Recent work has proposed that cognitive maps and their neural instantiations encode the predictive structure of experience. One fundamental implication of predictive theories is that the structure of cognitive maps depends not only on the spaces they map, but also on the way the agent explores those spaces. As a result, agents that explore the same space in different ways should have differently structured maps. Here, we leverage longitudinal miniscope calcium imaging of CA1 in mice navigating a multicompartment environment to test this prediction. We show that (1) different mice instantiate differently-structured maps, (2) this idiosyncratic map structure can be predicted from the navigational history of that mouse,  and even that (3) changing the navigational patterns of the mouse can change map structure as predicted. These results could be reproduced by a computational model which learns predictive structure over inputs tuned to environmental geometry. Together, these results demonstrate that the predictive structure of navigation is a key determinant of hippocampal map structure, consistent with predictive theories of cognitive mapping.

In previous work described below (Ajabi, et al., 2023; Nature), we found that experience with a rotating visual cue caused a recalibration of heading integration in the neural compass which persisted even after that cue was removed. This predicts that human heading judgements should also exhibit signatures of recalibration after experience with rotating visual spaces. Here, we use immersive virtuality to test this prediction, finding clear evidence of such signatures. These results provide important evidence of common cross-species mechanisms underlying heading integration.

Here we use chronic miniscope imaging to characterize the response of mouse CA1 to nine parametric deformations of a familiar square environment. Using the framework of Representational Similarity Analysis, we show that different mice exhibit common representational structure across spatial scales. Finallly, we compare this structure to the predictions of computational models, finding a subset of model parameterizations which are able to account for it. Beyond these insights, these data set the stage for future comparisons of representational structure across brain regions, species, and scales.

Here we show that the population dynamics of head direction lie on a neural manifold that includes an informative second radial dimension.  Variability along this dimension is correlated with the overall activity of head direction cells (the 'network gain'), is most pronounced during darkness, and bears signatures of previous experience with informative reference frames - be they static or dynamic. 

Here we show that although the CA1 spatial code for even the same environment changes over the course of weeks, it does so in a very particular way - one which preserve the relative representational structure of context.

Here we show that the firing rate of CA1 place cells is modulated by the recent history of the navigator, even in the absence of disambiguating sensory cues and explicit task demands. Moreover, we show that this firing rate modulation is causally mediated by the trisynaptic pathway (EC->DG->CA3->CA3).

Previous work has shown that the firing fields of place and grid cells are distorted in deformed environments, sparking a lot of debate about the representational content of these codes. Here we show that when a familiar environment is reshaped, the grid pattern undergoes dynamic shifts dependent on the recent history of the navigator, challenging previous interpretations of these data.

As a compliment to the work described above, we noted that the dynamic history-dependent shifts we observed in rodent grid cell patterns motivated an additional prediction - that spatial memory should also show such dynamic history-dependent shifts. Here, we demonstrated these predicted shifts in three separate experiments in humans, including when navigating in immersive VR. 

Previous work has shown that numerous organisms, from insects to toddlers, preferentially rely on the shape of the space around them when reorienting after becoming lost, even when other cues in the world are more informative. Here we show that this provocative behavior is reflected in the hippocampal maps of mice, which also preferentially reorient based on the shape of the surrounding space and predict the navigator's memory-driven behavior.