Bringing researchers from different areas together is vital in sharing expertise and generating opportunities for new collaborations and connections. Last week members of the Integrative Cancer Epidemiology Programme (ICEP) joined colleagues specialising in Mendelian randomization (MR) within the MRC Integrative Epidemiology Unit (IEU), for a day of discussion, networking and sharing their latest research.
Here, Senior Research Fellow Philip Haycock, who leads the Cancer Progression and Drug Target research theme in ICEP, summarises some the day’s discussions and overlapping research.
Discovering mutual research interests
The first ICEP – IEU programme 1 away day was held on 3 July, to explore opportunities to collaborate, to increase mutual awareness of our research and to meet people.
In the morning session, a series of lightning talks highlighted exciting areas of overlapping research interest across the two groups. In the afternoon, presentations on cancer biology (Emma Vincent), age of infection (George Davey Smith), the cancer data landscape for ICEP3 (Emma Hazelwood) and pathways to impact (Richard Martin) generated interesting discussions and new project ideas.
We learnt a lot of new, sometimes counterintuitive, things, e.g. that ADH1B does not make a good instrument for alcohol intake (Jasmine Khouja), that DNA-methylation is a better measure of usual CRP levels than a one-off CRP measurement (Paul Yousefi) and that Otto Warburg (discoverer of the “Warburg effect”) was a bit of a narcissist (Emma Vincent).
It also turns out that we should expect genetic effects to be similar across populations of different ancestries; deviations from this expectation will likely reflect differences in linkage disequilibrium or gene-environment interactions (Gibran Hemani).
Lightning talks also generated ideas for new projects, for example using MR to study causes of non-neoplastic co-morbidities in individuals with cancer (Sarah Lewis).
Developing ideas for collaborative projects across prediction, detection, risks, and prevention
The late afternoon was dominated by breakout sessions, to come up with ideas for collaborative projects. The pre-diagnostic breakout group discussed using “de-Mendelization” to identify predictive biomarkers (e.g. removing the genetic component of PSA to improve its predictive ability).
The post-diagnostic breakout group discussed using sequencing data in UK Biobank to identify somatic mutations (either cell-free, reflecting non-blood tissues, or white-cell based), and assessing correlations with time-to-cancer diagnosis or interactions with genetically proxied exposures to identify cancer promoters. This led to discussions about studying cell-free DNA in the context of non-invasive pregnancy testing (Michel Nivard).
Another idea was to conduct deep phenotyping of cancer patients after receiving a physical activity intervention but before receiving chemotherapy. This could be a way to identify mediators of the effect of physical activity on cancer survival.
It was noted by the early detection breakout session that passive detection methods (e.g. sensors embedded in toothbrushes, loo role, exhalation, etc.) are more popular than talking to doctors; and that non-causal biomarkers are better for early detection than causal biomarkers.
The role of environmental and infectious exposures on cancer incidence
One of the highlights for me was George Davey Smith’s talk on the role of ubiquitous exposures in cancer incidence. These are exposures that everyone in a population experiences, for example infectious agents, anthropogenic lighting and obesogenic environments.
It’s an important issue because we know that 75 to 80% of cancers are preventable in principle, but only 1/3 of the causes are known. The difficulty in identifying the remaining 2/3 could be down to their ubiquitous nature, which are not identifiable using traditional epidemiological studies.
Ubiquitous exposures could also explain cohort effects, such as the increasing incidence of obesity-related cancers in the young. Perhaps the most fascinating ubiquitous exposures are those involving infectious agents, such as Epstein-Barr Virus (EBV). More than 90% of us will be exposed to EBV over our lifetimes, although with substantial variation in exposure timing. Interestingly, exposure timing strongly affects its health impact. For example, EBV exposure after age 20 (but not before) increases the risk of multiple sclerosis.
Strategies for identifying risks
Considerable discussion was generated on strategies to identify ubiquitous exposures. A promising approach could be to study the genetics of exposure biomarkers.
For example, in a hypothetical population where everyone smokes 10 cigarettes a day for 20 years, the only source of variation in lung cancer risk would be genetic. Imagine that this is explained by the CHRNA5 gene, through an effect on depth of smoke inhalation. One could identify that smoking is the causal exposure through a study of cotinine – a biomarker of nicotine metabolism.
This thought experiment also points to cross-population comparisons as a strategy to identify ubiquitous exposures. For example, the magnitude of the CHRNA5 genetic effect on lung cancer risk would be expected to vary substantially across populations with different rates of smoking, being close to 0 in populations with little-or-no smoking and increasing in size with increasing smoking prevalence.
In a parallel universe where we didn’t know smoking was a cause of lung cancer, we could compare the effect of CHRNA5 on lung cancer risk amongst populations where lung cancer rates varied substantially. We would then look for population-level exposures that correlate with the population-level CHRNA5 effect size differences, which would presumably identify a very strong correlation with smoking rates.
However, being an ecological study, other population-level characteristics that correlate with smoking would also likely be identified, which would undermine causal interpretations.
Nevertheless, the identified candidate factors could be further investigated for causality through MR of exposure biomarkers. This discussion linked up very nicely with Gibran Hemani’s talk on multi-ancestry MR from earlier in the day, which highlighted the value of comparing effect sizes across diverse populations to identify gene-environment interactions.
About the programmes: Within the IEU, several methods have been developed for MR – a system of using genetic variants associated with specific modifiable exposures to explore the causal effects of these potential risk factors on health (and other) outcomes. ICEP focuses on cancers that are common, present late or have poor survival rates, including bowel (colorectum), brain (glioma), breast, head and neck, kidney, lung, ovarian, pancreatic and prostate cancers. Using statistical methods and the genetic data of many thousands of individuals, ICEP research provides evidence of the causes of cancer, factors influencing cancer progression, ways to predict who will develop cancer, and ways to prevent progression.