Cells spend energy to keep gene expression noisy, and that may be the point

On 13 July 2026, researchers published a paper arguing that the seemingly random on-off flickering of genes inside living cells is not cellular sloppiness. It is a feature, and an expensive one. The work, reported by Phys.org and originally described in Physical Review Research, models gene regulation as a system that pays an energetic premium for precision, and uses bursts of stochastic activity to get there.

The argument lands at an awkward moment for a field that has spent two decades treating biological noise as a nuisance term to be averaged out. If the new model holds up, the textbook picture of a gene switching cleanly between "on" and "off" states is closer to a marketing slide than a description of what actually happens inside the nucleus of a human, mouse, or yeast cell.

Precision, priced in ATP

The first mathematical models of gene regulation were developed in the 1960s, and they assumed a tidy world. Genes were either transcribed or they were not; the variables were binding affinities, promoter strengths, and the concentration of transcription factors in the neighbourhood. The cell was, in effect, treated as a small circuit.

That abstraction has been crumbling for years. Single-molecule imaging and fluorescence microscopy routinely show that identical cells, carrying identical DNA, make widely different choices about which genes to activate and when. Two neighbouring liver cells, biopsied seconds apart, can show a two-fold difference in the output of the same transcript. The technical term for this is gene expression noise, and the experimental literature has documented it across bacteria, yeast, fruit flies, mice, and human cell lines.

The new paper takes the next step. It treats noise not as a side effect but as a quantity the cell can budget for. Borrowing from statistical physics, the authors recast precision as a thermodynamic cost: each bit of reliable output requires a corresponding expenditure of metabolic energy, much as compressing a signal against thermal jitter costs power in an engineered communication channel.

A second look at the textbook

The dominant explanation for noise has been that it is an unavoidable consequence of small numbers. A cell carries only one or two copies of a given gene; transcription factors bind and unbind stochastically; mRNA is produced in random bursts. In this reading, the system is as precise as physics allows, and the residual flicker is waste.

The new model challenges that. If precision were the only objective, the authors argue, evolution would have selected for tighter regulatory machinery, more binding sites, more feedback loops, more redundant control. The fact that cells tolerate, and even engineer, a baseline level of randomness suggests that noise carries information of its own. A gene that switches on and off unpredictably is harder for a virus or a parasitic sequence to mimic, harder for a downstream circuit to lock into a single state, and easier for a population of cells to diversify in case the environment changes.

That interpretation is consistent with a wider shift in the field, in which stochasticity is increasingly treated as a design parameter rather than a defect. Studies of bacterial persistence, cancer cell dormancy, and immune cell diversification have all pointed in the same direction. The Phys.org coverage is the first to package the argument as a unified thermodynamic framework.

What it would take to falsify

The model is not difficult to challenge, and the paper's own authors list the obvious tests. If precision costs energy, then cells placed under metabolic stress, starved of glucose, deprived of oxygen, dosed with mitochondrial inhibitors, should show noisier gene expression. Conversely, cells engineered to express the same gene from extra copies, or with stronger promoter sequences, should show reduced noise at the cost of higher baseline energy consumption.

Both predictions are within reach of existing tools. Single-cell RNA sequencing can quantify noise directly; calorimetry can measure the energy bill. A clean experiment would put numbers on the price of precision in joules per bit and compare it across cell types.

The framework is also vulnerable to a simpler critique. It may be that gene expression noise is not a single phenomenon but several, each with its own cause and cost. The model treats noise as a unified budget item; the biology may resist that tidy accounting. Whether the abstraction survives contact with the data will be settled in labs, not in press releases.

Why the framing matters

For most of the post-genomic era, biological noise has been a footnote, something to be controlled for, batch-corrected, and averaged away in the search for the "real" signal. That instinct reflects the history of the field. Molecular biology was built on the assumption that cells are reliable machines, and the experimental infrastructure, microarrays, bulk RNA sequencing, quantitative PCR, was designed to deliver population averages.

Single-cell technologies broke that assumption in practice, and the new paper tries to do the same in theory. If the cost-of-precision reading holds, it would explain why cells have not evolved to be quieter than they are, why cancer cells in particular seem to revel in transcriptional noise, and why efforts to engineer perfectly predictable gene circuits in synthetic biology keep running into walls.

The stakes are concrete. Drug discovery programmes increasingly target gene-regulatory proteins, and cancer therapies in development aim to push tumour cells out of noisy, drug-tolerant states into well-defined ones. Knowing whether noise is a bug or a feature, and what it costs to suppress, changes which levers are worth pulling. It also reframes a sixty-year-old body of theory. The 1960s models did not get gene regulation wrong, exactly; they just left the energy budget out of the spreadsheet.

What the sources do not yet settle

The Phys.org summary does not specify which cell types the model's predictions have been tested against, nor does it name the lead authors or the institutional affiliations beyond the journal of publication. The energy-cost numbers, if any are reported in the underlying paper, are not reproduced in the available coverage. Readers who want to assess the framework against their own data will need to consult Physical Review Research directly, where the full equations and parameter choices will appear.

What the reporting does establish is the shape of the argument: precision in gene expression is not free, cells appear to pay for it in metabolic currency, and the residual noise is the change. Whether that framing displaces the older, cheaper picture of the cell as a tidy circuit remains an open question, one that the next round of single-cell experiments will be well placed to answer.

This piece frames gene expression noise as a thermodynamic cost rather than a nuisance term. Where wire coverage described the result as "surprising", Monexus reads it as the latest move in a longer convergence between statistical physics and molecular biology.