Structured navigation in a goal-directed task reveals flexible spatial coding

Technical Summary: Structured Navigation in a Goal-Directed Task Reveals Flexible Spatial Coding

Problem Statement
Current experimental approaches for studying the neural basis of navigation face a fundamental tradeoff between behavioral richness and experimental control. Traditional open-field tasks allow for natural, freely moving behavior but lack structured trials, repeated trajectories, and defined decision points, making it difficult to align neural activity with specific behavioral epochs or goals. Conversely, head-fixed virtual reality (VR) paradigms offer precise control over stimuli and repeatable trajectories but constrain natural movement, dampen vestibular feedback, and limit the range of expressible behaviors. Furthermore, recent evidence suggests that spatial representations in the medial entorhinal cortex (MEC) are state-dependent, changing significantly when navigation involves remembered reward locations compared to random foraging. There is a critical need for a platform that combines the ecological validity of freely moving behavior with the rigorous control of VR to study how neural maps support flexible, goal-directed navigation.

Methodology
The authors developed a novel behavioral and hardware platform designed to bridge this gap. The system integrates freely moving behavior with dynamic environmental control and high-density electrophysiology.

• Hardware Design: The core is a large, mechanically stable arena constructed from 80/20 aluminum with a satin-finish acrylic floor. This floor allows for optical bottom-up projection of visual stimuli via a short-throw projector mounted underneath. The arena features four translucent wells for liquid reward delivery, each equipped with infrared (IR) beam break detectors and internal LEDs for cueing. To manage the constraints of chronic recordings in a large space, the authors designed a motorized slip ring and a swinging arm suspension system. This setup allows rats to rotate freely and traverse the entire arena without cable tangling or excessive slack accumulation, facilitating long-term Neuropixels recordings.
• Software and Control: Task execution is managed by a central computer running the Linux Comodular Realtime Interactive Computation Engine (LiCoRICE). Real-time pose estimation is achieved using a DeepLabCut Live model with a MobileNet architecture, ensuring inference speeds exceed the camera frame rate to prevent behavioral lag. The system controls visual cues, auditory tones, and reward delivery with millisecond precision.
• Task Structure: To validate the platform, the authors designed a multistage spatial targeting task. Each trial consists of three phases:
  1. Targeting: The animal must locate and collide with a visual target projected onto the floor.
  2. Reporting: After a hold period at the target, the animal must navigate to a remembered reward well (indicated by a hint LED at the start of a block).
  3. Pre-initiation: The animal resets to an initiation well to start the next trial.
     The task enforces structured, repeated trajectories without physically constraining the animal, allowing for the collection of dense spatial data across defined behavioral epochs.
• Neural Recording: Chronic Neuropixels 1.0 probes were implanted in the MEC of three Long-Evans rats. Neural data were recorded during task performance, and units were sorted using Kilosort 2.5 and Phy.

Key Results

• Behavioral Performance: Rats successfully learned the multistage task, demonstrating high spatial coverage of the arena, particularly during the targeting phase. Trajectories were largely direct (low circuity), and animals showed the ability to perform memory-guided navigation to reward wells without continuous cues. Error rates were low, and the system successfully enforced task rules (e.g., beam break errors, timing out).
• Neural Viability: The platform supported stable chronic recordings from the MEC. The authors identified canonical cell types, including head-direction cells and spatially tuned cells (grid and non-grid cells), with tuning stability comparable to previous studies.
• Goal-Directed Coding: Population-level analysis revealed that MEC activity tracks distance to task-relevant goals. Multi-layer perceptron (MLP) models trained on neural data could predict the distance to the current target location and the remembered reward well.
  • Decoding accuracy for the distance to the report well remained low throughout the trial, with a notable drop in the middle of the reporting phase, suggesting the representation of the static reward location is maintained but may fluctuate.
  • Decoding accuracy for the distance to the target improved as the animal progressed through the targeting phase.
  • These findings indicate that MEC population activity encodes both stable (reward well) and transient (target) goals, potentially with distinct temporal dynamics.

Significance and Claims
The authors claim that this platform provides a powerful framework for examining the neural basis of flexible navigation under experimentally controlled yet naturalistic conditions. By combining unrestricted movement with precise trial structure and dynamic visual stimuli, the system overcomes the limitations of both open-field and head-fixed VR paradigms.

The paper asserts that this approach enables the study of how internal states and task-dependent goals influence spatial representations in the MEC. Specifically, the authors demonstrate that their setup can capture not only canonical spatial tuning but also population signals related to the distance to dynamic, task-relevant objects. They conclude that this platform bridges the gap between naturalistic navigation and experimental control, offering a new avenue to investigate how neural maps support goal-directed behavior in complex, changing environments. The authors modestly note that minor adjustments to the arena boundaries could further improve spatial coverage and that the system is extensible for more complex tasks involving prediction, planning, or uncertainty.