Читать книгу Knowing One's Medical Fate in Advance - Группа авторов - Страница 21

Andreas Papassotiropoulos

Division of Molecular Psychology, Life Sciences Training Facility, Biozentrum, University of Basel, Basel, Switzerland

When asked about genetics, many people think it is about prediction at the individual level, a kind of pre-determination, or in other words: ‘knowing’ for certain what is going to happen. The reality about human genetics is actually quite different. I will explain what genetics can do in the context of my discipline, what it cannot do, and perhaps, what it should not do. In my lecture I will focus on the science of molecular psychology, specifically on human memory.

I often am asked what psychology has to do with molecules. In fact, the right question would be: how is it possible that one could think psychology does not have anything to do with molecules? The organ which produces our emotions, thoughts, and cognition - is the brain. Of course, the brain and its functions are related to molecules.

I am not the only one to claim this. The giants have already thought of the idea of molecular psychology. For example, in 1859, Charles Darwin wrote in his On the Origin of the Species:

“In the distant future I see open fields for far more important researches. Psychology will be based on a new foundation, that of the necessary acquirement of each mental power and capacity by gradation.”

Implying that psychology would be a far more important research than others, Charles Darwin imagined in 1859 that someday human genetics will contribute to the understanding of human mental power and capacity.

I address 5 important points about my view of human genetics, while keeping it as simple as possible:

1 Genetics of complex human traits and phenotypes is a tool to understand biology.

2 Genome-wide association studies revolutionize our knowledge on complex traits relevant to neuropsychiatry.

3 Genetic clusters rather than single genes are potential biomarkers.

4 The use of human genetic information will lead to improved characterization of complex human traits.

5 The combination of genetics with other relevant sources of information such as functional brain imaging (fMRI) will increase biological knowledge.

In neuroscience, we mostly study diseases or physiological traits that are complex. We have to understand and appreciate the need for multifactorial models: several genetic factors apply, but there is also the environment, comorbidity (the presence or absence of additional diseases), personality, sociodemographic factors, lifestyle, life events, and medication. Thus, there are several factors, each with a certain weight, which contributes to the development of a certain trait or disease.

What do we mean when talking about ‘personalized medicine’? What do we want to do with ‘personalized medicine’ and how do we want to use human genetic information? Is it about prediction, whether one will develop a disease or not? Is it about diagnosis? Or is it pharmacogenomics? Irrespective of these issues, we have to realize this: human genetic information is related to probability. Furthermore, the way we want to use human genetic information does not only depend on probability but also on the consequences that this probability would have.

Let me give you an example related to Alzheimer’s disease. This complex disease at its end stage is marked by severe atrophy of the hippocampus, a brain region related to memory. We believe that specific molecular events lead to the erroneous degradation of the amyloid precursor protein. In order to escape or break free from the metabolic pathway(s) leading to this disease, there are possibly several steps or stages, all of which are modified by genes. In other words, genes contribute to the direction that this metabolic cascade may take. Only if we could understand the nature and function of all contributing genes, we could know who is likely to develop this disease.

Some very rare cases of Alzheimer’s disease are monogenic. For example, one of the first cases I saw when I was in Zürich was of a 32-year-old patient with full-blown Alzheimer’s. From his family pedigree, it looked like a Mendelian segregation, specifically an autosomal dominant mode of inheritance. It is known that these rare cases are related to mutations in at least 3 genes: APP, PS-1, and PS-2. In his family, he had the PS-1 mutation.

If an individual has this PS-1 mutation, the probability of developing the disease is almost 100%. So, in this case, it might be important to know whether one carries the mutation or not. However, these cases are very rare.

What we see in the general population is sporadic Alzheimer’s disease, which is much more common. The heritability of sporadic Alzheimer’s disease is 75%, i.e., the contribution of the genome to the development of the disease is 75%. Please understand that this does not imply that a first degree relative will have a 75% probability of developing the disease. Heritability is a population-based value, and it cannot be used at an individual basis. There are several genes, and other non-genetic factors, which are related to the risk for developing sporadic Alzheimer’s disease.

