Lessons from a Nature Method article

Although it is said that working too long hours works against creativity and efficiency, Balázs Rózsa's team is not aware of this. They work at KOKI, Pázmány, Brain Vision Center, Femtonics - to name just the most important places - and besides their microscope development work, mentioning only Nature journal and only "their" publication, four years after 2012 Gergely Szalay was the last author of a Nature Communication publication, and in 2022 Linda Judák was the first author of a paper published in the same journal.

- You seem to have moved up the "ladder". But what is really important, what really matters?

Linda

- Articles at Nature Methods, Nature Communication level are typically many years in the making, the publication of the article is just a long-awaited moment. Progress is felt in the process rather than in the end result. While we are of course more involved in the development of scientific questions and results, we both lead our own teams within the lab and, in addition to experimental work, we have to make independent decisions on many issues that are not necessarily related to the scientific part of the work, but are HR, grant applications or financial tasks. However, we believe that these are also essential, important and useful experiences for building a career in science.

- Are you satisfied with what you have achieved so far? 

Linda

- We are both pleased that, in parallel to developing the results, we have also put in place the team, infrastructure, measurement and analysis methods needed to create them. That's why we are confident that with these results and experiences behind us, we will be able to complete the next article in a shorter time. 

Gergely 

- Unfortunately, basic research as a career remains weak. I am very sorry that we have not been able to retain many enthusiastic and talented people who we have spent a lot of time teaching, and although they could have been very valuable members of our team, they have moved abroad or into industry. Although this is only a by-product of the work we do, it has led to important insights into the organisation of the team. We have tried to apply these insights to the development of the operational strategy of the newly formed BrainVisionCenter.

- In order for a "Nature" journal to send out a manuscript for review only, an infinite set of requirements must be met. Nature Method's requirements range from technical details to usability, and are the subject of a separate article. How long did it take to write your article?

Linda

- We started writing the manuscript in December 2022 and originally submitted it to Nature. At the suggestion of the editor, as the topic was indeed better suited to Nature Methods, we submitted it to Nature Methods in September 2023, where it was accepted in October 2024 and finally published on 12 December 2024. 

- The time between acceptance and publication is rather short for Nature, but a full year elapsed between submission and acceptance. I guess you have not been idle!

Linda

- We answered four rounds of questions from the reviewers. However, in the end (apart from the article itself and supplementary material), with over a hundred pages of detailed answers and figures, we managed to win the battle. Gergel and Balázs and I are a close-knit team, as this is not our first time working together, so the assignment and execution went smoothly.

- The 2012 Nat. Method article received - according to the internet - 434 citations. Even including self-citations, this is a significant number. Are you satisfied with this result?

Gergely

- As the first Nat. Methods article was the first work in the literature describing acousto-optical based 3D scanning, it was an important milestone. This may also explain the high citation rate, which we certainly consider a good result.

The success of the paper may have been helped by the fact that, shortly after its publication, two similar communications from other laboratories followed, so that it was no longer an isolated technique but a method described independently by several teams. 

- Has a similar coincidence occurred with any of your other developments?

Gergely

- Yes. Within a year, the mouse virtual reality (VR) technique was described by two other teams besides us, and this helps a lot to make this kind of technology more widespread. It is, of course, a competition on several fronts. Linda and I and the rest of the research team are working to be the first to describe new scientific results with this tool. In the meantime, however, the BCV developers have been working on refining our method to product level, so we hope to make the technique available in many other labs soon. 

- Let's stick to science! The use of mice for experimental purposes has many advantages. But their eyes are not at all like humans', so designing an experimental system for them to match was an extra challenge. You did an excellent job and we learned a lot about the mouse brain. But what use could this new knowledge have for the human brain? 

Gergely

- Not only the mouse eye, but the mapping, depth of field, retinal structure, neural processing in the eye, cortical representation, are all different in mice. In our experiments, we have modelled learning in mice that is relatively "self-evident" in humans. The improvements were needed precisely because of these many differences. Among other things, we have designed completely new optics that follow the large field of view and different optical resolution of mice compared to humans. 

- This allowed us to extract previously unknown data from the mouse brain, but it's different from the human brain!

Gergely

- That's true, but this system can also be used to learn what happens at the level of brain networks in the early stages of human learning. The technology is not available to study this at the level of individual cells in humans. Take the new discovery, also presented in the article, that plasticity during learning in primary sensory regions is more significant than expected. The same approach could be used to investigate whether there is a difference in this learning mechanism in, for example, Parkinson's or Alzheimer's disease models compared to what is found in healthy mice. This could bring us closer to the development of some therapeutic or complementary therapeutic method!  

- Mice have now been shown, or rather they themselves have been shown, to have much better visual learning abilities than we thought. But why did we think they were worse?   

- Over the last 20-30 years, countless virtual reality tools have been developed to test the vision of laboratory animals. But they have all used two-dimensional projections to represent virtual space, because it was assumed that, like humans, laboratory animals could understand and reconstruct the 3D reality around them from two-dimensional projections, such as a flat image on a TV screen. But recent research has shown that this assumption is wrong. For rodents, two-dimensional projections do not provide a realistic experience and this distorts the results. 

- So in this case, we were misled by our anthropomorphic view, but you have now taken into account the real vision of the mouse.

