On the use of genetic information in causal inference for human complex diseases: a collection of studies on cancers

Abstract

The introduction chapter (Chapter 1) starts with a summary of recent progress in genetic epidemiology. The advancement of genotyping technology made it feasible to revisit a 30 years old study design known as Mendelian randomization (MR) where germline genetic variants can be used as natural instruments to assess causality between human complex traits. Here the rationale and technical assumptions required to conduct feasible MR studies were explained. The overall thesis objective and study approaches used in the thesis were also outlined. The final subchapter provides details on the definition of disease phenotypes that will be used throughout the thesis, including the creation of a “pan-cancer” phenotype, a combined cancer outcome of whether a person is diagnosed with any cancer, which was subsequently used in Chapter 4 and 5.            In Chapters 2 and 3, I investigated whether MR can be used to infer causality between modifiable risk factors and specific cancers. For Chapter 2, I applied MR to evaluate the link between vitamin D and coffee intake on epithelial ovarian carcinomas (EOC). This was done in a two-sample MR framework where genetic instruments for the risk factor were obtained from public GWAS literature and the cancer GWAS statistics were obtained from the Ovarian Cancer Association Consortium (OCAC). While higher genetically predicted vitamin D reduced the risk of EOC, there was no evidence to support a link between genetically predicted coffee intake and EOC. For Chapter 3, I contrasted and compared the association between alcohol intake with breast and ovarian cancer using both MR and observational analyses. Genetic summary statistics were obtained from the OCAC and BCAC (breast cancer). Previous observational findings show an adverse relationship between alcohol intake and breast cancer susceptibility, however the association was protective on EOC. Our observational data replicate previous findings, with MR estimates showing consistent direction of effect for EOC; while no evidence to support a strong link between alcohol intake and breast cancer risk. Taken altogether, the effect of alcohol on these cancers is likely small if causative at all.            In Chapter 4 and 5, I performed MR to evaluate a series of modifiable risk factors on overall cancer outcomes. Using data from the UK Biobank, we constructed an aggregate phenotype allowing us to evaluate whether “modifying specific behaviour (or risk factor) will alter an individual’s risk of being diagnosed or dying from cancer”. The proof-of-principle MR study using height is shown in Chapter 4, where MR findings are highly concordant with observational findings that taller people have increased risk of developing cancer. Findings for obesity, vitamin D and coffee intake on overall cancer outcomes are further elaborated in Chapter 5.            In Chapter 5, the approach undertaken to evaluate these risk factors and cancer were each described in turn, followed by a subchapter to summarise these MR findings. Genetically higher BMI is associated with an increase of being diagnosed with any cancer, with a much larger effect observed for cancer deaths. There was no clear evidence via MR that vitamin D or coffee intake was associated with overall cancer risk/mortality. At the end of the chapter, I also presented a preliminary snapshot of findings for other putative risk factors that was evaluated but were not included in the thesis.            For Chapter 6, I explore the possibilities of using MR to infer causal factors on putative cancer risk factors like coffee. There is currently an abundance level of MR findings revolving the relationship between coffee and complex diseases; yet little is known on the causal factors that may explain coffee drinking behaviour. Here I explored whether differences in taste perception influence the consumption of coffee, tea and alcohol. The analysis revealed how the consumption of these beverages might be causally influenced by perception of several bitterness compounds. Although these findings have useful implications in addiction and nutritional epidemiology, the primary motivation of this chapter is to evaluate common modifiable risk factors as a “consequence” of changes in some preceding exposure itself.            Chapter 7 is a stand-alone chapter discussing some of the recent advancements in MR methodologies. Here I contrast two of the more recent approaches (MR-PRESSO and GCTA-GSMR) on assessment for horizontal pleiotropy – a source of bias for MR findings. The benefits and challenges for both methods were discussed, with several guidelines proposed on designing MR sensitivity analyses based on the underlying genetic architecture of the risk factor of interests. The commentary summarises a key aspect of utilizing biological insights in quantifying pleiotropy in MR when statistical approaches are not feasible.            In the final chapter, I highlighted some of the common pitfalls of drawing causality from MR. Inconsistencies of study findings was also discussed. The potential avenue to apply MR to various other disease was also briefly touched upon. Finally, I provide insights on some future direction of this field, relating to some of our ongoing work, to expand the methodology and application of MR as a promising alternative in disease epidemiology.