What do we know today? Not much more, but back in 1993 Alan Roses’s group published a very nice paper [1]. In this study, they identified a variant of a gene called APOE. This gene has 3 alleles: it is present in 3 variants. Possession of the variant ε4, is related to an increased risk for Alzheimer’s disease. What does this mean?

About 25% of a healthy control population carries the ε4 variant without suffering from Alzheimer’s disease. However, in a group of AD patients almost 70% are APOEε4 carriers. Which means if you are a heterozygous carrier of ε4 variant, if you have one copy, or two copies (homozygous), you have a three-fold and eight-fold, respectively, increased risk of developing Alzheimer’s disease. This finding has been broadly replicated and, therefore, it is considered stable.

However, if you look more closely at it - the sensitivity and specificity of this variant is only about 60%. This means there is absolutely no positive predictive validity when using this variant as a predictive factor. I remember when I was doing my residency in Bonn: there was this new lab sheet for clinical routine examination with a check box: ‘APOEε4, yes or no?’ I wondered why was that there? When I called up the lab the technician there told me it was for knowing whether the patient has Alzheimer’s or not. However, this is absolutely the wrong implication of a ε4 genetic testing result. Indeed, the first Alzheimer’s patient was not a ε4 carrier!

Remember, there might be a value in doing predictive genetics, but we need to understand how genetic information may be used and interpreted. Please keep also in mind, it is not only an issue of positive predictive values or sensitivity or specificity, there are other issues.

For instance, let us look at the interaction of the ε4 variant with the genetic background or ethnicity. If you are an ε4 carrier and you carry two copies and if you are of European ancestry: the odds ratio associated with this is 8, meaning you have about an 8-fold increased risk of developing the disease. If you are of sub-Saharan ancestry, you do not have any increased risk at all. If you are Japanese, and you are homozygous for ε4, you have a 33-fold increased risk of developing Alzheimer’s disease. This means for a Japanese individual it might be important to know, and it is actually very informative to know, whether he/she is homozygous for ε4, whereas for a European it is less important, and for a sub-Saharan individual, it is absolutely irrelevant.

When talking about complex genetics, I truly cannot overstate the importance of the phenotype. Remember, complex human genetics is also a statistical discipline; one can correlate anything with anything. One could correlate genes with religious feelings, for example. In my opinion, the so-called God gene is absolutely meaningless. Scientifically speaking, it is simply correlating genetic variation with something, which we don’t know what it is! This is not serious science.

Let us go back to understand why the choice of phenotype is important. I would like to stress that the value of genetics is realized only when it is combined with a suitable phenotype or trait.

What are the criteria for a suitable phenotype or trait? The trait should be heritable, be reliably assessable, and it should have a neuronal correlate (at least, in neuroscience) or a biological correlate. Human memory - the ability to remember episodes of our lives - is such a suitable biological trait. It is heritable, it can be reliably assessed and it does have a neuronal correlate.

Why do complex genetics on memory? Sometimes, we want to just know and understand the biology of this complex trait. If you want to understand in humans more about the biology of memory, not in model organisms, you can choose between two possible approaches. (1) The classical pharmacological approach: you think that a certain receptor or gene is important in memory, the first thing you have to do as a pharmacologist, you have to find a compound that will bind with the molecule that you are interested in, and then, to study humans to see whether they react and how they react when taking this drug, if their memory becomes better or not. This approach has its caveats (e.g. availability, safety, specificity). (2) The genetic approach: again, it might be hypothesis-driven for a certain receptor or gene related to human memory. Then, you scan in the human genome and look for genetic variants of this molecule. Some individuals would have a specific variant of the molecule, the other individuals have other variants, etc. You then can stratify it according to the groups with the different variants. If they significantly differ regarding the trait of interest (in this example, memory performance), then it is most probably due to the stratification that you did, i.e. it is due to the genotype. It is important to remember that these are group comparisons, not individual predictions.

