Emergence of stable sensory and dynamic temporal representations in the hippocampus during working memory

Jiannis Taxidis, Eftychios Pnevmatikakis, Apoorva L Mylavarapu, Jagmeet S Arora, Kian D Samadian, Emily A Hoffberg, Peyman Golshani  

Posted on Biorxiv,  November 20, 2018

https://doi.org/10.1101/474510

Note: Below is our review of the Biorxiv preprint article by Taxidis and colleagues, "Emergence of stable sensory and dynamic temporal representations in the hippocampus during working memory." There has recently been an interesting discussion about best practices in open peer review. Accordingly, we decided to reach out to the authors with our review prior to posting online. The authors were extremely receptive and engaged and responded to our review. Below our comments are in plain text and author responses are in boldWe are grateful to the Golshani lab, and in particular Jiannis Taxidis, for engaging in open dialogue. 
Summary
In this manuscript, Taxidis et al. recorded sequential activity in CA1 during a delayed olfactory non-match-to-sample task in mice over multiple days. They describe two populations of cells active during the task. One population, termed “odor-cells”, are active during odor delivery and another, termed “time-cells” are active at specific time intervals in the intervening delay between two odor presentations. They showed that classifiers trained on these cells can reliably decode the presented odor and also the time interval on the delay, suggesting that items might be held in working memory by these cells. They also found that odor cells generally are reactivated more across days than time cells, and that temporal and odor information can be reliably extracted (by Bayesian classifiers) up to 5 days later. In additional experiments, the delay was elongated and the authors found that odor cells generally retained their firing field whereas time cells shifted their fields. Over behavioral training, the number of time cells increased as well as odor decoding and temporal interval decoding fidelity, which all correlated to behavioral performance. On the other hand, the number of odor cells stayed relatively constant. Lastly, mice exposed to a passive version of this task with no memory demand showed far fewer time cells and no change in these metrics over days.
 
The manuscript is a technical tour de force involving complex behavior, learning data, and sophisticated analyses. The results add valuable knowledge to how the hippocampus produces and maintains neural sequences. Particularly interesting is how these sequences evolve while the mice learn the task. Overall, we are excited about this impressive body of data and believe that it would greatly enhance our understanding of encoding mechanisms in the hippocampus. We have several suggestions, though some may have been addressed in the supplemental figures, which we were unable access through the preprint servers.