Genetic Drift Overpowers Selection When the Population Is Small Enough — Epoche B2
Re-evaluating Genetic Drift: Beyond Neutral Noise For a long time, genetic drift was understood primarily as a random, minor force in evolution, overshadowed by natural selection. This perspective viewed drift as mere 'noise' in the genetic landscape, particularly in large populations, and relevant mainly to 'neutral evolution' [1] , where changes in gene frequency are not driven by selective advantage but by chance. Contemporary population genetics offers a more nuanced quantitative re-evaluation, treating drift as a stochastic process with a definite role in adaptive evolution. This essay examines when genetic drift becomes a powerful evolutionary force, capable of overpowering natural selection, and when its influence is relegated to background noise. The Stochastic Nature of Genetic Drift in Finite Populations Evolutionary change relies on changes in allele frequencies over generations. An allele is a variant of a gene, and its frequency is the proportion of that allele among all copies of the gene in the gene pool. In an idealised, infinitely large population, frequencies stay constant in the absence of selection, mutation, or migration, as the Hardy-Weinberg principle describes. All real populations are finite, so only a limited number of individuals contribute gametes to the next generation, and that finiteness introduces stochasticity into transmission. Genetic drift is this process of random fluctuation in allele frequencies from one generation to the next, purely due to chance in gamete sampling. Consider a diploid, panmictic population of $N$ individuals: individuals carry two copies of each gene, and mating is random. The total number of gene copies is $2N$. If an allele has current frequency $p$, the number of copies in the next generation is not exactly $2Np$ but a random sample from the current gene pool, modelled as a binomial draw. If $k$ copies are drawn, the new frequency is $p' = k/(2N)$, and $$ P(k \text{ copies}) = \binom{2N}{k} p^k (1-p)^{2N-k} $$ Here $p$ is the current allele frequency, $1-p$ the frequency of the alternative allele, and $2N$ the number of gene copies sampled. Because the process is probabilistic, $p'$ will almost certainly differ from $p$ even with no selective advantage. These random changes accumulate, and over long timescales drift inevitably drives an allele to fixation (frequency $1$) or loss (frequency $0$) in a finite population, even in the absence of selection. The Wright-Fisher Model and Effective Population Size ($N_e$) The Wright-Fisher model is the foundational idealisation of drift in a finite population with discrete, non-overlapping generations. Its assumptions are: Constant population size : the number of individuals remains fixed at $N$, all of them diploid. Panmixia : any individual is equally likely to mate with any other. Discrete generations : the whole adult generation is replaced by the next. No selection, mutation, or migration : drift is the only evolutionary force. Random sampling of gametes : the next gene pool is $2N$ gametes drawn with replacement from the current one, so the number of surviving offspring per individual is binomial with mean $2$ and variance $2(1-1/N) \approx 2$ — that is, approximately Poisson. This is the single idealisation of reproductive success used throughout this essay. Real populations rarely meet these assumptions. Fluctuating size, unequal sex ratios, variation in reproductive success and non-random mating all alter the intensity of drift. The effective population size $N_e$ absorbs these complications: it is the size of an ideal Wright-Fisher population that would show the same rate of drift — the same rate of loss of heterozygosity, or the same variance in allele frequency change — as the real population observed. If size fluctuates over $t$ generations, the harmonic mean is a good approximation: $$ \frac{1}{N_e} = \frac{1}{t} \sum_{i=1}^{t} \frac{1}{N_i} $$ where $N_i$ is the size in generation $i$. Bottlenecks therefore dominate $N_e$, because a small $N_i$ makes $1/N_i$ large. For many species $N_e$ is far below the census size $N$: the human census population exceeds $8 \times 10^9$, while estimates of human $N_e$ are around $10^4$, reflecting historical bottlenecks and unequal reproductive success. A new neutral mutation arises as a single copy among the $2N$ gene copies actually present, so its initial frequency is $p = 1/(2N)$ — the census size, not $N_e$. Because a neutral allele's frequency is a martingale, its probability of eventual fixation is exactly its current frequency: $$ u_0 = p = \frac{1}{2N} $$ Note that $N_e$ does not appear here. The effective size governs how long fixation takes (of order $4N_e$ generations) and how violently the frequency wanders on the way, but not the probability itself. In a smaller population a new neutral allele starts at a higher frequency and is correspondingly more likely to fix; in a larger one it is less likely. The two effects cancel exactly. If neutral mutations arise at rate $\mu$ per gene copy per generation, then $2N\mu$ of them enter the population each generation, and each fixes with probability $1/(2N)$, so the rate $k$ at which neutral substitutions accumulate is $$ k = 2N\mu \cdot \frac{1}{2N} = \mu $$ independent of population size. This cancellation is the molecular clock of the neutral theory, and it is the reason the rate of neutral substitution carries no information about how large a population has been. With the human mutation rate $\mu \approx 1.2 \times 10^{-8}$ per site per generation and a generation time near $30$ years, the neutral clock ticks at about $4 \times 10^{-10}$ substitutions per site per year. Quantifying Fixation: The Balance Between Drift and Selection Natural selection is the differential survival and reproduction of individuals according to their heritable traits; unlike drift it is systematic, raising the frequency of beneficial alleles and lowering that of deleterious ones. Its strength is measured