György Buzsáki is Board of Governors Professor at the Center for Molecular and Behavioral Neuroscience at Rutgers University. His recent book, Rhythms of the Brain, is a clear explication of the study of network-level dynamics in the nervous system, ranging from innovations in extracellular recording to theoretical solutions to the binding problem. Rather than list his numerous awards and accomplishments, I direct you to the hundreds of research articles and reviews which have placed him at the forefront of the rapidly progressing field of systems neuroscience. Chris Chatham of Developing Intelligence and I collaborated to bring you the following ten questions.
1. Modeling necessarily requires simplification. In your view, what features of biological neural networks are so important that they must be captured in any accurate artificial neural network model? For instance, do you believe it is enough to supplement firing rate-coding units with a parameter for “phase,” or do computational models need to simulate biological neural networks at a lower level?
Models can be useful in at least two different ways, inferential and inductive. Inferential class modeling is analogous to statistics. In the simplest scenario we measure two variables and compare their relationship by e.g., a t-test. When the system under investigation is complex and the multiple measured variables are not related to each other in an obvious way, testing the statistical validity of the individual and interactive contributions can become a daunting task. Here models can be invaluable because they may explicitly illustrate which variables we believe are critical. Such model-based summaries can be conveyed to others much more rigorously and effectively than using words and pencil drawings typical of the ‘old days’. Models in the second class can extrapolate from a limited set of observations, so that only a few test points are needed to verify the validity of the extrapolation. These models can (or rather should) address the issue of ‘scaling’. A good model is not about the reproduction of the observations but about its ability to predict how the network/system should grow to preserve the functions and timing of the smaller size network. For example, if a network of 100 neurons can generate gamma oscillation, what should be the rule to generate the same coherent rhythm when the network is scaled up to 1,000 or 10,000 neurons.
Every model, as any biological system, must have a ‘goal’, to be meaningful and interpretable. It is not the ‘biologically realism’ or ‘detail-equivalence’ that matters in most cases but the inductive power of the computational model. If I gain new insights from a model, I like it no matter its ingredients. I gained a lot more insights about oscillations from models over the past decade than from my own experiments. Experiments provide the constraints and the model should provide alternatives.
2. Contextual fear and simple context learning have become standard assays for hippocampal functioning and plasticity, but it is difficult to reconcile the use of this behavioral tool with the theories that arise from unit recording. What is the relationship of hypotheses concerning episodic and semantic memory and types of navigation to context or contextual fear memory? Is a context (or configural representation) a sequence or a map?
I believe, along with many other evolutionary biologists, that a computational algorithm introduced by nature in a small network does not change drastically when the network grows. I believe that the fundamental nature of computation is the same in the hippocampus of mice, rats and humans. This is why I keep working with rodents. I pointed out that the computations used for dead reckoning (or path integration) and map-based navigation in the rat are perhaps identical to the computations used for episodic and semantic memories in humans, respectively. Both dead reckoning navigation and episodic memory rely on self-reference and require a unique spatio-temporal context, in contrast to the self-independent (explicit) map and semantic information. Since fear is custom-tailored and does not exist outside the brain, and it is both contextual and self-referenced, the connections between these man-invented terms are perhaps not so remote as they presently appear.
3. You have invested much time and effort developing and refining the silicon probe for multi-unit recording. This investment seems to be paying off now in a series of remarkable findings only achievable through the recording of large numbers of neurons. What is the next level of refinement needed in this technology? For that matter, do we need to record and isolate more units or do you think that we have reached the point of diminishing returns?
With some effort several laboratories could record from a thousand or more neurons simultaneously even from a small brain with existing technologies. But this in itself is not interesting. One can place hundreds of wires in the neocortex and other structures and increase the n. The emphasis is not on gigantic numbers but on statistically representative samples of neurons that can provide insight into the nature of the computation. Accordingly, my strategy is to record from two or more representative populations of local neurons without inflicting detrimental damage to the network. This task cannot be achieved effectively with wire electrodes but silicon probes can provide progress. Multiple-site probes allow not only measuring the spike output of neurons but also provide information about intracellular and intradendritic events brought about by the inputs, and all this can be done in the behaving animal. Only when both inputs and outputs of the networks are monitored simultaneously can one hope to infer the underlying computation.
