Funded Grants

Network mechanisms of memory formation

Imagine trying to figure out the plot of a movie by watching the top right corner of a wide screen TV standing one inch away from the screen. Under these conditions you will only be able to observe a few flickering pixels making it very hard to tell anything about the global shapes on the screen. If an analogy can be drawn between the brain and a TV screen, with the activity of individual neurons corresponding to the changes in color and intensity of individual pixels, then until recently studies of brain activity could only be done under conditions similar to the ones described above. Such studies have led to very important progress towards understanding the response properties of individual neurons. However, it is the coordinated interactions across networks of neurons that hold the secrets of information processing in the brain, just like the plot of the movie in the above example can only be deciphered by observing the spatial and temporal structure in the activation of multiple pixels.

There are three basic ingredients to making further progress in understanding the organization of network activity in the brain: (1) reliably isolate the activity of individual neurons (make sure we are observing single pixels, rather than an out of focus mixture of several pixels, thus obtaining a sharp image), (2) obtain stable recordings from the same neurons across long-periods of time (make sure we are observing the same pixels over time, thus obtaining a stable image), and (3) simultaneously record the activity of multiple neurons in multiple brain areas (track many pixels from many parts of the screen at the same time, thus obtaining a larger, more representative image). The recent development of the technique of tetrode electrophysiological recordings has made experiments that satisfy all three of the above requirements possible. Using this technique we plan to study the network mechanisms underlying memory formation, a cognitive process that is fundamental to shaping perception and guiding future action.

The natural focus of our studies is the hippocampus, a brain structure that acts as a memory making unit in the brain. This surprising fact was discovered in a famous clinical case, the case of patient H.M., who had bilateral removal of his hippocampus in order to relieve epileptic seizures. This hippocampal damage resulted in profound and surprising memory deficits which manifested themselves in two major ways: (1) No new long-term memories of facts and events could be formed after hippocampal damage, and (2) memories before hippocampal damage were affected in proportion to their recency with very distant memories completely unaffected. These effects of hippocampal damage were confirmed in numerous other clinical cases and experiments with animals and led to two main conclusions about the role of the hippocampus in memory formation: (1) The hippocampus is indeed critical for forming new long-term memories, and (2) memories are not stored in the hippocampus, since distant past memories are not affected by hippocampal damage.

If memories are not stored in the hippocampus, where are they are actually stored? The predominant conjecture is that memories are distributed throughout the cerebral cortex with different cortical areas holding different aspects of the memory trace. For example, when watching a movie, the colors and shapes on the screen generate specific activity patterns in your visual cortical areas, while the movie sound generates specific activity patterns in your auditory cortical areas. To make a memory of this experience these activity patterns across the different cortical areas would have to be selectively linked into a coherent whole (memory) that can later be retrieved as a unit and used to guide behavior or be linked to novel activity patterns. The process of linking of activity patterns to form distributed memory traces across cortical areas is orchestrated by the hippocampus.

Hence, understanding how the hippocampus interacts with cortical brain areas is the key to understanding the process of memory formation. The direct experimental study of these interactions has only recently become feasible through the technique of tetrode recordings. Using this technique, we will record the simultaneous activity of large numbers of hippocampal and cortical neurons during learning as well as during the sleep periods preceding and following experience. Our goal is to characterize the structure in the language of cortico-hippocampal communication: the basic alphabet and words in this language and the fundamental rules of the underlying grammar. In our previous experimental work we have already begun identifying some of the basic elements in this language. We want to build on our understanding of the structure of cortico-hippocampal interactions to characterize their role in shaping neuronal activity patterns, and their relationship to learning and behavioral performance. To understand the link between network interactions, memory formation, and behavior in a rigorous and quantifiable way we will concentrate on the study of simple associations between sensory stimuli, such as a tone and an eye-blink inducing stimulus, under conditions that can be experimentally precisely controlled. Depending on how the stimuli are arranged in time the learning of these associations can be independent from or critically dependent on the hippocampus. Hence by manipulating the temporal relationships between the stimuli we can systematically study the hippocampal involvement in the establishment of memories of these associations.

Critical for understanding complex systems such as the brain is the ability to combine quantitative observations of the system with controlled manipulations of the underlying circuitry. To this end we will combine large-scale recordings with both pharmacological and genetic manipulations. Rodents provide an extremely powerful system for combining these different levels of analysis, from the molecular and cellular, to systems, and behavioral. Characterizing the effects of pharmacological and genetic manipulations on the organization of network neuronal activity can have important implications for understanding the etiology of neurological disorders affecting memory, such as Alzheimer's disease.

Memory formation is at the core of the mechanisms of information processing in the brain, which appear to be drastically different from the mechanisms employed in our current theories of computation. In particular, in current digital computers there is a conceptual and practical separation between computation and memory. This makes storing information fast as it just consists of simply writing it to disk or moving it to memory chips. What becomes extremely difficult is maintaining associations among heterogeneous yet related types of information and retrieving them in a content and context dependent fashion. This is the reason why, while we have computers that beat grand masters in chess, we still can't build them to effectively recognize faces as well as a small child can. In the brawn associations and context based retrieval of memories and computation are almost automatic. This is because the circuits that do the computations are the very circuits that are shaped to hold the memories that we extract from experience with all their associations to our previous knowledge. This makes certain types of computations extremely fast, at the cost of making it non-trivial to insert new memories since they have to be gracefully integrated with previously stored information. Nature has evolved an impressive process for integrating new memories in cortical networks that is orchestrated by the hippocampus. Understanding this process will undoubtedly have profound influences on information technology and on our understanding of the nature of computation itself.