The degeneration of Y chromosomes

Ychromosomes are genetically degenerate, having lost most of the active genes that were present in their ancestors. The causes of this degeneration have attracted much attention from evolutionary theorists. Four major theories are reviewed here: Muller’s ratchet, background selection, the Hill^Robertson e¡ect with weak selection, and the `hitchhiking’ of deleterious alleles by favourable mutations. All of these involve a reduction in e¡ective population size as a result of selective events occurring in a non-recombining genome, and the consequent weakening of the e¤cacy of selection. We review the consequences of these processes for patterns of molecular evolution and variation at loci on Y chromosomes, and discuss the results of empirical studies of these patterns for some evolving Y-chromosome and neo-Y-chromosome systems. These results suggest that the e¡ective population sizes of evolving Y or neo-Y chromosomes are severely reduced, as expected if some or all of the hypothesized processes leading to degeneration are operative. It is, however, currently unclear which of the various processes is most important; some directions for future work to help to resolve this question are discussed.


INTRODUCTION
Morphologically and genetically distinct X and Y chromosomes have evolved independently in many groups of animals and plants (Bull 1983;Charlesworth 1996a).A striking common feature is the almost complete erosion of genes from the Y chromosome (or W chromosome, in groups with female heterogamety: we shall use the term Y chromosome for both systems), except for some genes with functions speci¢c to the heterogametic sex.The Y chromosomes of groups such as mammals or Drosophila were presumably originally homologous with the ancestral X chromosomes and have subsequently lost genetic activity, giving rise to the familiar phenomenon of sex-linked inheritance.The possession of genetically eroded Y chromosomes in many groups with di¡erentiated sex chromosomes is also associated with dosage compensation, such that the activity of most X-linked genes is e¡ectively the same in males and females (Bull 1983;Charlesworth 1996a).
Understanding the evolutionary causes of these phenomena is a challenge for both theoreticians and experimentalists.The ¢rst step in the evolution of sex chromosomes must have involved the establishment of restricted recombination between a pair of proto-X and proto-Y chromosomes (Muller 1918), probably in response to the need to prevent recombination between genes with primary sex-determining roles, which would produce neuters or hermaphrodites (Lewis 1942;Charlesworth & Charlesworth 1978;Charlesworth 1996a).Although sex determination in many contemporary organisms may only involve segregation at a single major switch gene (Bull 1983), the initial evolutionary transition from hermaphroditism or environmental sex determination to separate sexes probably often involved several loci with e¡ects on sexual phenotype (Westergaard 1958;Charlesworth & Charlesworth 1978;Charlesworth 1996a).
Once genetic sex determination has evolved, selection for alleles that are advantageous in males but disadvantageous in females can lead to genetic di¡erentiation between the two sex chromosomes at other loci, and to selection for the suppression of recombinational exchange over most or all of their length (Rice 1996).This sets the scene for the further evolution of an incipient Y chromosome.We shall be mainly concerned here with the question of why most genes that were originally present on both the proto-X and proto-Y chromosome have been lost from the Y chromosome.The answers must lie in the special situation of a large block of non-recombining genes, which are restricted to one sex and can never become homozygous.
Several models of the degeneration of proto-Y chromosomes have been proposed and are described below, but the models are only just starting to be tested and we are far from being able to decide de¢nitively between them.Although other non-recombining systems are probably subject to the same evolutionary forces, relevant data are scarce for these too.However, there is some evidence for reduced levels of adaptation in clonally transmitted genomes in ¢shes (Leslie & Vrijenhoek 1978) and in organelle genomes (Lynch & Blanchard 1998), which probably re£ect the action of one or more of the processes that we discuss in ½ 2.
One reason for the dearth of tests is that the most familiar and thoroughly investigated sex chromosome systems, such as those of Drosophila and mammals, are very ancient, and betray few signs of their evolutionary origins (but see Lahn & Page 1999).We currently know little about Y chromosomes of more recent origin, such as those in dioecious £owering plants (Westergaard 1958;Charlesworth & Guttman 1999), or systems of neo-Y chromosomes in which a fusion or translocation between an autosome and a sex chromosome has created a new chromosomal arm that must experience the same evolutionary forces as Y chromosomes (Lucchesi 1978;Charlesworth 1996a;Steinemann & Steinemann 1998).These situations are suitable for testing models of Y-chromosome degeneration, as we shall discuss.

