Mitochondria, metabolism, and autonomic regulation are reshaping autism neuroscience. But energy is fuel — and fuel alone can’t explain why the same challenge lands so differently in different nervous systems. We’re missing a layer.
Consider three contrasts among autistic nervous systems under load. One child loses access to speech and daily-living skills during an illness and regains them only slowly; another moves through a comparable infection unchanged. One adult recovers from an overwhelming week after a weekend of rest; another, with a strikingly similar metabolic profile, drops into burnout that lasts weeks, months, or years. One person is transformed by an accommodation that barely moves the needle for someone who, on paper, looks almost identical. Nominally similar demands, radically different trajectories.
Energy-forward models sharpen an important part of that picture, but they leave the divergence unexplained — and seeing why points to what’s missing.
Autism neuroscience is in the middle of an energy revolution. Alongside the genetics, researchers are increasingly examining mitochondrial function, metabolism, immune activity, autonomic regulation, sleep, inflammation, and the energetic cost of building and running a developing nervous system. The Transmitter publication has devoted a continuing series, “The mitochondria candidate,” to the question of whether the cell’s powerhouse is a hidden factor in autism. In a recent paper, Robert Naviaux proposed a three-hit metabolic signaling model in which an ancient cellular stress response — the cell danger response — stays switched on through critical windows of neural development. Some transdiagnostic studies report partly overlapping mitochondrial signatures across autism and ADHD, though the findings remain heterogeneous. The through-line is unmistakable: energy matters, and it matters more than the field long assumed.
But energy is fuel, and fuel alone can’t tell you what a nervous system does with it. That’s the gap beneath those opening cases. If energy were the whole story, nervous systems with similar energetic profiles should behave alike. They don’t — not even close.
Here’s the distinction I think the field needs. Bioenergetic state constrains what a nervous system can do; it doesn’t, by itself, determine what that system will do. Fuel, demand, and conversion efficiency matter — but their effects are filtered through the developmentally assembled, continually plastic organization of the brain and body, and through the environment in which that organization functions.
So the vehicle metaphor only takes you so far. A race car and a pickup can both run out of gas, and no amount of fuel data tells you why one corners at speed and the other hauls gravel. But a nervous system is not a vehicle assembled before the fuel arrives. It is a living system built while running: metabolic, immune, and activity-dependent signals help shape its organization, and that organization, in turn, channels how later physiological stress is expressed. Fuel and structure are coupled, not separable — which is exactly why the same measured state can mean different things in differently organized systems.
That coupling runs across two timescales. Metabolism may help shape neural organization during development and later constrain what that organization can sustain under changing conditions — two distinct causal roles on different timescales, but both requiring a model of how physiological state meets an already-organized, still-adapting system.
Hold that up to the current findings, and a pattern appears. A recurring limitation of single-marker accounts is fragmentation. The Children’s Autism Metabolome Project found real bioenergetic signals, but individual markers identified small, partly overlapping subgroups while larger panels covered much more of the cohort — not one autism-specific signature so much as a layered pattern of overlapping biological variation. A recent genetic study identified four phenotypic classes with markedly different genetic architectures and developmental profiles. Many biology-forward accounts resolve into partly overlapping subgroups rather than one convergent pathway.
I don’t think this is merely a measurement problem that the field will eventually clean up. These patterns are what you’d expect if similar measured states arise from—and play out differently within—systems with different organization. The reading on the gauge means one thing in one system and something else in another.
This is where the missing layer comes in. Between physiological state and behavior sits an emergent property: what a given organization actually produces when it runs under a given energetic and environmental load. Call it regulatory capacity—the time-varying margin within which a person can absorb demand, retain access to skills, and recover. Adaptive flexibility, burnout, recovery time, and the loss and return of function—the patterns autistic people, families, and clinicians actually track—live here, at the meeting point of state, organization, and environment, and in none of them alone.
This is the layer I’ve spent the most time on; in recent work, I’ve tried to formalize one dimension of it as regulatory bandwidth. I won’t rehearse the formalism here — the point is more general than any one construct. Regulatory capacity overlaps with older ideas like physiological reserve, allostatic capacity, and resilience, but it places the emphasis on the dynamic coupling of brain, body, and environment, and on how function comes and goes unevenly under load. And it needn’t be an invisible essence: it can be inferred from how function bends and recovers — autonomic recovery, sensory tolerance, recovery time — as measurable demand changes. If this account is right, models that combine physiological state, developmental and network organization, environmental demand, and prior load should predict who loses function and who recovers better than metabolic markers, connectivity, or diagnostic labels alone. That’s a claim you can test. (Hogenkamp, 2026)
Naviaux’s model is interesting precisely because it already points toward this dynamic view. A cell danger response that stays switched on isn’t only an energy deficit; it’s a regulatory configuration — a system holding a stance. Held too long, in the wrong developmental window, that stance does different things depending on the organization holding it. The model reaches for the timing and the dynamics; the organization is what tells you where those dynamics land.
