Finding. In human cortex, the tissue that carries unique visual information is relatively poor in mitochondria and respiratory capacity. The tissue that carries unique semantic information is relatively rich in mitochondria. 7T fMRI during natural scenes, encoding models that partial out the two feature classes, careful spatial statistics, and postmortem mitochondrial atlases agree on the signs. Transcriptomes run opposite mitochondrial programs on the same two axes. Representation has a bioenergetic anatomy.
Why this paper matters
Cognitive neuroscience rarely touches organelle atlases. This paper does, and it does not find a generic “harder computation, more mitochondria” smear. Visual and semantic unique variances point opposite ways. That is a constraint on any story that treats cortical mitochondria as one map.
It is still a spatial correlation between living function and dead-tissue density. Treat it as alignment.
What they actually measured
7T scene-viewing fMRI, unique-variance encoding splits, autocorrelation-preserving inference, mitochondrial density and respiratory-capacity atlases, transcriptomic enrichment.
How to read the score
High seventies. Human, opposing signs, organelle atlases. Confidence is medium. Score 78. The heuristic 57 undersold it.
What to do with it
If you map cortical energy or semantic/visual systems, overlay mitochondrial density instead of assuming uniform OXPHOS. Do not claim that thinking “uses more mitochondria” without naming the axis. The directional implication is that visual-specific and semantic-specific cortex are provisioned with mitochondria in opposite directions.