World-Learning and Self-Learning: How Genes Play a Role in Learning and Behavior

Hello! Who are you and what are you working on?

I’m a neurobiologist working on the neurobiology of behavior. I study how brains generate spontaneous actions in the absence of external stimuli and how they use feedback to decide what to do next. I use largely invertebrate nervous systems for my experiments, such as flies, snails or leeches.

What’s your backstory and how did you come up with the idea?

I wanted to become a scientist when I started reading books from scientists about their discoveries in high school. I had spent countless hours observing not only the behavior of my pet amphibians, but also of all kinds of animals in the wild, so while all science was attractive, I started studying biology in 1991. I was always broadly interested, so it took a while until I decided I wanted to focus on developmental biology: I found (and still find) the question of how a fertilized egg knew where to grow a head and where a tail tremendously fascinating. However, I quickly realized that the actual lab work in developmental biology consisted largely of pipetting clear liquids and waiting for machines to complete their treatment of the samples. This was fun while it was new, but it grew old very rapidly.

After that, I got hooked on a topic that I had excluded as a specialization already in High School. Neuroscience seemed like it was impossibly complex, despite it also being very interesting, so I decided that Neuroscience was not for me long before I started studying at university. A lecture by Martin Heisenberg in 1994 changed all that. Starting his introduction into the neurogenetics of Drosophila with big questions such as free will and the organization of behavior, he had my undivided attention. While I was affixed to all the exciting things he was telling us about, he also managed to somehow convince me that Drosophila was just the right level of complexity that one could do interesting neuroscience without being completely overwhelmed by the complexity of its nervous system. Knowing what I know now, I’m not so sure any more of that last aspect. 🙂

During one of the courses at the Heisenberg lab, when we were doing operant conditioning experiments, I asked what the neurobiological difference was between operant and classical conditioning. I had learned about these two forms of associative learning in High School as forms of learning discovered by Pavlov and Skinner many decades before I was born. Given the time that had passed since then, I thought I would get a straightforward answer. To my surprise, nobody knew what the difference was or even if there was a difference. This answer got me started in my research career in neuroscience in 1995.

I started reading and became convinced that neurons store all experiences in the same way they store classical memories: synaptic weight changes depending on the evolutionary conserved cAMP-PKA cascades for the discovery of which Eric Kandel would receive the Nobel Prize in 2000. After all, this mechanism was perfectly sufficient to story anything, so why would a nervous system evolve more than one learning system?

I set out to test this hypothesis and in the course of more than a decade I not only learned that I had been wrong, but also that the many different learning systems that neurons possess, likely evolved so that they can interact with each other. These interactions form a dynamic learning web that allows the organism not only to check memories for accuracy and reliability but also to keep its internal model of the world up-to-date by constantly incorporating new experiences.

Please describe the process of learning, iterating, and creating the project

Over the course of several years, I learned that operant and classical conditioning experiments, when set up as analogously as possible (e.g., the animals learn to differentiate between the same two stimuli and are then tested identically after operant and classical training, respectively), do not leave any distinguishable differences in the behavior of the animals. I got the impression that the distinction between the stimuli was the only thing the animals had learned, whether they learned that in a classical or in an operant experiment. However, the process by which this memory was acquired was not equivalent: when equilibrated perfectly, classical conditioning required much more training to eventually yield lower learning scores than operant conditioning (i.e, a form of learning.-by-doing effect).

This was so puzzling that I designed a long series of experiments to test various hypotheses and improve the experimental design in case our results were due to a methodological flaw. However, over several years, I kept getting the same results, despite making the experiments more and more refined and making the behavior more and more equivalent to the learned stimuli: the animals only learned the stimuli and not the behavior, not matter how operant or classical the experiments were in their design.

By the time I finished my Ph.D. in 2000 and moved to work on operant learning in the marine snail Aplysia in the lab of Jack Byrne in Houston, Texas, I had a long list of candidate genes I wanted to test for their involvement in an operant paradigm that didn’t involve any stimuli and so didn’t include and confounding ‘classical’ components.

But this list had to wait until I could work in flies again. In Aplysia, I learned how cellular and not synaptic plasticity can be recruited to change the network dynamics such that operant learning leads to a change in behavior. This was the first neurobiological hint that my hypothesis about the single learning mechanism underlying all learning was wrong. The second hint was one of my mentors telling me that a student of his had tested one of the genes on my candidate list and found that this gene was not necessary for operant learning without stimuli.

