In vivo, in vitro, in silico

Wednesday, 13 July, 2022

To understand the complex processes of learning, we also need the tools of computational neuroscience. The results of Szabolcs Káli's group, published in eLife, have already attracted a lot of interest.

"Historia est magistra vitae". This saying of Cicero, perhaps the most famous of all orators, who lived in the first century BC, was quoted a long time ago. However, it has been lost in the course of a few hundred years that the saying was part of a longer sentence, and that the word 'historia' in that era meant more than the science of history, it meant 'the' science itself.

But even before Cicero, in the golden age of the ancient Greeks, science was a vast and varied thing. Scientists, even if they did not live in a Greek city-state, practically all of them knew several languages, so they could understand each other's work. Later, during the Hellenistic centuries, the mediating and leading role of Greek became clear, till this was taken over by Latin with the advance of the Roman Empire, and Latin dominated until the national languages were made suitable for this task.
But there was a period of a few centuries when Arabic was the main language of science - and poetry and literature - in the South-Western half of Europe.

The story began in the even more fabulous city of Baghdad in the fabulous East. Not Harun al-Rashid, the fabled caliph of the Abassids, but al-Mamun, the son of the learned caliph who gathered the world's greatest scholars of the day to his court, regardless of nationality or religion, and even had Aristotle translated into Arabic at his request. By then, in the early ninth century, the Arab conquest had already reached Europe for a hundred years, and it was only after 1491, when the magnificent Granada was conquered by the Spanish, that Latin's leading role as the language of science was consolidated on the Pyrenees peninsula.
After Latin, from the end of the 18th century, depending on the field of science and the geographical location, English, German and French were all used as languages of science, however, after the Second World War English took over the leading role.

No doubt this will not always be the case, but the article in question was published in English in the journal eLife and has been praised in previous articles. The article presents the results of a dynamic field of computational neuroscience, which has been developing for several decades and which can help us to understand the brain processes of learning.
Learning itself is not a simple process, nor should understanding brain processes be, but in Szabolcs Káli, the lead researcher on the paper, we have an initiated leader.

So let's step into the brave new world of computational neuroscience, and let Szabolcs Káli take the floor.

- What is your current article about and what was the result that made your work accepted for publication in this excellent journal?

- Since the 2000s, Norbert Hájos and his group have been investigating hippocampal slice preparations in which network activity very similar to activity patterns observed in vivo (e.g. sharp waves and gamma oscillations) can be measured. These in vitro experiments have also allowed detailed investigation of the cellular and synaptic mechanisms underlying these population activity patterns (including electrophysiological, pharmacological and optogenetic methods) and the construction of in silico models based on these data. Combining these tools, we have been working for quite some time with Norbert, Attila Gulyás and a number of talented young colleagues on understanding different hippocampal activity patterns and the transitions between them.
In the current paper, we have gone one step further, trying to link the in vitro data and the model with observations from live animals, and aiming to use the model to explore the links between hippocampal network dynamics and coding and learning.
This paper presents mainly the results of computer simulations.

- You can read and hear a lot about the hippocampus, but please start at the beginning in this topic, too!

- The hippocampus is known to play a fundamental role in learning and memory processes, as well as in spatial orientation. It is also known that synaptic changes during awake exploration and spontaneous reactivation of the hippocampal network during sleep during wakefulness, which is associated with the "replay" of activity sequences of hippocampal place cells observed during movement, play an important role in learning processes.

- The discovery of the role of nerve cells being activated at a particular point in space - hence the name place cells - has won a Nobel Prize, but what are sharp waves?

- Sharp waves were originally defined as characteristic field potential patterns ('brain waves') in the hippocampus, usually produced during slow-wave sleep and wakeful rest. The sharp waves themselves are transient (i.e. non-periodic) patterns lasting roughly 50-100 ms, but are usually accompanied by a very fast oscillation (roughly 200 Hz), which is usually referred to as a "ripple". The ripples originate in the hippocampus and are associated with a strong increase in synchronous neural activity, and are thus very effective in influencing the activity of the rest of the cortex, especially in states where external stimuli are weaker. A popular theory is that our experiences are first stored in the hippocampus, and then relevant information is transmitted to the rest of the cortex during sleep, via the brain waves. This hypothesis is also supported by the observation that during sharp waves, the hippocampus and other areas repeat ('replay') patterns of activity of previously observed populations of neurons, for example during exploration of the environment.

- Now that we understand what these sharp waves are and what their physiological role is, share with us what was the hypothesis on which you based your model?

- Our hypothesis was that the structure of the recurrent stimulus connectivity of the CA3 region during awake learning could be responsible for the subsequent spontaneous replay of activity sequences and could also influence global hippocampal dynamics, i.e. the generation of sharp waves and the accompanying fast ("ripple" frequency) field potential oscillations.

