It takes a village to do research. Scientific discoveries from the 23andMe Research Team are made possible both by our three million research participants and by our academic collaborators. Jennifer McCreight, Ph.D., a research communications scientist for 23andMe, sat down with Abraham Palmer, Ph.D., professor of psychiatry and vice chair for basic research at University of California, San Diego School of Medicine, to talk more about our recent collaboration on delay discounting that was published in Nature Neuroscience.
Jennifer McCreight: What is delay discounting, and what inspired you to study this particular trait?
Abraham Palmer: I have wanted to study delay discounting, which is the tendency to favor small immediate rewards, over larger delayed rewards, since I was an undergraduate. The concept resonates with me because being able to work for a larger delayed reward is so critical to the survival of many animals, including humans. The idea that pigeons, for example, are calculating, in real time, the value of a delayed reward is amazing. We do it too. Some of it clearly is at the conscious level, but I’m convinced that people also have innate differences that are due to genetic factors. So this is a trait that impacts health, but it is also an important concept in microeconomics.
From a practical perspective, I’m excited to use these and future results to identify specific genes. Part of my lab works on human genetics, and part of it works on mice and rats. If we can identify genes using human genome-wide association studies (GWAS), we can study them in rodents, where we can also measure delay discounting. That allows us to understand how those genes influence delay discounting, something that is hard or impossible to do in humans.
JM: But mice can’t exactly fill out a survey like a 23andMe research participant, can they?
AP: Ha, yes, the whole field of behavioral neuroscience is based on elaborate methods that get around the fact that we can’t just ask a mouse how it is feeling, or whether it wants 0.8 ml of water now, or 1.6 ml of water in 72 seconds. If we gave them a paper survey they’d sniff it for a few minutes and then shred it up and make a nest out of it!
JM: What are the benefits of using a cohort of humans versus a mouse model? Disadvantages?
AP: The problem with people is that they are too damn smart. So they might see these questions as pertaining to current interest rates, or their expectations about the return on their investments, or similar financial issues. Or they might have a cultural bias that it is always better to wait. Those factors obscure the “innate” genetic differences that we are trying to study.
But a really nice thing about humans is that they are so damn smart. We can ask them these questions in a minute or two. With rodents, we have to spend a couple of weeks “teaching” them how to answer these questions. Also, rodent studies do use a real reward, typically water or food, and demand for those rewards might change from subject to subject, or over the course of a session, or between sessions. But rodents don’t have bank accounts, and don’t know what the prime interest rate is. They’ve also been raised in very uniform environments, which minimizes variability due to environmental factors. In a paper written by Jerry Richards, we estimated that the heritability of delay discounting was about 60 percent. (A bit of trivia: Jerry is the one who first got me interested in delay discounting, when I was an undergrad and he was a postdoc in Lew Seiden’s lab at The University of Chicago).
One of my concerns is that delay discounting in humans may or may not be similar to delay discounting in rodents. There are many differences, including the abstract “reward” offered to humans versus the food or water “reward” offered to rodents. Another issue is the timescale: we’d never be able to train rodents to wait days or weeks or months for a reward, and I’ll bet that they wouldn’t wait that long even if they understood the question. I think that their time horizon is much shorter than ours.
So that raises a question: are decisions about abstract rewards months away really equivalent to a behavioral paradigm in which a rat chooses between water now or more water in 72 seconds? Quite frankly, no one knows; there is pharmacological and neuroanatomical evidence to support the model, but I think questions remain. Those are questions I hope to tackle in the coming years, and these results will be really helpful in that effort.
JM: What do we generally know about the genetics of behavior and personality traits? How heritable are they compared to other traits or diseases?
AP: My postdoc Sandra Sanchez-Roige and I just finished a review of the genetics of personality, so this information is fresh in my mind. Twin studies using various personality traits suggest heritability in the ~40 percent range. SNP heritability is estimated more like 5-18 percent (SNP heritability is expected to be lower).
JM: What are the benefits of using GWAS to investigate behavioral traits? Are there any limitations?
AP: I think of behavioral genetics as a broad field that includes diseases and also normal variation, which may contribute to disease in extreme cases. Personality provides a continuous (rather than case:control) measure, and can be studied in “normal” populations, yet may provide insights into diseases. The limitations that come to mind include power (sample size) and the obvious role of environmental (including but not limited to cultural) influences.
JM: Why is it important to look at an individual process (such as delay discounting), rather than a disease as a whole (such as ADHD)?
AP: I wouldn’t say it is an either/or proposition, but I think delay discounting can add a lot to studies of disease phenotypes. For example, ADHD can be dissected into various components; that is the goal of the Research Domain Criteria initiative, which is a big deal at the National Institute of Mental Health. Delay discounting is one component of ADHD, but I expect that some ADHD genes will have a big impact on delay discounting while others will not. If we know which is which, we know how to study the mechanism of those ADHD genes. That means that having data about both ADHD and delay discounting will allow us to better dissect the diseases into processes, neural pathways, and molecular mechanisms. For me, that is the goal of doing GWAS in the first place.
