An extended dataset We measured heterochiasmy as the log of the male/-to-female ratio (ρ) of autosomal recombination rate measured either with chiasma number or map length. We log-transformed the ratio to avoid bias due to measurement error in the denominator. Chiasma-count data for different species were compiled by Burt et al. [16], and we used their dataset, adding a few recent studies. We compiled genetic map data and linkage studies in animals and plants for which both a male and a female map were available. Only homologous fragments (i.e., between shared markers) in male and female maps were considered (especially in low-resolution maps). Heterochiasmy data were available for 107 species, with 46 sets of data based on genomic maps (Table 2). Table 2 Dataset Pooled by Species with Levels of Phylogenetic Grouping Used in the Analysis Note that references given in Burt et al. [17] were not repeated here a K, kingdom. Numeric indicators in this column are: 1, Animalia; 2, Plantae b P, phylum. Numeric indicators in this column are: 1, Arthropoda; 2, Chordata; 3, Embryophyta; 4, Platyhelminthes c C, class. Numeric indicators in this column are: 1, Actinopterygii; 2, Amphibia; 3, Magnoliopsidae (subclass asterids); 4, Aves; 5, Coniferopsida; 6, Insecta; 7, Liliopsida; 8, Mammalia; 9, Magnoliopsidae (subclass rosids); 10, Trematoda; 11, Turbellaria d Data refers to linkage map (LM) or chiasma count (CC) e Male and female indicate the value for the chiasma count or map length for each sex f Ratio refers to male/female recombination rate g  Vsc refers to the presence or absence of sex chromosome (see Materials and Methods, “Sex chromosome effect”) h Data were obtained from maps DBNordic2 and NIAIJapan (http://www.genome.iastate.edu/pig.html) [54,55] ND, no data Table 2 Continued Phylogenetic inertia Heterochiasmy may evolve so slowly that there is important phylogenetic inertia. Alternatively, it may be so fast-evolving that the amount of heterochiasmy takes on nearly independent values among related species. In the same way, heterochiasmy may be so variable between genotypes within a species that it may be difficult to measure and irrelevant to analyse species specific effects. In order to get a picture of phylogenetic inertia on heterochiasmy, we estimated the phylogenetic autocorrelation of ρ using Moran's I spatial autocorrelation statistic [24]. When standardized, values of Moran's I vary from −1 to 1. Positive values indicate that heterochiasmy is more similar than random within a taxonomic level, whereas negative values indicate that it is more different. Because a few species had multiple estimates of heterochiasmy, we also estimated the within-species correlation. The resulting correlogram is shown in Figure 1. We found that heterochiasmy is a fast-evolving trait: Genotypes tend to be correlated within a species (I/Imax = 0.38, p = 7.9%), but this correlation is lower among species within genera (I/Imax = 0.18, P-value = 13%), and very low when comparing genera within families (I/Imax = 0.039, p = 63%). This pattern is very different from the one observed for highly autocorrelated traits using the same method (for instance, mammalian body size [25]). This analysis indicates that there is very little phylogenetic inertia overall on heterochiasmy, but that the species level is appropriate for our dataset. However, this low level of inertia may nevertheless inflate type-I error while testing the effect of independent variables on heterochiasmy. In order to avoid this problem, we tested the association between different variables and heterochiasmy using a generalized estimating equations linear model correcting for the full phylogeny (see below) [26]. Figure 1 Phylogenetic Correlogram of Heterochiasmy and Selfing Rate The y-axis represents Moran's I rescaled to enable comparisons between each taxonomic level for heterochiasmy (ρ, solid line) and selfing rate (Vm, dashed line). The x-axis represents the taxonomic level: /S is the correlation within species, S/G is the correlation of species within genera, etc. F, family; O, order; C, class; P, phylum; K, kingdom. Filled points indicate significance at p = 0.05. Sex chromosome effect For each species, we reported the presence of sex chromosomes. We defined the variable Vsc with the following values: −1 for XY/XX species, −1/2 for XO/XX or XY/XX without pseudoautosomal regions (marsupials), 0 for species without sex-chromosomes, and +1 for ZZ/ZW species. We distinguished the −1 and −1/2 cases to reflect the fact that, in the latter, recombination does not occur between sex chromosomes, so we expect a lower current selection pressure to suppress recombination. Under the Haldane-Huxley hypothesis, the presence of sex chromosomes is supposed to favour reduced recombination rate in the heterogametic sex. We therefore expect a positive effect of the variable Vsc on ρ. We did not find such an effect in animals or plants (the linear effect of Vsc on ρ is not significantly different from zero [p = 0.75 in animals and p = 0.52 in plants], assuming species were independent), and this result is unchanged if the −1 and −1/2 cases are not