The missing layer: Physiological state, system organization, and emergent capacity give heterogeneous findings a place within the same model.
None of this rivals the energy program; it’s the layer that makes the program’s own results legible. Separate physiological state from organization from emergent capacity, and the subgroups stop looking like a scandal — they look like what a heterogeneous set of systems under variable load should produce. The metabolomic slices, the phenotypic classes, the transdiagnostic overlaps don’t get discarded; they get a place to sit.
It also reframes several familiar but poorly explained patterns. Under this hypothesis, autistic burnout isn’t a motivational failure or a character weakness; it emerges when cumulative demand repeatedly exceeds the margin a given system can sustain and restore—which is why rest that restores one person may leave another far from recovered. Loss of function during illness becomes more intelligible: constrain physiological state far enough, and systems with narrower margins may lose access to capacities that systems with wider margins retain. Comparable load, different capacity, different outcome.
None of this lessens the energy revolution. Energy is necessary — rest matters, metabolism matters, mitochondria may matter enormously — but necessary isn’t sufficient. The question worth asking is no longer whether energy is involved, but what it meets on the way to an outcome: the organization that turns a given state into capacity or into collapse.
We’ve learned to measure the fuel. The harder task ahead is understanding the living system that spends it.
That is the translation challenge the Center for Adaptive Stress is taking on: building a shared language across metabolism, stress physiology, interoception, network biology, neurodiversity, and lived experience so that findings generated at different levels can become part of a coherent and usable picture.
Autism is an especially revealing place to begin, but the implications extend far beyond it. Burnout, chronic illness, post-infectious conditions, fluctuating disability, and stress-related disease all raise versions of the same question: How does a living system maintain—or lose—access to capacity under changing demand?
This is an exciting moment to watch—and help shape—these fields as they converge. The coming revolution may be as much conceptual and translational as technological. The next major advance may not come from one decisive biomarker or pathway, but from learning how markers acquire meaning within living systems that are adaptive, history-dependent, context-sensitive, and organized across levels.
The concepts are still being assembled. The relevant disciplines do not yet fit neatly together. But that is often what an intellectual transition looks like from inside it.
The energy revolution has opened the door. The work ahead is to make the missing layer visible—and the emerging science usable.
Selected References
Al-Kafaji, G., Jahrami, H. A., Alwehaidah, M. S., Alshammari, Y., & Husni, M. (2023). Mitochondrial DNA copy number in autism spectrum disorder and attention deficit hyperactivity disorder: A systematic review and meta-analysis. Frontiers in Psychiatry, 14, Article 1196035. doi:10.3389/fpsyt.2023.1196035.
Hogenkamp, L. (2026). Regulatory bandwidth: A theoretical integration of present multisystem stress-regulatory capacity. Psychoneuroendocrinology, 192, Article 107955. https://doi.org/10.1016/j.psyneuen.2026.107955
Hogenkamp, L., Sanghavi, D., & Natri, H. (2026). Toward an emergent paradigm for neurodiversity and health. Autism in Adulthood. Advance online publication. https://doi.org/10.1177/25739581261433443
Litman, A., Sauerwald, N., Green Snyder, L., Foss-Feig, J., Park, C. Y., Hao, Y., Dinstein, I., Theesfeld, C. L., & Troyanskaya, O. G. (2025). Decomposition of phenotypic heterogeneity in autism reveals underlying genetic programs. Nature Genetics, 57, 1611–1619. https://doi.org/10.1038/s41588-025-02224-z
Moisse, K. (2023, June 26). The mitochondria candidate: Is the cell’s powerhouse a hidden factor in autism? The Transmitter. https://www.thetransmitter.org/the-mitochondria-candidate-is-the-cells-powerhouse-a-hidden-factor-in-autism/
Naviaux, R. K. (2026). A 3-hit metabolic signaling model for the core symptoms of autism spectrum disorder. Mitochondrion, 87, Article 102096. https://doi.org/10.1016/j.mito.2025.102096
Smith, A. M., Natowicz, M. R., Braas, D., Ludwig, M. A., Ney, D. M., Donley, E. L. R., Burrier, R. E., & Amaral, D. G. (2020). A metabolomics approach to screening for autism risk in the Children’s Autism Metabolome Project. Autism Research, 13(8), 1270–1285. https://doi.org/10.1002/aur.2330
Leave a Reply