TL;DR
This study reveals that mouse posterior parietal cortex (PPC) activity during visually-guided navigation is primarily driven by the animal's spatial position and heading, rather than abstract decision variables. These simple spatial attributes accurately predict neural activity and the formation of choice-selective sequences, suggesting PPC's core role in representing the animal's trajectory.
What Problem Does This Paper Address?
The posterior parietal cortex (PPC) is a brain region known to be involved in a variety of functions, including sensorimotor integration, movement control, decision-making, and spatial navigation. However, the precise role of PPC has been debated, with different research groups emphasizing different aspects of its function. Some studies highlight its involvement in motor control, others in decision-making based on visual cues, and yet others in spatial navigation. This paper aims to reconcile these views by investigating PPC activity in mice performing a virtual reality task that integrates visual decisions with navigation. The central question is whether PPC neurons encode abstract decision variables, visual information, motor commands, or a combination of these, or if simpler spatial representations are sufficient to explain their activity.
Methods
The researchers trained head-fixed mice to navigate a virtual reality maze using visual cues to make decisions. The mice ran on a ball that translated their movements into virtual locomotion. During this task, the activity of hundreds of PPC neurons was recorded using two-photon calcium imaging. To analyze the neural data, they developed models that predicted neuron activity based on various factors:
- Position-Heading Model: This model posits that PPC neuron activity is a function of the animal's spatial position (z) along the virtual corridor and its heading angle (θ).
- Position-Decision Model: This model considers the animal's position (z) and its final choice (d) (left or right).
- Combined Models: Models incorporating combinations of position, heading, and decision were also tested.
They used cross-validation techniques to compare the predictive power of these models, assessing how well they could explain the observed neural activity. Additionally, control experiments were conducted where mice passively viewed visual stimuli (playback condition) to isolate the role of visual input, and models based on motor variables (angular and linear velocity) were also evaluated.
Datasets
The study utilized data collected from seven mice performing the virtual navigation task. The behavioral and two-photon imaging data from these experiments have been deposited in the Dryad Digital Repository and are publicly available. No new benchmarks were introduced; rather, existing methods for neural data analysis and modeling were applied.
Key Insights and Ideas
- Spatial Representation Dominates: The most striking finding is that simple spatial variables – the animal's position along the corridor and its heading angle – are remarkably powerful predictors of PPC neural activity. This suggests that PPC's primary role in this context is to represent the animal's current location and orientation within its environment.
- Decision is Secondary: While PPC neurons do exhibit choice-selective activity (firing differently for left vs. right choices), this selectivity can be largely explained by the animal's heading and position, which naturally differ depending on the intended choice. Explicitly including "decision" as a predictor offered only marginal improvements over models based solely on spatial attributes.
- Trajectory Encoding: The formation of "choice-dependent sequences" observed in PPC, previously thought to reflect abstract decision-making, can be understood as a consequence of PPC encoding the animal's navigation trajectory. Different trajectories lead to different position-heading combinations, which in turn activate specific neuronal populations.
- Vision and Motor Control are Less Predictive: Models based purely on visual stimuli (during a playback condition) or motor variables (like velocity) were significantly less effective at predicting PPC activity compared to the spatial models. This suggests that while vision provides the input for decisions, PPC's internal representation is more focused on the spatial consequences of those decisions.
Key Results
- The position-heading model explained a significant portion of PPC neuron activity, with median correlations between model predictions and actual neural data around 27% (cross-validated).
- Adding "decision" as a predictor to the position-heading model improved correlations by only 0.4%, while adding "heading" to a position-decision model improved correlations by 6.1%, highlighting heading's greater predictive power.
- Models based on visual stimuli or motor variables performed worse than the position-heading model.
- A population decoder, using the recorded PPC activity, could accurately predict the animal's position in the virtual environment and its trajectory, demonstrating the richness of spatial information encoded in PPC.
- The decoded trajectories also allowed for accurate prediction of the animal's final choice, with predictability increasing as the animal progressed through the corridor.
Limitations and Future Work
The authors acknowledge that their spatial models might covary with visual and motor factors, as position and heading are influenced by both. While control experiments (playback, motor models) helped to disentangle these factors, it's difficult to completely rule out subtle influences. They suggest that future work should design tasks specifically to disentangle these correlates more precisely. Additionally, they note that their findings in mice, while informative, may not directly translate to the more complex parietal cortex in humans. The study also highlights a potential discrepancy with previous findings (Harvey et al., 2012), suggesting that task differences might account for the differing emphasis on decision versus spatial representations. Future research could explore these differences further.
Summary updated