Contributions of cortical neuron firing patterns, synaptic connectivity, and plasticity to task performance

Michele N. Insanally, Badr F. Albanna, Jade Toth, Brian DePasquale, Saba Shokat Fadaei, Trisha Gupta, Olivia Lombardi, Kishore Kuchibhotla, Kanaka Rajan, Robert C. FroemkeNature Communications 2024 · OpenReview

TL;DR

This paper introduces a novel spiking recurrent neural network (RNN) model that incorporates biologically realistic synaptic plasticity rules to investigate how diverse neuronal firing patterns contribute to task performance. The study reveals that both "classical" (highly responsive) and "non-classical" (less responsive) neurons play distinct but crucial roles, with plasticity shaping their synaptic connections to optimize network function and behavior.

What Problem Does This Paper Address?

The brain exhibits a remarkable diversity in how individual neurons respond to stimuli and engage in behavior. While some neurons show clear, "classical" responses to stimuli, others exhibit more irregular or "non-classical" firing patterns, often appearing less responsive. It's unclear how this diversity arises from underlying synaptic mechanisms and how these different response types contribute to complex cognitive functions like perception and decision-making. This paper aims to bridge this gap by exploring the synaptic origins and functional roles of these diverse neuronal response profiles within a computational model that mimics biological learning.

Methods

The authors developed a novel spiking recurrent neural network (RNN) model. Key features include:

  • Spiking Units: The network is composed of leaky integrate-and-fire neurons, mimicking the discrete nature of neuronal firing.
  • Synaptic Plasticity: The model incorporates biologically inspired spike-timing-dependent plasticity (STDP) rules for both excitatory-to-excitatory and inhibitory-to-excitatory synapses. This allows the network's connections to change over time based on the relative timing of pre- and post-synaptic spikes.
  • FORCE Training: A FORCE (Functional Optimization with Recurrent Error) learning algorithm was adapted for spiking networks to train the output weights of the RNN to perform a go/no-go auditory frequency recognition task.
  • Heterosynaptic Plasticity: Homeostatic mechanisms (heterosynaptic balancing and enhancement) were included to prevent individual synapses from dominating and to maintain synaptic diversity.
  • In Vivo Data Integration: The model's predictions were validated against whole-cell recordings from rodent auditory cortex during a similar behavioral task.

Datasets

  • Experimental Data: Whole-cell recordings from the auditory cortex of behaving rats and head-fixed mice performing a go/no-go frequency recognition task.
  • Model Training: The spiking RNN was trained on a simulated go/no-go auditory frequency recognition task, mirroring the experimental paradigm.

Key Insights and Ideas

  • Synaptic Plasticity Shapes Diversity: STDP rules are not just for learning but are fundamental in shaping the distribution of neuronal response types, leading to a higher proportion of non-classically responsive units.
  • Differential Contributions: "Classical" neurons primarily influence task performance through their output connections to the readout layer, while "non-classical" neurons contribute more significantly through their recurrent connections, shaping network dynamics.
  • Inhibition Constrains, Excitation Shifts: Excitatory plasticity shifts the distribution of neuronal responses, while inhibitory plasticity helps constrain the diversity, leading to improved overall task performance.
  • "Synaptic Signatures" Predict Response Types: Specific patterns of local synaptic connections (motifs) can predict whether a neuron will be classically or non-classically responsive, suggesting that a neuron's wiring dictates its functional role.
  • Local Connectivity is Key: The model demonstrates that local synaptic structure, particularly monosynaptic connections, is a strong predictor of individual neuron response properties, and STDP enhances this "hyper-local" characteristic.

Key Results

  • The spiking RNN model successfully recapitulated the diverse range of neuronal response profiles observed in vivo.
  • Inactivating non-classically responsive units had a larger initial impact on task performance than inactivating classically responsive units, highlighting their critical role in network dynamics.
  • STDP significantly improved task performance in the model compared to networks without plasticity.
  • The model's predictions about the relationship between synaptic inputs and neuronal response types were validated by in vivo whole-cell recordings, showing that the RNN-derived synaptic motif model accurately predicted in vivo firing rate modulation.
  • Non-classically responsive neurons in vivo exhibited significantly weaker excitatory and inhibitory synaptic inputs compared to classically responsive neurons.

Limitations and Future Work

  • Species Differences: While the model captured general trends, there were some differences in response distributions between simulated and mouse experimental data, suggesting potential species-specific nuances in plasticity or network architecture.
  • Model Simplifications: The model uses simplified integrate-and-fire neurons and specific STDP rules. Future work could explore more complex neuron models and a wider range of plasticity mechanisms.
  • Multi-Region Architectures: The current model focuses on a single cortical area. Extending the model to incorporate multi-region RNN architectures could provide insights into how different brain areas interact and contribute to complex behaviors.
  • Dendritic Geometry and Intrinsic Properties: The authors acknowledge that other factors like dendritic geometry and intrinsic neuronal properties could also contribute to neuronal diversity, which were not explicitly modeled here.

Summary updated

Abstract

Neuronal responses during behavior are diverse, ranging from highly reliable ‘classical’ responses to irregular ‘non-classically responsive’ firing. While a continuum of response properties is observed across neural systems, little is known about the synaptic origins and contributions of diverse responses to network function, perception, and behavior. To capture the heterogeneous responses measured from auditory cortex of rodents performing a frequency recognition task, we use a novel task-performing spiking recurrent neural network incorporating spike-timing-dependent plasticity. Reliable and irregular units contribute differentially to task performance via output and recurrent connections, respectively. Excitatory plasticity shifts the response distribution while inhibition constrains its diversity. Together both improve task performance with full network engagement. The same local patterns of synaptic inputs predict spiking response properties of network units and auditory cortical neurons from in vivo whole-cell recordings during behavior. Thus, diverse neural responses contribute to network function and emerge from synaptic plasticity rules. How synaptic plasticity rules lead to diverse neural responses in a spiking neural network model remains poorly understood. Here, the authors show how a diversity of response types contributes to network function and task performance.
BibTeX
@inproceedings{insanally2024contributions,
  author = {Michele N. Insanally and Badr F. Albanna and Jade Toth and Brian DePasquale and Saba Shokat Fadaei and Trisha Gupta and Olivia Lombardi and Kishore Kuchibhotla and Kanaka Rajan and Robert C. Froemke},
  title = {{Contributions of cortical neuron firing patterns, synaptic connectivity, and plasticity to task performance}},
  booktitle = {Nature Communications},
  year = {2024},
  url = {https://openreview.net/forum?id=s3TLGbWQGw},
}
Figure