Plant traits correlated with generation time directly affect inbreeding depression and mating system and indirectly genetic structure.

BACKGROUND: Understanding the mechanisms that control species genetic structure has always been a major objective in evolutionary studies. The association between genetic structure and species attributes has received special attention. As species attributes are highly taxonomically constrained, phylogenetically controlled methods are necessary to infer causal relationships. In plants, a previous study controlling for phylogenetic signal has demonstrated that Wright's FST, a measure of genetic differentiation among populations, is best predicted by the mating system (outcrossing, mixed-mating or selfing) and that plant traits such as perenniality and growth form have only an indirect influence on FST via their association with the mating system. The objective of this study is to further outline the determinants of plant genetic structure by distinguishing the effects of mating system on gene flow and on genetic drift. The association of biparental inbreeding and inbreeding depression with population genetic structure, mating system and plant traits are also investigated.
RESULTS: Based on data from 263 plant species for which estimates of FST, inbreeding (FIS) and outcrossing rate (tm) are available, we confirm that mating system is the main influencing factor of FST. Moreover, using an alternative measure of FST unaffected by the impact of inbreeding on effective population size, we show that the influence of tm on FST is due to its impact on gene flow (reduced pollen flow under selfing) and on genetic drift (higher drift under selfing due to inbreeding). Plant traits, in particular perenniality, influence FST mostly via their effect on the mating system but also via their association with the magnitude of selection against inbred individuals: the mean inbreeding depression increases from short-lived herbaceous to long-lived herbaceous and then to woody species. The influence of perenniality on mating system does not seem to be related to differences in stature, as proposed earlier, but rather to differences in generation time.
CONCLUSION: Plant traits correlated with generation time affect both inbreeding depression and mating system. These in turn modify genetic drift and gene flow and ultimately genetic structure.

Background

One achievement of empirical population genetic studies is the survey of the genetic diversity of thousands of plant and animal species using molecular markers, first on the basis of proteins and then of DNA. This "immense outpouring of data on genetic variation" has been considered to be both a milestone, because any type of organism could be investigated using standardised molecular methods, and a "millstone around our neck", because it resulted in the "depauperization of the diversity of empirical work in evolutionary genetics" [1]. Notwithstanding possible side-effects on the development of the field, such a large dataset of species-level microevolutionary approaches should allow addressing important issues in comparative biology. However, few studies have taken advantage of this rich but heterogeneous data for drawing inferences on patterns of genetic variation and on their relation with species attributes. Among the few attempts to relate genetic diversity statistics to species attributes, some influential studies have dealt with plants (e.g. [2-5]). Unfortunately, these early studies neither accounted for phylogenetic inertia in the data nor for potentially confounding covariates, making it difficult to identify causal mechanisms. We recently showed that when accounting for phylogenetic signal and covariation between traits, the only species trait that consistently predicted genetic structure was the mating system: selfing species tend to build up more genetic structure than outcrossing species [6]. Mating system was also shown to be the best predictor of plant genetic diversity [7]. However, mating system was not quantified precisely in these studies [2-4,7]. Furthermore, plant traits could affect genetic structure indirectly, via their effect on mating system, so predicting mating system itself on the basis of plant traits is of interest. Here we revisit the interrelations between population genetic structure, mating system and plant traits using a quantitative estimate of the mating system (i.e. outcrossing rate), while controlling phylogenetic signal and using traits as covariates.

Species genetic structure reflects the balance between divergence processes and cohesion processes among populations. In particular, gene flow limits genetic structure, whereas genetic drift increases genetic structure. Hence, traits affecting mating system and dispersal should affect genetic structure by modifying the drift/migration equilibrium. Since traits affecting genetic structure often evolve conservatively during evolution, measures of genetic structure typically present a strong phylogenetic signal (see e.g. [6]). This does not imply that FST [8], which measures the mean proportion of the total genetic variance contained in a population, is an intrinsically 'heritable' feature of species. In fact, it is expected to approach equilibrium on a relatively short time scale [9], compared to speciation events. Its strong phylogenetic inertia is due to the fact that related species share similar traits influencing the distribution of genetic diversity. Once phylogeny is controlled for when testing relationships between plant traits and genetic structure, FST (as measured at biparentally-inherited nuclear markers) has been shown to depend mostly on the inbreeding coefficient (FIS) and the mating system, whereas life form or perenniality appear to affect genetic structure only indirectly, through their effects on the mating system and inbreeding [6].