If you study a group of students and measure their memory capacity, you will find a physiological phenotypic variability. Some students have good memory, and some do not. Importantly, studies in twins indicate that inherited factors account for about 50% of this variability. So, we know that this trait is heritable.

Imagine you are a student, and you are told to watch 30 words at a rate of 1 word per second, with the instruction to learn and recall them. In my presentation, I will now give an example of 10 words, and it is still quite challenging!

In any case, what you get is a normal distribution of performance. Some of the students remember nothing 5 minutes later; some remember 19 out of 30 words. This normal distribution is a good way to begin with a human genetics study. We conducted a hypothesis-based study first: we looked at the variants of the serotonin 2a receptor gene, either tyrosine or histidine. When we stratified the population we observed that tyrosine carriers had worse memory performance than histidine carriers.

Does this mean you have a bad memory if you are a tyrosine carrier? Absolutely not! However, what we learn from this study is that the serotonin 2a receptor is important for human memory function.

Thus, this genetic approach allows us to know and understand more about the biology, not about knowing one individual’s fate. Today it is possible not to study just one variant but rather 900,000 variants of the genome in one individual. The standardized processing with genetic arrays and validated equipment really opens up a new era where we can study large populations. We now have the ability to scan the entire human genome and to identify variants, without a prior hypothesis, in order to understand more about biology and to not only look where the light is, but to look also where perhaps it is dark.

Can we use this strategy in Alzheimer’s disease to identify molecules related to the pathological memory symptoms of this disease? In a genome-wide association study, we recruited 1,411 individuals and scanned their entire genome for 500,000 variants [2]. We identified a very high peak that was statistically significant; again, it was the e4 variant of APOE that we saw before. In addition, it is important to understand that when you do 500,000 statistical tests as in this case, you have to correct for false-positive results. With this correction, one can be more certain that the association observed is not a spurious one.

Another group adopted a similar strategy with genetic information from 10,000 individuals; they identified two variants in two genes that were very highly significant, CLU and PECAN [3]. Does this mean that someone with these variants will develop Alzheimer’s disease? Absolutely not! Remember, again, these are group comparisons not individual predictions. Let us look at the odds ratio related to these genes: 0.9, as reported in a similar study [4]. At an individual basis, these variants are not relevant for the risk of developing Alzheimer’s disease. Why are they so significant in terms of the analysis done in this genome-wide association study? This is because of the large sample size of 10,000 individuals; however, as we learned previously, this doesn’t mean that the risk (or odds ratio) might be high as well.

With this data and other confirmation studies, we now know that the CLU and PECAN variants are related to the pathways that may lead to Alzheimer’s disease. Somehow, these variants interfere with the physiological metabolic pathway. Again, we learn more about biology - not about the individual risks -when analyzing genetic data in a population.

Perhaps we still don’t know more about individual risk because we are analyzing our studies erroneously. The used methodology appears like this: we analyze each variant independently, one at time, and associate it with a trait od interest. This is simple, straight-forward, but often wrong. In reality there are so many genes and interactions in the genome, which are simply not accounted for, in the way we are currently doing the studies.

Whether complex genetics will be really useful also for individual predictions or not depends therefore on the development of new analytical methods and algorithms. It is still uncertain whether this will be ever possible - whether someday we will be able to predict who is going to develop what complex disease and when by just using genetic information. Personally, I think it is not possible, at least not with any meaningful accuracy at the individual level.

Another interesting approach is taking data from individuals and creating genetic maps, for example, all over Europe [5]. In this study, the authors scanned the genome as usual, with 900,000 variants per individual. In doing so, one can identify based on genetic information which is the ethnicity of the respective individual. This is simply a genetic map of ancestry, which absolutely corresponds to the geographical map of ancestry! We are able of course to identify many things with genetics, but to predict whether something is going to happen is scientifically a completely different story.