(a) General features of models of Y-chromosome degeneration
Once low recombination between proto-X and proto-Y chromosomes has evolved, any genes on the nonrecombining part of the proto-Y with homologues on the proto-X (and which are not sex speci¢c in function) are in a very peculiar genetic situation.Muller (1918) was the ¢rst to point out that the permanent heterozygosity of such genes implies reduced selection against deleterious mutations.Even mutations eliminating the functions of vital genes might have only minor e¡ects on ¢tness when heterozygous: experiments on Drosophila suggest that recessive lethals reduce the ¢tness of heterozygotes by only 1^2% on average (Crow 1993).
Another di¡erence is that there are one-third as many copies of Y-linked genes as X-linked genes in populations, and one-quarter as many copies as autosomal loci.Thus, compared with X-chromosomal genes, Y-linked genes experience a smaller e¡ective population size, N e .Ylinked deleterious mutations are therefore particularly likely to rise to high frequencies, or to replace wild-type alleles, by genetic drift (Nei 1970).In species with male heterogamety, the high variance in male reproductive success associated with sexual selection could further reduce the e¡ective population size of the Y chromosome and (to a smaller degree) of the autosomes, relative to that of the X chromosome (Caballero 1995).N e -values can be estimated from data on variation in silent DNA sequences, if mutation rates for neutral or nearly neutral changes are accurately known (Kimura 1983).Evidence for sex di¡erences in the variance of reproductive success can therefore be obtained by measuring levels of silent variability; strong sexual selection on males should bring the X-linked value close to, or even higher than, the autosomal value (Charlesworth 1996b).
Initial explanations for degeneration were focused on these properties of Y chromosomes, ignoring other consequences of the lack of recombination between Y-linked loci.However, they encountered several di¤culties (Charlesworth 1978(Charlesworth , 1996a)).Fixation probabilities for deleterious mutations are extremely low unless the product of N e and the selection coe¤cient, s, is close to 1 (Fisher 1930, chapter 4).Accelerated ¢xation of Y-linked loss-of-function mutations thus requires very small population sizes, or a very high variance in male reproductive success.Mutations with extremely slight ¢tness e¡ects, such that N e s is not much bigger than 1, could nevertheless accumulate.However, a reduction in N e increases the proportion of loci with high frequencies of deleterious alleles only over a narrow range of N e s (see, for example, ¢g. 1 of McVean & Charlesworth 1999), such that the e¤cacy of drift is delicately balanced against selection at a large number of genomic sites, many of them polymorphic.Under these conditions, a lack of recombination leads to a reduction in the ability of the genome to increase its level of adaptation under natural selection, and to resist forces that reduce adaptation (Maynard Smith 1978;Kondrashov 1993;Barton & Charlesworth 1998).

(b) Models of mutation and selection
To understand the in£uence of the lack of recombination on Y-chromosome degeneration, it is necessary to examine some general features of the evolutionary dynamics of mutation and selection at many loci.Consider a large but ¢nite random-mating population.For locus i, with deleterious mutation rate u i , the equilibrium frequency of mutant alleles with selection co-e¤cient s i 4 4u i is q i ˆui /s i (Haldane 1927).In the context of the Y chromosome, s i re£ects the ¢tness reduction experienced by heterozygous carriers of the mutant allele.If di¡erent loci a¡ect ¢tness independently, this generates a Poisson distribution of numbers of deleterious alleles carried on di¡erent Y chromosomes in the p opulation (Kimura & Maruyama 1966;Johnson 1999), whose mean is the sum of u i /s i over all sites subject to mutation and selection.This is equal to the ratio of the net mutation rate, u ˆiu i , for deleterious alleles at all relevant Ychromosomal loci, to the harmonic mean selection coe¤cient, s h .In a ¢nite population with relatively free recombination and N e s i 4 41 for all sites, this equilibrium is closely approached (see, for example, Charlesworth et al. 1993).The frequency of mutation-free individuals is then f 0 ˆexp(7u/s h ); this has a critical role in several of the processes discussed below.

