Cell, Vol. 101, 581–584, June 9, 2000, Copyright 2000 by Cell Press Evidence for the Adaptive Minireview of Rates

David Metzgar*‡ and Christopher Wills*† under starvation conditions in Escherichia coli (Rosen- *Department of Biology berg et al., 1998). Loss-of- in genes † Center for Molecular Genetics required for growth can back-mutate, allowing survival University of California, San Diego and reproduction. The rate of these back-mutations (as La Jolla, California 92093 well as of other, nonadaptive mutations) increases greatly with starvation. This mutator is not itself heritable—the survivors produce progeny cells Adaptive evolution has long been regarded as the result with normal mutation rates. of postmutational sorting by the process of natural se- The SOS response has been implicated in such adap- lection. Mutations have been postulated to occur at ran- tive mutation (Rosenberg et al., 1998). The SOS system dom, producing genetically different individuals that is a DNA repair mechanism induced when the E. coli then compete for resources, the result being selection genome suffers physical damage such as double-strand of better adapted genotypes. Molecular biology has breaks, and can be triggered by starvation. It includes demonstrated, however, that the rate and spectrum of alternative DNA polymerase enzymes, such as DinB and mutations is in large part under the control of genetic the UmuDЈ2C complex, which are capable of replicating factors. Because genetic factors are themselves the badly damaged sequences. In the process of repair, subject of adaptive evolution, this discovery has brought these alternative polymerases generate mutations at a into question the random nature of mutagenesis. It high rate. DinB is known to generate mutations at un- would be highly adaptive for organisms inhabiting vari- damaged sites when SOS is induced. It has been sug- able environments to modulate mutational dynamics in gested that DinB and similar enzymes have evolved to ways likely to produce necessary adaptive mutations in produce genome-wide under stressful a timely fashion while limiting the generation of other, conditions, allowing organisms to adapt rapidly when probably deleterious, mutations. necessary (Radman, 1999). Homologs of these alterna- Have genetic systems emerged that tune mutation tive polymerases with similar functions have been identi- rates and spectra in adaptive ways? Such tuning would fied in Saccharomyces cerevisiae, and homologs have produce mutations with a greater chance of being adap- also been identified by sequence similarity in mice and tive than if they were completely random. This possibility (for review see Friedberg and Gerlach, 1999). has met with resistance from theorists who point out that Another potential mechanism for generating variation such systems may violate a basic tenet of evolutionary under stressful conditions has been recently described theory—evolution does not involve foresight (Dickenson in a . Some of the genetic variation that is and Seger, 1999). Here, we review recent evidence for capable of affecting development in Drosophila is masked the existence of adaptively tuned mutation rates. We under normal circumstances. This masking is accom- conclude that these mechanisms do not require any plished by the chaperone properties of the heat-shock special foresight. Instead, they must have been selected protein Hsp90. This protein stabilizes signal transduc- for repeatedly in the past for their ability to generate tion factors that regulate development. Mutational or genetic change. Mutational tuning does not require the pharmacological impairment of Hsp90 reveals otherwise specific generation of adaptive mutations (nonran- unexpressed developmental variation (Rutherford and domness with respect to function) but rather the con- Lindquist, 1998). These laboratory manipulations mimic centration of mutations under specific environmental the potential effects of heat stress or other protein- conditions or in particular regions of the genome (non- damaging conditions in the natural environment, condi- randomness with respect to time or location). Given a tions under which Hsp90 is diverted from the proteins predictably variable environment, adaptively tuned mu- it normally stabilizes to other environmentally denatured tation rates can evolve in ways completely consistent proteins. Exposed developmental variants can be fixed with the modern synthetic theory of evolution. in the population by selection. Changes in mutation rate can be either environment Environment-dependent mutators operate when ge- dependent or heritable. Environment-dependent changes netic change is necessary (when the organism is mal- result from induction or suppression of mutator mecha- adapted to its current environment). Moreover, the off- nisms that are present in all individuals and that have spring are not saddled with an increased rate of global (genome-wide) effects on mutation rate. Heritable deleterious mutation. The essential evolutionary ques- mutator mechanisms are independent of the environ- tion is this: did systems such as SOS and Hsp90 evolve ment. They may act by altering the global mutation rate to produce or reveal genetic variation under stress, or or they may be local, taking the form of gene sequences is this potentially adaptive property an accidental by- that experience unusually high or low rates of specific product of other factors? types of mutation. Such an evolutionary accident has been termed a Environment-Dependent Mutators spandrel, by analogy with the spaces (spandrels) that Environment-dependent global mutators are responsi- result whenever a round is supported by a square ble for adaptive mutations of the type that are produced formed by four arches. These spaces were secondarily used as aesthetically stunning venues for mosaic por- ‡ To whom correspondence should be addressed (e-mail: traits in St. Mark’s Cathedral and other buildings. Al- [email protected]). though they provide an optimal form for these mosaics, Cell 582

