Adult dentate gyrus neurogenesis: a functional model

Olivia Gozel, Wulfram GerstnerbioRxiv 2019 · DOI

TL;DR

This paper presents a computational model of adult neurogenesis in the dentate gyrus, demonstrating how the developmental "GABA-switch" from excitatory to inhibitory neurotransmission is crucial for integrating new neurons and enabling them to learn and distinguish similar, but not identical, patterns. This work offers a mechanistic explanation for the dentate gyrus's role in pattern separation and memory.

What Problem Does This Paper Address?

The dentate gyrus (DG), a brain region critical for learning and memory, continuously generates new neurons throughout adulthood. While this neurogenesis is known to be important for behavioral pattern separation (distinguishing similar memories), the precise mechanisms by which these new neurons integrate into the existing network and contribute to this function have remained unclear. Specifically, how the changing neurochemical environment of a developing neuron, particularly the shift in GABAergic signaling, influences its functional integration and selectivity is a key question. This paper aims to bridge this gap by developing a theoretical model that explains both the integration process and the functional outcome of adult neurogenesis in the DG.

Methods

The authors developed a computational model of a small patch of the dentate gyrus, comprising dentate granule cells (DGCs) and GABAergic interneurons. Their model incorporates several key features:

  • Biologically Plausible Plasticity Rule: They employ a Hebbian learning rule for synaptic weight updates, incorporating both Long-Term Potentiation (LTP) and Long-Term Depression (LTD), as well as heterosynaptic plasticity to prevent unbounded weight growth.
  • The "GABA-Switch": A central element of the model is the simulated "GABA-switch," where the GABAergic input to developing DGCs transitions from excitatory in the early maturation phase to inhibitory in the later phase. This switch is driven by changes in chloride cotransporter expression (NKCC1 and KCC2).
  • Two-Phase Maturation: Newborn DGCs undergo two distinct maturation phases:
  • Early Phase (Excitatory GABA): Characterized by cooperative learning, where indirect excitation via interneurons drives synapse growth, making newborn cells sensitive to stimuli similar to those that activate mature cells.
  • Late Phase (Inhibitory GABA): Characterized by competitive learning, where inhibition leads to specialization, pushing synapses towards stimuli that are different from previously stored ones.
  • Network Architecture: The model uses a recurrent network where DGCs are connected to interneurons, and interneurons provide feedback inhibition to DGCs.

Datasets

  • MNIST Dataset: Used for training and testing the network's ability to classify handwritten digits (specifically digits 3, 4, and 5). The patterns were preprocessed to a 12x12 pixel resolution.
  • Artificial Data Set: A custom-generated dataset designed to control the similarity between clusters of patterns. This allowed for a more precise investigation of how similarity influences pattern separation.

Key Insights and Ideas

  • The GABA-Switch is Crucial: The transition of GABAergic signaling from excitatory to inhibitory is not just a developmental marker but a critical computational mechanism for integrating new neurons and enabling them to learn and specialize.
  • Cooperation to Competition: The early, cooperative phase of maturation allows newborn neurons to broadly tune into existing patterns, while the later, competitive phase forces them to specialize and differentiate, crucial for distinguishing similar stimuli.
  • Selective Learning of Novelty: Newborn DGCs become selective for novel patterns only if those patterns are similar to already learned ones. They fail to learn or represent distinct novel stimuli.
  • Pattern Separation Mechanism: The model provides a concrete mechanism for how the dentate gyrus achieves pattern separation: new neurons learn to represent novel, similar patterns by averaging inputs during the cooperative phase and then specializing during the competitive phase, effectively creating distinct representations for similar inputs.
  • Unresponsive Units are Replaced: The model implicitly handles "unresponsive units" (neurons that don't develop strong connections) by assuming they are replaced by new, plastic newborn DGCs, ensuring continuous learning capacity.

Key Results

  • Successful Integration: The model successfully demonstrates how newborn DGCs integrate into the existing network through the described two-phase maturation process.
  • Pattern Separation of Similar Stimuli: The model shows that newborn DGCs become selective for novel patterns that are similar to pre-existing ones, leading to improved classification performance for these similar patterns.
  • Failure to Separate Distinct Stimuli: Conversely, the model demonstrates that newborn DGCs do not become selective for novel patterns that are distinct from pre-existing ones, and thus do not improve discrimination for such stimuli.
  • Classification Performance: The model achieved high classification performance for digits 3 and 4 after pretraining (99.25%), and when introducing a novel digit 5, performance for digit 5 was significantly higher (94.56%) when learned sequentially by newborn DGCs compared to simultaneous learning.
  • Robustness: The model's results were shown to be robust to changes in network architecture, such as the number of neurons.

Limitations and Future Work

  • Simplified Network: While the model captures key biological processes, it is a simplification of the complex dentate gyrus circuitry. For instance, Mossy cells were omitted for simplicity.
  • Timing of Novel Stimulus Introduction: The paper suggests that the timing of introducing novel stimuli during the maturation phase is important for successful learning. Further investigation into optimal timing could be explored.
  • Experimental Validation: The authors propose several testable predictions, such as the effect of NKCC1 knockout mice on discrimination, which could be experimentally validated.
  • Online Learning Scenario: The model's ability to learn novel patterns sequentially suggests potential for an "online learning" scenario where new information is continuously integrated, which could be further explored.

Summary updated

Abstract

Summary In adult dentate gyrus neurogenesis, the link between maturation of newborn neurons and their function, such as behavioral pattern separation, has remained puzzling. By analyzing a theoretical model, we show that the switch from excitation to inhibition of the GABAergic input onto maturing newborn cells is crucial for their proper functional integration. When the GABAergic input is excitatory, cooperativity drives the growth of synapses such that newborn cells become sensitive to stimuli similar to those that activate mature cells. When GABAergic input switches to inhibitory, competition pushes the configuration of synapses onto newborn cells towards stimuli that are different from previously stored ones. This enables the maturing newborn cells to code for concepts that are novel, yet similar to familiar ones. Our theory of newborn cell maturation explains both how adult-born dentate granule cells integrate into the preexisting network and why they promote separation of similar but not distinct patterns.
BibTeX
@misc{gozel2019adult,
  author = {Olivia Gozel and Wulfram Gerstner},
  title = {{Adult dentate gyrus neurogenesis: a functional model}},
  year = {2019},
  doi = {10.1101/704791},
  howpublished = {bioRxiv},
}
Figure