(c) Muller's ratchet
Various processes can cause a non-recombining population, such as a proto-Y chromosome, to depart from this equilibrium and accumulate deleterious mutations.One such process is `Muller's ratchet' (Muller 1964;Felsenstein 1974).This is the stochastic loss, from a population of ¢nite size, of the class of chromosomes carrying the fewest deleterious mutations.In the absence of recombination and back mutation, this `least-loaded' class of chromosome cannot be restored, once lost.The next best class then replaces it as the least-loaded class, and is in turn lost, and so on, in a fairly steady process of successive irreversible losses of the current least-loaded class.In a haploid genome (such as a proto-Y chromosome), each such loss is quickly followed by the ¢xation of one deleterious mutation on the chromosome (Higgs & Woodcock 1995;Charlesworth & Charlesworth 1997).This ¢xation process can be many orders of magnitude faster than in a freely recombining population.The ratchet could thus lead to the cumulative impairment of gene function at sites scattered over a proto-Ychromosome.
The speed at which the ratchet moves (reviewed in Gordo & Charlesworth 2000) is clearly critical for its plausibility in relation to the degeneration of the Y chromosome.Other things being equal, the ratchet should move faster as the population size becomes smaller.For mammal populations, whose N e -values, estimated from variability in silent-site DNA sequences, are probably often only in the tens of thousands (Nachman 1997(Nachman , 1998;;Wang et al. 1998), the ratchet might be plausible.For various species of Drosophila (Powell 1997) and £owering plants (Charlesworth & Charlesworth 1998), sequence variation data suggest e¡ective p opulation sizes in the millions, so that the potential ability of the ratchet to in£uence the degeneration of the Y in these taxa has been questioned (Charlesworth 1996a).However, recent work suggests that the ratchet can move rapidly in a very large population if there are many mutations with very small e¡ects on ¢tness, so that f 0 is small (Gessler 1995;Gordo & Charlesworth 2000, 2001).To determine whether this is likely will require a better knowledge of the properties of spontaneous mutations (Keightley & Eyre-Walker 1999;Lynch et al. 1999).
Simulation results on very large populations with 500 000 haploid genomes (Gordo & Charlesworth 2001) indicate that the speed of the ratchet in large populations can be approximated by an explicit equation, assuming the same selection coe¤cient s for each site.With u ˆ0.015 and s ˆ0.0015, the mean and 95% con¢dence interval of the number of generations between successive clicks of the ratchet is 732 § 66 from simulations; with u ˆ0.03 and s ˆ0.005, the estimate is 12 819 § 5681.These examples indicate that the ratchet can operate at a biologically relevant rate even in a very large population, provided that mutations with small selection coe¤cients occur su¤ciently frequently.
How fast does mean ¢tness decline under Muller's ratchet ?On a logarithmic scale, ¢tness decreases approximately with the product of the rate of the ratchet and the mean selection coe¤cient.A ratchet driven by the accumulation of mutations with very small selection coe¤cients (less than 0.001) will clearly produce a low rate of decline of mean ¢tness, even though the ratchet moves rapidly (Gessler 1995;Gessler & Xu 1999).With mutations of larger e¡ect, the ratchet might be capable of explaining Y-chromosome degeneration over plausible time periods (see ½ 3(b)).The two numerical examples described above give expected mean ¢tnesses of the populations after ¢ve million generations of 3.6 £10 75 and 0.14 of the initial value, re£ecting the ¢xation of an average of 6830 and 390 mutations, resp ectively.

(d) Background selection
`Background selection' (Charlesworth et al. 1993;Charlesworth 1994), or `background trapping' (Rice 1996), is closely related to Muller's ratchet.A neutral or weakly selected mutation that arises in a large nonrecombining population has a non-zero chance of survival only if it arises on a chromosome free of strongly deleterious mutations (for which N e s4 41) (Fisher 1930, chapter 4;Charlesworth 1994;Peck 1994;Barton 1995).Chromosomes carrying one or more strongly selected deleterious mutations are soon eliminated from the population, carrying any new variants with them.The e¡ective population size of a non-recombining chromosome is therefore reduced to f 0 N e .The ¢xation of mildly deleterious mutations with selection coe¤cients of the order of 1/f 0 N e will therefore be accelerated (Charlesworth 1996a), and the ¢xation of mildly advantageous mutations will be retarded (Peck 1994;Barton 1995;Rice 1996;Orr & Kim 1998).Over long periods of evolutionary time, therefore, the mean ¢tness of the Y chromosome should decline, relative to that of the X. Background selection can operate with selection co-e¤cients and population sizes where the ratchet is inoperative, and can lead to a signi¢cant reduction in the ¢tness of Y chromosomes in very large populations over a biologically plausible time-scale (Charlesworth 1996a).
In addition, the e¡ective population size experienced by weakly selected sites will be reduced by the presence of mutations with selection coe¤cients su¤ciently large that they will not accumulate by the ratchet.This could facilitate a ratchet at the more weakly selected sites.Simulations show that the rate of the ratchet for weakly selected mutations is well predicted by substituting f 0 N e for N e in the relevant equations, where f 0 is determined by using the parameters of the the strongly selected mutations (I.Gordo and B. Charlesworth, unpublished data).The two processes might therefore operate in concert, with a considerably accelerated ratchet when strongly selected deleterious mutations are causing background selection.
Background selection also causes reduced genetic variability at neutral or very weakly selected sites (Charlesworth et al. 1993), because equilibrium diversity depends on the e¡ective population size (Kimura 1983;McVean & Charlesworth 1999).When a ratchet is operating at a moderate rate, reduced neutral or nearly neutral variability will also probably result, because a ratcheting population has originated largely from the currently least-loaded class over a relatively short timescale, compared with the coalescence time of a pair of alleles in a freely recombining population of comparable size (Rice 1987;Higgs & Woodcock 1995;Charlesworth & Charlesworth 1997).This remains to be studied quantitatively.