Mating system and inbreeding have long been known to influence species population genetic structure [2,4,6,10,11]. But only qualitative classifications of plant mating systems (selfed, mixed and outcrossed) have been used to study this relationship [2,4,6]. The classification of mating system on the basis of floral morphology can lead to false attributions and does not provide any quantitative measure of actual outcrossing rates. Direct analyses of offspring genotypic data at several gene loci allow a much more precise quantification of multilocus outcrossing rates tm[12]. Such quantitative measures of mating system should help study the association between plant traits, mating system, inbreeding and FST.

Mating system can affect the distribution of genetic diversity in different ways, in particular by its effect on pollen-mediated gene flow and by its effect on genetic drift within population. This is because selfing can (i) lower gene flow among populations by limiting pollen-mediated gene flow (as pollen used for self-pollination does not contribute to gene flow) and (ii) increase inbreeding, thereby reducing effective population size Ne and increasing genetic drift [13]. Which of these two effects predominates in plants is typically not known. The definition of an alternative measure of FST not affected by the reduction of Ne due to inbreeding would help to evaluate their respective influence.

Mating system is related with inbreeding in a complex, interdependent way: while outcrossing rate should have a strong impact on inbreeding, this factor, in turn, directly influences the evolution of the mating system. Indeed, the main factor controlling mating system evolution appears to be inbreeding depression, the detrimental effects of close inbreeding on fitness [14]. In plants, inbreeding depression affects inbred (especially selfed) progeny, particularly during the earliest stages of the plant's life cycle [15]. Hence, the inbreeding coefficient (FIS) of adult plants integrates information not only on selfing rate [16] but also on inbreeding depression. The selective removal of inbreds limits genetic drift and hence indirectly genetic structuring. When pollen and seed dispersal are limited, resulting in intra-population structure, FIS can also be affected by biparental inbreeding. Hence, comparing the expected inbreeding coefficient due to selfing, Fe, with the FIS observed at the adult stage allows testing the impact of both inbreeding depression and biparental inbreeding on population genetic structure.

Inbreeding depression could depend on plant traits and is typically stronger in perennials than in annuals [17-19]. Among the models proposed to explain the relationships between plant traits and inbreeding depression, the Τ-model of plant mating system evolution states that the per-generation mutation rate of a plant is a function of the number of mitoses that occur from zygote to gamete. High-Τ plants (typically, high-stature plants) should accumulate more mutations during their lifetime and their offspring should therefore show more inbreeding depression than low-Τ plants (low-stature plants) [17], thereby selecting for outcrossing. Alternatively, mating system could depend not so much on the per-generation mutation rate but on generation time per se [20]. This is expected if inbreeding depression had particularly detrimental effects in long-lived plants (for instance because it has multiplicative effects across years, [21]). Finally, generation time rather than stature could drive mating system evolution if the survival of short-lived plants but not of long-lived ones would depend on reproductive assurance conferred by selfing.

In this study, we have gathered data on the partitioning of diversity within and among populations (FST), on individual inbreeding (FIS), on the mating system (based on a quantitative estimate, the outcrossing rate tm), and on some key plant traits (stature, growth form, perenniality, mode of pollen dispersal) for 263 plant species. A characterization of the correlation structure among these variables using phylogenetically-controlled regression methods should provide insights into the determinants of plant genetic structure. In addition, using an alternative to FST that controls for the impact of inbreeding on genetic differentiation, we will attempt to distinguish the effects of the mating system on gene flow and on genetic drift. Finally, we will test the occurrence of biparental inbreeding and of selection against inbreds to assess their relevance in explaining the relationships between population genetic structure, mating system and plant traits.

Methods

The database

A systematic search in the peer-reviewed literature [17,22] yielded 263 plant species for which the following information were available: (i) genetic structure at nuclear markers, as measured by the FST index, (ii) inbreeding (or heterozygote deficit), as measured by the FIS index, (iii) mating system, as measured by the multilocus outcrossing rate tm (additional file 1). Estimates of outcrossing rate that exceeded 1 (which is technically possible using unbiased estimators) were set to 1 because all reported values did not use the same method of estimation. FST estimates were based on allozymes for 216 species and on nuclear microsatellites for 33 species. Although the higher mutation rate at microsatellite loci might sometimes violate the assumption that mutation is negligible relative to migration [23], no difference in FST due to the type of marker was detected (results not shown).