Figure 1. General Model of Functional Variability Driven by Hypermutable Microsatellites in Prokaryotic Contingency Loci Antigen- and phase-determinant genes containing nontriplet microsatellites turn on and off at high rates due to replication slippage mutations that cause repeat-number changes in the microsatellite. These mutations occur at very high rates (␮ Ϸ 10-3). it was not their original purpose. In evolutionary termi- Yersinia (Najdenski et al., 1995) have demonstrated posi- nology, the adaptive use of a function selected for an- tive correlations between global mutation rates and viru- other purpose is known as an . lence. Bacterial mutator have been traced The DNA polymerases involved in SOS-associated to mutations in DNA repair genes, particularly the mutagenesis are, by necessity, less specific than their methyl-directed mismatch repair (MMR) system. Muta- constitutive counterparts. They must be able to copy tors have also been found in HIV reverse transcriptase. DNA that is so severely damaged that it interferes with The fixation of heritable global mutator systems re- the action of higher-fidelity constitutive polymerases. quires second-order selection; the mutator can only be Indeed, the mutations introduced by the SOS-associ- spread through the population through the production of ated UmuDЈ2C are the result of repair activities. The linked advantageous mutations (Weber, 1996). Second- tendency for DinB to introduce mutations at undamaged order selection is facilitated by asexual reproduction, sites may also be a consequence of low fidelity that which maintains linkage. evolved to facilitate repair. Thus, the apparently adap- Both laboratory experiments and modeling studies tive hypermutagenesis of the SOS response may simply have shown that increased rates of mutation are only be a spandrel—the byproduct of functions evolved for favored when the normal mutation rate is the limiting repair. factor in and when the selective advantage Similarly, it is impossible to tell whether the Hsp90 of possible adaptive mutations is high (deVisser et al., system evolved under selective pressure to limit devel- 1999). Pathogenic prokaryotes may be in a particularly opmental variability in the absence of stress, to provide good position to take advantage of mutator phenotypes. variability in the presence of stress, or both. It is not Their populations experience extreme bottlenecks that clear whether any of the variability revealed under stress reduce existing genetic variability, and individual muta- through the action (or inaction) of Hsp90 is of a poten- tions can have exceptional advantages. Nonetheless, in tially adaptive nature nor has the ability of heat-stressed asexual populations there is a limit to the contribution flies to adapt more rapidly to environmental changes that increases in mutation rate can make to the rate of been tested under circumstances designed to mimic adaptation. This is because multiple adaptive mutants natural adaptation. coexisting in the same asexual population but in differ- Heritable Global Mutators ent individuals cannot be simultaneously brought to fixa- Heritable global mutators have been detected in short- tion. They compete with each other for fixation, a phe- and long-term laboratory selection experiments (Snie- nomenon known as clonal interference (deVisser et al., gowski et al., 1997; Miller et al., 1999), where they tend 1999). to outcompete nonmutators. Surveys of pathogenic Heritable global mutators, like environment-depen- strains of E. coli and Salmonella reveal much higher dent mutators, would appear to be adaptive under cer- proportions of mutator strains in natural populations tain circumstances. While heritable mutators are likely to than would be expected to arise by chance (LeClerc et arise entirely by chance, they may be carried to high fre- al., 1996). Studies of HIV (Wainberg et al., 1996) and quency through second-order selection acting through Minireview 583

Figure 2. Two Ways in which Have Modified the Rate at which They Pro- duce New Genetic Variants (A) Karl Ho¨ gstrand and Jan Bo¨ hme have found that regions of mouse MHC genes that are unusually rich in CpG doublets preferen- tially act as donors and recipients for gene conversion. CpG doublets are generally re- pressed in eukaryotic genomes and may be retained at these locations specifically for their ability to increase the mutation rate. Conversions in such genes account for a sub- stantial fraction of the new MHC alleles in mice. (Modified from Figure 1 of Ho¨ gstrand and Bo¨ hme [1999].) (B) The expressed variant surface glycopro- tein (VSG) genes of trypanosomes change continually during an infection. Expressed antigen alleles (often mosaics) are generated using genetic information (cassettes) stored elsewhere in the genome. The cassettes themselves remain unchanged.