There must be other ways to perform our calculations. For example, what we have done in the past, we tried to identify clusters of genes related to certain diseases [6], instead of looking at only one variant at a time. In Alzheimer’s disease we are, again, able to identify clusters of genes related to this disorder. But do they give us more information in terms of prediction? No, they don’t: it looks better than one marker, but the sensitivity and specificity are still very low. So, still this cluster information is useful only for understanding biology, not for individual prediction.

In another study, we identified a cluster of genes related to ‘good’ human memory [7]. With this information, one can calculate an individual ‘genetic score’, which is how many of these genes or alleles are you possessing, weighted by their effect size. So, it’s now possible to calculate your individual ‘genetic score’ for good memory! However, what does this really mean? It doesn’t automatically mean that you have a good memory. We also observed that the higher the genetic score, the more activity in memory-related regions of the brain we can measure by functional magnetic resonance imaging (fMRI). However, this positive correlation is measurable only in groups. At an individual basis it is not relevant. We still have much to learn about the underlying biology.

We can use genetic information to understand the neural mechanisms of, for example, emotional memory. You often remember the more emotional events of your life than the neutral ones. One can test this ability and quantify a related phenotype by showing standardized photographs; most of the subjects are able to remember the more positive or negative pictures versus the neutral pictures. When we do this, we get a normal distribution of phenotypic variability and we can stratify the phenotype according to the genetic information. Therefore, we can learn about the biology of emotional memory.

We quantified this phenotype in 435 healthy Swiss participants and found a gene related to emotional memory, called the a2B-adrenergic receptor [8]. Carriers of a specific deletion variant of this gene have -as a group, not individually, better emotional memory than non-carriers.

We can do this analysis also for memories related to traumatic experiences, such as those leading to post-traumatic stress disease. To study this, we went to Rwanda and studied subjects who suffered from traumatic memories. These people were survivors of the genocide in 1993-1994 [8]. Again, the specific deletion variant of the a2B-adrenergic receptor gene was related to strong traumatic memories. This is a nice example of the real power of human genetics - you learn a lot about biology and sometimes also pathophysiology.

We can use the genetic approach to identify drug targets as well. It is a very valuable source of information. For example, we did a hypothesis-free scan of the entire genome to identify genes related to good human memory. We identified one gene, called KIBRA, which is related to memory performance; KIBRA is expressed in the human brain, and also it is related to differential brain activation on MRI [9].

When we looked more closely at KIBRA, we realized that KIBRA is related to a pathway called ROCK [10]. There are compounds that interfere with this pathway, such as hydroxyfasudil, and we know now that in aged rats, hydroxyfasudil improves memory. Today, hydroxyfasudil is being tested in phase I human clinical trials. This knowledge would not have come about without the ability to do hypothesis-free genetic studies simply because we didn’t know about the existence of KIBRA before.

However, when talking about human genetics and the quest for ‘personalized medicine’, there are huge caveats with the current approaches and the results are prone to erroneous interpretations. Remember that the difference between group statistics and individual prediction is enormous. In addition, by exploring the human genome in this manner, we produce data sets that are huge. We receive new information from the Human Genome Project, which we use to help us understand more about biology, diseases, and health. Someday, this information will come to the healthcare society through education and/or training. However, it is still dangerous to leave it the way it is now: several other ‘pillars’ need to be considered before coming to an interpretation at the society level, such as ethics, legal, and safety issues. One has to avoid erroneous interpretations of human genetics, but the reality is, at the moment, these wrong interpretations are difficult to avoid.

Several companies offer services for ‘personalized medicine’, sometimes without any medical advice. The largest company right now is called 23andMe; on their website, you can either ‘Fill in your family tree’, ‘Take charge of your health’, or ‘Choose to have it all’. If one focuses on ‘Take charge of your health’, the options are: ‘Upgrade your health records with your carrier status’, ‘Live well at any age’, or ‘Get the treatment that’s right for you’.

