Energy efficient network activity from disparate circuit parameters

Michael Deistler, Jakob H. Macke, Pedro J. Gonçalves

TL;DR

Neural circuits can achieve the same function with vastly different internal parameters, a phenomenon known as parameter degeneracy. This paper shows that even when considering energy efficiency and temperature robustness as additional constraints, this degeneracy persists. The authors demonstrate that energy-efficient neural circuits can be found across a broad range of parameters, and that temperature robustness doesn't necessarily come at an increased metabolic cost.

What Problem Does This Paper Address?

Neural systems are remarkably robust and adaptable, capable of producing consistent activity patterns despite variations in their underlying components. This flexibility, termed "parameter degeneracy," allows circuits to maintain function even when individual channels or synapses change. However, biological systems operate under additional constraints beyond just functional output, such as minimizing energy consumption (metabolic cost) and maintaining performance across environmental changes (like temperature). The key question this paper addresses is: how do these additional constraints of energy efficiency and temperature robustness influence the range of permissible circuit parameters, and does this lead to a significant reduction in parameter degeneracy? In essence, can a circuit be both functional, energy-efficient, and robust, and if so, what does the parameter space look like under these combined pressures?

Methods

The authors leverage a sophisticated machine learning framework called Sequential Neural Posterior Estimation (SNPE) for simulation-based inference. A key innovation is the integration of a "rejection classifier" that efficiently identifies and discards simulations producing non-sensible outputs (e.g., no bursting activity). This significantly speeds up the process of finding parameter sets that match experimental data. They then employ linear regression to analyze the relationship between circuit parameters and energy consumption, identifying which parameters most strongly influence metabolic cost. Further analysis involves exploring pairwise parameter interactions and using Markov Chain Monte Carlo (MCMC) to search for optimal configurations. For temperature robustness, they extend their model to include temperature-dependent parameters (Q10 values) and use SNPE again to infer parameters that match experimental data at different temperatures.

Datasets

The study utilizes experimental data from the stomatogastric ganglion (STG) of the crab Cancer borealis. Specifically, they use extracellular recordings of motor neurons involved in the pyloric rhythm, collected at two different temperatures (11°C and 27°C). This data serves as the ground truth for training and validating their computational models. No new benchmarks are introduced; rather, they leverage existing experimental datasets to constrain their model parameters.

Key Insights and Ideas

  • Parameter degeneracy persists even under multiple constraints: The most striking finding is that even when optimizing for both functional activity and energy efficiency (and temperature robustness), a wide range of circuit parameters can still achieve these goals. This suggests that biological systems might have considerable flexibility in their underlying structure.
  • Energy efficiency is not a niche parameter space: Contrary to what one might expect, energy-efficient solutions are not confined to a tiny, highly specific region of the parameter space. Instead, they are found across a significantly reduced, but still substantial, range.
  • Temperature robustness doesn't necessarily increase metabolic cost: The study debunks the intuition that being robust to temperature changes would inherently require more energy. They show that temperature-robust circuits can be just as energy-efficient as those that are not.
  • Specific currents drive energy consumption: While many currents contribute to neural activity, the sodium and calcium currents (specifically Na+, CaT, and CaS) are identified as the primary drivers of energy consumption. This suggests that evolutionary pressure might focus on regulating these specific channels for metabolic efficiency.
  • Individual neuron optimization can lead to circuit efficiency: The research demonstrates that optimizing individual neurons for energy efficiency can contribute to overall circuit-level energy savings without sacrificing network function.

Key Results

  • The developed SNPE framework with a rejection classifier significantly speeds up the inference process, requiring half the simulations compared to previous methods while achieving higher accuracy.
  • They identified a database of 35,939 model configurations that closely match experimental data. Within this database, the most energy-efficient circuits consumed up to nine times less energy than the least efficient ones, despite producing similar network activity.
  • Linear regression analysis revealed that energy consumption is linearly predictable from circuit parameters, with sodium and calcium conductances being the strongest correlates.
  • Pairwise parameter modifications (e.g., reducing sodium and increasing potassium) could lead to significant energy savings (up to 55% for the entire circuit).
  • Analysis of temperature robustness showed that circuits matching experimental data at both 11°C and 27°C exhibited similar energy consumption distributions to those only matched at 11°C, indicating no inherent metabolic cost for temperature robustness.

Limitations and Future Work

  • Simplifying assumptions: The study uses single-compartment neuron models, which is a simplification compared to more complex multi-compartment models. Additionally, their energy measure is based on the Hodgkin-Huxley model, which doesn't capture the full complexity of ATP consumption.
  • Limited scope of constraints: While energy efficiency and temperature robustness were investigated, the authors suggest future work should explore the interplay with other perturbations like pH fluctuations or neuromodulation.
  • Parameter space exploration: The authors acknowledge that their parameter search might not have covered all possible configurations, and that even more energy-efficient solutions might exist in unsampled regions of the parameter space.
  • Generalizability: While the pyloric network is a well-studied model, future research could investigate whether these findings generalize to more complex neural circuits, especially those with more extensive synaptic connectivity.

Summary updated

Abstract

Abstract Neural circuits can produce similar activity patterns from vastly different combinations of channel and synaptic conductances. These conductances are tuned for specific activity patterns but might also reflect additional constraints, such as metabolic cost or robustness to perturbations. How do such constraints influence the range of permissible conductances? Here, we investigate how metabolic cost affects the parameters of neural circuits with similar activity in a model of the pyloric network of the crab Cancer borealis . We use a novel machine learning method to identify a range of network models that can generate activity patterns matching experimental data, and find that neural circuits can consume largely different amounts of energy despite similar circuit activity. Furthermore, a reduced but still significant range of circuit parameters gives rise to energy-efficient circuits. We then examine the space of parameters of energy-efficient circuits and identify potential tuning strategies for low metabolic cost. Finally, we investigate the interaction between metabolic cost and temperature robustness. We show that metabolic cost can vary across temperatures, but that robustness to temperature changes does not necessarily incur an increased metabolic cost. Our analyses show that, despite metabolic efficiency and temperature robustness constraining circuit parameters, neural systems can generate functional, efficient, and robust network activity with widely disparate sets of conductances.
BibTeX
@inproceedings{deistler2022energy,
  author = {Michael Deistler and Jakob H. Macke and Pedro J. Gonçalves},
  title = {{Energy efficient network activity from disparate circuit parameters}},
  booktitle = {Proceedings of the National Academy of Sciences},
  year = {2022},
  doi = {10.1073/pnas.2207632119},
}
Figure