valuable mutations that they generate. Modeling indi- tion. The best examples come from investigations of cates that this process can fix mutators (Tenaillon et al., pathogenesis, presumably because pathogenesis re- 1999). In the real world, however, complete fixation of sults in very strong selective pressures favoring high mutators is rare. Bacterial global mutators are usually rates of specific types of mutations. Hypermutable se- found in populations consisting primarily of nonmutators. quences generate antigen diversity in pathogens and HIV provides further evidence for the optimization of antibody diversity in hosts. The best known of these are global mutation rates through selection. When HIV-1 the “contingency loci” of pathogenic prokaryotes, which infections are treated with the nucleotide analog 3-TC, produce highly specific tuning of mutation rates in par- the virus evolves to a drug-resistant state through muta- ticular genes. tions that increase the fidelity of reverse transcriptase Contingency loci are antigen- and phase-determinant and lower the mutation rate of the virus. 3-TC resistance genes in pathogenic bacteria that switch between func- arises readily, suggesting that the mutations responsible tional (on) and nonfunctional (off) states at very high for resistance (and therefore for decreased rates of mu- rates (Figure 1). In many cases, the genes being tation) are common, yet these mutations are neither switched on and off are primary antigenic determinants, fixed nor maintained in the absence of 3-TC selection resulting in a pathogen that can vary its antigenic signa- (Wainberg et al., 1996). In fact, there is evidence for a ture abruptly without suffering an increased rate of mu- selective force that maintains a nonminimal mutation tation elsewhere. This switching behavior results from rate in HIV-1. Strains resistant to 3-TC are less virulent the mutational properties of tandem repetitive DNA (mi- and less able to evolve resistance to other antiviral drugs crosatellites) located within the gene or its associated than are 3-TC-sensitive strains. Even in this case, the controlling elements (Moxon et al., 1994). Microsatellites causal link between a need for rapid adaptation and the experience high rates of single-motif insertion and evolution of nonminimal mutation rates is circumstantial. deletion mutations through replication slippage. This Heritable Local Mutators generates a high rate of alternating loss-of-function and The evidence for the adaptive optimization of mutation reversion mutations, which is highly adaptive for the rates is much stronger in the case of heritable local pathogens that carry them. mutators. All known local mutators are the result of Although the genes that are controlled by contingency unique sequence characteristics that predispose spe- repeats in different organisms are often involved in anti- cific regions of the genome to particular types of muta- genicity, the proteins they code for are very diverse, and Cell 584

the repetitive DNAs driving hypermutability are diverse generated by these hypermutable repeats or by hitch- as well (Moxon and Wills, 1999). Most contingency re- hiking with strongly selected generated in- peats are the longest microsatellite DNAs in their re- dependently of the microsatellites but tightly linked to spective prokaryotic genomes, suggesting that they did them. It may soon be feasible to distinguish between not arise at these locations by chance. Thus, repetitivity these possibilities, through comparisons between the itself is evolving in these sequences, and microsatellites and chimpanzee genomes. have independently been selected for their mutability Local mutators do not predict the future, even though many times in response to similar selective pressures. they do preferentially produce the very types of muta- Local mutator systems have evolved in eukaryotic tions likely to increase survival. Hypervariability evolved organisms, but some are very different from contingency in antigenic contingency genes because it allowed the loci (Deitsch et al., 1997). Many of these systems do ancestors of today’s pathogens to repeatedly escape not generate heritable, germline mutations but rather immune clearance by their hosts. Because host immune employ mutational mechanisms that effectively decou- responses confront pathogens with variable environ- ple the consequences of hypermutation from the pro- ments that have strong elements of predictability, muta- cess of long-term evolution. The best characterized ex- tional mechanisms in the pathogens have evolved to ample is mammalian antibody diversification