- Because the system we developed is optimized for mouse vision, it generates spatiotemporal patterns of brain activity that encode specific visual elements of their environment orders of magnitude more accurately. The visual learning process is thus significantly shortened.

- The difference between stereo and 3D images is striking, so you can see why stereo images are more lifelike for the mouse. Is it essential to use such a system to study the human brain?

- The human brain is much better at 3D abstraction than the mouse. For example, if a 2D image has the right perspective and shading, the human brain can interpret it in 3D. As far as we know, the mouse cannot. Obviously, it is difficult to test this directly, but indirect results point to this, as in our real 3D VR system, mice can perform tasks that they cannot when projecting simple 2D projected spaces. 

- Are there other similarly significant - spectacular differences?

- Unlike mice, humans can report on learning, subjective experience of perception, etc., so such complex techniques are unlikely to be needed for human tests. 

- So...

- In humans, learning itself can probably be achieved by simpler means, and the marking and measurement techniques that are available for mice are unlikely ever to be used in humans, if only for ethical reasons.

- What do we know about the virtual learning capabilities of humans?

Linda

- In an age of technological advances and the digitalisation of teaching methods, people's virtual learning abilities are receiving increasing attention. Virtual learning includes the use of virtual reality (VR), augmented reality (AR) and online learning platforms. For understandable reasons, its impact on human learning abilities is already being investigated by a wide range of disciplines.

- Recently, there has been growing attention to the dangers of overuse of virtual tools. Young children may develop "virtual autism".

- Our subjective view is that for humans, VR is not necessary for learning, as new knowledge can be acquired through conventional methods along human abstractions, but that undivided attention can help concentration and learning in full immersion, when the entire field of view is filled by the virtual experience.

It's worth noting that the same effect is at work in our mouse tests, with the fact that the visual experience being taught fills the mouse's entire field of view certainly improving behavioural outcomes. In this respect, we can find parallels between the two species. 

- To what extent have they been able to investigate changes in the virtual learning abilities of humans in different diseases or disorders, and to what extent can the results be considered authoritative? 

Gergely

- Since the retina is part of the nervous system, there is literature evidence that neurodegenerative diseases with anatomical changes also affect the retina and thus vision. However, these changes are minimal and difficult to detect, for example, they can only be observed during coordination tasks involving downward movement. In theory, a VR system can help diagnose such changes, as it gives you full control over both the projected image and the position of the eye. 

However, we do not know how the representation of rapid learning at the level of the cellular network changes in some diseases. This is precisely one of the important areas where we believe the device could bring new results.

 - Articles are also published on anticipatory (predictive) signals observed during learning. What are these signals and why should we pay attention to them? 

Linda

- As their name suggests, these signs indicate a future event, occurrence or stimulus based on past experience and learning. 

In effect, the brain is constantly making predictions about the environment around us to minimise errors and maximise our reaction time, and the information we gain from our experiences helps us to anticipate, in the literature's term, the anticipation process. It is good not only to be aware of this, but also to pay attention to it, because the disruption of the anticipatory mechanism plays a significant role in the development of several neurological and psychiatric diseases. Examples include motor anticipation disorders in Parkinson's disease, or predictive coding errors in schizophrenia.

- Apart from noticing this dysfunction of the predictive system, can we do anything to improve it - our own health?

Linda

- After a stroke, it can be helpful to improve our anticipatory skills, as it helps to restore the patient's motor coordination, but also improves responsiveness in everyday activities!

- The tool has also been used to uncover new, previously unknown neural network mechanisms that are created during visual learning. How many types of learning are there, and how do they differ?

- Probably not much in terms of their mechanism, since learning is essentially the formation of synapses (connections between neurons) and temporary or permanent changes in their strength. However, billions of such synapses can form complex functional mechanisms. A good analogy is the principle of the operation of a computer, where complex tasks are solved by the parallel operation of billions of logic gates. Everything depends on the inputs received at the right rate and in the right order. To continue the analogy, our discovery is not novel in the mechanism of action of learning, but in how these effects are integrated at the system level. This is a bit of speculation, we will talk about this in a future article!

- Okay, but how and why did your Moculus tool allow you to make new discoveries in your experimental mice that you have been studying in visual learning mechanisms, with a single training session instead of a long training period? Have you been able to measure something that others have not? 

- In essence, yes. Firstly, it is important that our system for measuring cellular activity is compatible with our 3D imaging system, so that the behaviour of cells and cell networks can be studied in the learning process. In most of the work to date, learning has taken place in an independent system and only the end result of learning has been examined, so that the increased activity that occurs over time cannot be captured. On the other hand, even if they could test activity during learning, it took several days, the mouse had to be put back into the system every day, and the result could be influenced by several external and internal factors. 

- Poor mice. Although they were given a statue in honor of their role as laboratory animals, it was widely believed that they were dumb. At least compared to rats. Why were they said to be slow learners? 

Linda

- I would say that with other VR systems, visual learning was mostly seen as just at the end, when there was already a rearrangement of brain activity patterns. With our equipment we can follow the whole process of visual learning, from the first minute. With the systems we have used so far, it took 5 to 9 days for the mice to learn the task. In our experiments, we can see that the mouse's visual learning process is complete after 30-40 minutes and remains stable thereafter.