Masculino , Análise Multivariada , Polimorfismo de Nucleotídeo Único , Escalas de Graduação Psiquiátrica , Análise de Regressão , Medição de Risco , Reino UnidoRESUMO
Mendelian randomisation (MR) is a method for establishing causality between a risk factor and an outcome by using genetic variants as instrumental variables. In practice, the association between individual genetic variants and the risk factor is often weak, which may lead to a lack of precision in the MR and even biased MR estimates. Usually, the most significant variant within a genetic region is selected to represent the association with the risk factor, but there is no guarantee that this variant will be causal or that it will capture all of the genetic association within the region. It may be advantageous to use extra variants selected from the same region in the MR. The problem is to decide which variants to select. Rather than selecting a specific set of variants, we investigate the use of Bayesian model averaging (BMA) to average the MR over all possible combinations of genetic variants. Our simulations demonstrate that the BMA version of MR outperforms classical estimation with many dependent variants and performs much better than an MR based on variants selected by penalised regression. In further simulations, we investigate robustness to violations in the model assumptions and demonstrate sensitivity to the inclusion of invalid instruments. The method is illustrated by applying it to an MR of the effect of body mass index on blood pressure using SNPs in the FTO gene.
Assuntos
Teorema de Bayes , Análise da Randomização Mendeliana/métodos , Causalidade , Simulação por Computador , Humanos , Desequilíbrio de Ligação/genética , Modelos Estatísticos , Polimorfismo de Nucleotídeo Único/genética , Fatores de RiscoRESUMO
One of the features that distinguishes modern humans from our extinct relatives and ancestors is a globular shape of the braincase [1-4]. As the endocranium closely mirrors the outer shape of the brain, these differences might reflect altered neural architecture [4, 5]. However, in the absence of fossil brain tissue, the underlying neuroanatomical changes as well as their genetic bases remain elusive. To better understand the biological foundations of modern human endocranial shape, we turn to our closest extinct relatives: the Neandertals. Interbreeding between modern humans and Neandertals has resulted in introgressed fragments of Neandertal DNA in the genomes of present-day non-Africans [6, 7]. Based on shape analyses of fossil skull endocasts, we derive a measure of endocranial globularity from structural MRI scans of thousands of modern humans and study the effects of introgressed fragments of Neandertal DNA on this phenotype. We find that Neandertal alleles on chromosomes 1 and 18 are associated with reduced endocranial globularity. These alleles influence expression of two nearby genes, UBR4 and PHLPP1, which are involved in neurogenesis and myelination, respectively. Our findings show how integration of fossil skull data with archaic genomics and neuroimaging can suggest developmental mechanisms that may contribute to the unique modern human endocranial shape.
Assuntos
Evolução Biológica , Hibridização Genética , Homem de Neandertal/anatomia & histologia , Crânio/anatomia & histologia , Adolescente , Adulto , Idoso , Idoso de 80 Anos ou mais , Animais , Feminino , Fósseis , Humanos , Masculino , Pessoa de Meia-Idade , Países Baixos , Fenótipo , Adulto JovemRESUMO
Mendelian randomization (MR) requires strong assumptions about the genetic instruments, of which the most difficult to justify relate to pleiotropy. In a two-sample MR, different methods of analysis are available if we are able to assume, M1 : no pleiotropy (fixed effects meta-analysis), M2 : that there may be pleiotropy but that the average pleiotropic effect is zero (random effects meta-analysis), and M3 : that the average pleiotropic effect is nonzero (MR-Egger). In the latter 2 cases, we also require that the size of the pleiotropy is independent of the size of the effect on the exposure. Selecting one of these models without good reason would run the risk of misrepresenting the evidence for causality. The most conservative strategy would be to use M3 in all analyses as this makes the weakest assumptions, but such an analysis gives much less precise estimates and so should be avoided whenever stronger assumptions are credible. We consider the situation of a two-sample design when we are unsure which of these 3 pleiotropy models is appropriate. The analysis is placed within a Bayesian framework and Bayesian model averaging is used. We demonstrate that even large samples of the scale used in genome-wide meta-analysis may be insufficient to distinguish the pleiotropy models based on the data alone. Our simulations show that Bayesian model averaging provides a reasonable trade-off between bias and precision. Bayesian model averaging is recommended whenever there is uncertainty about the nature of the pleiotropy.