Flexible theta sequence compression mediated via phase precessing interneurons

Angus Chadwick, Mark C. W. van Rossum, Matthew F. Nolan

TL;DR

This paper proposes a novel computational model of hippocampal circuits that explains how the brain compresses sequences of events occurring over seconds into rapid neural firing patterns within theta oscillations. This mechanism, driven by interneurons, is crucial for associative learning and memory, allowing the brain to flexibly link past, present, and future experiences.

What Problem Does This Paper Address?

The hippocampus is known to encode experiences as sequences of neural activity, particularly within the theta rhythm (a ~4-12 Hz oscillation). These "theta sequences" are thought to be vital for memory and navigation, compressing longer behavioral timescales (seconds) into millisecond neural timescales. However, the precise neural mechanisms generating these sequences and their diverse observed properties (like speed-dependence and variations across the hippocampus) have remained elusive. Existing models often struggle to account for these complexities while remaining biologically plausible. This paper aims to bridge this gap by proposing a biophysically constrained model that explains how interneurons, in conjunction with spatial and pacemaker inputs, generate flexible and high-capacity theta sequences.

Methods

The authors developed a computational model of hippocampal circuits, starting with a minimal interneuron-pyramidal cell pair and progressively scaling up to larger networks. Key techniques include:

  • Reduced Model of Interneuron Phase Dynamics: A mathematical model (based on phase response curves) was used to analyze the dynamics of an interneuron driven by both a constant excitatory current (representing spatial input) and a weak oscillatory pacemaker current (from the medial septum). This allowed for analytical solutions to understand phase locking and phase precession.
  • Integrate-and-Fire Neuron Network Simulations: For more complex scenarios, the authors simulated networks of leaky integrate-and-fire neurons with conductance-based synapses. This allowed for testing predictions in more biologically realistic settings, including varying running speeds, spatial scales, and network connectivity.
  • Quantitative Metrics for Sequence Compression: Two metrics were introduced: a "single-cycle theta sequence metric" to measure the similarity between slow input sequences and individual theta sequences, and a "population phase precession metric" to assess the robustness and coherence of phase precession across a population of cells.

Datasets

Not discussed in the paper. The study relies on simulations and comparisons to existing experimental data.

Key Insights and Ideas

  • Interneurons as Key Sequence Compressors: The model highlights the critical role of spontaneously active interneurons in integrating spatial and pacemaker inputs to generate phase-precessing action potentials. This is a departure from models that primarily focus on pyramidal cell dynamics.
  • Dynamic Generation of Phase Precession: Phase precession isn't just inherited; it's dynamically generated by the interplay of interneuron intrinsic properties, pacemaker drive, and spatial inputs. This allows for flexibility and adaptation.
  • Speed and Location Dependence Explained: The model naturally accounts for how running speed and dorsoventral location in the hippocampus influence phase precession rates and patterns, linking them to variations in synaptic inputs and pacemaker amplitude.
  • High Capacity for Representation: The proposed circuit organization, particularly the non-overlapping place field constraint for cells coupled to the same interneuron, allows for an "unlimited" capacity to generate distinct spatial maps, cell assemblies, and sequences, far exceeding an animal's lifetime experiences.
  • STDP for Associative Learning: The compressed theta sequences provide a temporal reference frame that enables Spike-Timing Dependent Plasticity (STDP) to form associations between events (conditioned stimuli) and outcomes (unconditioned stimuli), crucial for learning.

Key Results

  • The model successfully reproduces experimentally observed features of phase precession, including its dependence on running speed, place field size, and dorsoventral location.
  • It predicts that phase precession can occur over a full 360 degrees of the theta cycle, unlike some previous models limited to 180 degrees.
  • The model demonstrates that network configuration (e.g., place field overlap) significantly impacts sequence compression quality, with sparse and organized maps yielding better compression.
  • It shows that even with dense random maps, a substantial proportion of individual cells can still exhibit robust phase precession, aligning with experimental observations of population variability.
  • The proposed mechanism allows for flexible association of past or future events with outcomes by modulating the spike phase of downstream neurons.

Limitations and Future Work

  • Experimental Validation: The authors emphasize that their model's predictions need to be tested experimentally, particularly regarding subthreshold phase precession in silent pyramidal cells and the necessity of non-overlapping place fields for cells coupled to the same interneuron.
  • Network Complexity: While the model scales to larger networks, the precise mechanisms for organizing place field maps to avoid interference and the role of plasticity in this reorganization remain areas for future investigation.
  • Dorsoventral Gradient Mechanisms: While the model explains dorsoventral gradients, alternative mechanisms (e.g., gradients in pacemaker input phase) are acknowledged as possible.
  • Role of Other Interneuron Types: The model primarily focuses on a specific type of interneuron; future work could explore the roles of other interneuron populations.
  • Learning Rules: The paper suggests potential plasticity mechanisms for reorganizing network activity but acknowledges that the precise learning rules and circuit mechanisms underlying decorrelation require further study.

Summary updated

Abstract

Encoding of behavioral episodes as spike sequences during hippocampal theta oscillations provides a neural substrate for computations on events extended across time and space. However, the mechanisms underlying the numerous and diverse experimentally observed properties of theta sequences remain poorly understood. Here we account for theta sequences using a novel model constrained by the septo-hippocampal circuitry. We show that when spontaneously active interneurons integrate spatial signals and theta frequency pacemaker inputs, they generate phase precessing action potentials that can coordinate theta sequences in place cell populations. We reveal novel constraints on sequence generation, predict cellular properties and neural dynamics that characterize sequence compression, identify circuit organization principles for high capacity sequential representation, and show that theta sequences can be used as substrates for association of conditioned stimuli with recent and upcoming events. Our results suggest mechanisms for flexible sequence compression that are suited to associative learning across an animal's lifespan.
BibTeX
@inproceedings{chadwick2016flexible,
  author = {Angus Chadwick and Mark C. W. van Rossum and Matthew F. Nolan},
  title = {{Flexible theta sequence compression mediated via phase precessing interneurons}},
  booktitle = {eLife},
  year = {2016},
  doi = {10.7554/elife.20349},
}
Figure