When Evolution Mixes Its Rules: New Study Reveals How Blended Update Dynamics Shape Fixation on Networks

Evolution is often imagined as a simple contest: individuals reproduce, offspring replace the less fortunate, and the fittest lineages gradually take over. But in real populations, the rules governing who replaces whom are rarely uniform. Some replacements begin with a death, leaving a vacancy that neighbors compete to fill; others begin with a birth, whose offspring pushes into an adjacent territory. A new study published in PLOS Computational Biology by David A. Brewster, Yichen Huang, Michael Mitzenmacher, and Martin A. Nowak examines what happens when a population blends these two replacement rules, and the results overturn the intuition that mixing rules should simply average their effects. The work, situated in the field of evolutionary graph theory, shows that even a modest amount of mixing can produce fixation probabilities and fixation times that rise, fall, or swing non-monotonically as the blend shifts from one rule to the other.

Evolutionary graph theory models a population as a network, or graph. Each of the N individuals occupies a vertex, and the edges connecting vertices define who can interact with whom for the purpose of competitive replacement. This framework, developed over the past two decades, has revealed that population structure is not a neutral backdrop: it can amplify or suppress selection, favor cooperation, and dramatically alter the fate of a new mutant. The two canonical update rules sit at the heart of this framework. Under death-Birth updating, abbreviated dB, an individual is chosen uniformly at random to die, and its neighbors compete to fill the vacant spot, with reproduction proportional to fitness. Under Birth-death updating, abbreviated Bd, an individual is chosen for reproduction with probability proportional to fitness, and its offspring replaces a randomly chosen neighbor. These two rules can yield strikingly different evolutionary outcomes on the same network, which is precisely why the choice between them has been treated as a fixed modeling decision.

The new study challenges that convention by introducing mixed updating. In each time step, the simulation performs a death-Birth update with probability δ and a Birth-death update with the remaining probability of 1 − δ. The parameter δ becomes a dial: at one extreme the population evolves entirely by dB dynamics, at the other entirely by Bd, and in between by an interleaving of both. The authors study two central quantities as functions of this dial. The first is the fixation probability, the chance that a single mutant with a given fitness advantage ultimately spreads to every vertex of the graph. The second is the fixation time, the expected number of updates required for that takeover, or for the mutant’s extinction under neutral evolution where all individuals have equal fitness.

The central surprise is that neither quantity behaves predictably as δ varies. On some graphs, fixation probabilities increase steadily with δ; on others they decrease; and on still others they trace non-monotonic curves, rising and then falling or the reverse. Fixation times show the same zoo of behaviors. This means that a researcher who models a population with a single update rule may be making claims that are artifacts of that choice rather than properties of the underlying population structure. A mutant that appears favored on a star graph under one rule may be disfavored under a blend, and the direction of the effect can flip as the mixture parameter moves through intermediate values. The study provides exact formulas demonstrating these sensitivities on specific structures, including cycles, stars, and more elaborate composite graphs, and classifies which structures are most and least sensitive to the mixing parameter.

Stars illustrate the drama vividly. In a star, one central hub is connected to many leaves, and the asymmetry between center and periphery makes the graph exquisitely sensitive to update rules. Under Birth-death updating, a mutant that reaches the center can rapidly colonize the leaves, while a mutant stranded on a leaf is likely to be eliminated. Under death-Birth updating, the random death event tends to strike a leaf, and the center then competes for the vacancy with its fitness-weighted advantage. When the two rules are mixed, the fixation probability becomes a function of δ that the authors derive in closed form, revealing exactly how the balance of hub control and leaf vulnerability shifts with the mixture. Cycles, by contrast, where every vertex has the same two neighbors, behave far more regularly, and the exact fixation formulas there show a different, gentler dependence on δ.

