Personalised medicine: prediction is paramount
Successful implementation of personalised medicine requires accurate predictive tests to help physicians select the optimum treatment for a patient. This article discusses how genomic technologies will impact this predictive accuracy.
It may seem self-evident, but it is still worth spelling out: the key to the successful implementation of personalised medicine is the predictive power of the measurements used to individualise treatments. There will be valid concerns about other aspects of personalised medicine, such as: cost-effectiveness, availability of healthcare infrastructure and information delivery, physician training, inequality of access to healthcare between different socioeconomic groups. However, these are irrelevant unless the tests or other measurements provide useful predictive information that can impact therapeutic decision-making.
“…the key to the successful implementation of personalised medicine is the predictive power of the measurements used to individualise treatments.”
Limited predictive power has been a major barrier to the adoption of a number of personalised medicine approaches. Generating the clinical data to support a predictive test can be long and expensive, and promising tests often fail to show the same impact in different ethnic groups. This article looks at how emerging scientific understanding and technologies will impact – and hopefully improve – the predictive nature of tests in future.
There are an increasing number of DNA tests being implemented, although almost all are in oncology, where mutations with strong genetic effect have been detected. Outside oncology, DNA markers show much more limited predictive power. One issue is that in these more complex settings, several (or many) gene variants combine to affect the overall risk. To date, genetic technologies have struggled to identify these gene variants which exhibit weak effects. This is seen most starkly in disease genetics genome-wide association studies (GWAS) where the predictive power of the identified genes represents only a small fraction of the disease genetic burden calculated from family and twin studies.
There are a number of hypotheses to explain this ‘missing heritability’, including rare variants of strong effect, and gene-gene interactions. A major driver behind the current explosion in DNA sequencing is the assumption that rare variants will account for much of this missing heritability. This assumption has yet to be fully tested, although a recent publication from Morris et al. in Nature Genetics suggests that this might not be the case, at least for type 2 diabetes. Nevertheless, until the genetic technologies can account for most of the genetic burden in a disease or drug response phenotype, DNA tests may struggle to be sufficiently predictive outside oncology.
“…DNA tests may struggle to be sufficiently predictive outside oncology.”
The growing area of epigenetics may provide another way of addressing the problem, as our understanding grows of the modifications to DNA that occur in the nucleus. These approaches however suffer from some of the drawbacks of RNA-based tests discussed below, in that the modifications are dynamic (so sampling time is critical) and cell-specific (so appropriate tissue samples may not be readily accessible).
RNA-based tests have lagged behind other types of analysis in clinical use, for a number of practical reasons. Firstly, RNA samples are more difficult to process, although modern approaches stabilise RNA species remarkably effectively. Secondly, RNA species show at least as great a diurnal variation as proteins, thus there are more constraints on the timing of patient sampling. Finally, only a limited number of cell types in the body can be examined for most disease types (oncology is an exception here), and not every disease will leave an RNA signature in blood-born leukocytes.
The recent research on non-coding RNA (ncRNA), such as miRNA, provides additional opportunities for RNA-based tests. These ncRNAs have critical regulatory activities and as such are prime candidates for the dysregulation that underpins disease states. Our understanding of these molecules is advancing rapidly, although the potential for predictive testing is not established.
From a practical perspective, protein-based tests are still the easiest to implement: the technology is well established, a significant proportion of proteins are secreted – or released – from cells so that they can be measured in readily accessible tissue fluids, and healthcare professionals are experienced in interpreting the results. However, as discussed above, there is an extraordinary amount of biological regulation that is not dependent on proteins. This will inevitably put constraints on the predictive power that can be delivered by protein-based tests: the metabolic dysregulation that underlies diseases may not be detectable solely in the proteome.
The above sections show that even with the anticipated developments in technology and understanding, there are constraints on the predictive power available from the different testing paradigms. This will be less of an issue in oncology, where tissue is more readily accessible, and gene variants may have stronger genetic effects than is usual in other diseases. In these other diseases, it is perhaps inevitable that the necessary predictive power will only be possible when the different technologies are combined, similar to the way that family history, blood biochemistry and blood pressure is used to estimate the risk of cardiovascular disease.
“… such a combined approach to biomarkers will place more emphasis on healthcare infrastructure…”
However, combining these different approaches provides a number of challenges. Firstly, there are few large patient datasets where all the measurements are collected to permit the necessary clinical research. So it will be important for biobanks and other large patient-based studies to ensure that a wide range of samples (with appropriate consent) are available for researchers. Secondly, most methods to co-analyse different types of data are not optimal for integrating the different measurements: development of additional analytical methods that can combine different data types is required. Finally, such a combined approach to biomarkers will place more emphasis on healthcare infrastructure, and training of healthcare professionals to ensure that this more predictive information is available to physicians making treatment decisions.
About the author:
Alun McCarthy is President and CEO of PGXIS Ltd, and Chief Scientific officer of CytoPathfinder Inc. PGXIS is a UK-based company which has developed innovative technology to analyse large, complex genomic datasets. This approach – called Taxonomy3 – has a number of benefits over conventional analysis methods. These include: more statistical power to find associations, the ability to identify gene-gene interactions, greater predictive power, and integration of different data types in a single multivariate analysis.
PGXIS Ltd, Aston Court, Kingsmead Business Park, Frederick Place, High Wycombe, Bucks, HP11 1LA, UK
Tel: +44 (0) 1494 616035
Is prediction the key to personalised medicine?