Sex differences in exploration–exploitation strategies during home-cage decision making

Reviewer #1 (Public review):

Summary:

This paper examines potential sex differences in the conflict between exploitation, pursuing food and rewards in previously-associated locations/paradigms, or exploration of new locations that might result in better outcomes. Dysregulation of this conflict may be an underlying behavioral modality of psychiatric diseases. They used four distinct tasks: a two-armed Bandit 100:0 task, a standard fixed ratio 1 task, a two-armed Bandit 80:20 task, and a closed-loop economy PR1 task that allows for the assessment of motivational breakpoint.

Male mice show significantly higher accuracy under conditions of high probability known rewards, sticking with an action that just resulted in a reward or "win-staying". This was demonstrated in multiple paradigms, and there was a predictive nature of this behavior that could predict animal sex with modest accuracy. Under probabilistic environments, males were no longer more accurate than females but still used a higher win-stay strategy. A closed-loop PR1 task showed that there were no inherent differences in motivational breakpoint between sexes. Finally, the authors use simulations to determine an appropriate number of animals needed to detect these differences.

Strengths:

The manuscript attempts to resolve inconclusive sex differences that have heretofore been neglected or inconclusive due to insufficient power. The most impressive aspect of this paper is its scale, assaying 62 female mice and 74 male mice in identical exploration-exploitation tasks using high-throughput and noninvasive operant feeding via FED3. Very few labs can achieve this scale, which is necessary to detect sex differences with a small effect size.

The authors use some sophisticated modeling approaches and analysis of data from the 136 mice to investigate the significance of these sex differences and interrogate other conditions. They also use simulations to model the likelihood of replicating these differences given a sample size. This is extremely helpful for other researchers as they consider sex as a biological variable.

Weaknesses:

The study is largely descriptive in nature and does not pursue any mechanism of the underlying differences, like hormones, neuromodulators, or circuits. The lack of estrous cycle tracking is acknowledged as a limitation.

Reviewer #2 (Public review):

Summary:

Murrell and colleagues examine sex differences in mouse decision-making tasks, using the FED3 device, which allows for continuous data collection in the home-cage. Mice performed four tasks across two weeks, which provided all of their food. Across tasks, male mice were more likely to repeat a rewarded choice than females, which benefits decision accuracy in deterministic tasks. This work complements existing results for decision-making differences in males and females, affirming that this domain of cognition is particularly sensitive to sex differences. However, there are some specific features of the FED3 device, such as single housing, closed economy feeding, and 24-hour access that can uniquely influence decision-making in a (likely) sex-dependent manner that encourage considering these data as examining sex differences in a particular context, rather than as a generalized finding. At the same time, these data could offer new insights about nuances of behavior like circadian rhythms or bout analysis, uniquely enabled by the extended availability of the FED3 devices. The analyses in this paper also make an important point, encouraging researchers to use methods that allow for much larger N's to provide clearer and more robust results.

The FED3 devices are an innovative new way to approach behavior, and have allowed the authors to test many dozens of mice in a battery of tasks, over which they see similar patterns of increased win-stay behavior in male C57b6 mice (wildtypes from several knockout lines). The authors point out that there are discrepancies in prior literature across tasks and species in terms of how sex differences influence decision making, but there are some particular ways that sex differences could interact with the FED3 devices that it would be interesting and important to consider further. In particular, the fact that the animals live with the device, singly housed, may be an underrecognized contributor to sex differences. Changes in social interaction and dominance arising from long-term single housing are very likely to impact males and females differently, for example.

Continuous data collection is a fascinating way to look at learning and decision-making, but it also raises interesting questions about whether these dynamics are impacted as a function of continuous access to the device. In addition to summary metrics over the whole task, it might be valuable to look at learning across the task each day, and within circadian periods of each day. For example, it seems based on the example sessions for the 100-0 bandit task that animals take at least a few reversals to learn the task structure. How many trials does it take for a male or female mouse to reach some criteria of success? Do the sex differences exist at all time points? Does the light cycle affect the accuracy or trial counts? There are numerous such analyses that could particularly inform future use of the FEDs across laboratories, and identification of similar or distinct patterns of sex differences or behavior in other apparatuses, and would be a benefit to the field.

The authors employed several computational techniques to identify parameters or features of behavior that might explain the sex differences they observed, and this is a strength of the manuscript. However, the win-stay lose-shift agent may not be an ideal match to make conclusions about exploitation, as it is unclear how win-stay and lose-shift strategies map onto explore/exploit tradeoffs. If an animal were exploiting an option, they may win-stay *and* lose-stay, if the task is probabilistic. Indeed, the model fit is weaker for the 80-20 bandit, suggesting this model may not reflect the actual strategies mice are engaging in, potentially in both bandit tasks. This point is particularly worth considering in light of the criteria for shifts in the tasks being based not on the number of trials completed, but on the number of pellets earned. When shifts are tied to reward collection rather than trials, it can amplify differences in behavior driven by reward consumption.

Reviewer #3 (Public review):

Summary:

The manuscript by Murrell et al. describes a high-throughput approach for evaluating food foraging strategies in mice. Building on their prior publication describing the technical aspects of the Feeding Experimentation Device 3 (FED3), this study demonstrates the utility of the FED3 in evaluating decision-making in mice. The authors identify key differences in male and female foraging strategies that could not be accounted for by total food consumption or overall food motivation. Given the rapid adoption of the open source FED3 platform, this work is likely to be of broad interest and utility in the field.

Strengths:

(1) The use of cost-effective, open-source devices like FED3 provides substantial value to the scientific community. Validation of appropriate conditions for using this equipment is an important step toward broad adoption.

(2) The authors implement a simple but elegant experimental design for studying food-motivated decision-making behavior. This approach could be applied to a wide range of preclinical disease models in future studies.

(3) The study is well-powered to evaluate sex as the primary experimental factor (62 females, 74 males), allowing the authors to make convincing claims about differences in strategy. Additionally, the dataset provides a useful benchmark for power analyses in future studies involving more complex experimental designs.

(4) The figures are clear and generally easy to interpret the primary findings.

(5) The conclusions are appropriate and not overstated

Weaknesses:

(1) A major strength of this study is the potential utility for new investigators trying to implement cognitive behavioral tasks in mice. However, the present version provides limited background on the rationale for selecting the bandit task and on prior work applying it in similar contexts. Including additional background and discussion would better contextualize the approach for other groups considering adopting it in their own studies.

(2) Some methodological details surrounding the initiation of the experiment could be clarified. Specifically, it is unclear if mice transitioned directly from standard housing conditions (group housed, standard chow) to the study conditions (single housed, FED3-based probabilistic learning), or intermediate acclimation/training steps were used, such as autoshaping, free access to new food pellets, or FR1 training. A more detailed experimental timeline (for example, see Figure 1 from PMID 39710132) would address this concern.

(3) The authors evaluated multiple probabilistic conditions (100%, 90%, 80%, 70%, 60%), but ultimately focused on the 80% condition for this study. A more detailed explanation for how this conclusion was reached would be useful for future researchers working under different experimental conditions (i.e., age, strain, genotype, disease model) where other probabilistic conditions may be more appropriate.