Sex-chromosome evolution: recent progress and the influence of male and female heterogamety.
]. Both cases involve a chromosome that is heterozygous (the Y or W) and present in only one sex. Although a broad range of situations has been described, in many cases, most of the chromosome has stopped recombining and has degenerated considerably. Current theory, which we term “degeneration by selective interference,” (DSI) has been substantially refined since the 70’s, but its core idea—degeneration caused by selective interference—has remained unchanged [
Genetic Draft, Selective Interference, and Population Genetics of Rapid Adaptation.
]. DSI involves a sequence of steps that occur after the arrest of recombination between the Y and X chromosomes (all our arguments also apply to Z/W chromosome system): (1) degeneration of Y-linked genes by “selective interference” (also known as the “Hill-Robertson effect”), due to processes such as Muller’s Ratchet, hitchhiking, and background selection [
Y-chromosome evolution: emerging insights into processes of Y-chromosome degeneration.
], (2) facultatively, adaptive silencing of Y-linked genes, and (3) evolution of dosage compensation. A variant of this theory proposes that the accumulation of deleterious alleles in regulatory sequences by selective interference leads to reduced Y gene expression [
Chromosome-wide gene silencing initiates Y degeneration in Drosophila.
In this paper, we propose a new “degeneration by regulatory evolution” (DRE) theory to explain Y chromosome degeneration. The main differences from the DSI model are that our theory does not require selective interference and that steps 1–3 occur simultaneously after recombination suppression. We previously showed that, for autosomal genes, a “cis-regulator runaway” process occurs that leads stronger cis-regulators to become associated with chromosomes with fewer deleterious mutations [
Enhancer runaway and the evolution of diploid gene expression.
]. This favors the stronger cis-regulatory alleles, provided that they are tightly linked to their coding gene. We also showed that cis-regulators diverge in asexuals, where diploid expression is unstable and quickly becomes “haploidized” [
]. DRE theory involves such divergence of cis-regulators, but with an asymmetry between the X and Y chromosomes (preventing the suppression of gene expression on the X). We investigate DRE using individual-based stochastic simulations of a population of Npop diploid individuals, with XY males and XX females. In order to capture the essence of the mechanism, we first study a minimal system with only four loci (Figure 1): a gene G, its cis-regulator C, and two trans-regulators T (we will then extend the model to the case of a non-recombining region comprising a larger number of genes). Trans-regulators, such as transcription factors, are not closely linked to their target gene and influence expression on both homologs, whereas cis-regulators, such as enhancers, control the expression of the closely linked gene and influence only the copy carried on the same chromosome as themselves [
The evolution of transcriptional regulation in eukaryotes.
]. We assume that G and C are present on both sex chromosomes and that they recombine (in females only) at a rate Rc. We include trans-regulators in order to examine whether, and over what time-scale, dosage compensation will evolve when expression of the Y-linked allele decreases. With these trans-regulators, overall expression can be maintained, i.e., dosage compensated, even if cis-regulators change and diverge between the X and Y. Dosage compensation cannot evolve if the trans-regulators act in the same way in both males and females or equally on the X and the Y. Hence, we do not consider all potential trans-regulators, only those that could influence dosage compensation. Specifically, we focus on the simplest symmetrical case with one trans-regulator expressed in males Tm, and one in females Tf. Both cases have been described empirically (C. elegans dosage compensation works by halving X expression in females, whereas in Drosophila, it works by doubling X expression in males [
Sex chromosome dosage compensation: definitely not for everyone.
]). For simplicity, we assume that these T loci are autosomal and that they recombine freely with each other and with the G and C loci.
This initial model is later extended to nLCGTmTf quadruplets of genes (where nL = 1, 50, and 500). In these models, we assume that the C and G loci are uniformly spaced on the sex chromosomes, with two adjacent genes G recombining at a rate Rg in females, and where each C locus is assumed to be closer to the G gene it regulates than to the next G gene (ten times closer in the simulations,); again, recombination is assumed not to occur in males, representing a non-recombining region including the sex-determining locus, while all T factors are again assumed to recombine freely with the sex-determining region. In this model, each CG pair is influenced by its own TmTf pair, which represents the lowest degree of pleiotropy of these trans-acting factors, but involves a very high number of trans-regulators. We also considered a model where only one TmTf pair controls all the G and C loci, representing the other extreme case where trans-regulators are maximally pleiotropic and influence many (here all nL) genes.
