A long list of genetic variants with abstruse names (rs9712235_A, rs4602255_ T…). Figures and percentages associated, incomprehensible. An explanation, abstract: “By adding the effect of the highlighted variants, we calculated your polygenic score for luminal B/HER2 breast cancer”. And a brutal conclusion: “You are in the 98th percentile. This is a very high genetic predisposition”. “In reality, these data made no sense, but the young woman who had received this result after buying a DNA test on the internet was panicked, and had asked us for an emergency appointment”, sighs Antoine de Pauw, adviser in oncogenetics at the Institut Curie, in Paris.
A great classic: geneticists and doctors regularly receive this kind of worried call. Even if predictive genetic tests remain prohibited in France, the temptation can indeed sometimes be great to carry them out anyway, by sending a little saliva to a company based across the Atlantic, where they are authorized. Despite the risks – 3,750 euros fine. Despite, above all, their lack of reliability. Because even if the promise is great – predicting risks to better prevent them – no health authority has yet validated this technology for medical use. And it’s not for nothing.
“These tests still have many biases and limitations, the interpretation of their results remains difficult, which makes their use premature for individual predictions”, assures David-Alexandre Trégouët, researcher in genetic epidemiology. In recent years, however, scientists have not stood idly by. While statisticians and geneticists continue to improve their models, doctors have launched clinical trials to validate them and understand how best to use them. It will take a few more years to see more clearly. But a country like the United Kingdom, at the forefront in this field, is beginning to consider the possibility that doctors can use it in certain specialties.
The genomes of hundreds of thousands of patients analyzed
Of course, these are not analyzes intended to look for rare pathological mutations, which undoubtedly lead to a disease, as in cystic fibrosis or myopathies. In the event of suggestive symptoms or a family history, patients can of course already benefit from DNA sequencing. “Polygenic scores”, which aim to predict an individual’s risk of developing one or other of the most frequent diseases (non-hereditary cancers, Alzheimer’s, diabetes, cardiovascular pathologies, etc.), come under a completely different logic.
To construct these tests, the scientists first crossed, pathology by pathology, the genomes of thousands, or sometimes hundreds of thousands of patients and uninjured subjects, in order to find genetic markers specific to the patients. These pan-genomic association studies, or GWAS (pronounced “ji-ouase”, for “genome-wide association studies”), were initially intended to better understand our biology and the causes of these multifactorial diseases. “The variants point to genes or sets of genes that must then be analyzed to see if they are really involved in the pathology or in the trait studied”, explains researcher Jean-Charles Lambert, who works at the Pasteur Institute of Lille on Alzheimer’s disease. Because an association does not necessarily mean that there is a causal link: this must be demonstrated.
But in 2007, an Australian researcher, Peter Visscher, had the idea of using a mathematical model invented at the beginning of the 19th century to calculate, from these data, the risks of individuals or groups of individuals to develop this or that pathology. It is basically a matter of adding up the variants found in a person, then looking at how it is compared to the average of the population studied in the gwas. If her genetic profile is close to that of sick people, she will be considered at greater risk of falling ill.
The complex functioning of the genome
Almost fifteen years later, this model continues to arouse many criticisms. “It assumes, among other things, that each genetic variant has only a small effect, and that genes do not interact with each other or with the environment, and that parents only pass on genetic factors to their children. However, these hypotheses are in total contradiction with the biological knowledge accumulated for more than a century”, insists Françoise Clerget-Darpoux, specialist in statistical and human genetics. In fact, the functioning of our genome turns out to be much more complicated than a simple addition of markers. “To give you an idea of the complexity, it may turn out that depending on whether a variant is transmitted by the father or by the mother, it will not have the same effect. Current models do not know how to take this into account. “, details David-Alexandre Trégouet. And this is just one example among many…
This model also ignores the effects of the environment (in the very broad sense: diet, lifestyle, exposure to pollutants, etc.) on our biology. “Each pathology has a genetic component and a more or less strong environmental component. The more the environment plays an important role in the disease, the less the polygenic score will give relevant results”, emphasizes Jean-Charles Lambert. There may also be confounding factors, which geneticists are still struggling to eliminate completely. “In France, preferring salted butter to sweet butter is associated with numerous genetic variants without signifying that genetic factors play a role in this eating habit. Similarly, breast cancer is more frequent in women with a high body mass index and certain associations found for this cancer may simply reflect differences in the way of eating”, notes Françoise Clerget-Darpoux.
