Review Papers Claim by Claim

Introduction

This paper challenges the prevailing view that representational drift (RD) in neural populations is driven exclusively by synaptic weight changes. Instead, the authors propose that changes in intrinsic neuronal excitability can account for observed alterations in individual neuron tuning while preserving stable decoding at the population level. They support this claim via detailed simulations using spike coding networks (SCNs) across different scales, from a toy model with 6 neurons to a high-dimensional network with 2592 neurons. Core concept of excitability-driven drift

Key Findings and Methodology

• Mechanistic Proposal: The paper argues that minor fluctuations in cellular excitability, which can be modeled as effective changes in neuronal thresholds, cause apparent shifts in tuning properties. These shifts include alterations in the gain (amplitude) of neural responses and subtle changes in preferred tuning directions, without requiring a reorganization of synaptic weights. Mechanism explanation
• Spike Coding Network Simulations: Utilizing SCNs, the study simulates both two-dimensional and high-dimensional input spaces. These models incorporate Gaussian noise to emulate excitability drift over days. Key metrics like population activity vector correlation, decoder performance, and tuning curve shifts are analyzed to establish quantitative matches with experimental data. Simulation methods
• Decoder Performance: The authors contrast decoders trained on single-day data versus those trained on multi-day heterogeneous datasets. They find that while day-specific decoders fail to generalize over time, a general decoder trained across many days can closely approximate the optimal performance, thereby reconciling apparent behavioral stability with underlying drift. Decoder findings

Critical Evaluation

Novelty: The work is highly novel as it shifts the focus from synaptic plasticity to intrinsic excitability fluctuations as a driver of RD. This perspective offers a parsimonious explanation for experimental observations which has rarely been explored in such depth. Novelty assessment

Scientific Quality: The technical quality is high with robust model formulations and detailed parameterization (as shown in their simulation parameters and interpolation schemes). However, as a computational study, its predictive power depends on experimental validation in biological systems, which remains a potential limitation. Quality critique

Generality: The mechanism proposed by the authors has broad potential implications, as the interplay between excitability and recurrent dynamics is a ubiquitous feature in neural circuits across brain areas. This enhances the paper's generality. Generality evaluation

Usefulness: From a practical viewpoint, understanding that excitability changes can drive RD without compromising behavior could streamline future experimental designs and data analysis strategies, particularly when interpreting longitudinal neural recordings. Usefulness insight

Reproducibility and Explanatory Depth

Reproducibility: Although the models are detailed with explicit parameter settings (e.g., Gaussian noise scaling, interpolation steps in threshold changes), direct reproducibility hinges on access to the simulation code and experimental replication. The authors state that code is available on request, which is positive but may hinder immediate validation by others. Reproducibility review

Explanatory Depth: The paper provides a deep mechanistic insight by linking changes in excitability to both single-cell tuning and network-level phenomena. The use of both small-scale and large-scale network simulations enhances confidence in the proposed mechanism's robustness. Explanatory depth analysis

Summary and Suggestions for Improvement

Overall, this paper is a significant contribution to our understanding of representational drift. It presents a novel, parsimonious model that shifts focus from synaptic plasticity to excitability changes. To further strengthen the work, additional experimental validation is recommended, including direct physiological measurements of excitability changes correlating with RD, and open-source release of simulation code to bolster reproducibility.

How to improve response:

Provide more experimental validation and code accessibility details.

Ratings