Deleterious mutations occur within genes G at a rate UG per gene. Their fitness effect s is drawn from an exponential distribution with mean smean. The effects of multiple mutations in the same gene are assumed to be additive, but with a maximum effect per gene, smax (which may be interpreted as the fitness effect of a full gene knockout). Their dominance depends on the strength of their associated cis-regulator (see STAR Methods). The effects of alleles at the cis– (C) and trans-regulators (Tm, Tf) are modeled as quantitative traits denoted by c, tm, tf, respectively, and control the level of expression Q of the gene, which is under stabilizing selection with intensity I (STAR Methods). The different events of the life cycle occur in the following order: diploid selection, meiosis with recombination, mutation, and syngamy. Simulations are initialized with no polymorphism present, and the optimal gene expression level (no deleterious allele, all c and tm, tf alleles fixed to 1). After a burn-in phase in which the chromosome evolves with recombination in both sexes, we stop XY recombination in males to create a sex-linked region and follow the frequencies and effects of deleterious mutations on the X and Y, as well as the evolution of the regulatory genes. These outputs are averaged over different numbers of replicates depending on the variance in the process under different parameter values (STAR Methods; Table S1). At regular intervals, we compute Phalfsilent, the probability (across replicates) that Y allele-specific expression
decreased by two fold from the initial value of 0.5 to
(see STAR Methods). We also compute Psilent, the probability that
becomes close to zero (below 0.01), so that alleles on the Y become nearly entirely recessive. The quantity Phalfdead then refers to the probability that, by a given number of generations after the Y-linked region stopped recombining, deleterious mutations on the Y gene copy have reduced fitness by an amount
, and Pdead that they reduced fitness by an amount
, indicating that the gene has entirely degenerated on the Y.
Figure 2 illustrates the process with nL = 1 (one gene, one cis-regulator, one male limited and one female limited trans-regulator). The system does not generate any male-female or X-Y asymmetry before recombination arrest (STAR Methods; Figure S1). After recombination arrest, the gene carried by the Y degenerates: it becomes progressively recessive, as
changes from h = 0.25 to zero, and accumulates deleterious mutations (the overall fitness effect of mutations present on the Y copy increases up to smax), despite there being very limited selective interference (at most only occurring between the gene and its cis-regulator). Silencing occurs first, and the accumulation of deleterious mutations follows later in the process (the curve representing Phalfsilent is ahead of the one showing the accumulation of deleterious mutation on the Y, as measured by Phalfdead). Degeneration also occurs with full dosage compensation, and overall expression never departs from the optimum in either sex (Figure 2B). Compensation typically involves, at least initially, a mixture of upregulation of X gene copies in males and downregulation in females (STAR Methods; Figure S3).
What is the underlying cause of this asymmetrical degeneration? Once X and Y stop recombining, diploid expression becomes unstable. Cis-regulators on the X and Y can diverge, eventually leading to the haploidization of expression in males. This is not prevented by stabilizing selection on expression levels as long as trans-regulators can coevolve to maintain near optimal total expression in both sexes. When the strength of cis-regulators on the Y starts decreasing, the process is accelerated by a “haploidization” positive feedback loop. Indeed, weak Y cis-regulators become associated with coding sequences carrying more deleterious mutations, as they cause a reduction in dominance. They are then selected to weaken further in order to mask those deleterious mutations, which leads to the accumulation of even more deleterious mutations, and so forth. By contrast, the other “haploidized” situation (in which the X is silenced in males) is reversible, as X chromosomes with weak cis-regulators and a higher load of deleterious mutations cannot fix, as they become homozygous and selected against in females when too frequent (unlike partially silenced Y genes, which can spread as they stay heterozygous in males). Therefore, the regulatory system has only one stable equilibrium, in which the Y is silenced and degenerate.
Selective interference plays no role in this process, explaining why degeneration occurs even for a single Y-linked gene. However, the process is stochastic, as it is initiated by a random departure from diploid expression with a sufficiently weak Y cis-regulator to trigger the “haploidization” feedback loop. In individual simulation replicates, degeneration is indeed very abrupt, but occurs at varying time points (Figure 2A). Because of this stochasticity, the process is slowed down in larger populations: it is
10 times slower in
10 times larger populations (Figure 2D). With the same parameters but without mutation in the cis-regulator (Uc = 0, Figure 2C), as in DSI theory, degeneration does not occur, as expected, since there is neither selective interference nor cis-regulatory divergence. Degeneration does not occur either in the absence of mutation in the trans-regulators (Ut = 0, Figure 2C) and for the same parameter values, since cis and trans-regulators have to coevolve to maintain total expression levels: if trans-regulators cannot evolve, the divergence of cis-regulators is prevented and Y degeneration cannot occur. However, if the intensity of stabilizing selection on expression levels is weak enough, degeneration evolves but is not dosage compensated (Figure 2C). Otherwise, the intensity of stabilizing selection on dosage only plays a marginal role in DRE (Figure 2C). Control simulations without mutation in the coding gene (UG = 0, Figure 2C) show that cis-regulatory divergence and Y silencing can occur even in the absence of deleterious mutations, but as expected, this silencing is slower, not being accelerated by the “haploidization” feedback loop, and reversible (STAR Methods; Figure S2).
As expected, DRE and DSI combine when more than a single Y-linked gene is considered. The effect is strong: Figure 3 shows that a 50-fold or 500-fold increase in the number of loci results in degeneration being 5-fold and 10-fold faster, respectively: a larger non-recombining Y-linked region degenerates faster than a small one. Without mutation in regulators (i.e., with only DSI), degeneration occurs but is 23–36-fold slower with 50 loci (depending on the control used for the comparison; STAR Methods; Figure 4). With 500 loci, however, the comparison with and without regulators is problematic, as very quickly, a modest accumulation of deleterious mutations on the many Y-linked genes causes an important reduction in male average fitness so that male fitness reaches unrealistically low values (of the order 10−17). Even if a proportion of genes affecting male fitness may be under soft selection, it seems unlikely that a population with such a low male fitness would survive.
