In preparation for OHBM 2020, we talked to Dr Tomas Paus, who will be giving a keynote lecture on Friday, June 26th. Dr. Paus is Director of the Population Neuroscience & Developmental Neuroimaging Program at the Holland Bloorview Kids Rehabilitation Hospital, and Professor of Psychology and Psychiatry at the University of Toronto.
Roselyne Chauvin (RC): Thank you for taking the time to chat with us. In your talk you will be speaking about “population neuroscience and the growing brain.” There are a few ongoing longitudinal big data initiatives, such as ABCD or generation R. Those projects are now starting to think about the current pandemic situation. On one side, the situation is affecting everyone without discrimination; on the other, government responses create different experiences (from full to partial lockdown, to no restrictions), and of course, individuals show different stress responses. How do you think this might affect longitudinal datasets? And what are the questions that will need to be investigated out of this situation with regard to psychiatry and genetics?
Tomas Paus (TP): You can look at COVID as a natural disaster. There are studies where natural disasters have been used in the past as pseudo-experimental designs, i.e., to study the effects of a perturbation, because in most of our observational studies, we can really only look at associations between x and y and so cannot infer causality. In most cases, we don't know anything about the directionality of those relationships. But natural disasters provide an opportunity to study before and after and try to attribute the observed changes to those events.
A key component in the context of brain development and psychiatric disorders is social distancing and what has happened with social relationships. For children in particular there are two elements that I think really stand out. One is homeschooling, which, depending on a country, may last for several months. I don't know how it is in the Netherlands, but in Canada, it will last for at least three more months, if not more. And then the other element is the family, so it depends on what's happening at home. Unfortunately, in some cases, that means bigger exposure to adversity, adversity as bad as family violence. So there the stressor may be huge for some children.
Studies that have acquired detailed phenotypes, whether it's behaviour or brain phenotype on children before the event are in a unique position to go back when it will be possible and study the change in behaviour or in the brain. Generation R is certainly one such cohort, ABCD is another one. There are others. Even birth cohorts that may not be at the most relevant age from the perspective of child development but able to study the relationship between exposure to COVID-19 and events related to the disease and health in general. Of course, UK Biobank is the biggest one of all, right?
Now, one more thing in terms of children. Unfortunately, we do know that the most vulnerable segment of the population in terms of mortality are older people. And so there will be an increase in the number of grandparents dying. That is again, of course, a highly stressful life event and that will, one way or another, affect those children. Finally, we know already that at the level of mortality, COVID-19 is more frequent in disadvantaged populations, mostly in the context of socio-economic position. So there may also be an interaction between the pre-COVID conditions of those children and COVID-related stress.
RC: You’ve been involved in many different types of big data projects, from the acquisition and study of local communities like the SYS (saguenay youth study, ~1000 adolescents and their parents, from the genetic founder population of the Saguenay Lac St Jean region of Quebec, Canada) to the ENIGMA consortium (ENhancing Imaging Generic through Meta Analysis, a worldwide collaboration with more than 40 countries involved). How have you found carrying out these projects, and what advice would you give for those wanting to carry out these big data projects?
TP: It's a very good question and makes me reflect on my own path from the Saguenay to now. Over time I have increasingly become involved with collaborative work in the context of Enigma, and CHARGE, the other consortium that we work very closely with. I started this Saguenay study with my wife, Dr. Pausova, and others almost 20 years ago. That gives our team a lot of hands-on experience in carrying out big data projects. We learned what it takes to set up a cohort, to set up the protocol, to carry out quality assurance. All those different steps, on a relatively small scale. Even though 20 years ago, 1000 individuals was a fairly large scale for us. But I think that hands-on experience with a cohort is very, very important once you enter collaborations with others, and also once you start using data that had been produced by others. Of course, in a consortium, you share that experience and that's a currency.