JM: What advantages does the size and variety of 23andMe’s dataset offer a researcher? Did it help speed up the work you were doing or allow you to research something you couldn’t otherwise?
AP: Well, size matters, a lot. 23andMe was able to collect phenotypes from 23,217 subjects in about four months. That’s fantastic. My lab has none of the infrastructure needed for that scale of data collection. Having genotypes already available was also key, and the research participants were willing to do this work on a volunteer basis, which meant a lot to me.
JM: One can imagine that financial security plays a large part in an individual’s ability to delay receiving a reward – do you control for this?
AP: We had hoped to include socioeconomic status (SES) as a covariate, but because 23andMe only had that data for a fraction of the subjects we decided not to. If we had to exclude research participant for which we had no SES data, it would have reduced our sample size by about 25 percent. However, one advantage of doing this project with 23andMe was that we knew that research participants were reasonably affluent; for example, they could afford to pay to have themselves genotyped, which is a sign that they are not living “hand to mouth.”
In other cohorts/populations, this is a major concern. For example, someone who is a drug addict might urgently need the smaller amount of money that is available now because they need to buy more drugs. Moreover, Carl Hart from Columbia once pointed out to me that delay discounting may also measure how much the subject trusts the researcher to actually give them money later. If you have learned that you can’t trust other people to make good on their commitments, or if you don’t find the scientists trustworthy, you might assume there was a degree of risk in waiting for a delayed reward. So delay discounting may measure different things, depending on the population used. I saw 23andMe as an advantage because these factors were less likely to be in play, but also possibly a limitation because highly impulsive people might not be well represented because they might be less likely to join 23andMe in the first place.
JM: How accurate are self reported measures of delay discounting compared to direct measures of the behavior? Do people overestimate their ability to wait for greater reward because of how “impulsivity” is perceived?
AP: That is a great question. When we were designing this project, I wanted to include some tasks that are intended to measure impulsivity. These can be implemented as Java applets, and would have allowed us to measure behavior directly. I still hope to do that study one day. We considered trying to test behavior by telling the research participants that one of their decisions would have real consequences; that is, they would actually get $10 now or $20 in three months. Other papers have suggested that this has little effect on the answers people give, which made us think we could forgo giving any real rewards. In hindsight, I’m glad we did, because giving ten dollars (or more) to 23,217 people would have been expensive!
JM: Your most significant SNP was on the X chromosome — do you think this is why women tend to show more delay discounting compared to men?
AP: I wouldn’t read too much into that. Although it was the most significant finding, that locus has a very small overall effect. I was surprised when we saw that women in this cohort put less value on future rewards. I actually panicked when I first saw that, because I had assumed (and thought I knew) that men would discount more. So Sandra and I had this panicked evening where we thought we’d made some critical error in the processing of the data, but then we realized that our results were consistent with prior studies. It is true that men are more impulsive when other measures/definitions are used, but delay discounting is different, for whatever reason. So all that panic was for nothing!
JM: What are the next steps for your research?
AP: This is a down payment. I’m hoping that by this time next year I’ll have at least 100K genotyped subjects who have answered these questions. We are working with at least half a dozen groups already to collect that data, and we are very open to more collaborations. To be honest, I probably won’t be satisfied until I have a million. Another next step is to translate these results into rodents studies, as we just discussed. A third direction is to get more measures of impulsivity, hopefully including behavioral measures. And of course all of this requires endless grant writing, which I mostly enjoy, but I’ve been financing this project out of discretionary funds so far. Now I need to reload, and that means grants of possibly philanthropy.
JM: Where do you see the field of psychological genomics going in the future?
AP: I ask people that question all the time – what will we be talking about in five years. I don’t even ask about 10 years in the future because it is laughable; no one can predict that far in the future, the field is developing so quickly. I think digital mental health is a big deal. One aspect of digital mental health is collecting “big data” from people to monitor, predict, and possibly improve aspects of mental health. That big data, especially when it is from genotyped research participants, will be a huge resource. I suspect we could reframe questions about impulsivity by measuring aspects of behavior using smart phones or watches or credit card activity or whatever.
Another huge area is the use of electronic medical records in genotyped subjects. I’ll bet that polygenic prediction of delay discounting will predict all sorts of good and bad health outcomes; we are already looking at that with Lea Davis and Nancy Cox and others at Vanderbilt University. Of course every year more and more people will be whole genome sequenced, which will allow us to examine rare variants and their impact on all phenotypes. Model systems, both using whole animals and cellular systems, will be extremely important in realizing the promise of molecular insights from GWAS. In terms of how this will impact patients, I hope to see new and better drugs developed, and we all hope that drug treatments will be personalized based on an individual’s DNA.
JM: Who is the mastermind behind the science limericks on your lab webpage?AP: Aren’t those cute? I gave a talk years ago at University of Illinois Urbana-Champaign — I think Gene Robinson was the one who invited me — and they had a tradition that the audience would write limericks about the seminar. When I finished taking questions, one of the graduate students got up and read them. I was so impressed I got copies.