(e) Weak selection Hill^Robertson e¡ect
A third, distinct but related, process is the `weak selection Hill^Robertson' e¡ect (McVean & Charlesworth 2000).Closely linked selected alleles interfere with each other, inhibiting the spread of favourable alleles and the elimination of deleterious ones (Fisher 1930, chapter 5;Muller 1932;Hill & Robertson 1966;Felsenstein 1974;Birky & Walsh 1988).Muller's ratchet and background selection can be viewed as examples of this principle for situations when the sites creating the e¡ects have N e s4 41, so that the loci subject to mutation and selection would be near their deterministic equilibria if recombination were not restricted.If, however, N e s is of the order of one or less, genetic drift and back-mutation from deleterious alleles to wild-type become more important.Models incorporating the joint e¡ects of selection, forward and backward mutation and genetic drift are commonly used in the study of biased codon usage, in which alternative triplets coding for the same amino acid might be subject to weak but statistically detectable selection in insects, nematodes and plants (Li 1987;Bulmer 1991;Akashi 1994;Duret & Mouchiroud 1999;McVean & Charlesworth 1999).Some classes of mutations causing amino-acid changes might also fall into this category.Many sites will then segregate for favoured and disfavoured alleles at intermediate frequencies, so that the assumptions of the ratchet and background selection models break down.
Finite population-genetic models with large numbers of closely linked segregating sites are complex, and Charlesworth and D. Charlesworth 1565 investigation of this model requires Monte Carlo simulations (Li 1987;Comero¨n et al. 1999;McVean & Charlesworth 2000).With very large numbers of sites subject to mutation and selection, and assuming multiplicative selection, the mean level of adaptation (measured by the frequency of sites carrying favoured alleles among chromosomes sampled from the population) is strongly reduced when recombination is absent (see ¢gure 1); genetic diversity is also reduced.Because the mean size of a eukaryote gene is about 500 codons, and a Drosophila or plant chromosome might have about 2000^3000 genes, the number of third coding positions at which silent mutations can occur on a proto-Ychromosome is of the order of 10 6 , ten times the number for the most extreme example shown in ¢gure 1.For a population size of 5 £10 5 , the parameters used in ¢gure 1 imply a selection coe¤cient of 0.4 £10 75 ; with multiplicative ¢tnesses and only 10 5 sites, the ¢tness of the population at statistical equilibrium is ca.5% of the value with free recombination.
Hill^Robertson e¡ects with weak selection therefore have considerable power in the long term to erode the ¢tness of an evolving Y chromosome.However, the time taken for ¢tness to be substantially reduced could be considerable, because new mutations might have to approach ¢xation at numerous sites along the chromo-some.The upper limit to the rate of ¢xation is given by the mutation rate (Kimura 1983); thus, if there are 10 6 sites with selection coe¤cients of 0.4£10 75 , and a mutation rate of 10 79 per site, the maximum per generation rate of decline of mean ¢tness would be of the order of 4 £10 79 .With ¢ve generations per year, as might be expected for Drosophila miranda (see } 3(b)), it would take a duration of the order of ¢ve million years for ¢tness to be reduced by 10%.Weak Hill^Robertson e¡ects therefore require a long period to reduce the ¢tness of an evolving Y chromosome signi¢cantly.This casts some doubt on their relevance to cases in which degeneration has occurred over a much shorter period, as with D. miranda discussed below.

(f) Hitchhiking e¡ects of favourable mutations
The ¢nal class of model involves hitchhiking e¡ects during the spread of selectively favourable mutations.Maynard Smith & Haigh (1974) were the ¢rst to study the e¡ect of the spread of a selectively favourable mutation on variability at linked neutral sites.With complete linkage between the two loci, as for genes on a proto-Y chromosome, it is intuitively obvious that a `selective sweep' of a favourable allele will eliminate all variation at the neutral locus; neutral or nearly neutral variability will recover only slowly, as new variants arise and drift to high frequencies.Selective sweeps in a non-recombining genome can also drag to ¢xation any deleterious mutant alleles associated with the favourable mutation, so that successive adaptive substitutions on a proto-Ychromosome could lead to the ¢xation of deleterious mutations at many loci, contributing to its degeneration (Rice 1987).However, the favourable mutations must be as strongly selected as the deleterious mutations segregating in the background if they are to have a good chance of ¢xation when there is no recombination (Peck 1994), so that the e¤cacy of this mechanism is questionable (Charlesworth 1996a).