In the CHARGE Consortium we have weekly conference calls. It's amazing how much you learn during one hour given there are between 20 to 40 people on the call. In one hour, we pick a topic, usually a study that is being carried out, and it's being discussed from the beginning to the end. You benefit, of course, from the expertise of people who have done many of those studies before. And you benefit from informal expertise that is very hard to get from reading the paper. In the same way that I can share my 20 years of experience with the Saguenay study with this group, every member of CHARGE group shares her or his experience back. So that's a huge plus.
In these consortia, it's not only about accessing data, you're really sharing knowledge; not only expertise in designing studies and acquiring data, but you’re also learning about the latest in genetics, epidemiology and statistics. So you’re keeping up-to-date with developments across many different fields. That's a huge benefit of working within a consortium.
The last point is about the diversity of the group. The group is diverse not only in terms of the disciplines, but also cultural backgrounds: it includes researchers from different countries, different educational systems. So for us, it means that there is a diversity of perspectives and I think that that's what you want. If you want to create new knowledge, you don't want everyone to have exactly the same background; you want to see things in many different ways and from many different perspectives.
RC: That also makes me think of sharing experience and trying to find the best way to maintain high quality. I mean, there are many initiatives to standardise scientific practices, for example using the BIDS format to organise data - that type of knowledge came from a consortium. Do you think we could extract some guidelines to help big database initiatives?
TP: I'm not sure about that. I mean there is a whole science of data harmonization of origin - there are experts who work on that. I'm somewhat sceptical about coming up with guidelines or toolboxes to be imposed on investigators when they are starting a new study. I think that there is a danger there. Yes, it would get easier then to harmonize across cohorts, but there is a danger that it would stifle innovation and new discoveries. If everyone is doing everything the same way, then where is the novelty? Where is the potential for new knowledge?
What I've seen is that, basically, it's a democracy of the scientists and the trainees voting indirectly by adopting certain tools more often than others. And then all of a sudden that tool emerges as the most commonly used tool. Freesurfer is an example of that, right? There are different ways to extract information about cortical thickness and surface area, but I must say that in the majority of studies Freesurfer became the main tool that everyone uses and so now you have a sort of natural emergence. So harmonisation has emerged in a natural way.
RC: In a similar vein, neuroimaging has faced a reproducibility crisis, just like genetics did before. There is increasing recognition that studies need to use larger sample sizes to produce more representative and reproducible findings. OHBM sessions have reflected these improvements in working, creating best practices for methods, promoting transparency via open publications, code, and data. The OHBM open science room grows every year and now the announcement of Aperture, their publication platform. What has been your experience and your change in practice? What advice do you give your lab members or early career researchers to improve the quality of their science?
TP: Well, that's a difficult one. I think that the starting point is critical thinking and that's what I'm trying to convey to my students. We need to question conclusions, to question reliability and that's maybe one of the reasons that even though we do use functional imaging, I do put more emphasis on multimodal imaging of brain structure because we know that structural imaging has higher reliability. Even though I started with imaging with PET with blood flow activation studies, I moved into that field from my interest in brain behaviour relationships, in a way. The relatively low test-retest reliability of functional measures and behaviour in general made me shift my focus to features that are easier to measure, such as the structural properties of the brain. That's probably one of the reasons why I changed my way of doing science in those large numbers - test-retest reliability becomes crucial if you are interested in a trait and if you are doing genetics, if you are running epidemiological studies where you are interested in influences of environment, you need to have that measure with a quality of a trait. That is, if I measure a trait today, and I measure it again two weeks later, I get more or less the same number. That's really crucial. I started by saying that one has to be critical, and I think, that that's kind of the simplest advice.
Another key for quality of science is replication. Let's say functional imaging studies, split the sample, analyse the data in one half and then see whether you find the same thing in the other half. Don't trust p-values. That would be my other advice. P-values will not guarantee reproducibility; replication would.
R.C.: So you said, you started with PET and then moved more towards structural MRI. Now that we are on the advice side, what do you think would be the next big topic in neuroimaging? Would you advise a young neuroscientist to follow the trend or look for their own niche? If you had to start something new, what would you go for?