distinguished. Given this negative result, there was no need to do a phylogenetic correction. Gametic selection In animals from our dataset, there is no female haploid phase because the completion of meiosis occurs only at fertilisation (sperm triggers the end of meiosis). In male gametes, very few genes are expressed, and sperm phenotype is determined mostly either by the diploid genotype of the paternal tissue or by its mitochondrial genome. Imprinted genes, which can also affect the evolution of heterochiasmy [18,21], may be as numerous as haploid-expressed genes and act as a confounding factor while evaluating the “opportunity” for male or female gametic selection. As a consequence, we did not attempt to evaluate the opportunity for haploid selection in animals. Rather, we focused on plants, in which there is both a male (pollen) and female (ovule) haploid phase and during which many genes are expressed (e.g., as many as 60% of genes may be expressed in the male gametophyte [27,28]). In order to evaluate the effect of the “opportunity for selection” for male haploid phase on ρ, we used selfing rate as an indirect variable estimating the degree of pollen competition. We assume that with high selfing rates, there is less genetic variation among competing pollen grains and, therefore, less scope for haploid selection. We defined Vm (the degree of male gamete competition in plants) using three values depending on the amount of selfing: 0 for dioecious, self-incompatible or largely outcrossing (less than 5% selfing reported) species; 1 for species exhibiting low selfing rates (less than 30% reported); and 2 for other species. We used these three broad categories to reflect the fact that selfing rate is often variable within species and that it is often measured indirectly and with low precision. We therefore expect a positive effect of the variable Vm on ρ if the opportunity for male gametic selection favours smaller ρ values, as predicted by the modifier model [18]. We tested this effect using the 57 species for which we were able to estimate Vm (Table 3). We used a linear model in R [29] assuming that all species are either independent or phylogenetically related. In the latter case, we used a generalized estimating equations linear model [26] with a plant phylogenetic tree to the family level using data from Davies et al. [30], and several calibration points, including the Picea/Pinus divergence approximately 140 million years ago [31], that are not included in the Davies et al. dataset. We found an effect in the right direction with or without correcting for the phylogeny (linear effect of ρ on Vm, p < 0.0002 in both cases, Figure 2). The fact that selfing plants exhibit higher recombination rates than their outcrossing relatives has been mentioned previously in the literature [32,33]. However, in most cases, recombination was measured only in male meiosis. It would be valuable to reexamine this trend in the light of our results that recombination in male meiosis is typically greater than in female meiosis among selfers. Figure 2 Logarithm of Male-Female Ratio in Recombination Rate in Plants Mean and 95% confidence interval of ρ is shown for different groups of plants, assuming normality and independent data points The number of species in each group is indicated next to the mean. Table 3 Plant Species Used to Test the Effect of Male and Female Opportunity for Selection a Ratio refers to male-to-female recombination rate LM, linkage map; CC, chiasma count; n, haploid number of chromosomes; Vm, measure of male opportunity for haploid selection; Vf, measure of female opportunity for haploid selection In order to evaluate the effect of the “opportunity for selection” during the female haploid phase on ρ in plants, we contrasted angiosperms with gymnosperms. In angiosperms, ovules do not compete much with each other on a mother plant, because resource accumulation starts after fertilisation (i.e., during fruit development in the diploid phase). In Pinus (three species in our dataset; see Table 2), male meiosis, female meiosis, and pollination occur in the year prior to fertilisation, but the pollen tube stops growing until the next spring, while the female gametophytes continue to accumulate resources and compete with each other over the course of the year. The same situation occurs in Picea, although the period between female meiosis and fertilisation is only 2–3 mo [34]. Perhaps more importantly, the endosperm (which is the organ managing resources for the zygote) is haploid in Pinaceae, in contrast to the double fertilisation that occurs in angiosperms to produce at least a diploid (typically triploid) endosperm [35,36]. We therefore expect that ρ should be greater in Pinaceae, compared to angiosperms. We assigned Vf (the degree of female gamete competition in plants) the values 1 for gymnosperms and −1 for angiosperms. We expected a positive effect of the variable Vf on ρ according to the modifier model. An effect in the right direction was indeed detected (linear effect of Vf on ρ, p = 0.011 and p = 0.0001, with and without correcting for the phylogeny as above, respectively; see Figure 2).