TL;DR
This paper investigates whether the cerebral cortex operates in a "fluctuation-driven regime" by analyzing neuronal activity in mice performing a decision-making task. While standard models capture spiking patterns, they fail to explain the observed heterogeneity in sub-threshold membrane potentials. By incorporating dendritic morphology into these models, the researchers demonstrate a more accurate representation of cortical activity, suggesting a crucial role for dendrites in neuronal selectivity and task performance.
What Problem Does This Paper Address?
The brain's cerebral cortex is characterized by neurons that fire irregularly and at vastly different rates. A leading hypothesis to explain this is the "fluctuation-driven regime," where neuronal activity is primarily shaped by the constant barrage of excitatory and inhibitory inputs, rather than a strong net input. However, it's been debated whether real cortical circuits truly operate this way. Specifically, while models of this regime can explain the spiking statistics of neurons, they struggle to reproduce the observed sub-threshold voltage variations across different neurons. This discrepancy suggests a missing piece in our understanding of how cortical circuits function.
Methods
The researchers employed a multi-pronged approach:
- Experimental Data Analysis: They analyzed electrophysiological recordings (both extracellular and whole-cell) from the anterior lateral motor cortex (ALM) of mice performing a decision-making task. This allowed them to measure both spiking activity and sub-threshold membrane potentials.
- Standard Network Modeling: They first simulated a network of simplified "point neurons" (integrate-and-fire models) connected in a fluctuation-driven regime. This served as a baseline to see if it could replicate the experimental data.
- Morphologically-Informed Modeling: To address the limitations of point neuron models, they developed an "extended-like" point neuron model. This model phenomenologically incorporated the effects of dendritic morphology by introducing parameters (attenuation and visibility) that capture how synaptic inputs on dendrites influence somatic currents and conductances. This was achieved by first simulating detailed multi-compartment neuron models and then deriving these parameters.
- Analytical Modeling: Mean-field theory was used to analytically investigate the sub-threshold voltage heterogeneity in both standard and extended models.
Datasets
The study utilized electrophysiological data collected from mice performing a delayed-response task (auditory and tactile versions) and a Go/NoGo task. These datasets were previously published and are referenced in the paper. No new benchmark datasets were introduced.
Key Insights and Ideas
- Dendrites Matter for Sub-threshold Heterogeneity: The most significant insight is that simple point neuron models, even when operating in a fluctuation-driven regime, fail to capture the wide range of sub-threshold membrane potentials observed in real cortical neurons. This heterogeneity is crucial for neuronal function.
- Clamping Effect of Point Neurons: In standard models, the dense synaptic input effectively "clamps" the membrane potential of all neurons to a similar average value, masking individual differences.
- Dendritic Morphology as a Solution: Incorporating the spatial distribution of synapses on dendrites, through the "extended-like" point neuron model, effectively breaks this clamping effect. This allows for a much wider distribution of sub-threshold voltages across the neuronal population.
- Functional Role of Dendrites: The improved model, which includes dendritic effects, not only better matches experimental data but also leads to increased neuronal selectivity and a reduction in decision-making errors. This suggests dendrites play a direct functional role in cognitive tasks.
- Fluctuation-Driven Regime Confirmed (with caveats): The study supports the idea that the ALM operates in a fluctuation-driven regime, but emphasizes that this regime's full implications for sub-threshold activity are only revealed when neuronal morphology is considered.
Key Results
- Standard fluctuation-driven network models accurately predicted the spiking statistics (e.g., log-normal firing rate distributions, coefficient of variation of inter-spike intervals close to 1) of ALM neurons.
- However, these models failed to reproduce the observed wide distribution of time-averaged membrane potentials across neurons. The simulated neurons had very similar average voltages, clustered tightly around the spike threshold.
- The "extended-like" point neuron model, incorporating dendritic morphology, successfully reproduced the wide sub-threshold voltage heterogeneity seen in experimental data.
- In the improved model, increasing neuronal selectivity was observed, and the number of error trials in a simulated decision-making task decreased as dendritic effects (modeled by the visibility parameter) were incorporated.
Limitations and Future Work
- Data Limitations: The authors acknowledge that their estimates of voltage heterogeneity from experimental data might be upper bounds due to potential biases in recordings (e.g., selection of high-firing neurons, variations in network state, movement artifacts).
- Model Simplifications: While the "extended-like" model is an improvement, it still simplifies complex neuronal structures. Future work could incorporate non-linear NMDA synapses, active dendritic ion channels, and different GABAergic cell types to further increase voltage heterogeneity.
- External vs. Recurrent Fluctuations: The study primarily focused on recurrent network dynamics. Future research should disentangle the contributions of external and recurrent fluctuations to membrane potential variability.
- Broader Applicability: The study focused on ALM. Future work should investigate whether these findings generalize to other cortical areas and how the "partially-balanced" regime observed in some regions (like vS1) differs functionally.
- Technological Advancements: The authors anticipate that future advancements in simultaneous multi-neuron sub-threshold recordings will be crucial for further understanding cortical dynamics.
Summary updated