Abstract
Hippocampal place cell sequences serve diverse purposes, from synaptic plasticity induction to navigation and planning. During spatial behaviors, "theta sequences" encode running trajectories as 1D behavioral sequences. However, in 2D space, locations can be visited from arbitrary directions, creating a topological conflict for memory models. We propose a computational model of the CA3 and Dentate Gyrus (DG) where sensorimotor input drives direction-dependent (extrinsic) sequences in CA3, while CA3-DG projections produce direction-independent (intrinsic) sequences. These intrinsic sequences act as stable landmark cues, potentially reflecting replay/preplay activity during non-theta states.
Introduction: The 2D Topological Challenge
In the hippocampus, place cells fire in rapid sequences during theta oscillations (~8Hz). While these sequences are well-understood in 1D environments like linear tracks, 2D navigation presents a problem: a single location can be approached from any angle. If neural sequences were purely driven by movement (extrinsic), they would change with every trajectory change. Conversely, purely hard-wired (intrinsic) sequences would lack the flexibility needed for novel paths. This study investigates how the brain reconciles these two mechanisms to maintain a reliable spatial map.
Methods: Modeling the CA3-DG Circuit
The researchers developed a spiking neural network model using Izhikevich neurons to simulate the interaction between the CA3 and Dentate Gyrus (DG) subregions.
- Extrinsic Mechanism: Sequences in CA3 are generated through short-term synaptic plasticity (facilitation and depression), making firing patterns dependent on the animal's velocity and sensory input.
- Intrinsic Mechanism: A CA3-DG feedback loop with fixed, asymmetrical synaptic weights "hard-wires" specific sequences, making them independent of the animal's trajectory.
- Analysis: The model was evaluated using cross-correlation analysis to distinguish between extrinsic and intrinsic components. A Tempotron—a bio-inspired supervised learning algorithm—was used to test if downstream neurons could decode locations from these sequences.
Results: Stable Landmarks and Theta Compression
The model demonstrates that intrinsic sequences function as "temporal landmarks." While extrinsic sequences shift according to movement, intrinsic sequences provide a stable pattern recognizable from any approach angle (0° to 360°).
The Role of the Dentate Gyrus
Simulating a DG lesion revealed its critical role in theta compression—the process of encoding seconds of experience into milliseconds of neural activity. In control simulations, the slope of the relationship between spike phase and position was ~0.18 rad/cm. In the "lesioned" model, this dropped to ~0.05 rad/cm, aligning with experimental observations in rats.
Decoding and Phase Precession
The Tempotron achieved near-perfect decoding accuracy across all directions when intrinsic sequences were present, whereas it failed to generalize using only extrinsic sequences. The model also successfully reproduced "phase precession," showing how this phenomenon becomes direction-dependent in 2D environments.
Discussion and Future Directions
While the model provides a robust framework for understanding 2D spatial sequences, it uses a simplified CA3-DG anatomy. Real-world connectivity involves complex interneuron networks and hilar mossy cells. Furthermore, the model assumes pre-configured intrinsic weights; how these 2D topologies are learned through experience remains an open question. Future research could explore how these fixed loops interact with "global remapping" when an animal enters a completely new environment.
Summary updated