TP: I wasn't really thinking too deeply about what I want to do in five years. I went with the flow and was always driven by curiosity, by novelty, by something unexplored. Often I was critical of a finding that I didn't believe and that triggered a line of thoughts: “I don't believe it's this way. Let's prove that it's the other way and what do I need to prove it.”
I do like to combine different levels of analysis. That's partly because of my initial educational background in medicine, human physiology, anatomy, etc, combined with deep interest in behaviour and psychiatric disorders. So you have both the systems level and molecular level, and integrating across systems, across levels, and I think it did work for me.
If I was going to do it again, I would probably again try to get a broad education that gives me at least some understanding of the different levels, rather than one very deep understanding of a particular approach, like the details of DNA structure. That just doesn't work for me, but it may work for someone else.
RC: Multidisciplinarity is at the core of cognitive science.
TP: It wasn't like that when I was starting! The fact that I got that broad education really prepared me for that interdisciplinarity and for working in large teams. When I was starting, the labs were small and there was little data sharing, even in genetics, and particularly in genetics of Mendelian traits. There were fierce competitions between people in terms of discovering disease genes, so people did not share. They competed with each other and that is a dramatic change over the past 30 years, possibly the biggest change I've seen in science and the social aspects of science.
Now, even with the amount of sharing there is always competition. Competition is good, we need it. But the competition doesn't interfere, as it did in the past, with generating data, with access to data because open science puts everyone on an equal playing field. So now it's not about someone having access to these data and blocking us from having access. It's not the case any longer. You really have to share data in some form.
RC: Yes. The evolution of the field is towards being open, being collaborative and getting experience from those that know how to acquire data and those that have strong expertise in methods.
TP: Also, when you look at institutions that support this kind of approach - they are successful. Institutions that are supporting open science and developing platforms for data sharing and open science in, for example, bringing different bioinformatics databases to communicate with each other, etc. An example is MIT Broad Institute in genetics.
RC: What are the findings that you are most proud of?
TP: There are two different types of things that I am proud of. I told you that I like innovation. I like doing things in a new way. In that context, I'm proud of two innovations. One is when we put together brain imaging and brain stimulation, our combined studies with transcranial magnetic stimulation and PET. Technically it was quite a challenge and I think we did it the right way. That approach eventually did not take off on a large scale. But I think in the mid 90s, when I worked on it, it was really exciting to be putting together TMS and PET in the way that we did. I'm definitely proud of that aspect.
Then, I think about what I'm doing now in terms of the combination of epidemiology, genetics, and neuroscience. I'm glad that I was able to put it together into that framework and I wrote a little book about it. I'm happy about it.
In terms of findings, I think two, for me, stand out. One goes back to the late 80s, to my PhD when I noticed some very interesting deficits associated with lesions of the anterior cingulate and then I followed up those findings with my first PET studies in Montreal. I came up with some discoveries about the function of the anterior cingulate cortex and its role in the interface between intention and motor control. Those early studies I still like.
The second finding is more recent and relates to what we have done in teenagers. The observation that testosterone has something to do with the radial growth of the axon. So, basically, the thickness of the axon, in particular in male adolescents, and how this may relate to axonal transport. That is a slight shift away from myelin and toward axon and I think it's important. We are pursuing that finding. I think that it's the axonal transport element that becomes very important for function. I personally believe that the link between axonal diameter and axonal transport will inform new studies of individuals, also mental illness. So that's the second finding that made a difference in my research.
RC : Are you going to talk about that during your OHBM lecture? Can you give us a sneak peak?
TP: I will talk mostly about big data and some findings from our work in the context of ENIGMA and CHARGE consortium, relating to the developing brain. This will illustrate the power of big data. But I will start with a bit of history on how we got where we are now and how important observations are, going back to my mentor Brenda Milner.
RC: Thank you for your time and for chatting with me!
TP: Thank you, it was really enjoyable.
RC: I am really looking forward to your lecture. This year is going to be a different format, as OHBM is happening online. So I hope this teaser will attract a lot of digital attendees and that everyone will enjoy your lecture and the meeting safely from home.