About one in four positions in folded globular proteins were described as having significant dark energy, above a fitted threshold of about 1 kilocalorie per mole. In the review's framework, dark energy is the difference between evolutionary free energy and physical folding free energy. It represents functional constraints that remain after folding stability is considered.
That is the central argument of a review that treats protein evolution as a joint negotiation between folding stability and biological function. The two pressures can conflict, the review says, and the difference between evolutionary and folding energies offers a quantitative way to describe the cost of function beyond stability.
The gap between folding and function
The signal was not described as evenly spread through a protein. The review reports that dark energy localizes around enzyme catalytic centers and declines smoothly with distance from them. It also reports that dark energy can map functional and allosteric regions, or sites linked to the regulation of a protein's activity.
That distinction is especially important at catalytic sites. The summarized evidence describes functional constraints as stronger than stability constraints there. At binding sites, stability constraints were described as secondary, but not coupled to function in a general way. The review therefore does not support the idea that function and stability generally move in opposite directions.
How the signal is estimated
The review brings together experimental assays and computational methods that pair a fitness proxy with a stability proxy. One approach, deep mutational scanning, measures nearly every possible single-site sequence variation under the same conditions. Other tools use protein language models to predict how a point mutation changes sequence likelihood, while inverse-folding models predict stability changes from a specified backbone. The comparison is designed to expose functional effects that a folding-only measure can miss.
The document is a review rather than a single experiment. It cites site-saturation measurements spanning 500 human protein domains and discusses missense variants across the complete human proteome. The range of material is part of the story: the reported result is assembled from different experimental and computational approaches.
What the pattern says about mutations
In one reported analysis, FunC-ESMs classified missense variants in the complete human proteome into three categories. Sixty percent were WT-like, 16% were classified as total-activity-loss, and 24% as stable-but-inactive. Those categories make the framework's key separation concrete: a mutation may leave a protein stable while its activity is lost.
Around one half of protein sequence variants associated with loss of function were reported as not being explained solely by folding effects. The review uses that gap to motivate measuring functional costs beyond folding stability, rather than treating stability as the whole account of a variant's effect.
A useful signal, with a calibration catch
That conclusion comes with a calibration warning. Raw evolutionary scores are not comparable across protein families unless differences in selection temperature are taken into account or protein-specific calibration is applied. A dark-energy estimate therefore needs to be interpreted in its protein-family context, rather than treated as a universal scale.
The review's broader catalogue includes dark-energy patterns in functional and allosteric regions, distinctive signals in disease-related, oncogenic and gain-of-function variants, and both conserved and lineage-specific features across homologous proteins. The authors present the framework as a possible aid for functional-site prediction, disease-variant interpretation and protein engineering.
The work's message is therefore narrower than a claim that folding and function always pull in opposite directions. It is that some positions carry an additional functional cost beyond folding stability, and that the cost can be studied by comparing evolutionary and folding signals. The review says the framework could help locate functional regions and interpret variants, but those uses still depend on calibration of the underlying scores.
The acknowledgments name support from the European Union, the Barcelona Collaboratorium for Modelling and Predictive Biology, CONICET and the Universidad de Buenos Aires.
Paper data and sources
Original title: Dark energy: the cost of function in protein evolution
Authors: Ezequiel A. Galpern, Federico Caamaño, Ignacio E. Sánchez, Diego U. Ferreiro
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-27
DOI: Not available
Original paper · Full text