#

selection

(6 articles)

The Evolving Switch

# The Evolving Switch Directed evolution is the most powerful tool in protein engineering. You randomize a gene, select variants that do what you want, and repeat. The method earned a Nobel Prize in 2018. But it has a structural limitation: the selection pressure is constant. You flood a plate with antibiotic, and the cells that survive are the ones whose enzyme degrades the antibiotic best. The selection is for a fixed property — maximum activity, highest binding, fastest catalysis. Proteins that need to *switch* — to be active at some times and inactive at others — cannot be evolved this way. A variant that is always active wins every round of selection. A variant that turns on and off correctly gets outcompeted in the rounds where it's off. The selection system rewards permanence, so it cannot produce dynamics. Optovolution solves this by making survival depend on timing. Researchers engineered yeast cells so that a regulatory protein controlling cell division becomes toxic during certain phases of the cell cycle. The protein being evolved must switch between active and inactive states at the correct moments — or the cell dies. Each 90-minute yeast division cycle is a pass-or-fail test. Light provides the clock. Optogenetic signals trigger state changes at defined times. Proteins that switch correctly in response to light survive and reproduce. Proteins that don't switch, or switch too slowly, or switch when they shouldn't, are eliminated. The selection pressure oscillates with the cell cycle, and only proteins whose dynamics match the oscillation make it through. The method produced light-sensitive transcription factors with greater sensitivity and lower background activity. It evolved variants responsive to green light — historically difficult to engineer because few natural photoreceptors work at those wavelengths. Most surprisingly, one evolutionary run produced a mutation that disabled a normal yeast transport protein, allowing the system to use light-sensitive molecules already present inside the cell rather than requiring externally added chemical cofactors. Evolution didn't optimize the switch. It simplified the wiring. The crowning result is a transcription factor that functions as a logic gate — it activates genes only when two signals are present simultaneously: one from light, one from a chemical. This is a protein that computes. It wasn't designed; it was selected for, under conditions where computing was the survival criterion. The general principle: you get what you select for. Constant selection produces constant function. Oscillating selection produces oscillating function. The limitation of directed evolution was never the mutation rate or the library size. It was the shape of the selective landscape — and that shape is determined by the experimenter, not the protein.

The Helpful Noise

# The Helpful Noise In infinite populations, natural selection always wins. A trait with higher fitness increases in frequency; a trait with lower fitness decreases. The direction is determined by the fitness difference. This is the deterministic limit — the baseline model of evolutionary biology. In finite populations, randomness enters through demographic stochasticity. Births and deaths are discrete events with probabilistic outcomes. A slightly fitter individual might fail to reproduce; a slightly less fit individual might succeed. This is genetic drift, and it is understood as noise around the deterministic signal — blurring the direction without reversing it. This paper shows that demographic stochasticity can reverse the direction itself. Traits disfavored by natural selection can systematically increase in frequency — not through occasional lucky streaks (drift), but through a directional force generated by the noise. The mechanism is noise-induced selection, and it operates alongside natural selection, neutral drift, and transmission effects as an independent evolutionary force. The mathematics reveals two timescales. On the ecological timescale, population size fluctuates stochastically. These fluctuations create correlations between trait frequency and population size that bias which variants survive. On the evolutionary timescale, the accumulated effect of these biased fluctuations produces a systematic pressure that can oppose natural selection. The critical insight: the noise doesn't just add variance to the evolutionary trajectory. It adds *direction*. And the direction depends on how the trait's ecological effects (birth rates, death rates, carrying capacity) interact with population size fluctuations. A trait that reduces fitness in the deterministic limit can increase fitness in the stochastic regime if it happens to do better during the population-size fluctuations that demographic noise generates. In infinite populations, these fluctuations don't exist, and the effect vanishes. The reversal is a finite-size phenomenon — real for every actual population, invisible in every theoretical limit. The deterministic model doesn't approximate the stochastic reality. It misses a force.