by somatic generate an appropriately variable antigenic repertoire. hypermutation. Other systems, such as the cassette- The hosts’ immune systems have also evolved local switching system involved in trypanosome antigenic mutators of various kinds, in response to pathogen vari- hypervariability, do give rise to heritable mutations, but ability. The key is the element of predictability in these these mutations too are strictly confined to certain parts interactions. If general environmental change also has of the genome (Figure 2). elements of predictability, global mutation rates may These systems are much more complex and highly have evolved toward optimal levels, but such optimiza- regulated than bacterial contingency genes, and it is tion has not yet been clearly demonstrated. clear that they have evolved specifically to increase the rate of mutation in localized parts of the genome. Other Selected Reading eukaryotic local mutators, such as the CpG island-asso- ciated gene conversions generating diversity at the MHC Deitsch, K.W., Moxon, E.R., and Wellems, T.E. (1997). Microbiol. Mol. Biol. Rev. 61, 281–293. locus, generate germline genetic changes in a manner more similar to that seen in contingency loci (Figure 2). deVisser, J.A.G.M., Zeyl, C.W., Gerrish, P.J., Blanchard, J.L., and Lenski, R.E. (1999). Science 283, 404–406. The existence of all of these systems demonstrates that Dickenson, W.J., and Seger, J. (1999). Nature 399, 30. adaptive evolution can produce highly specific mutators in eukaryotes. Friedberg, E.C., and Gerlach, V.L. (1999). Cell 98, 413–416. The high mutation rates experienced in prokaryotic Gould, S.J., and Lewontin, R.C. (1979). Proc. R. Soc. Lond. Ser. B 205, 581–598. contingency genes and in their eukaryotic counterparts are determined only by the unique sequence character- Haraguchi, Y., and Sasaki, A. (1996). J. Theor. Biol. 183, 121–137. istics of the genes themselves. The evolution of local Hogstrand, K., and Bohme, J. (1999). Immunogenetics 49, 446–455. mutators does not depend on linkage or population size, King, D.G., Soller, M., and Kashi, Y. (1997). Endeavour 21, 36–40. factors that can interfere with fixation of global mutators. LeClerc, J.E., Li, B., Payne, W.L., and Cebula, T.A. (1996). Science Local mutators also carry no cost resulting from in- 274, 1208–1210. creased rates of deleterious mutation elsewhere in the Miller, J.H., Suthar, A., Tai, J., Yeung, A., Truong, C., and Stewart, J.L. (1999). J. Bacteriol. 181, 1576–1584. genome. The evolution of adaptively variable antigenic determinants in pathogens generates tremendous se- Moxon, E.R., and Wills, C. (1999). Sci. Am. 280, 72–77. lective pressure on their hosts to evolve adaptively vari- Moxon, E.R., Rainey, P.B., Nowak, M.A., and Lenski, R.E. (1994). able immune defenses, and vice versa. This is an optimal Curr. Biol. 4, 24–33. situation for the evolution of mutators (Haraguchi and Najdenski, H., Iteman, I., and Carniel, E. (1995). Contrib. Microbiol. Immunol. 13, 281–284. Sasaki, 1996). Factors other than host-pathogen interactions may Radman, M. (1999). Nature 401, 866–869. be spurring the evolution of local mutators. Coding- Rosenberg, S.M., Thulin, C., and Harris, R.S. (1998). Genetics 148, 1559–1566. region triplet-repeat microsatellites, which are both common and hypervariable in higher eukaryotes, may Rutherford, S.L., and Lindquist, S. (1998). Nature 396, 336–342. provide genetic variability that allows selective fine tun- Sniegowski, P.D., Gerrish, P.J., and Lenski, R.E. (1997). Nature 387, 703–705. ing of protein function (King et al., 1997). Single-repeat length mutations in these microsatellites alter the length Tenaillon, O., Toupance, B., Nagard, H.L., Taddei, F., and Godelle, B. (1999). Genetics 152, 485–493. of repetitive amino acid tracts within protein products Wainberg, M.A., Drosopoulos, W.C., Salomon, H., Hsu, M., Borkow, without generating frameshift mutations. Among hu- G., Parniak, M., Gu, Z., Song, Q., Manne, J., Islam, S., et al. (1996). mans, some of the longest coding triplet-repeat micro- Science 271, 1282–1285. satellites occur in or near proteins that affect neural Weber, M. (1996). Biol. Phil. 11, 67–88. function, and some of these are long enough to become highly unstable, resulting in genetic disorders such as Fragile X syndrome, Huntington’s disease, and spino- cerebellar ataxia. Because the human brain has recently undergone considerable directional selection, it is pos- sible that the deleterious effects of these long microsat- ellites have been counterbalanced, either by second- order selection acting through highly adaptive mutations