The Hidden Evolutionary Cost of Death Regulation

When How Populations Control Their Size Matters More Than You Think

Every population on Earth faces the same basic problem: grow too fast and you'll exhaust your resources; stay too small and you might vanish entirely. So every species has evolved mechanisms to keep its numbers in check. Some, like elephants, simply reproduce slowly—fewer offspring, less often. Others, like salmon, flood the world with eggs but let most of their children die. In the language of biology, one regulates through birth (suppressing reproduction) and the other through death (elevated mortality).

For decades, evolutionary biologists assumed this distinction was cosmetic. Whether a population keeps its numbers stable by making fewer babies or watching more die, the thinking went, shouldn't matter to the long-term evolutionary trajectory. The average behavior would be the same.

A new paper in arXiv by Yunbei Pan and Tom Chou of UCLA shows this assumption is wrong—not just a little wrong, but profoundly, fundamentally wrong. In a sweeping mathematical analysis spanning dozens of species and hundreds of pages of derivations, they demonstrate that the mechanism of population regulation leaves a deep imprint on evolutionary fate. Death-regulated populations experience stronger demographic noise—larger, more erratic swings in numbers—and as a result, they drift toward extinction even when natural selection has no preference whatsoever. Birth-regulated populations, by contrast, smooth out those dangerous fluctuations and persist. The mechanism of regulation isn't a footnote to evolutionary dynamics. It is the dynamics.

"A population may be regulated through reduced fecundity, elevated mortality, or some combination of both," the authors write. "These mechanisms are biologically distinct." And their consequences, Pan and Chou show, are enormous.

The Science

The paper's starting point is a question that sounds deceptively simple: what happens to a multi-species ecosystem when you impose a hard carrying capacity—a ceiling on how many individuals can exist at once? This is a universal constraint. No population grows forever. Trees don't fill the sky; bacteria don't consume the oceans. Something always brings growth to a halt.

Biologists have modeled this for decades using what are called birth-death processes: mathematical descriptions of how populations change as individuals are born (adding to the population) or die (subtracting from it). The canonical framework is the Moran model, named after the British statistician Patrick Moran, which treats births and deaths as coordinated events that keep the total population exactly fixed—a useful simplification that lets researchers focus on how the frequencies of different species change over time, rather than worrying about the total number fluctuating.

Pan and Chou extend this framework in a way that had not been systematically explored: they allow the regulation of population size to be asymmetrically distributed between birth and death events. They introduce a parameter they call ${\alpha }_i$ (alpha-i) for each species $i$, which controls how that species' population is regulated. When ${\alpha }_i=1$, the species is purely birth-regulated—density suppresses reproduction, but deaths occur at a constant background rate no matter how crowded things get. When ${\alpha }_i=0$, the species is purely death-regulated—everyone reproduces freely, but crowding increases the death rate. Any value in between represents a mixture of the two mechanisms.

They work with a $d$-species birth-death process and project it onto a $(d-1)$-dimensional Moran process—a mathematical trick that separates the fast dynamics of total population size from the slower evolutionary dynamics of species frequencies. They then apply a diffusion approximation, converting the discrete, stochastic birth-death events into a continuous Fokker-Planck equation that describes how probabilities evolve over time. This is the same mathematical machinery physicists use to describe Brownian motion, and evolutionary biologists have used it since Motoo Kimura's pioneering work in the 1950s and '60s.

The critical move is the introduction of a diffusion tensor—a mathematical object that captures the amplitude and structure of demographic noise, the random fluctuations that arise from the inherent stochasticity of birth and death events. Pan and Chou show that this diffusion tensor depends explicitly on the regulation parameters ${\alpha }_i$. That dependency is the paper's central result.

To make their analysis concrete, the researchers examine three progressively more general scenarios. The first, which they call the fully neutral limit, assumes all species have identical intrinsic birth rates $b$, death rates $d$, competitive interactions $\gamma$, and regulation parameters $\alpha$. In this regime, there is absolutely no selection, no advantage conferred by any trait—the species are perfectly equal in every deterministic sense. The second scenario, semi-neutral, maintains equal $b$, $d$, and $\gamma$ but allows different species to have different ${\alpha }_i$ values—meaning they regulate their populations in different ways. The third scenario, quasi-neutral, allows small perturbations in all parameters simultaneously, capturing the realistic case where species differ slightly but not dramatically.

