Task and Behavior-Related Variables Are Encoded by the Postrhinal and Medial Entorhinal Cortex During Non-Spatial Associative Learning

Reviewer #1 (Public review):

Summary:

Nysten et al. use in vivo 2-photon calcium imaging in behaving mice learning a visual associative memory task to understand how neural dynamics in the postrhinal cortex and medial entorhinal cortex evolve over task learning and through reversal learning. Using a combination of analyses to measure trial-averaged neural responses, regression models, and population decoding methods, the authors argue that both POR and MEC dynamics evolve over learning, with relatively more neurons in MEC becoming responsive. The impact of this study comes from comparing neural dynamics across multiple medial temporal lobe circuits to show how different aspects of task structure are differentially encoded. Below, I have listed several major concerns that need to be addressed to ensure the findings are robust.

Strengths:

(1) The study employs a well-controlled behavioral paradigm alongside powerful cellular-resolution two-photon imaging, enabling high-throughput recordings of hundreds of neurons simultaneously in deep brain structures.

(2) The simplicity of the task allows for a detailed examination of learning dynamics across multiple stages, including early and late learning in the main task, as well as during reversal learning.

(3) The use of sophisticated analysis methods to compare and contrast learning dynamics in large neuronal populations strengthens the study, though additional steps are needed to ensure their robustness (detailed below).

(4) Two-photon imaging enables the investigation of functional topography, further supporting previous findings of functional clustering in MEC across different task and behavioral domains.

Weaknesses:

(1) GLM Robustness & Behavioral Attribution: The current GLM design may misattribute neural activity by lacking appropriate time lags for velocity and not accounting for distinct neural states (e.g., rest vs. run). Given MEC's known speed-invariant coding, the observed decrease in speed-modulated neurons may be an artifact rather than a true learning effect. Additionally, gradual behavioral stabilization over training could influence neural dynamics in ways not fully accounted for.

(2) Licking vs. Movement Encoding: The increase in lick-modulated neurons raises questions about whether these neurons encode reward anticipation or motor execution. Without a detailed analysis of error trials and the timing of licking vs. movement adjustments, it remains unclear whether MEC activity reflects predictive coding of reward or simply motor feedback.

(3) Clustering Interpretation Issues: The functional clustering approach does not control for correlations between behavioral features, making it difficult to determine whether speed modulation plays a role in cluster assignments. The anatomical analysis in Figure 6 relies heavily on clusters that may be predominantly defined by a single regressor, requiring further clarification.

(4) Data Presentation & Statistical Support: Some key claims, particularly the increase in task-modulated neurons with learning (Figure 3), lack statistical quantification.

Reviewer #2 (Public review):

Summary:

The authors examine medial entorhinal cortex (MEC) and postrhinal cortex (POR) responses using Ca imaging during a non-spatial, Go/No-Go visual association task. The authors specifically consider whether MEC encodes stimulus information, as previously seen and hypothesized in POR, as well as other task elements such as reward, and whether and how these responses evolve with learning in both regions. The authors find that, in general, POR encodes task-related information more strongly compared to MEC. In particular, POR encodes the stimulus even before the animal reaches expert performance, whereas MEC shows considerably weaker stimulus encoding that emerges with learning. Both regions also display licking-related coding, although notably this activity reflects choice or licking-preparation, which emerges with learning. Further, despite its overall reduced coding, MEC exhibits greater anatomical clustering of cells with similar functional properties compared to POR.

Strengths:

These data are generally well-presented, both in the description of the experimental paradigm - which is simple yet highly informative - and in the individual results for each section. A major strength is the dataset, which includes many cells, including a subset that are tracked across learning. I found the core findings - (1) that POR has robust stimulus encoding while MEC develops weaker stimulus information with learning, and (2) that both POR and MEC exhibit an increase in lick-modulated cells, although POR has more, and stronger, lick-modulated cells - to be generally well-supported by the data presented. The general question of whether and how MEC encodes non-spatial task-relevant features and how these responses (if they exist) emerge with learning is of general interest. In addition, how MEC activity contrasts with activity in an upstream region, thereby indicating what information MEC gets and what it does with it, is also of broad interest.

Weaknesses:

I perceived two primary weaknesses.

The first was that it was not entirely clear to me what was expected of MEC and POR responses, and whether the observations the authors made were surprising or entirely in line with what would've been predicted based on prior work. In some ways, the results seem expected - POR had visual signals, MEC had few visual signals but some reward signals.

The second is that it took me a long time to extract what I perceived to be the core results of the paper, and in some places, it was a little hard for me to understand all the analyses and results together as one cohesive step forward in our understanding of MEC and POR coding properties.

I think this was most evident in the results presented in Figure 4. Up until Figure 4, it seemed to me that the core results were:
(1) visual (stimulus information) is present in POR responses from very early learning, whereas weak stimulus information develops in MEC with learning, and in both cases, there is a preference for the plus stimulus.
(2), both POR and MEC show an increase in lick-modulated cells with learning, although more cells encode licking at all stages in POR.

This is nicely summarized in my view by Figure 3e. However, I became confused when Figure 4 entered the picture. Here, it seems that by far the most predominant coefficient in the model is the lick response, with stimulus features playing a smaller role - specifically, at the end of learning, 60% of POR cells were characterized as predominantly lick/non lick, compared to 25% defined by their coding to the stimulus. I can appreciate that there might be nuances to these and previous analyses such that all the results sit cohesively together, but I think that needs to be clarified.

A second example - Figure 2b - shows that many (75%) of MEC neurons seem to be selectively active for the plus stimulus, but when doing the GLM analysis with the plus stimulus (and reward/licking) as features, many fewer neurons (35%) are determined to be encoding task information. It was not clear to me what was contributing to the discrepancy between these two results - is it that MEC activity often increases with learning, but doesn't increase by that much?

I think in general this can be helped by specifically pointing out how the results of these different analyses relate to each other, including specifically mentioning where the results might seem unaligned (at least on the surface).