TESTING THE MODELS
(a) Detectable consequences of the di¡erent processes Some limits to the extent to which the above processes might operate could be established if we had accurate information on the e¡ective population sizes of natural populations, together with estimates of mutation rates to deleterious alleles and selective e¡ects of mutations.As noted in ½ 2, variation in silent DNA sequences can provide estimates of N e , which must be corrected for the di¡erent e¡ective population sizes experienced by autosomal, X-linked and Y-linked loci (Caballero 1995).However, population subdivision complicates the estimation of N e because total neutral diversity values for alleles sampled species-wide will often be in£ated over the value (N e P ) for a panmictic population with the same total number of breeding individuals.N e P is probably the value most relevant to the processes discussed above; in simple migration models, such as the island or stepping-stone models, it can be estimated from the mean withinpopulation diversity value, although this is not true in general (Nagylaki 1998).Estimates of mutational parameters for deleterious mutations are currently even more problematical, with di¡erent investigations yielding widely divergent values (Keightley & Eyre-Walker 1999;Lynch et al. 1999).
Given these di¤culties, we can reach only very tentative conclusions about the possible roles of di¡erent mechanisms from this kind of approach.Another approach is to study patterns of molecular variation and evolution at loci on newly evolving Y chromosomes, and to relate them to the predictions of the models (Charlesworth 1996a).A starting point is to note that they all predict that the realized e¡ective population size of a proto-Y or neo-Y chromosome should be greatly reduced below that predicted by demographic considerations.Estimates of silent variability at loci on such chromosomes should therefore be consistently lower than would be expected from measurements of variability at X-linked or autosomal loci, after the necessary adjustments for demographic factors and sexual selection (Nunney 1993;Caballero 1995).
A predicted consequence of a reduction in N e is an accelerated rate of ¢xation of slightly deleterious substitutions, and a reduction in the rate of ¢xation of slightly advantageous mutations (Kimura 1983).Thus we might expect an increased rate of amino-acid replacement relative to silent substitutions on evolving Y chromosomes (provided that su¤cient time has elapsed since the origin of the sex-chromosome system), and more ¢xations of non-preferred versus preferred synonymous substitutions at silent sites in exons (Charlesworth 1996a).If, however, Y chromosomes predominantly experience slower adaptive evolution relative to the X, as suggested by Rice (1996) and Orr & Kim (1998), the rate of replacement substitutions on the Y should be less than for the loci elsewhere in the genome.If both processes have a role, there should at least be increased variation between sites, some having faster substitution rates and others slower rates.Other tests for the signatures of the various processes are listed in table 1.Although in principle there are several criteria that could be useful, statistical power to discriminate between the di¡erent models requires extensive sequence information.
Very few studies have been conducted with the explicit aim of testing the observable consequences of di¡erent models of Y-chromosome degeneration.We review these below, together with relevant information from work performed for other purposes.

