The evolutionary trajectories of organisms are paved with mutations, which generate the raw material (genetic variation) essential for evolutionary change. In biology class, we learn that mutations are random: the probability that a mutation occurs is independent of its fitness effect (i.e., its impact on individual survival or reproduction). As Luria and Delbruck famously showed [1], bacterial mutations that confer resistance to a virus continually arise in a population before exposure to the virus; under subsequent viral infection, these mutations spread in the population. Thus, mutations do not occur in response to a specific stress once it is encountered, and the fact that we observe beneficial mutations after imposing selection is simply an outcome of natural selection. Clearly, genetic variation is critical for adaptation, so why don’t mutation rates continually increase? In fact, very few mutations are beneficial; most have no phenotypic effect or are deleterious (see [2] for a recent review). Thus, increasing the total mutation rate would introduce a “genetic load” in the population, whereby most individuals will acquire deleterious mutations. The problem is exacerbated in asexual populations by Muller’s ratchet (the irreversible accumulation of deleterious mutations): without sex, mutation-free regions of two genomes cannot recombine and generate mutation-free offspring. Indeed, in laboratory experiments, bacteria with high mutation rates (“mutators”) may have a growth advantage in the short term, but as they accumulate deleterious mutations, they are eventually outcompeted by nonmutators (e.g., [3]). Hence, selection should generally act against increasing mutation rates. Additionally, adaptation may not always be limited by the supply of mutations: standing genetic variation (i.e., already existing mutations, expected in large populations) may include beneficial genotypes that allow for rapid adaptation to environmental change [4].

However, a large body of work demonstrates stress-induced mutagenesis (SIM)—a transient increase in mutation rates under stresses such as antibiotic exposure or starvation—via specific pathways that are typically suppressed under rapid growth [5,6]. This has been best studied in bacteria, where stress-associated DNA damage triggers the SOS response, a large and complex network of genes and proteins that pauses cell division and repairs damaged DNA [7]. The response includes upregulation of the RpoS protein, a global regulator of gene expression. The regular high-fidelity, methyl-directed mismatch repair pathway (MMR) is suppressed, and error-prone DNA repair machinery (involving DNA polymerase IV and V) is upregulated, ultimately increasing the mutation rate (Fig 1A). Analysis of single genes suggest that SIM can also alter the mutation spectrum [8]—the relative frequency of different types of mutations (e.g., base pair substitutions or large insertion sequences)—thus, determining the available set of mutations upon which natural selection can act. E. coli strains lacking MMR genes also show distinct mutation spectra [9,10], indicating that the suppression of MMR induced by stress may have a similar effect. In fact, inducing high RpoS expression in nonstressed E. coli also significantly changed the mutational spectrum [11], suggesting that the perception of stress (despite growing in rich media) is sufficient to alter mutation spectra. SIM occurs in diverse organisms and contexts, including natural bacterial isolates [12] and cancer cells [13]; in many of these cases, the mechanisms underlying SIM are well understood. However, the potential implications and interpretations of SIM for evolutionary dynamics have been controversial. Does SIM represent an adaptive mechanism to increase sampling of beneficial mutations under stress, or is it a by-product of a tradeoff between efficient and accurate DNA repair as suggested by the crystal structure of the error-prone polymerase [14]? At the heart of this debate are various open questions (Fig 1B) whose answers are critical to determine the evolutionary implications of SIM. For instance, which stresses induce mutagenesis, and are the mutational consequences consistent across stresses?