What can the exposome tell us about ageing and mortality?

A recent article in Nature Medicine has created quite a stir. The article takes a very broad approach in developing an understanding of ageing and ill-health (see BioNews 1278).

The study is huge in every sense. It involves a large dataset, using data from almost half a million participants in the UK Biobank, data on 164 different environmental exposures (using 'exposure' in the broad epidemiologists' sense, from smoking and intake of various foods, to how plump they were at age ten, to their ethnicity) and (for some participants) genetic and blood measures too. The researchers related these measures to records of deaths of participants, and diagnoses of several serious diseases.

It's big data, and the researchers use some big-data methods. These include splitting the data into two parts, fitting a statistical model to one part and then using the other part to validate the findings from the fitted model. That's good practice.

The aim was to quantify the contributions of environmental exposures and genetics to ageing and premature death, considering many aspects of people's environment rather than concentrating on a few risk factors decided in advance. And, broadly, the researchers conclude that environmental exposures can tell us more about population levels of mortality and age-related ill health than can genetic factors.

The results are interesting, and I think they do support the researchers' view that we can learn more by looking at many environmental exposures together rather than trying to pick them off one at a time.

It would be easy to dismiss this new research by saying that all they have really found is that, if you want to be healthy in old age, you need to give up smoking, do some exercise and not be poor, and we already knew that. But that's not the important finding at all. The researchers conclude that you get a broader understanding by looking simultaneously at many more aspects of the environment, if you have enough good data.

Like all observational studies, the findings are about correlations and associations, not about cause and effect. The statistical methods used by the researchers can't determine whether the associations between exposures and ill health and mortality that they observed are there because the exposures cause the ill health and mortality. They might be, or they might not.

The researchers filtered out exposures that might have showed up as associated with ill health only because they were correlated with other exposures, or because the exposure was actually caused by ill health (reverse causation, as it's called). They also used a biological ageing clock based on blood proteins in choosing the set of exposures to consider. All this does make it rather more likely that the associations they mainly report on are ones of cause and effect. But no study of this kind can confirm that the associations are definitely demonstrating cause and effect.

The researchers conclude that interventions based on environmental exposures are possibly the best starting point for improving age-related health. But they add that 'future causal modelling [research methodology that specifically looks at cause and effect] will be needed to study specific exposures of interest.'

Entirely unsurprisingly, the main factors associated with variation in overall mortality between individuals are their age and biological sex, though their smoking history is very important too. The paper bases its main conclusions on additional mortality variation, in addition to the variation explained by age and biological sex.

Age and biological sex explain (statistically) almost half the variability in mortality from all causes. The extra variation explained by also including 25 measures of environmental exposure (including smoking) was about 19 percentage points, so that age, sex and environment explained 66 percent of variability in all. As these things go, that's a large percentage. But including data on polygenic risk scores for 22 major diseases in the model, instead of the environmental exposures, increased the percentage of variation explained by only two percentage points.

The researchers emphasise the importance of looking at a wide range of exposures – the exposome, as they call it – by pointing out that their statistical model included no less than 25 exposures, even after filtering out exposures that were less likely to reflect cause and effect. Though a few of the 25 exposures, particularly smoking, had strong correlations with mortality on their own, most of them were much less important taken one at a time. But they do, together, add up to the strong association between the exposome and ageing and death.

There is no implication that the 25 independent environmental factors that were identified in this research are the most important environmental factors, or the only important ones. The filtering process may have removed some exposures that were in fact important to health. (I'm not saying that they should not have been removed, in the light of the overall aims of this study – just that removing them could have led to something being missed.)

In addition, the polygenic risk scores used in these analyses are not the only possible measures of genetic variation. Using different estimates of associations with environmental exposures, genetic factors, or both would have given different numerical results. But the difference between the measures of extra variability explained by the environment and explained by the genetics seems to me to be so great that it would not change its direction if other measures had been used.

The paper also presents the same type of finding for 25 major diseases. For many of them, the environmental exposures again explained considerably more of the variation in disease prevalence and incidence than did the genetics, though there were other diseases (including some important cancers, and dementias) for which the role of the genetic measures was statistically more important than environmental exposures.

But what should be done about environmental causes of ill health? Public health can't operate by dealing with all exposures at once – so despite the authors' pleas to look right across the exposome, we have to come down at some point to considering individual factors.

The researchers state that 23 of the 25 independent environmental exposures, identified as contributing to the association between environmental exposure and ill health, are 'potentially modifiable'. What does that actually mean?

Smoking is modifiable, even if it can be hard for individuals to make that modification. How do you modify things so that you are living with a partner, if you currently aren't? How do you modify how often you feel fed up, or how often you feel unenthusiastic? It's important to understand that some modifications would be possible only by changes in society – it's not just a question of individuals choosing what to do.

This research, then, is likely to be valuable in giving a broad overview and suggesting the most fruitful directions for future research, but a lot more research must be done before it can lead to practical public health advice.

---

The work of UK BioBank – and also the work of Genomics England and Our Future Health – will be discussed at the free-to-attend online event Our Future Health, UK BioBank, Genomics England: Exploring the Impact, taking place on Wednesday 30 April 2025.

Find out more and register here.