Adaptive Dynamics and the Evolution of Virulence

Why doesn’t natural selection make every pathogen either perfectly benign or maximally deadly? The answer is a trade-off: the very transmission that spreads a pathogen is often tied to the harm it does its host, and selection settles on an intermediate, optimal virulence.

Adaptive dynamics and invasion fitness

Adaptive dynamics studies long-term evolution as a sequence of rare mutants trying to invade a resident population sitting at its ecological equilibrium. The key quantity is invasion fitness: the per-capita growth rate of a rare mutant strain in the environment set by the resident. If invasion fitness is positive the mutant spreads and may replace the resident; if negative it dies out. Evolution proceeds as a succession of successful invasions that gradually change the trait — here, the pathogen’s virulence.

For a pathogen introduced into a susceptible host population, invasion fitness is governed by the basic reproduction number: a mutant strain invades the resident’s disease-free environment exactly when its own $R_0$ exceeds one, so selection favours the strain with the largest $R_0$ .

The transmission–virulence trade-off

Consider a standard SIR-type infection with transmission rate $\beta$ , recovery rate $\gamma$ , disease-induced host mortality (virulence) $\alpha$ , and background mortality $\mu$ . An infected host stays infectious for an average duration $1/(\gamma + \alpha + \mu)$ , and transmits at rate $\beta$ throughout, so

$R_0(\alpha) = \frac{\beta(\alpha)}{\gamma + \alpha + \mu}.$

The crucial biological assumption is that transmission is not free: higher $\beta$ requires higher pathogen replication, which also raises host mortality $\alpha$ . We encode this as an increasing, decelerating trade-off function $\beta(\alpha)$ . Now virulence faces opposing pressures. Raising $\alpha$ increases the numerator $\beta(\alpha)$ , but it also shortens the infectious period $1/(\gamma+\alpha+\mu)$ in the denominator by killing the host sooner. The pathogen “wants” high transmission but not at the price of a host that dies before infecting others, so the strain that maximises $R_0(\alpha)$ has some intermediate optimal virulence $\alpha^*$ — neither the avirulence of $\alpha \to 0$ nor unbounded harm.

Singular strategies and invasibility

The value $\alpha^*$ that maximises $R_0$ is a singular strategy of the adaptive dynamics, found by setting $\mathrm{d}R_0/\mathrm{d}\alpha = 0$ . Finding it is an optimization problem, and because the winning strain is simply the one with the highest $R_0$ , this optimum is both an ESS (uninvadable once established) and an attractor of the evolutionary dynamics. Adaptive-dynamics analyses visualise this with a pairwise-invasibility plot (PIP): for every resident trait on one axis and mutant trait on the other, the plot shades where a mutant can invade. An evolutionarily stable $\alpha^*$ appears where no nearby mutant can invade the resident — the resident’s row lies entirely outside the invasion region.

Worked example: an optimal virulence

Take the common phenomenological trade-off $\beta(\alpha) = a\sqrt{\alpha}$ , so that transmission rises with virulence but with diminishing returns. Then

$R_0(\alpha) = \frac{a\sqrt{\alpha}}{\gamma + \alpha + \mu}.$

Write $d = \gamma + \mu$ and differentiate. Using the quotient rule,

$\frac{\mathrm{d}R_0}{\mathrm{d}\alpha} = a\,\frac{\tfrac{1}{2}\alpha^{-1/2}(\alpha + d) - \alpha^{1/2}}{(\alpha + d)^2}.$

The denominator is always positive, so the optimum occurs where the numerator vanishes: $\tfrac{1}{2}\alpha^{-1/2}(\alpha + d) = \alpha^{1/2}$ , i.e. $\alpha + d = 2\alpha$ , giving

$\alpha^* = d = \gamma + \mu.$

The optimal virulence equals the host’s total background loss rate (recovery plus natural death): the pathogen should harm its host at roughly the rate the host would leave the infectious pool anyway. With $\gamma = 0.5$ , $\mu = 0.1$ , and $a = 3$ , we get $\alpha^* = 0.6$ and $R_0(\alpha^*) = 3\sqrt{0.6}/1.2 \approx 1.94$ , higher than the $R_0$ of a more benign ( $\alpha = 0.2$ , $R_0 \approx 1.68$ ) or more aggressive ( $\alpha = 1.5$ , $R_0 \approx 1.75$ ) strain.

Simulation

We compute $R_0(\alpha)$ across a range of virulence and locate the optimum numerically, confirming $\alpha^* = \gamma + \mu = 0.6$ .

R

gamma <- 0.5; mu <- 0.1; a <- 3
R0 <- function(alpha) a * sqrt(alpha) / (gamma + alpha + mu)

opt <- optimize(R0, c(1e-6, 10), maximum = TRUE)
opt$maximum     # ~0.6  = gamma + mu
$opt$objective   # ~1.936
$
curve(R0, 0, 3, xlab = "virulence alpha", ylab = "R0")
abline(v = gamma + mu, lty = 2)

Python

import numpy as np
from scipy.optimize import minimize_scalar

gamma, mu, a = 0.5, 0.1, 3.0
R0 = lambda al: a * np.sqrt(al) / (gamma + al + mu)

res = minimize_scalar(lambda al: -R0(al), bounds=(1e-6, 10), method="bounded")
print(res.x, R0(res.x))    # ~0.6 (= gamma + mu), ~1.936

0.6000014256626104 1.9364916731023418

Julia

using Optim

γ, μ, a = 0.5, 0.1, 3.0
R0(α) = a * sqrt(α) / (γ + α + μ)

res = optimize(α -> -R0(α), 1e-6, 10.0)   # Brent's method on the interval
Optim.minimizer(res)     # ~0.6 = γ + μ
-Optim.minimum(res)      # ~1.936

Why it matters

The trade-off theory of virulence explains why pathogens are neither harmless nor uniformly lethal, and it warns that interventions can shift the optimum: imperfect vaccines, treatments that extend the infectious period, or crowding that eases transmission can all select for higher virulence. It rests on computing and maximising $R_0$ — the same threshold that, in structured host populations, is obtained as the dominant eigenvalue of the next-generation matrix — and it links pathogen evolution to broader evolutionary game theory, since competing strains playing off transmission against host survival mirror the frequency-dependent contests and competition for coexistence seen throughout ecology.