Learning Shapes the Energy Cost of Neural Tasks.

Area of Science:

• Neuroscience
• Bioenergetics
• Computational Neuroscience

Background:

• The brain's high efficiency is notable compared to AI, yet the energetics of neural computation are poorly understood.
• Measuring the energy cost of specific neural tasks in vivo has been challenging.
• Neural design principles often cite efficiency, but direct circuit-level measurements are limited.

Purpose of the Study:

• To define circuit-level energy costs associated with learning using neuronal glucose consumption.
• To investigate the bioenergetic changes during learning-based behavioral tasks in mice.
• To explore the relationship between neural plasticity mechanisms and energy expenditure during learning.

Main Methods:

• Simultaneous in vivo measurement of intracellular glucose and calcium dynamics in behaving mice.
• Utilizing neuronal glucose consumption as a proxy for circuit-level energy costs.
• Employing hippocampus- and cortex-dependent learning models.

Main Results:

• Post-learning fuel cost per task was significantly lower than pre-learning levels.
• Reduced energy cost was observed across multiple learning models, independent of bulk calcium dynamics.
• This fuel cost reduction stemmed from decreased intracellular glucose consumption, not extracellular transport.
• The effect depended on canonical plasticity mechanisms like NMDAR signaling and protein synthesis.

Conclusions:

• Attenuation of task-specific energy cost may be a general bioenergetic trajectory of learning and plasticity.
• The findings support an "energy minimization" hypothesis for biological brain efficiency.
• This perspective complements existing neural activity-centered frameworks for understanding brain function.