Furthermore, these tests will give even less reliable results if they are applied in populations that are too different from those in which the Gwas were carried out. “It is also necessary that the disease or the trait studied be well defined”, underlines Jean-François Deleuze, director general of the National Center for Research in Human Genomics. “Studies concerning pathologies such as schizophrenia or depression are, for example, highly discussed in the scientific community, because the patients present a very strong heterogeneity, with probably very varied origins to their illness”, continues this expert. Under these conditions, it is not certain that the markers found by the association studies are relevant.
An imperfect picture of reality
“Because of the many biases, the risk scores give us an image, but this image is only an imperfect reflection of reality”, summarizes Jean-Charles Lambert. As a result of the falling price of DNA sequencing, using this tool to predict risks has nevertheless proved very tempting for many researchers. With variable results depending on the pathologies. At the Institut Pasteur, Professor Thomas Bourgeron, who was the first to discover mutated genes explaining certain forms of autism, uses it to try to better understand the role of genetics in all of these disorders. “At first, I was skeptical. Then I saw in several cohorts of well-characterized autistic patients, in France, in Europe and in the United States, that there were many statistical differences between the genetic variations of people with autism. and control subjects, which are reflected in the polygenic scores”, explains the researcher, author of the book Genes, synapses, autisms published by Odile Jacob. However, it is not yet possible, from the genome of an individual, to say whether he is autistic or not: “We are only at the beginning of this work”, he underlines.
In Alzheimer’s disease, the predictions turn out to be somewhat better. “We took the 83 genetic markers most associated with this pathology, and we looked at whether they made it possible to differentiate between individuals who had a greater or lesser risk of developing the pathology,” says Jean-Charles Lambert. In fact, in a population of people who already had mild cognitive impairment, those who emerged as the most at risk had a 40% “chance” of developing the disease within three years, compared to only 20% for those classified as the least at risk. risk. “It gives information, but for a given subject, the estimate is not precise enough and depends on many other factors. Two people may have the same score, but one will actually get sick and the other will not”, notes the doctor.
Numerous clinical trials
In cardiovascular diseases, on the other hand, the United Kingdom is already considering offering GPs the use of polygenic scores in their routine practice. The NHS, the British health system, evaluated on 1,000 people aged 45 to 64 the addition of this calculation to the usual indicators (weight, tobacco, history, blood pressure). “In view of the results, the doctors modified the care of 13% of their patients, by offering them statins for prevention”, indicates the company. Genomics, which deployed the test. But this perspective remains highly debated in the country: “Polygenic scores in themselves will always have limited predictive power, because individual risk is determined by factors that these scores do not measure”, experts in epidemiological genetics immediately warned. in the british medical journal.
Many studies are nevertheless underway around the world, particularly in the field of cancer. Some concern women carrying mutations in the BRCA1 and BRCA2 genes which, like star Angelina Jolie, expose them to very high risks of breast tumors. “Even for this population, we would like to be able to specify their situation thanks to polygenic scores, to personalize their care even more. But as long as the results of the studies are not available, there is no question of using them in practice. running,” explains Antoine de Pauw, at the Institut Curie.
Same analysis at Gustave Roussy, where oncologists have launched a large study (MyPebs) in the general population, to assess whether a polygenic score could improve the effectiveness of breast cancer screening. “The objective is to offer the women most at risk the opportunity to have an annual mammogram and not every two years like today, and to offer the least at risk a slightly lighter rhythm”, explains Dr Suzette Delaloge, behind this project. But there is no question, here either, of using a single indicator: the genetic score will be combined with family and personal history, age, hormonal exposure, or even breast density. “Today, no learned society considers that there is enough evidence to deploy it routinely. But within five years, we will have the first results of prospective studies like MyPebs, and we may then have enough data to make it interesting,” hopes the researcher. No need, therefore, to rush to send a little saliva across the Atlantic.