(b) Drosop hila neo-Y-chromosome systems
Given the lack of crossing over in male Drosophila, an autosome that forms a neo-Y chromosome by fusion or translocation to a sex chromosome will immediately be exposed to forces that cause degeneration.Several such systems, of various ages, are known in the genus Drosophila.Given the highly conserved gene content of Drosophila chromosome arms (Muller's `elements' (Muller 1940)), the opportunity exists to use data on the molecular variation and evolution of neo-X and neo-Y chromosomal genes to test the predictions of the models (Lucchesi 1978;Charlesworth 1996a;Steinemann & Steinemann 1998).
A prediction common to all models is that the process of evolution of a completely degenerate neo-Y chromosome, and of full dosage compensation of the genes on the neo-X chromosome, should be very slow.The comparative evidence from Drosophila supports this prediction (Charlesworth 1996a).The D.pseudoobscura neo-X chromosome was formed by the fusion of element D to the true X chromosome ca.13 million years ago.There is apparently complete dosage compensation of genes on the neo-X, and loss of activity of all genes on the neo-Y.The related species, D. miranda, which diverged approximately two million years ago, has a neo-Y chromosome formed by the fusion of element C with the true Y chromosome.DNA sequence data indicate that the neo-X and neo-Y chromosomes have been diverging for approximately one million years (Yi & Charlesworth 2000).Although many of the genes on the neo-Y still retain their function, a substantial fraction have become non-functional or completely lost, and their homologues on the X have become dosage-compensated (Steinemann & Steinemann 1998, 1999).
The neo-Y of D. americana americana is a very recent fusion between element B and the X chromosome, which are unfused in its  a This refers to the sp ectrum of frequencies of segregating changes at silent sites.b This refers to negative linkage disequilibrium betweenpairs of synonymous variants representing preferred codons (Akashi 1994).c These include amino-acid replacement changes at weakly constrained sites, and changes to synonymous variants representing unpreferred codons.d These include amino-acid replacement changes at strongly constrained sites, deletions of parts of coding sequences, or frameshift mutations.
e Q is the ratio of interspecies divergence to intraspecies diversity value for slightly deleterious changes, divided by the ratio for neutral changes (Charlesworth1994).
only very small genetic distances, and the Adh locus on the neo-Y (which is very close to the centromere of the fused chromosome) shows no ¢xed sequence di¡erences from its neo-X counterpart (McAllister & Charlesworth 1999).These data suggest that the neo-Y is either only a few tens of thousands of years old or is still recombining with the neo-X chromosome (McAllister & Charlesworth 1999).Genetic experiments indicate that most loci on this chromosome are active (Charlesworth et al. 1997).
Another prediction, again common to all models for Ychromosome degeneration, is a reduced e¡ective population size for genes on the Y chromosome.Reduced Y-chromosomal silent-site variability is thus exp ected, and suggests the operation of some process or processes constantly reducing Y diversity.The ideal comparison would be between sequences of orthologous X-and Ylinked genes; autosomal data are also needed, to test whether diversity is reduced at Y-linked loci, or unusually elevated at X-linked loci.It is not necessary to study the loci at which the hypothesized mutational and/or selective processes are actually happening.All that is needed is a set of data from loci on the Y and X chromosomes, because (owing to the absence of recombination) all Y-linked loci will be a¡ected by the processes outlined above.In D. a. americana, the neo-Y-linked Adh locus does indeed have signi¢cantly reduced variability compared with its neo-X counterpart, suggesting the action of one or more of the forces discussed in ½ 2 (McAllister & Charlesworth 1999).
Among the potentially useful Drosophila systems for investigating the degeneration of neo-Y chromosomes, the D. miranda system seems the most promising, from the data on its age reviewed above.The Lcp1 and Lcp3 loci are present on both the neo-X and neo-Y chromosomes; the neo-Y copies apparently have reduced levels of expression (Steinemann & Steinemann 1998).Sequence comparisons indicate that silent variability at these loci on the neo-Y is ca.29% of the neo-X value, which is similar to the null expectation of 33% (Yi & Charlesworth 2000).However, the con¢dence interval for this ratio is large, owing to the low polymorphism levels observed for both chromosomes; indeed, the neo-X Lcp loci seem to have lower silent diversity than loci on the true X chromosome, p ossibly because of hitchhiking e¡ects associated with the evolution of dosage compensation (Yi & Charlesworth 2000).A survey of seven microsatellite loci that are present on both chromosomes indicates substantially reduced variability on the D. miranda neo-Y (ca.3% of the neo-X value (Bachtrog & Charlesworth 2000)), which is consistent with a severely reduced e¡ective size.
Because low mutation rates could also account for low Y-chromosomal variability, this must be tested for by comparing divergence between species, for Y-linked and other loci.Patterns of replacement and synonymous substitutions in the D. miranda lineage, with the use of D.pseudoobscura as an outgroup, show that amino-acid replacements have occurred on the neo-Y at the same rate as at silent sites, but at a much lower rate on the neo-X (Yi & Charlesworth 2000).This suggests relaxed selective constraints on the neo-Y, either because of reduced e¡ective size or inactivation of neo-Y loci, and is the opposite of what would be expected if adaptive evolution were occurring more rapidly on the neo-X than the neo-Y (Rice 1996;Orr & Kim 1998).
Transposable elements have accumulated at a very high abundance on the D. miranda neo-Y chromosome, and might have been involved in causing a loss of gene activity (Steinemann & Steinemann 1998).Muller's ratchet is a possible mechanism for the accumulation of elements in non-recombining genomic regions (Charlesworth & Langley 1991), although other population-genetic processes can also cause the accumulation of transposable elements and other types of repetitive DNA in such regions (Charlesworth et al. 1994).To evaluate the ratchet's possible contribution to the degeneration of the D. miranda neo-Y chromosome, we would need to have estimates of the net rate of insertion of elements into this chromosome, and the mean selection coe¤cient on insertional mutations, neither of which is available.Estimates of the net rate of movement of elements in D. melanogaster are consistent with a minimum of around 0.005 insertions per generation into a genomic region of the size of the D. miranda neo-Y chromosome; the actual rate could well be three or four times this (Maside et al. 2000).As noted in } 2(c), with a mutation rate of 0.015 the ratchet can cause a substantial decline in mean ¢tness over a realistic period, even in a very large population, provided that selection coe¤cients are small enough (Gordo & Charlesworth 2001).The role of the ratchet in causing the accumulation of deleterious transposable element insertions in this and other Y-chromosome systems (Bull 1983;Charlesworth et al. 1994) is therefore an important open question.

(c) Comparisons of sequence diversity between homologous Y-and X-linked loci in plants
A comparison of variability between single-copy Y-and X-linked loci in the dioecious £owering plant Silene latifolia (white campion) also suggests a reduced Y-chromosomal e¡ective population size (Filatov et al. 2000).The sex chromosome system of this species probably evolved between eight and 20 million years ago, on the basis of estimates of the divergence times of dioecious species from their closest relatives (Desfeux et al. 1996).However, the Y chromosome retains some expressed genes, including the recently discovered SLY1 locus with an X-linked homologue SLX1 (Deliche¨re et al. 1999).In a 2 kb region of these genes, sequenced from plants sampled from a number of natural populations, there were 54 ¢xed nucleotide di¡erences, and no shared polymorphic sites, between the X and Y genes, showing that the SLY1 gene must be in the non-recombining part of the Ychromosome.Net silent divergence between the coding regions of SLX1 and SLY1 is ca.3%, suggesting a time of ca.2.5 million years since the cessation of recombination between the two sets of alleles.This assumes a molecular clock of 0.6 silent-site di¡erences per million years, which is widely used for plants (Gaut 1998); however, this calibration is quite uncertain (Bremer 2000).
When SLY1 and SLX1 sequences from 12 males were compared, the SLY1 diversity was about one-twentieth of that for SLX1 (Filatov et al. 2000).The X^Y di¡erence is most probably due to low SLY1 sequence diversity, not to unusually high variability between the X-linked alleles.Only one autosomal gene has so far been studied in this species, but it shows signi¢cantly higher diversity than SLY1 and does not di¡er signi¢cantly in diversity from SLX1 (V.Laporte, D. Filatov, C.Vitte and D. Charlesworth, unpublished data).The low SLY1 diversity cannot be explained by a low Y-chromosomal mutation rate, because comparisons with outgroup species in the genus suggest a higher divergence rate than for the X-linked gene; the low Y-chromosome diversity is thus even more striking than the raw data suggest (D.Filatov, V. Laporte, C. Vitte and D. Charlesworth, unpublished data).
A further interesting aspect of the results from the S. latifolia Y chromosome is that they do not readily ¢t a recent selective sweep.Such an event would eliminate all diversity, so that existing SLY1 polymorphisms must have accumulated since the selective sweep and should thus mostly be at low frequencies (see table 1).However, all six SLY1 polymorphic sites found are non-singletons, and no signi¢cant deviations from neutrality are detectable by standard test statistics (Fu & Li 1993).However, this does not take into account the probable population subdivision in this species.Although a selective sweep eliminating variability throughout the worldwide population of this species seems unlikely, e¡ects of selective sweeps within local populations cannot be excluded.