An alternative measure of FST not affected by the reduction of Ne due to inbreeding

In a neutral island model, FST/(1-FST) = 1/(4Ne.m) at equilibrium, where m is the migration rate among populations. Therefore, selfing is expected to increase FST for at least two reasons: (i) selfing can reduce m by limiting pollen-mediated gene flow among populations (in strictly selfing species, only seed dispersal contributes to the overall gene dispersal), and (ii) as a consequence of inbreeding, selfing should reduce Ne by a factor of 1/(1+FIS) [13]. Hence, when the within-population heterozygote deficit FIS is positive, Ne = Ne'/(1+FIS), where Ne' is the expected effective population size in the absence of inbreeding (Ne' is the reciprocal of the probability that two gametes contributing to random separate adults come from the same parent). To distinguish the two effects of selfing on genetic structure, we define an alternative measure, FST' = FST/[(1+FIS)(1-FST)], closely related to the parameter ρ [24]. Like ρ, FST' is not affected by the reduction of Ne due to inbreeding because FST'/(1-FST') = 1/(4Ne'.m) in an island model at equilibrium. Hence, under a neutral model, selfing will increase FST' only if it reduces m by limiting pollen-mediated gene flow.

Plant traits

For each species (additional file 1), information on the following traits was gathered (either from the original publications, by contacting directly the authors of the original articles, in floras or by searching in specialised web sites):

• Species stature, i.e. plant size, expressed in meters. We expect a positive relationship between the stature of the plant and the dispersal ability (through pollen and seeds).

• Species growth form, classified as woody or herbaceous. Succulent species like Hedysarum carnosum and Melocactus curvispinus, which are rare in our database, were grouped with woody species.

• Species perenniality, corresponding to the distinction between short-lived species (annual and biennial species) and long-lived species (perennials). Note that growth form and perenniality, although strongly correlated (all woody plants are perennials), group different species, since herbaceous plants can be either short-lived or long-lived. Those species described in the literature as either short-lived or long-lived species (e.g. depending on the environment) were classified as long-lived species.

• Species mode of pollen dispersal, coded as biotic (insects, bats, birds) or abiotic (wind or water). Both biotic and abiotic pollen dispersal were described for Cocos nucifera and for Castanea sativa; they were grouped together with species having abiotic mode of pollination.

The analysis of the relationship between genetic structure (or mating system) and plant traits depends on the identification of the most relevant traits. In this respect, it would have been preferable to include generation time rather than some of its surrogates such as perenniality in the analyses. Similarly, the assignment of mode of pollination into discrete categories (abiotic and biotic) can be criticised as it is not always based on field observations and ignores the potential for mixed dispersal factors [25]. However, such information is generally not available in the literature.

Phylogenetic tree

A phylogenetic supertree (additional file 2) was built using as backbone (up to the family level) the updated version of the plant phylogenetic tree published by the angiosperm phylogeny group [26]. The details of the topology at lower taxonomic levels were then grafted according to phylogenetic information obtained from various sources, generally from specific phylogenetic studies at the family level or below [27-45]. Nevertheless some phylogenetic relationships were left as soft polytomies when it was not possible to resolve the topology due to lack of published information. Given the different sources of phylogenetic information it was not possible to determine branch lengths which were therefore set to 1, an assumption that generally performs well for phylogeny-controlled comparative studies [46].

Statistical analyses

Transformation of the variables

Qualitative variables were coded as dummy: for growth form, herbaceous = 0 and woody = 1; for perenniality, short-lived = 0 and long-lived = 1; for mode of pollen dispersal, abiotic = 0 and biotic = 1. To improve normality of the continuous variables (FST, FST', FIS, tm and stature) we used the option "Box-Cox-Bartlett transformations" included in the R Package for Multivariate and Spatial Analysis Version 4.0 [47].

Transformation of the data into phylogenetically-independent contrasts

To control for the phylogenetic relationships among species, data were transformed into phylogenetically independent contrasts (PICs, [48]) using the PDAP module [49] included in Mesquite software [50]. Phylogeny controlled methods are statistically conservative and tend to perform better in cross-species comparisons than non-phylogenetically controlled methods, even when their assumptions are violated (like the non-Brownian mode of evolution of the characters or inconsistencies of the chosen phylogenetic hypothesis) [51].

