PerspectiveNeuroscience

Neural representations across species

See allHide authors and affiliations

Science  29 Mar 2019:
Vol. 363, Issue 6434, pp. 1388-1389
DOI: 10.1126/science.aaw8829

A plethora of studies in rodents have described spatially tuned neurons, including place cells in the hippocampus and grid cells in the medial entorhinal cortex (MEC), suggesting a crucial role of the hippocampal formation in spatial navigation (1). Human studies have, in turn, shown that the hippocampal formation is involved in declarative memory (memories of facts and events) (2). What, then, is the function of the hippocampus? Is it involved in memory or in spatial navigation, or does it have a more general function that encompasses both? Several studies have shown that place cells remap, changing the location at which they respond, following geometrical changes in the environment, and that they can be modulated by nonspatial factors, according to the animals' specific tasks (3). These findings highlight the coding of cognitive components, challenging the notion that place cells represent an invariant spatial map. Although grid cells also realign with physical changes in the environment (4), the geometrical structure of their fields has been considered to provide a more invariant spatial representation. But do grid cells encode cognitive aspects as well? On pages 1443 and 1447 of this issue, Boccara et al. (5) and Butler et al. (6), respectively, show that this is the case, providing compelling evidence that grid cells are modulated by reward location. Complementing these studies, Baraduc et al. (7) showed that variations in spatially tuned neurons in the monkey hippocampus are accompanied by a “schema” of the task. Together these studies provide insights on how neurons in the hippocampal formation go beyond purely spatial representations and are modulated by cognitive factors.

Boccara et al. trained rats to search for three hidden rewards in a cheeseboard maze, while recording neural activity in the hippocampus (area CA1) and MEC. Critically, the location of the rewards was changed from one day to the next, thus varying the cognitive valence of local points in the environment. They found that the place of CA1 neurons' firing accumulated around the reward locations and that a vast majority of grid cells also altered their mapping, moving toward the rewarded areas and distorting their characteristic grid-like pattern of responses. Moreover, the shifts toward the goals progressed during learning and were more transient in CA1 compared to MEC, suggesting a more stable, though still malleable, representation by grid cells compared to place cells. Furthermore, analysis of the temporal structure of such transitions showed that, rather than smooth changes with intermediate states, there was a flickering between the competing representations.

In line with these findings, Butler et al. show alterations in the spatial representation of MEC grid cells (and other spatially tuned cells in this area) when rats performed different tasks in two arenas: In the first, they foraged for scattered food rewards, whereas in the second they navigated to a remembered reward location after hearing an auditory cue and also foraged to scattered food rewards in-between trials. Grid cells reorganized between environments, showing shifts in the grid patterns as well as rotations and changes in their spacing and geometrical features. By contrast, only translation differences were observed when random foraging was performed in both environments, indicating that the modulations were mostly due to the different cognitive demands. As in Boccara et al., the activity of spatially tuned cells increased near the reward zone.

Together, these two studies reveal that task-dependent factors can modulate the responses of spatially tuned cells in the hippocampal formation, going beyond modulations linked to geometrical changes that could, in principle, be attributed to a distorted perception of space. Therefore, both place and grid cells encode nonspatial cognitive components integrating contextual features into spatial representations.

Baraduc et al. performed recordings in the monkey hippocampus while the animals navigated a star-shaped virtual environment and searched for a hidden reward location signaled by the relative position of visual landmarks. Once the animals became familiar with the environment, they were also tested in new environments, containing landmarks that were changed every day. As in the studies by Boccara et al. and Butler et al. with rats, Baraduc et al. found a remapping of the spatial hippocampal representations, but a subpopulation of the spatially tuned neurons showed an invariant activation when realigned to reward location. Moreover, the correlation between the realigned familiar and the new maps increased with learning, and animals needed weeks of training to reach high proficiency levels in the initial, familiar environment, but only a dozen trials to reach such levels in the new ones. Besides the initial training to pay attention to the screen, navigate using the joystick, and so on, this effect can be attributed to learning a schema of the task that is generalized across environments. These “schema cells” then provide a high-level conceptual knowledge of the task structure, which is invariant to visual features and is reminiscent of concept cells in the human hippocampal formation—neurons that fire selectively in response to specific concepts (i.e., different pictures and even the name of a person) and have been proposed to be involved in declarative memory (8).

Neuronal coding in the hippocampal formation

Rodents demonstrate a conjunctive coding to represent context, with neurons firing in response to specific places AND in specific conditions. Humans show an invariant coding, with neurons firing in response to concepts in one OR another condition. Monkeys display a mixed coding, with neurons being modulated by context but also showing an invariant representation.

GRAPHIC: N. DESAI/SCIENCE

These findings in rats and monkeys show modulations of spatial representations in the hippocampal formation given by the reward locations. These modulations encode cognitive aspects that provide contextual information about the experience of the animals in each environment, in line with the memory function attributed to this area based on human studies.

It has been argued that the hippocampal formation has a general “relational memory” role, linking together the elements of experiences (3). Within this framework, spatial location is one of several components that constitute a memory, which is prevalent in rodents, given the behavioral importance of knowing their precise location and routes to reach safety and that they acquire information about the environment through exploration, whereas primates rely mainly on vision and eye movements to explore and navigate their surroundings. Although represented differently, rodents also have notions of concepts—a cat is a cat, and cheese is cheese—and humans also have spatial representations that enrich their memories and help avoid interference (for example, I do not confuse the conversation with a colleague in my office, with another one we had at a conference).

A particular location in space, represented by place cells or by other spatially tuned neurons, can therefore be considered a concept that is associated with different experiences. Place cells and concept cells share some similarities, most notably their high selectivity, firing in response to precise locations and specific persons, respectively (8). There is, however, a major difference. Whereas place cells and spatially tuned neurons in rodents show a conjunctive representation and their firing is determined both by the particular location and specific conditions (such as the reward location), human single-neuron responses to concepts show an abstract representation with a much higher degree of invariance. Indeed, using a pair association paradigm, it was shown that neurons initially firing in response to a given person started firing in response to a location that was associated with them (9). But the response to the first person remained the same in different tasks and conditions: during passive presentation of the person, while subjects learned the associations, during recall, and when shown with the associated stimulus. In other words, whereas in the rodent hippocampal formation, the associations that comprise the memory of experiences are encoded by modulations (or complete remapping) of the activity of single neurons, human hippocampal neurons fire in response to a concept irrespective of any particular association with it or the task at hand (i.e., neurons did not fire in response to a person AND a place, or a person AND a task).

The hippocampal formation in rodents can be considered as encoding contextual associations by performing a logical AND function (responding to a particular place AND a particular reward), which tends to orthogonalize memories and might be ideal to avoid interference, whereas in humans the computation seems to be an OR function (responding to a concept in one OR another condition), which, in turn, might be ideal for generalization and fast learning when changing context. In this case, the coding of associations and context is not represented at the single-neuron level but is given by the coactivation of invariant assemblies representing each concept. The representation in the monkey hippocampus seems to fall between these two cases, with spatially tuned neurons being modulated by the task but also showing an invariant high-level representation of it (see the figure). Future experiments and computational models should give more insights into the function of the hippocampal formation in spatial and nonspatial processing. Memory functions, such as the encoding of context and associations, seem to be implemented with different coding strategies in rodents, monkeys, and humans—something that could also be attributed to different types of neocortical inputs and that may eventually explain the different cognitive abilities of these species.

References and Notes

Acknowledgments: I thank E. Kropff and M. Okun for comments.
View Abstract

Subjects

Navigate This Article