For the fully neutral and semi-neutral cases with just two species, Pan and Chou derive exact analytical expressions—closed-form solutions that give precise predictions without approximation. For higher-dimensional systems with three or more species, they use perturbation theory and numerical methods to characterize what they call the spectral gap, which measures how quickly the population loses diversity over time.

The mathematical heavy lifting here is formidable. The authors develop a conjugate representation of the frequency simplex using a logarithmic coordinate transformation, converting the simplex's boundaries (where species might go extinct) into infinities in the new coordinate system—a trick that makes the diffusion tensor strictly positive definite and numerically stable. They decompose the diffusion tensor into an isotropic background and rank-one perturbations, enabling direct application of spectral methods. This structure lets them compute the spectral gap analytically even when regulation is heterogeneous across many species.

What They Found

The first major finding concerns the deterministic dynamics—what happens to the average population over time, ignoring random fluctuations. Pan and Chou show that the net growth rate of each species,

$${\beta }_i(\mathbf{n})-{\delta }_i(\mathbf{n})=(b_i-d_i)n_i-\frac{b_i}{K}\sum _{j=1}^d{\gamma }_{ij}{n}_j{n}_i,$$

is completely independent of ${\alpha }_i$. Two species that regulate their populations through entirely different mechanisms—say, one through reproductive suppression and one through mortality—will follow exactly the same trajectory at the deterministic level. The mean-field prediction makes them indistinguishable.

But the total event rate—the raw frequency at which births and deaths occur—is

$${\beta }_i(\mathbf{n})+{\delta }_i(\mathbf{n})=(b_i+d_i)n_i+(1-2{\alpha }_i)\frac{b_i}{K}\sum _{j=1}^d{\gamma }_{ij}{n}_j{n}_i,$$

which depends linearly on ${\alpha }_i$. When ${\alpha }_i=0$ (pure death regulation), the total event rate is elevated by an amount proportional to the competition term. When ${\alpha }_i=1$ (pure birth regulation), that extra term vanishes. This is the key physical distinction: a death-regulated population experiences more total turnover events per unit time, which amplifies demographic noise.

To see what this means concretely, consider the two-species case. Pan and Chou derive the exact fixation probability $q(p_0)$—the probability that species 1 eventually takes over the entire population, starting from an initial frequency $p_0$:

$$q(p_0)=\frac{p_0(1-{\alpha }_2{u}^{\ast })}{1-u^{\ast }[(1-p_0){\alpha }_1+p_0{\alpha }_2]},u^{\ast }=1-\frac{d}{b}.$$

When ${\alpha }_1={\alpha }_2$, this simplifies to the classical neutral prediction $q(p_0)=p_0$: if you start at 50% frequency, you have a 50% chance of eventual fixation. But when ${\alpha }_1\ne {\alpha }_2$, something remarkable happens. Even with zero fitness differences between the species, the regulation asymmetry generates a stochastic bias. A species with higher $\alpha$—more birth-regulated, less death-regulated—has a fixation probability $q(p_0)>p_0$ for all interior frequencies. It is stochastically favored, not because it is genetically superior, but because it experiences weaker demographic noise.

This bias peaks at intermediate frequencies (the numerator $p_0(1-p_0)$ vanishes at the absorbing boundaries, as it must—complete dominance or extinction are irreversible). But at $p_0=0.5$, the maximum bias can be substantial. For biologically reasonable parameter values where $b/d$ is large (a species that reproduces much faster than it dies), the effect is pronounced. The heatmap in Figure 1b shows the maximum fixation bias $q(1/2)-1/2$ across the full range of regulation parameter pairs $({\alpha }_1,{\alpha }_2)$, revealing a clear diagonal structure: the species with the larger $\alpha$ value always enjoys the stochastic advantage.

The researchers confirmed these exact predictions against stochastic simulations using the Gillespie algorithm, the gold standard for exact numerical solution of stochastic reaction networks. The matches are exact—the analytical predictions and simulation results are indistinguishable.