I went to the website myself after a colleague asked me whether I was in collaboration with this company (no, I am not). I wanted to know what he was talking about and focused on ‘Upgrade your health records with your carrier status’. Then, I realized there is a complete list of traits and phenotypes, where supposedly you can know your genes related to this phenotype. They do not only include diseases but also, for example, measures of intelligence and memory. In other words, the company wants to help you know, whether you possess genes that are related to good memory.

At first, I thought who did this work on genes and human memory, I thought we did? The 23andMe website, in fact, says if you genotype this polymorphism and if you are carrier of the KIBRA T allele, then you have slightly increased episodic memory; they even cited our study! This is outrageous. First of all, you do not need a genetic analysis to know whether you have good episodic memory, you can test this yourself. Secondly, the interpration on this website is absolutely wrong! Thirdly, unfortunately, we can not do anything about this: it is published data, they just take this information and produce a test, with a false interpretation.

Another company called Psynomics tells you whether you have bipolar disease or not. This is wrong: as it was with memory, because bipolar disease is related many alleles, its of which with very little predictive validity. Then there is Neuromark, providing genetic markers of suicide ideation emerging from certain anti-depressants. All of these companies are based on mostly ‘solid scientific data’; however, they have misinterpreted, or taken out of context, the meaning of these studies.

For example, Neuromark cites one well-conducted study that was published a couple of years ago and which clearly states in the abstract: “If replicated, these findings may shed light on the biological basis...” [of suicidality] “. . . and help identify patients at increased risk”. If replicated. This is not the case, as this study has not been replicated. Still, this company claims that one can do this test to predict suicidality after the use of anti-depressants.

All of these companies address important issues, this is true, but the implications are not supported by the scientific data. There are some other issues. First of all, are the tests diagnostic or predictive? Is it a direct-to-consumer or direct-to-doctor test, who informs you about the implications? Mostly it is direct-to-consumer. This is dangerous - the consumers are a vulnerable population; furthermore, doctors often do not know what to do with this information, there is the possibility of a stigma for some of these disorders, and negative reactions can occur. There are many, many issues.

The so-called ‘personalized medicine’ offered by these companies only addresses one issue and does not explain everything; furthermore, it definitely should not be the case of a money-making scheme - like slick oil company executives - by taking advantage of consumers and oversimplifying what are truly complex issues.

We do need personalized medicine to understand biology and perhaps to help where ever we can. Eric Green, the Head of the National Human Genome Institute in the U.S., was recently asked about diagnosing the future of genomics, specifically, what advice would he have for people who are considering buying personal genomic services from a company to find out their genetic risk for common diseases? [11]:

“I haven’t yet gone to get that information, because I think that the amount of information available at this time wouldn’t really change anything that I am doing. A lot of what I know about my own health is based on family history - I think that understanding family history, and making sure your physician knows that, is incredibly valuable, and that’s where I would put my priority at the moment. But, it is a changing landscape, so I don’t think any advice I would give today would be the same a year from now.”

I cannot agree more with Dr. Green’s advice.

In my opinion, the real value of genetic research is using it as a tool to understand biology, everything else is secondary. So, what do you need ‘personalized medicine’ for? Do you need it for prediction? Might be, depending what you want to predict and depending on the consequences of this prediction. Do you need it for diagnosis? Again, depending on what you want to diagnose and the implications. Do you need it for pharmacogenomics? Yes, for sure, but here, the implications are slightly different. It is a very different story to say: ‘You have an 80% risk of developing breast cancer’ versus ‘You have an 80% risk of reacting bad to this medication’. As a doctor, if I know that the drug has an 80%-vs-20% efficacy, then this is okay, because the consequences - for the possibility of making a wrong choice - are often not as heavy with regards to a medication and its reactions (balancing risk and benefit), versus the consequences of having a wrong diagnosis or prediction, leading to wrong treatment and worse.