ATP energy margin and the MEA signal floor

What the work claims

This is a theoretical and computational paper, not an experiment on live animals or organoids. The authors propose that the physical origin of mental fatigue is a decline in the Gibbs free energy available from ATP hydrolysis, which in turn degrades two things: the cooperative transport of amide I exciton energy inside protein alpha-helices, and the transmembrane ion concentration gradients that set the Nernst reversal potentials for sodium and potassium.¹ The result, they argue, is neuronal hyperexcitability combined with a higher risk of depolarization block, producing the lower signal-to-noise ratio and slower, less accurate information processing that fatigue studies report.

The paper is explicit about its method: it uses published intracellular metabolite concentrations to compute |Delta G_ATP| for liver, brain, and heart, then inserts the brain value into existing computer models of Davydov soliton transport in proteins and of a morphologically complete CA1 pyramidal neuron simulated in NEURON 8.2.7.¹ It therefore makes predictions rather than direct measurements, and its value lies in the mechanistic connection it draws between metabolism and excitability.

How it works

The Gibbs free energy of ATP hydrolysis is not a fixed number. It depends on the reaction quotient Q = [ADP][Pi]/[ATP], so the same ATP concentration can deliver different usable energy depending on ADP and inorganic phosphate levels. Using published values, the authors compute |Delta G_ATP| = 0.57 eV for liver, 0.62 eV for brain, and 0.68 eV for heart.¹ The brain value is high enough, in their model, to excite three amide I exciton quanta in protein alpha-helices, a cooperative state they associate with efficient molecular soliton transport.

When |Delta G_ATP| drops below about 0.6 eV, the model predicts only two quanta can be excited, the soliton lifetime and reliable transport distance shrink, and protein function becomes thermodynamically marginal. The paper gives a concrete threshold: below 0.6 eV the maximum reliable transmission distance for molecular solitons falls to roughly 19 nanometers, about one fifth of the rested range.¹

The second arm is electrophysiological. In the rested state the model uses E_Na = +71 mV and E_K = -89 mV. In the fatigue state the gradients are partially dissipated and the potentials shift toward equilibrium: E_Na = +60 mV and E_K = -70 mV.¹ With these shifted potentials the simulated CA1 pyramidal neuron becomes hyperexcitable and, under strong injected current, enters depolarization block more readily. At an injected somatic current of 1.15 nA the rested neuron maintains firing below 40 Hz, while the fatigued neuron shows a qualitatively different f-I curve.¹ The authors also note that extracellular potassium can exceed 10 mM during just 10 seconds of repetitive stimulation in rat hippocampus, a change that would accelerate the drift they model.

Where a skeptic should push

The most load-bearing assumption is that the Davydov soliton model is the right description of energy transfer in neuronal proteins under physiological conditions. Davydov's theory is elegant and influential, but it remains controversial: many biophysicists argue that thermal fluctuations at body temperature destroy quantum-coherent solitons too quickly for them to be biologically relevant. The paper does not settle that debate; it uses the model as a computational framework. If the soliton mechanism is wrong, the specific 0.6 eV threshold and the 19 nm distance collapse, even though the broader link between ATP depletion and ion-gradient failure would remain.

A second caution is that the fatigue potentials are chosen as representative endpoints, not measured from a graded fatigue experiment. The authors set E_Na = 60 mV and E_K = -70 mV to represent the fatigued state, but they do not show that these exact values correspond to a specific duration of mental workload or a specific drop in |Delta G_ATP| in human cortex. The CA1 neuron model is also a single, well-studied cell, not a cortical circuit or an organoid.

Third, the link to signal-to-noise ratio is asserted more than quantified. The abstract says fatigue decreases cortical SNR, and the mechanism makes that plausible, but the paper does not produce a numerical SNR value or a receiver-operating curve. For instrumentation purposes, that means the size of the effect on a real recording is still unknown.

Finally, the paper is about brain fatigue in intact, behaving humans, not organoids or acute slices. Organoids lack a vascular system, a sleep-wake cycle, and the systemic metabolic regulation that buffers human brain ATP, so they may be more vulnerable to the gradients the paper describes, but that is a hypothesis, not a result.

What metabolic drift means for the recording chain

The non-obvious implication is that a microelectrode array does not record a static biological signal; it records a signal whose statistics are coupled to the metabolic state of the tissue. Most array design treats the biology as a source with stable statistics and asks how to preserve every microvolt. This paper turns that around: the source itself drifts, and the drift is thermodynamically driven. If the tissue's ATP free-energy margin falls, the same neuron can move from sparse, well-timed firing to hyperexcitable bursts or into depolarization block, and the extracellular waveform can change amplitude and shape as ion gradients shift.

For long recordings, especially organoids kept in a dish for weeks or months, this is a genuine threat. A culture has no circulatory system to wash away extracellular potassium or replenish glucose and oxygen at the tissue core. Local ATP depletion can shift E_K and E_Na in the direction the paper models, raising excitability until neurons silence themselves in depolarization block. An array would see that as a rising event rate followed by a sudden collapse, and without metabolic monitoring it would be easy to misread the collapse as electrode failure or culture death rather than a reversible metabolic limit.

The opportunity is to close the loop. If metabolic state is a hidden variable that moves the signal source, then the instrumentation should measure it or control it. That means integrating the electrical record with proxies for energy state: oxygen, glucose, lactate, pH, or extracellular potassium. It also means designing stimulation and recovery protocols that do not push the tissue past its ATP margin. The paper's finding that extracellular potassium can rise past 10 mM in 10 seconds of repetitive stimulation is a direct warning for closed-loop stimulation schemes: the stimulus that elicits reliable spiking early in a session can drive the same neurons into block later, as the gradients that power their firing are depleted.

There is also a subtle implication for spike sorting. Sorting algorithms assume that each unit has a stable waveform template. If fatigue changes ion-channel availability and membrane conductance, the waveform can drift on the timescale of minutes to hours. A sorter that does not allow template adaptation would start misattributing spikes or splitting units as the tissue tires. The biophysics here gives a reason to expect that drift and a target variable, metabolic state, against which to model it.

The dual-use caution is about interpretation. A recording from a fatigued or metabolically compromised culture could show apparently pathological hyperexcitability or synchrony that is actually an artifact of ATP depletion. In drug screening or disease-modeling studies, that could produce false positives for compounds or mutations that perturb metabolism rather than intrinsic circuitry. The instrumentation fix is not better electrodes alone; it is better metabolic context.

The bottom line

Established: published brain metabolite concentrations give |Delta G_ATP| near 0.62 eV, and existing biophysical models predict that reducing this free-energy margin and the associated ion gradients shifts neuronal f-I curves toward hyperexcitability and depolarization block. Not established: the exact mapping between hours of mental workload and these electrophysiological endpoints, the magnitude of the effect on real extracellular recordings, or its specific behaviour in organoids. What would confirm the MEA-relevant reading is simultaneous electrical recording and metabolic monitoring during long organoid or slice sessions, showing that waveform drift, firing-rate changes, and sorting degradation correlate with falling ATP-related indices. What would break it is evidence that intact tissue buffers |Delta G_ATP| so tightly that the modeled gradients never reach the fatigue values in practice.