Beyond cataloging these sensitivities, the paper delivers two results of broad practical importance. First, the authors prove that nearly all unweighted undirected graphs have short fixation times. This is a reassuring statement about the tempo of evolution on networks: even though particular pathological structures can trap a mutant in long excursions before it either fixes or dies out, such slow dynamics are the exception rather than the rule. For the overwhelming majority of network topologies, the evolutionary contest resolves quickly, which matters both for the biological interpretation of graph models and for the computational feasibility of simulating them. Short fixation times mean that Monte Carlo estimates converge rapidly and that empirical populations modeled this way will not linger indefinitely in limbo between extinction and takeover.

Second, the study provides an efficient algorithm for estimating fixation probabilities on mixed-update dynamics across general graphs. Exact analytical formulas are available only for special structures such as cycles and stars; for arbitrary networks, the state space of possible mutant configurations grows exponentially with N, making exact computation intractable. The new algorithm exploits the proven structure of the dynamics to deliver accurate estimates without exhaustive enumeration, giving researchers a practical tool for analyzing real-world interaction networks, from microbial metapopulations to social networks of behavioral adoption, where neither the topology nor the update mechanism can be assumed uniform.

The theoretical significance of the work lies in how it reframes the role of update rules. In classical population genetics, the Wright-Fisher and Moran processes differ in detail but often agree on qualitative conclusions, so modelers felt licensed to choose whichever was convenient. Evolutionary graph theory shattered that comfort: on structured populations, dB and Bd updating can disagree not just quantitatively but qualitatively, flipping whether a graph is an amplifier or a suppressor of selection. By introducing δ as a continuous parameter, Brewster and colleagues turn a discrete modeling dilemma into a continuum that can be analyzed, classified, and, where data permit, measured. The non-monotonic behaviors they prove are particularly instructive, because they demonstrate that intermediate mixtures are not interpolations between the extremes; the mixed process has its own character, with fixation outcomes that cannot be guessed from either pure rule alone.

The biological motivation for mixed updating is easy to appreciate. In a microbial biofilm, some replacement events may follow the death of a cell that opens a niche for its neighbors, while others follow the dispersal of a proliferating cell into adjacent territory. In ecological communities, disturbance events create vacancies, while dispersal creates colonists, and both processes operate simultaneously at different rates. In social evolution, imitation dynamics can resemble death-Birth updating, where an individual abandons a behavior and copies a neighbor, while migration dynamics resemble Birth-death, where a successful individual spreads its strategy outward. A mixed-update model with a tunable δ offers a more faithful description of such systems than either pure rule, and the new results supply the mathematical machinery to interpret what the mixture does to the fate of a rare variant.

What emerges from the study is a richer map of evolutionary dynamics on networks than the field has previously possessed. Fixation probabilities and times, the twin currencies of evolutionary graph theory, are revealed to be functions not only of the fitness advantage and the topology but of the replacement protocol itself, in ways that can be computed exactly for key structures and estimated efficiently for general ones. The proofs that nearly all graphs admit short fixation times, together with the classification of δ-sensitivities on cycles, stars, and composite structures, give both theorists and experimentalists concrete handles on a problem that was previously handled by fiat. As evolutionary graph theory continues to inform questions from cancer progression to the spread of cooperation, the message of this work is clear: how a population replaces its members is not a technical footnote but a first-order determinant of evolution, and the space between the standard rules deserves to be explored, not skipped.

Subject of Research: Mixed death-Birth and Birth-death updating in evolutionary graph theory and its effects on fixation probabilities and times

Article Title: Mixed updating in structured populations

Article References: Brewster, D. A., Huang, Y., Mitzenmacher, M., & Nowak, M. A. (2026). Mixed updating in structured populations. PLOS Computational Biology, 22(9), e1014829. https://doi.org/10.1371/journal.pcbi.1014829

Image Credits: AI Generated

DOI: 10.1371/journal.pcbi.1014829

Keywords: evolutionary graph theory, fixation probability, fixation time, death-Birth updating, Birth-death updating, population structure, networks, neutral evolution, constant selection, stars, cycles, PLOS Computational Biology