Perhaps the most elegant result in the paper is the derivation of an $\alpha$-dependent effective population size $N_e(\alpha )$. In classical population genetics, effective population size is the parameter that relates the discrete-generation Wright-Fisher model to the continuous-time Moran model by matching their diffusion coefficients. Pan and Chou extend this concept to account for regulation mechanism:

$$N_e(\alpha )=\frac{\frac{K}{2}\left(1-\frac{d^2}{b^2}\right)}{1-\alpha \left(1-\frac{d}{b}\right)}.$$

This formula encodes a remarkable physical picture. At $\alpha =1/2$—when regulation acts equally on births and deaths—$N_e=K\left(1-d/b\right)$, which is essentially the carrying capacity scaled by the net growth rate. But as $\alpha$ increases toward 1 (more birth regulation), $N_e$ increases. As $\alpha$ decreases toward 0 (more death regulation), $N_e$ decreases. The ratio between the pure birth-regulated and pure death-regulated effective population sizes is

$$\frac{N_e(\alpha =1)}{N_e(\alpha =0)}=\frac{b}{d},$$

which can be a large number for rapidly reproducing organisms. A bacterial population that is birth-regulated (slow replication under stress) has an effective population size larger by a factor of $b/d$ compared to an otherwise identical death-regulated population (rapid death under stress). This factor directly controls the strength of genetic drift: larger $N_e$ means weaker drift, slower loss of genetic diversity, and greater stochastic resilience.

For multi-species systems with more than two types, where the diffusion tensor no longer reduces to a scalar, Pan and Chou turn to spectral gap analysis. The spectral gap is the smallest positive eigenvalue of the Fokker-Planck operator—it determines the rate at which the population loses diversity over time. In a fully neutral system of $d$ species with identical regulation mechanisms, all decay rates are degenerate. But when regulation is heterogeneous, this degeneracy breaks. The initially uniform spectral modes split into distinct levels, each corresponding to a pair of species with a characteristic turnover rate $W_k=(b_k+d_k)/u^{\ast }+(1-2{\alpha }_k)b_k{\bar{\gamma }}_k$. The global rate of diversity loss is governed by the pair with the lowest aggregate turnover rate—the bottleneck of the system. The perturbation theory predictions match simulation results to within 10% across a wide range of regulation heterogeneity parameters.

Why This Changes Things

This result has implications that ripple far beyond the mathematics. The first and most fundamental is conceptual: it revises a deeply held assumption in population genetics. The classical Wright-Fisher and Moran models, which form the bedrock of modern evolutionary theory, treat birth and death events as interchangeable apart from their effect on total population size. They implicitly assume regulation is symmetric between the two processes. Pan and Chou show this assumption has consequences. The effective population size, long treated as a property of total population magnitude, turns out to also depend on how that magnitude is maintained.

This matters for how we interpret real evolutionary data. When a population geneticist estimates effective population size from observed genetic diversity—using, for instance, the standard formula $N_e=\theta /(4\mu )$ where $\theta$ is a measure of heterozygosity and $\mu$ is the mutation rate—they are implicitly estimating an effective $N_e$ that combines both the census population size and the regulation mechanism. Two populations of the same physical size but different regulation modes will show different levels of genetic diversity, not because they differ in selection or mutation, but because death-regulated populations experience more drift. Untangling these effects from genetic data alone is difficult, which means historical estimates of $N_e$ from natural populations may systematically underestimate the influence of regulation mechanism.

The second major implication concerns immune system dynamics. Pan and Chou explicitly motivate their analysis with the example of T cell clones—distinct populations of immune cells, each expressing a different antigen receptor. The adaptive immune system maintains a diverse repertoire of T cell clones, allowing it to recognize a wide variety of pathogens. This diversity is essential: a person whose T cells have collapsed into a narrow set of clones is vulnerable to infections their immune system simply cannot see.