4. What is 1/f organization and what does it imply about a system?
It shows that a system is organized at multiple temporal levels, none of which is unique when assessed over large time periods but at any instance some temporal scale dominates, and that the pattern at each time point is a function of the past history of activity. Interestingly, the brain seems to generate these dynamics from a finite number of discrete oscillators with a unique, asymmetric relationship between them: slow oscillators affect faster ones but the reverse relationship is much weaker. We learned about the properties of 1/f systems from other disciplines but appreciation of these features in the brain is quite recent. These properties imply that e.g., cortical networks can be ‘sensitized’ to environmental inputs with extreme efficacy but this tuning depends strongly on the self-organized (ongoing) brain activity. At the same time, the dynamics can shift transiently to a dominant oscillation which, in turn, allows for precise timing and, therefore, prediction of events.
5. In the Science review of your book, Pascal Fries noted that Hungarian neuroscientists were highly represented. You, in fact, became an honorary member of the Hungarian Academy of Sciences in 2001. Is there a special emphasis on the study of neuroscience in Hungary, and if so, what is the reason?
Admittedly, part of it is just cultural chauvinism. But as in any self-organized system, weak links can have large effects. If you read the recently published ‘Martians of Science’ you will realize that these five men of physics were as diverse as any five can be. But they were linked by similar experiences: E.g., they were forced to change countries multiple times, shared an exceptional degree of enthusiasm about science, and it also helped their interactions – and relative isolation from others – that they were fluent in Hungarian and much less so in German or English. The result of mutual information exchange among them might explain why the knowledge they generated as a group exceeded so much the sum of their individual contributions.
But even if you are aware of this truism, you do not rationally form networks. You just happen to be in one of the participants in a spontaneously emerging web of links. In my early life, high school and science education was strong in Hungary under the communist regime and there were limited channels of communications with the West. Hard sciences were all strongly linked to the military technology of the Soviets. Neuroscience (apart from Pavlovianism) was a new and non-partisan field. After the war, only two individuals (János Szentágothai and Kálmán Lissák at the University of Pécs) in the entire country had the necessary connections at home and abroad, a unique protection from the political system and the personal charisma to form active neuroscience groups. These seeds attracted all motivated students who wanted to carry out brain-related research and have become parts of the same web. What also helped our generation is that by ending up living in different countries and continents we were not competing for the same limited sources of funding so we could ‘afford’ to share some complementing views and technical abilities.
6. Your discussion of the brain’s first rhythm could make one feel that we are close to understanding when meaningful cognition begins. Does your knowledge of EEG patterns and their underpinnings influence your thinking about beginning-of-life, end-of-life, or even animal rights debates?
I believe that cognition begins once the 1/f features of cortical rhythms emerge because this dynamics represents global (i.e., distributed) computation and only structures with these features appear to generate conscious experience. The ontogenetic appearance of 1/f dynamics coincides with the emergence of long-range cortico-cortical projections. In the newborn human the 1/f global feature of the EEG is already present. On the other hand, in preterm babies, depending on the gestation age, long seconds of neuronal silence alternate with short, spatially localized oscillatory bursts (known as “delta brush”), like in sharks and lizards. These localized intermittent cortical patterns in the premature brain, and similar ones in the strictly locally organized adult cerebellum, cannot give rise to conscious awareness, no matter the size. From this perspective, the structure-function relations between the small world network-like features of the cerebral cortex and the resultant global rhythms appear as necessary conditions for awareness. Earlier developmental stages without these properties simply do not have the necessary ingredients of the product we call cognition.
7. You’ve suggested that sleep disturbances associated with psychological disorders might be a cause rather than a symptom. Aside from the disturbance of circadian rhythm, do you think differences in the magnitude or frequency of other oscillations could be at the root of any particular psychological disorders?
Timing and network synchronization are the essence of all cortical computation, and the timing ability of cortical networks is reflected in the rhythms they produce. We have shown that deterioration of synchrony of hippocampal assemblies, e.g., induced by the active ingredient of marijuana, is reflected quantitatively by the field rhythms. In turn, the degree of impaired hippocampal oscillations is correlated with the deterioration of memory performance. Alterations of gamma oscillations observed repeatedly in schizophrenic patients may also reflect impaired assembly synchronization. Oscillations constitute a robust phenotype that reliably ‘fingerprint’ an individual and expected to alter in most psychiatric disorders. Often such changes are most pronounced in sleep.