- Can you give a lab-jargon-free and slightly easier to understand example of this?

- Many neuroscientists also believe that patterns of activity at the level of large populations of neurons (the "brain waves") provide a kind of fixed background that supports both information encoding and learning. We reasoned that since the strength and pattern of connections between neurons change during learning, and we know that these network connections play an important role in the generation of brain waves (e.g., sharp waves and oscillations of different frequencies), it is possible that changes during learning are necessary for the generation of, for example, sharp waves.

- How could you represent these two states and their interaction in your model?

- We created a detailed, albeit simplified, CA3 network model with cellular and synaptic parameters tuned based on in vitro experimental results, while during simulated learning we used activity patterns that are typical of hippocampal place cells during wakeful exploration.
By detailed examination and selective manipulation of the model, we found that there is a close link between the generation of physiological sharp waves and the replay of activity sequences - both of which require a chain-structured functional connectivity between CA3 pyramidal cells, which in turn can be generated by a plasticity rule dependent on cell activity during exploration. We have also shown that, in order for the observed replay of sequences to occur in both directions (forward and backward), a specific learning rule, symmetric in terms of timing of pre- and postsynaptic activity, is required, which is precisely what is observed in the CA3 stimulatory recurrent collateral system and which differs from the symmetric (causal) learning rule that is characteristic of other connections.

- Place cells can be found not only in the CA3 region of the hippocampus, but also elsewhere. Could what you found be true for those as well? And if not, what does this affect?

- Our specific results within the hippocampus are applicable primarily to the CA3 region because it has the large number of stimulus recurrent connections required, and it is here that the symmetric learning rule was observed, instead of the temporally asymmetric plasticity rule typical of other brain areas, which created the possibility of bidirectional sequence replay in the model.
However, the general conclusion that plasticity processes during learning and individual development lead to structured synaptic interactions and, through them, to structured activity patterns (e.g. sequences), and that this has a fundamental impact on the network activity (e.g. sharp waves) observed at the population level, may be important for understanding the functioning of other brain areas.

- Your article has been up on the web for perhaps 3 months now, and I have seen over a thousand readers, which is quite a nice readership in this topic, moreover your work has been cited once. Did you expect something like that? How important is it to you what other people think about your work?

- Of course, it is also important for us that others - preferably not only modellers but also experimenters - find our results interesting and important, use and cite them, and through this contribute to the development of neuroscience. I was therefore also pleased to see that the article was widely read and even cited. At previous conferences and workshops where I have presented this material, there has also been considerable interest and generally positive feedback. Of course, the extent to which this work will ultimately be known and cited will only really become clear in years to come.

- Some paths lead far and wide, but they cannot be continued, other paths must be sought. In music, Wagner is a famous example. How does this apply to your work?

- Fortunately, this work has natural continuations and extensions in several directions, and some of these have already begun with my current students.
Firstly, in this paper we have focused primarily on modelling sharp waves and ripple oscillations, and we would like to extend this to other hippocampal network states, in particular gamma oscillations, in the generation of which the CA3 region also plays a key role. We are greatly assisted in this by in vitro data showing how cellular and network-level changes accompany hippocampal state transitions that can be observed in brain slices during e.g. cholinergic agonist administration. On the other hand, we plan to extend the model to other hippocampal and even other cortical regions to understand the role of sharp waves and sequence replay in the generation, retrieval, and long-term retention and integration of memory traces.

-There's a list, a table at the end of your article of what can be interpreted. Is this a feature of the journal for articles on similar subjects, or was it your idea?

- This table contains the main assumptions of our model. It was András Ecker's idea and is intended to make absolutely clear the essential simplifications and assumptions that we have used in building the model. Although we think that most of these simplifications certainly do not affect the main conclusions of our work, we thought it important to list them, both so that each reader can decide how important he or she thinks they are, and so that anyone who is interested can try out what happens if other assumptions are used. For the same reason, we have published the full source code and all parameters of the model and simulations.

- You mentioned Andras Ecker. But other students were involved in the work.

- Not just involved - several exceptionally talented students from the University of Padua played a key role in the development of the model! In addition to András, Eszter Vértes and Bence Bagi, who worked as student researchers at KOKI during their BSc studies, continued their studies at elite universities abroad (EPFL, UCL, University of Berlin) after completing their BSc. This was of course a positive development for them, but the project progressed very slowly. Fortunately, one of them, András Ecker, who became the first author of the paper, continued the work in his spare time - doing the remaining simulations, creating spectacular diagrams and writing the first draft of the manuscript - while pursuing his Masters and PhD. He played a huge role in the eventual publication of this article.