But T cell diversity does not last forever. With age, or under certain disease conditions, the immune system undergoes a process called thymic involution—the thymus shrinks and produces fewer naive T cells—combined with chronic immune activation that drives proliferation and apoptosis in antigen-specific clones. The result is a progressive loss of diversity, a narrowing of the immune repertoire, which is associated with increased susceptibility to infection and cancer in older adults.

Pan and Chou's model maps directly onto this biology. Each T cell clone is a "species" in their framework, subject to homeostatic regulation. Under some conditions (acute infection), signaling drives replication, increasing the birth rate. Under other conditions (immune activation with cytokine withdrawal), signaling increases apoptosis, effectively increasing the death rate. The $\alpha$ parameter captures this physiological heterogeneity across clones. Clones that happen to be more death-regulated will experience stronger stochastic fluctuations and drift toward extinction over time, even if they are immunologically equivalent in terms of their deterministic dynamics. Clones that are more birth-regulated—those whose regulation is dominated by reduced proliferation rather than increased death—are buffered against this drift and persist longer.

This offers a quantitative framework for understanding why the immune repertoire narrows with age and disease, and why different immune challenges might accelerate that narrowing in different ways. A pathogen that drives T cell proliferation followed by activation-induced cell death (high turnover, more death-regulation) would, in this framework, accelerate clonal loss more than one that suppresses T cell replication at the bone marrow level.

The third implication is methodological. Many ecological and evolutionary models invoke a carrying capacity and implicitly assume regulation acts symmetrically on births and deaths. Pan and Chou's analysis suggests that the choice of how to implement density dependence is not just a modeling convenience—it is a parameter with measurable consequences. Different implementations yield the same deterministic dynamics but different stochastic ones. For modelers, this means the apparent "equivalence" of different regulation mechanisms at the mean-field level is a trap: the models will make different predictions that become visible only over evolutionary timescales, when genetic drift has had time to act.

What's Next

Several open questions emerge from this work. The first is empirical: can the regulation parameter $\alpha$ be measured in real biological systems? For T cells, it might be estimated by correlating the turnover rates of clones (measurable through deuterium labeling or T cell receptor sequencing over time) with their survival probabilities. For bacterial populations, single-cell time-lapse microscopy could in principle track both birth and death events under density-dependent stress and estimate the regulation mix directly. This is challenging, but not infeasible.

The second open question concerns non-linear regulation. Pan and Chou work with Verhulst-type linear density dependence, where the regulation term is proportional to the population density. Real biological regulation is often more complex—Allee effects, sigmoid functional responses, spatial structure. Extending the analysis to non-linear regulation would broaden its applicability but likely destroys the clean analytical results that make this paper so compelling.

The third question is about the spectral gap in high dimensions. For $d\ge 3$ with heterogeneous regulation, the authors can compute the spectral gap numerically or perturbatively, but there is no simple scalar effective population size. The diversity loss is characterized by a spectrum of timescales, each corresponding to a different pair of species and each reflecting that pair's specific noise characteristics. Whether this multi-scale picture admits a clean biological interpretation—analogous to the role that $N_e$ plays in simple models—remains to be seen.

Finally, the paper opens a door to studying non-equilibrium dynamics. The analysis focuses on populations near a quasi-stationary equilibrium with the total population hovering around its carrying capacity. Real populations, of course, frequently experience growth phases, bottlenecks, and expansions. How regulation mechanism interacts with non-equilibrium demographic history is an important direction that the authors acknowledge but do not pursue here.

What makes Pan and Chou's paper particularly striking is its intellectual economy. The core insight—regulation mechanism matters for stochastic dynamics—emerges from a single parameter ${\alpha }_i$ embedded in otherwise standard models. You don't need exotic biology or complex population structures. You need two ways of enforcing the same carrying capacity, and the stochastic dynamics immediately diverge. In a field where effective population sizes are estimated from genetic data, where conservation plans hinge on the long-term viability of small populations, and where immune system aging is emerging as a central biomedical challenge, understanding how regulation mechanism shapes evolutionary fate is not an academic curiosity. It is foundational.

The effective population size is not just about how many individuals you have. It is about how those numbers change—birth by birth, death by death—and whether the fluctuations that result leave your population drifting toward uniformity or maintaining the diversity that makes survival possible.