Regulating genomic risk tests

The idea of gene testing is familiar to most people nowadays. Tests for disorders such as Huntington’s disease, phenylketonuria and breast cancers are readily understood as tests for mutations in  particular genes, with a known pathway from gene variant to protein to pathological phenotype. Conditions such as these, in which a clear line of causation runs from single variant to pathology are dubbed "monogenic". They’re often simple to trace in family trees and easy to comprehend in the "gene for" terms of red top newspapers. The reality is more complex, because genes do not function alone, but in a genomic/cellular environment that includes the products of other genes. Understand the dynamics of this relationship, and we could derive new diagnostic tests that are more genomic than genetic in character. How we regulate these tests, and whether current regulations are fit for the purpose of regulating them are matters of growing importance.

As so often in our work, considering such issues begins with a close understanding of the science. That’s why I recommend reading this Nature article by Michael Eisenstein before thinking about regulation. It’s a highly readable primer about using “polygenic risk scores” (PRS) to calculate heart disease risk.

A “polygenic” condition is one in which more than one gene is responsible for, or perhaps linked to, a given pathology. As knowledge accumulates, the character and severity of some disorders that are demonstrably inherited and "monogenic" may be significantly modified by the presence or otherwise of a small number of “modifying genes” in an individual's genome. Cystic fibrosis (CF), a “monogenic” condition linked to recessive mutations in the CFTR gene, is a good example. Because CFTR’s modifying genes may have been inherited from a non-CF-carrying parent or be de novo mutations, the way in which CF manifests itself over the generations can appear erratic. Testing for modifying genes as well as for CFTR therefore provides important data for prophylactic or therapeutic intervention. Other genetic conditions, having no obvious pattern of inheritance or core single variant, may arise when a specific combination of gene variants appears within the same unlucky genome. In these cases, tests for indicative combinations may provide a more valuable diagnostic biomarker than the variants per se.

We only know about the polygenic nature of conditions such as CF because of large scale studies involving diverse populations. Our ability to conduct such investigations, so as to identify variants with statistically significant phenotype associations, has increased with the speed, accuracy and affordability of whole genome sequencing. It’s lead to a huge and growing mountain of genomic data. For the present, however, our ability to sequence genomes far outpaces our capacity to analyse this data in order to understand their dynamics and function in health and disease. Pathology-linked variants may be identified by comparative sequencing, but without gaining any understanding about their role, if any, in the disease phenotype.

Using data to infer a patient's pathological risk without understanding causation may contrast starkly with the explanatory foundations of orthodox genetic testing, but the practice is hardly new to medicine. Further, the lack of sufficient brainpower to scan the burgeoning data mountain for variant-pathology correlations and to suggest avenues for empirical research may be compensated for by the use of AI and machine learning systems, which can progressively improve PGS as larger and larger databases are interrogated by ever-more sophisticated algorithms. The potential is obvious, but the regulatory framework is not designed to cope with test devices that change themselves in the light of their own discoveries. There is also a danger of inadvertent off-label use, by scoring patients using risk data from unrepresentative populations (what makes a population representative?), or of attributing greater weight to PRS than to other diagnostic criteria of potentially greater significance, such as age, family history, environmental factors and medication.

Will polygenic risk scores catch on?  There is certainly some division of clinical opinion (well discussed in this excellent PHG Foundation paper), but the likelihood is that systems that generate them are here to stay, invoking new regulatory challenges. We’re monitoring these closely, and will discuss them in future posts.