(d) Y-chromosome variability in humans
and Drosop hila In contrast with neo-Y and recently evolved chromosomes, Y chromosomes that have degenerated fully have few active loci and will therefore not be much a¡ected by the processes we have described.Given enough time for variability to accumulate by mutation, sequences in old Y-chromosome systems should therefore have diversity levels similar to those expected from their relative e¡ective population sizes.Disagreement with this prediction might suggest that the processes discussed in ½ 2 are not the only forces a¡ecting Y chromosomes.In Drosophila melanogaster and D. simulans, a comparison of total sequence variability in a Y-linked £agellar dynein gene with several X-linked and autosomal loci indicated a reduction in diversity, even after correcting for the di¡erent e¡ective population sizes for the di¡erent chromosomes (Zurovcova & Eanes 1999).However, so far only this Y-linked gene has been characterized in detail.If other Y-chromosomal genes are extremely large, as is true of the Y-linked £agellar dynein gene, they would present large mutational targets, and hence a role is possible for e¡ects of background selection or selective sweeps.
From the few published results on human microsatellite and nucleotide sequence variability, it is not yet clear whether human populations have Y-linked diversity so low as to be inconsistent with expectations based on purely demographic considerations.In humans and mice, X-chromosomal microsatellite loci have lower diversity than autosomal ones ( Jarne et al. 1998), but extensive comparisons of autosomes with the Y chromosome have not yet been published.The best available microsatellite comparison seems to be between a set of eight human Ylinked loci (Pritchard et al. 1999) and 20 other loci (18 autosomal and two X linked (Di Rienzo et al. 1998)).The loci are not directly comparable in terms of factors that a¡ect mutability, such as repeat length and base composition (Estoup & Cornuet 1999).Nor are the population samples readily comparable; the non-Y-linked loci data came from three populations in di¡erent continents, whereas the Y-linked loci were sampled from much larger geographic units.However, both biases (narrower sampling region, and inclusion of X data in the non-Y loci) should reduce the di¡erence between the Y and non-Y loci.The observed di¡erence is nevertheless striking.The variance in repeat size for the Y-linked loci ranges from 0.50 (South America) to 1.18 (Africa), whereas the non-Y-linked values are 5.26 (South America), 7.49 and 8.30.Only one published comparison exists for a microsatellite locus that is present on both X and Y chromosomes, and it shows no consistent di¡erence between them (Karafet et al. 1998).Because this gene has only recently been moved to the Y chromosome (Schwartz et al. 1998), there might have been too little time for diversity di¡erences to develop.
The di¡erence between the human Y-linked and other microsatellite loci is qualitatively consistent with data on other kinds of variability.It is rarely possible to compare levels of single nucleotide polymorphism in Y-chromosomal loci with homologous X-linked loci, simply because few such gene pairs are known (Lahn & Page 1997), so that the available data include di¡erent kinds of genes and sequence types.A large study of sequence-tagged site (STS) sequences from a worldwide Y-chromosome sample gave a nucleotide site diversity value of 3.6 £10 74 (Underhill et al. 1997), and a DNAchip-based study of a large European sample of autosomal and X-linked STSs gave a similar value (4.5£10 74 (Wang et al. 1998)).Data from several regions of the human Xq22 region, from a worldwide sample of 24 men, yielded a diversity value of 1.7 £10 74 (Anagnostopoulos et al. 1999), whereas an 18.3 kb region of the Y chromosome from the same men had a diversity one-sixth as great (only three segregating sites).Human and chimpanzee divergence comparisons, obtained by sequencing regions that amplify from a chimpanzee by using human primers, suggest a Y-chromosome sequence divergence of ca.1.33%, roughly twice that of X sequences.As for Silene, the estimated mutation rate is therefore certainly not lower in males than females (Anagnostopoulos et al. 1999).
Very few studies of Y-and X-linked sequence variability have compared homologous genes or regions.Much lower variability was found in a 676 bp sequence of the tenth intron of the human ZFY gene, compared with 1089 bp of the corresponding ZFX intron, on the basis of single-strand conformational polymorphism studies of large samples from several di¡erent populations ( Jaruzelska et al. 1999).Although the probably higher mutation rate in males than in females makes this a striking di¡erence, it might not be typical of X and Y genes, because the extreme ZFX diversity is due largely to the presence or absence of an Alu element.Excluding ZFX alleles containing Alu, the di¡erence in variability levels was not signi¢cant.Overall, 22 pairwise comparisons between Y-linked sequences and other loci yielded only one signi¢cant test, after accounting for di¡erences in interspeci¢c divergence levels and multiple comparisons (Nachman 1998), so that the mean human XY di¡erence in variability is perhaps no more than threefold.
Y chromosomes B. Charlesworth and D. Charlesworth 1569