Please describe the process of launching your project

Upon my return to Europe in 2004, I confirmed that the classical fly  learning mutant rutabaga, which is the central gene in the Kandelian cAMP-PKA cascade of synaptic plasticity, indeed showed no defect in operant learning without stimuli, while it could not learn any of the stimuli we trained it on, classically or operantly. My hypothesis from ten years earlier had been thoroughly falsified. I tested other fly mutants that have been shown to be defective in classical conditioning and so far all of them could learn in the operant task without stimuli. But if operant learning relies on a different set of biochemical processes than classical learning, which are those? There had been hints in the literature that a gene called PKC might be involved, but the evidence was tangential and seemed far-fetched. However, when testing flies with manipulated PKC genes, they could learn stimuli just fine but failed to learn in the operant experiment without stimuli.

Since launch, what has worked to make your project grow/successful?

After having shown that these experiments doubly dissociate on the genetic level, I needed to convince the colleagues of my discovery. This proved very difficult as we were always using two different kinds of operant experiments, one with stimuli and one without. As both were operant experiments but the one with stimuli was dependent on other genes than the one without stimuli, it was confusing to speak of ‘operant’ and ’classical’ genes or learning systems. As the primary determinant for which system was going to be engaged was the content of the memory (i.e., stimuli or behavior) and not the procedure (classical or operant) by which it was acquired, we started using the terms world-learning (i.e, where the organism learns about the stimuli in the world around it) and self-learning (i.e., where the animal is learning about its own behavior) to describe the experiments engaging the Kandelian cAMP-PKA synaptic plasticity or the PKA-dependent neuronal plasticity, respectively. This distinction between the procedure and the content proved very valuable for explaining our complex findings to a wider audience.

It also helped to design new experiments tackling the interaction between these learning systems. These experiments explain my initial findings that the presence of stimuli always seemed to dominate the entire experiment, whether it was operant or classical. The presence of conditioned stimuli always engaged the world-learning mechanism in a way that would suppress the self-learning system. Fifteen years later, my very first experiments were starting to make sense.

How is everything going nowadays, and what are your plans for the future?

Now we are starting to discover the neuronal pathways that mediate the suppression of self-learning while stimuli are present. We have also discovered that the self-learning process is the process that is engaged when we humans learn to speak, as both the fly FoxP gene is required for self-learning in flies and a human orthologue, FoxP2, is required for the acquisition of speech and proper articulation.

The Kandelian learning mechanism thus appears embedded in a web of other learning processes that interact. These interactions can take the form of facilitation, as in learning-by-doing, or in suppression as in the dominant stimuli in operant experiments.

Through your science, have you learned anything particularly helpful or advantageous?

Science advances by being wrong and learning to always adapt to a new set of data and by changing one’s hypotheses and theories to best match current knowledge. To speak with Russel: no matter how beautiful or elaborate you think your hypothesis may be and how attached you are to it, it can be slain at any moment by the ugliest of experiential facts.

Our readers would be glad to know more about you. Please let us know what is your morning routine like.

Get up at six; breakfast; get the kid to school; drive to the lab; read the latest papers, research news, and general news, and get experiments ready.

And how does a typical day look for you?

If it’s not a teaching day, I’m doing experiments all day and since they are mostly computer controlled, I get to do theoretical work (preparing teaching, coding, data analysis, admin, etc.) on the side.

What does your workstation look like?

Big, overclocked, water-cooled tower (makes it easy to access everything in there); 4k screen further than arm’s length away, running an alternative shell instead of Windows Explorer with customized interface. While I have always the latest Windows version running on it (I have used Linux and iOS as well, intermittently), my interactions with the computer haven’t really changed since 1996 or so. Important: 9 virtual desktops and a custom right-click menu.

What platform/tools do you use for your professional life?

Windows, Android, Linux.

What secondary software and apps do you use daily?

Slack, Gdocs+PaperPile, Firefox, R

How do you stay up to date on news and resources?

Feedly, Paper.li, Twitter and major newspapers

What have been the most influential podcasts, or other resources?

No particular order: Swedish Vetenskapsradion, German Forschung Aktuell, Guardian Science Weekly, Nature Podcast, Science Podcast, Science Friday, Scientific American Science Talk, This Week in Science, BBC Science Hour.

What tools do you use in your personal life? Cook? Self Care? Hobbies?

Family, sports, fly fishing.

Advice for other scientists who want to get started or are just starting out?

The most important psychological characteristic of a scientist is frustration tolerance. Nearly everything else can be compensated.

Thank you very much for your time, Björn. Where can we go to learn more?

The scientific literature or at brembs.net

Youtube channel: https://www.youtube.com/brembs/