The drop in male fitness is less dramatic with regulatory evolution, as those mutations become progressively more recessive as the Y degenerates. There is nevertheless a transient drop in male average fitness, which can be quite large (e.g., 3% and 85% reduction for a 50-gene and 500-gene Y-linked region, respectively; Figure 3). Data on divergence between sex chromosomes indicate that Y degeneration is often sequential in chiasmate species, with several regions of various sizes, termed “strata” [
Four evolutionary strata on the human X chromosome.
], having stopped recombining at different time points. This high transitory fitness drop may prevent large strata from occurring in small populations and may bias toward scenarios involving multiple small strata as, for example, in humans [
Detecting evolutionary strata on the human x chromosome in the absence of gametologous y-linked sequences.
]. Comparatively, these scenarios involving small strata are more difficult to explain with DSI, as selective interference is weak on small non-recombining regions.
Last, degeneration is initially slower but faster overall when there is only one Tm and Tf controlling all cis-regulators. Despite the large pleiotropy constraint, having only two autosomal trans-regulators precipitates degeneration: the tipping point where it is worth fully silencing the Y is quickly reached when many very weakly deleterious mutations have accumulated on the Y. This is consistent with the observation that dosage compensation can occur locally on a gene-by-gene basis or by chromosome-wide trans-acting effects [
Contingency in the convergent evolution of a regulatory network: Dosage compensation in Drosophila.
]) may be worth investigating but are likely to behave similarly, as long as trans-regulators only target X genes that have a copy on the Y non-recombining region.
Simulations for lower values of mutation rates and strength of selection against deleterious alleles are shown in Figure S4 (for nL = 1 and nL = 50; STAR Methods). Unsurprisingly, reducing mutation rates slows degeneration, while the effect of s is more complicated. However, the acceleration of degeneration caused by regulatory evolution still holds. Figure S4 also shows that a scaling argument from diffusion theory indicates that larger populations with weaker mutation and selection should also behave similarly, albeit on a longer timescale.
The DRE theory proposes a different view of sex-chromosome evolution compared to the DSI theory that has been developed over the past 40 years. In both cases, degeneration starts after the arrest of recombination in a genome region completely linked to the sex-determining locus. In both cases, degeneration is slower in larger populations, but this is considerably less so in the DRE model. However, there are important differences. With very few exceptions [
Muller’s ratchet and the degeneration of Y chromosomes: a simulation study.
], DSI was developed without explicitly modeling regulatory evolution. With DSI, regulatory evolution (Y silencing and dosage compensation) is supposed to occur only after deleterious mutations have accumulated on the Y [
Chromosome-wide gene silencing initiates Y degeneration in Drosophila.
]. This is certainly possible, as selective interference applies to all functional sequences and may contribute to the fixation of many kinds of deleterious mutations, including those maintaining adequate expression levels. DRE is based on a reverse causality: regulatory evolution initiates the degeneration process. Contrary to the standard model in which compensation is needed because degeneration damages genes’ function in males, compensation evolves here from the very beginning of the process, by reducing the proportion of transcripts from Y-linked relative to X-linked alleles and maintaining an almost constant overall level of expression in both sexes. However, compensation may not occur when expression levels are under weak stabilizing selection (Figure 2C). Whether compensation occurs on a gene-by-gene basis or chromosome wide depends on the availability of the corresponding trans-acting factor, but both can occur in DRE and, surprisingly, at approximately the same rate (Figure 3).
The specific mode of dosage compensation depends on the type of trans-regulators. Here, we considered trans-regulators with sex-limited expression, which can mimic several well-known dosage compensation systems (with female-limited factors corresponding to C. elegans or mammal systems, while male-limited factors would be more similar to a Drosophila-like situation [
Sex chromosome dosage compensation: definitely not for everyone.
]; STAR Methods). The symmetric DRE model that we used often led to a Drosophila-like compensation, where the X in males was eventually expressed twice as much compared to the situation at the recombination arrest (Figure 2B). However, this is certainly dependent on the relative mutation rates on the different types of trans-regulators. The theory should be extended to examine the diversity of dosage compensation mechanisms, including sex-of-origin effects [
Genomic imprinting mediates dosage compensation in a young plant XY system.
Overall, we have presented an alternative theory for the degeneration of sex chromosomes. Although many underlying parameters are still poorly known, this theory could be tested quantitatively, as it works faster than current theory, on smaller non-recombining regions, and it does not require small population sizes or recurrent beneficial mutations causing hitchhiking effects. It does not exclude selective interference, which will necessarily co-occur as long as many genes stop recombining simultaneously. However, a hallmark of DRE is that regulatory changes occur very early. This is consistent with recent studies showing that dosage compensation evolves early on [