DISCUSSION
Given the di¡erent processes that can operate in large non-recombining genomic regions, it is not surprising that Y chromosomes and neo-Y chromosomes of su¤cient age have, in fact, degenerated.The processes described in ½ 2 are not, of course, mutually exclusive.Further exploration of the expected patterns of molecular variation and evolution produced by the various processes is needed to determine whether the conjectural predictions noted in table 1 are correct, and whether other testable properties of the models might exist.The overlapping predictions of the various models about patterns of molecular evolution and variation (see table 1) make empirical testing to assess the relative importance of the various processes a formidable task.However, it might be possible to answer some of the simplest questions, such as whether ¢rm evidence exists for reduced genetic diversity in systems of evolving sex chromosomes (especially in comparison with established Y-chromosome systems) or for the distortions in allele frequency distributions associated with hitchhiking events.
It is important that several independent examples of evolving Y chromosomes should be studied, because the lack of recombination means that all genes on the Yof any given species share a common genealogy, which limits the amount of information that can be obtained from data on within-species variability.Both evolving Y chromosomes and neo-Y-chromosome systems need to be investigated; whereas the latter might have better-de¢ned evolutionary histories, there is a possibility that their degeneration might be accelerated relative to that of truly newly evolved Y chromosomes, by co-opting pre-existing mechanisms for dosage compensation (Steinemann & Steinemann 1998).
This raises the further issue of whether Y-chromosome degeneration is driven mainly by the accumulation of deleterious mutations on the Y or is due to processes turning down the expression of genes on an evolving Y.We have focused on the ¢rst of these possibilities, but it might well not completely explain Y degeneration.If mutations of relatively minor ¢tness e¡ects, with N e s not much bigger than 1, become ¢xed more rapidly on a degenerating Ychromosome than on the X, selection pressure would develop to reduce the transcription of Y genes, provided that this was compensated for in the heterogametic sex by an increased activity of their counterparts on the X.This could lead to the evolution of dosage compensation in parallel with the loss of gene activity on the Y chromosome (Charlesworth 1978(Charlesworth , 1996a;;Jegalian & Page 1998).More rapid adaptive evolution of X-relative to Y-chromosomal loci would also favour the silencing of Y-linked genes (Rice 1996;Orr & Kim 1998), as would the hypothesis due to Hamilton (1967) that segregation distortion by Y chromosomes has created selection to repress their genetic activity.Active evolutionary modi¢cation of patterns of gene activity (with ¢xation of the relevant DNA sequence changes at regulatory sites) should leave traces of hitchhiking events at sweeps at X-chromosomal loci whose Y counterparts are inactive (Yi & Charlesworth 2000).If their Y-linked homologues have been actively downregulated, we should also detect distorted allele frequency distributions on the Y chromosome.Thus, despite the di¤culties, it seems possible to shed some light on the initially extremely puzzling process of genetic degeneration.
The degeneration of Y chromosomes illustrates the critical importance of recombination for maintaining the ¢tness of large genomic regions.With this view, it is one example of the wider pattern of a reduced ability of selection to maintain adaptation under an essentially asexual mode of reproduction (Maynard Smith 1978;Kondrashov 1993;Barton & Charlesworth 1998).As should be apparent from what we have written, John Maynard Smith has made many in£uential contributions to our understanding of the processes underlying this pattern.

Figure 1 .
Figure 1.Weak selection Hill^Robertson e¡ects in a nonrecombining haploid genome.The mean diversity p er site and mean p roportion of sites with the selectively favoured allele are shown, relative to their values for a freely recombining genome, as functions of the number of sites in the genome.Reversible mutation at rate u between a favoured and a disfavoured allele at each site is assumed, such that 2Ns ˆ4 close relatives, D. a. texana and D. virilis.D. a. americana and D. a. texana are separated by Y chromosomes B. Charlesworth and D. Charlesworth 1567 Phil.Trans.R. Soc.Lond.B (2000)

Table 1 .
Statistics of DNA sequence variation and evolution useful in testing for the action of processes involved in Y-chromosome degeneration (Symbols: + + , a strong e¡ect; + , a weak e¡ect; 7, no e¡ect; (?), a lack of rigorous theoretical analysis.)