8.What paper or presentation has most impressed you in the past 6 months, and can you explain why?
Jan Born and my ex-postdoctoral fellow Lisa Marshall from Lubeck, Germany reported in November that by applying weak electrical fields through scalp electrodes at 0.75 Hz during slow wave sleep enhanced the retention of hippocampus-dependent declarative memories in student volunteers. They speculated that the effect is due to the enhancement and regularization of slow (< 1 Hz) cortical oscillations. Since we have shown earlier that the cortical slow oscillations can trigger hippocampal sharp waves and possibly determine the neuronal content of these events, their findings provide support for the active role of these sleep patterns in memory consolidation. This is good news for us, of course. However, what fascinates me most about the work is that such a weak stimulation was able to entrain a cortical oscillator. The effect of the stimulus-induced electrical field in the brain must be extremely weak since the current is strongly shunted by the parallel resistance of the skin, subcutaneous tissue and cerebrospinal fluid. If you had asked any able biophysicist (or me) whether such an experimental plan would make sense, they would have told you that it would never work. Yet, if the finding is confirmed, it is a perfect demonstration that oscillators can indeed synchronize at an extremely low cost of energy that may not exert any measurable effect on anything else. It also implies that the electrical fields produced by the synchronously active neurons may exert a temporal constraint to the same population that gave rise to the field. The broader implication of this study is even more exciting. The effect of stimulation on memory retention was detected in young students who have large amounts of slow oscillations during sleep. However, the power slow of oscillations decreases rapidly after forty years of age. Thus, in individuals like me the density of slow oscillations is quite low and their effect on memory consolidation, therefore, must be quite limited. The cheap and simple method of electrical field-induced entrainment may revert sleep patterns to the young adult form with the hope that the induced field effects can bring about even larger improvement of memory compared with young subjects. So I can become as smart again as my postdocs and students. Isn’t this fascinating?
9. You seem to endorse Mountcastle’s idea that the cortex is relatively homogenous and “uniformly organized”. You do this by suggesting that Mountcastle’s claim is well supported, and by pointing to various features like the scale-free and small-world nature of cortical networks. However, the book also spends a lot of time discussing the diversity of interneuron types, and various other details that can make the cortex seem very heterogenous. Can these characterizations be reconciled?
Many years ago, while I was a postdoc in Canada, a friend of mine of Chinese origin planned to visit Europe. I prepared a list of things for him to see in Budapest and Vienna. When he returned from the trip, I learned that he spent only a few hours in Vienna and headed back to the train station after concluding that Vienna was just the same as Budapest. Back then I was shocked by his statement as would be any citizen of either Budapest or Vienna. This story nicely illustrates the important point that boundary problems, typically reflected by our terms of similar and different or integration and segregation, in fuzzy systems like the brain are hard to define because the boundaries can dynamically shift depending on function. Numerous scientists are interested in the common or similar features of cortical circuits and computation. E.g., the similarity of cortical computation in the visual, auditory and somatosensory cortices is probably more striking than the differences among these regions. Others look for differences and keep finding them. Thus, the issue of functional integration and segregation always depends on the context and perspective. With regard to the rich family of interneurons, their diversity seems to be quite similar in all parts of the cerebral cortex.
10. If you were back at the undergraduate or graduate level, would you choose the same course of study or would you make different choices now that you’re older and wiser?
We do not quite understand where our curiosity and motivation come from. Occasionally, we dream up an ideal life with constant happiness and success but attempts to define universal happiness and success always fail. Even if I confine your question to the â€œmost effective road to systems neuroscienceâ€™, it is hard to make up an ideal curriculum. Perhaps, I wish I had learned more math and engineering, and got exposed to a world-class laboratory environment from the beginning. But whereas possession of tools is useful in answering questions, the critical factors in science seem to relate to asking an important question and building up a sufficiently intense motivation to solve it. Living in a suppressive regime at the time when my interest in the brain emerged made me focus on inhibition. This may not have happened under other conditions. Hardship and failure can be as formative of character and creativity as a barrage of positive feedback and supportive advisors.