Students' Research Circle    
» 2022
Call for papers
The conference
Veterinary Session
Home » Archive » 2022

TDK conference 2022

The properties of environmental representations in the hippocampus
Kelemen Atilla Botond - year 3
University of Veterinary Medicine Budapest, Department of Ecology
Supervisors: Balázs Ujfalussy, Eszter Berekméri


It is well-established that the hippocampus is critical for successful completion of spatial memory tasks and that hippocampal pyramidal neurons show location dependent activity. However, it is not known how the hippocampal code adapts to changes in the environment to enable flexible behavior. Here we analyzed data from two-photon Ca2+-imaging experiments from head restrained mice, running to collect water rewards in virtual corridors under two experimental setups, recorded in the Laboratory of Neuronal Signaling (KOKI). In the first experiment one out of the two virtual corridors had a water reward in it, and the animals were imaged during learning. While in the second experiment there were again two corridors, but this time both had reward in it, at different locations, and the animals were expert at the task. Our aim was to understand how hippocampal neuronal population encodes the external variables relevant in various tasks. Specifically, we wanted to test whether the representation of the position is specific to each corridor, or some aspects of the code is shared across different contexts? We were also interested understanding how the decodability of position and corridor identity change during learning. We applied deconvolution and temporal smoothing on the recorded Ca2+ signal and divided the position into discrete bins. For decoding position or corridor identity, we binarized the inferred spike data and used a static Bayesian decoder assuming Bernoulli likelihood (SBB) with 10-fold cross validation and downsampling. For dimensionality reduction we used principal component analysis (PCA) and Isomap. We found that at the beginning of learning the identity of the corridor could not be decoded, while as performance increased the representation of the two corridors separated and could now be decoded with relatively high accuracy (0.9). We also found that in the rewarded corridor position decoding accuracy increased with learning, but only near the reward zone. Importantly, in the unrewarded corridor decoding accuracy increased as the animals got better at the task. In the second experiment both corridor identity and position could be decoded. However, even after the performance of the animals was constant decoding accuracy kept increasing with experience. To test the generalizability of positional mappings across context we used a decoder trained in one of the corridors to decode the position in the other corridor. We saw that in the first experiment decoding error increased with learning. In the second experiment we found that the relative distance of the animal from the reward zone could be accurately decoded. First, we conclude that during learning the representation of different environments diverges, the accuracy of the positional code flexibly adapts to the task. Second, we conclude that changes in the hippocampal code still occur in expert animals. Furthermore, parts of the positional mapping generalize across different corridors.

List of lectures