Regression analyses

Regression analyses were conducted under the following assumptions, based on population genetics theory. (i) The outcrossing rate (tm) can be influenced by plant traits. (ii) Individual inbreeding within population (FIS) can be influenced by plant traits and by tm. (iii) Genetic differentiation (FST and FST') can be influenced by plant traits, tm, and FIS. Regressions analyses were performed on the raw data (TIPs analyses: comparison of the values taken at the tips of the tree) and on the independent contrasts using SYSTAT, version 10.2.05 [52]. For PICs analyses, regressions were forced through the origin [48]. Multiple regression analyses were performed with the forward stepwise regression option to integrate in the model the independent variables as a function of their importance (i.e. their degree of relatedness with the dependent variable).

Species were considered mixed mating species when 0.1 0.9 (N = 129). Only five selfed species (tm < 0.1) were present in the database, so no selfed category was used.

Assessing biparental inbreeding and inbreeding depression

For the subset of non-outcrossed species for which both tm and ts (ts is the mean of single locus outcrossing rate) were available (80 species), we apply a sign test on the difference tm -ts to check for the presence of biparental inbreeding. A positive difference (tm >ts) will demonstrate the presence of biparental inbreeding as ts should be more affected by biparental inbreeding than the multi-locus estimator tm (due to an over-estimation of the selfing rate when polymorphism is low, [12]).

The expected inbreeding Fe would correspond to the fixation index (FIS) if selfing were the sole factor causing deviation from Hardy-Weinberg equilibrium. For a mixed mating model in the absence of inbreeding depression, Fe is estimated as: Fe = (1-tm)/(1+tm) [16]. Differences in the levels of inbreeding coefficient between seed and adult cohorts would suggest that other factors than selfing influence genotypic frequencies [53]. High FIS estimates relative to the expectation could be caused by biparental inbreeding, whereas low estimates could reflect the impact of inbreeding depression. To test for statistical trends in the distribution of Fe - FIS for all species and then separately for each growth form by perenniality category (short-lived herbaceous, long-lived herbaceous and woody species), we applied a sign test over all species except outcrossed ones. Predominantly outcrossed species were excluded from these analyses because the seeds produced are not inbred, so that inbreeding depression, even if it exists, cannot manifest itself.

Under inbreeding depression (δ ≡ 1-wi/wo where wi and wo are the fitness of inbred and outbred individuals, respectively), the expected inbreeding is [53]:

For each growth form by perenniality category, an average value of δ was estimated by least square fitting of this equation on the observed FIS and tm values. This approach neglects inter-species variation but captures general trends of the intensity of inbreeding depression according to growth form and perenniality.

Results

Taxonomic coverage

The 263 species are distributed into 155 genera and 73 families; there are 63 gymnosperms genera (all conifers) and 92 angiosperms genera (seven magnoliids, 22 monocots, 63 eudicots). Four botanical families are particularly represented (families that contain more than 5% of the species included in the database): Asteraceae, Fabaceae, Myrtaceae (mainly the genus Eucalyptus) and Pinaceae; together they include 47% of the species of the data set.

General patterns

The mean values of tm, FIS and FST as a function of growth form, perenniality, a combination of perenniality/growth form and pollination mode are provided in Figure 1 and 2 (and see additional file 3). The distribution of tm, FIS and FST clearly depends on plant traits. This is particularly remarkable for tm: long-lived woody species are generally outcrossed, in contrast to short-lived herbaceous species, for which the mating system ranges from fully selfing to fully outcrossing. Values of FST, FIS and tm covary: when FST decreases (for example FST is lower for woody species than for herbaceous species), then FIS also decreases (woody species present a lower inbreeding coefficient than herbaceous species) while tm increases (woody species are more outcrossed than herbaceous species). Nevertheless most of the statistically significant differences among categories in TIPs analyses disappear when the phylogenetic signal is controlled for (PICs analyses). The only relationships that remain significant are those between growth form and FST, between perenniality and FIS, and between growth form/perenniality and tm.

Figure 1

Mean values of For each category the P-value associated with the statistical test is indicated by *** P < 0.001, ** 0.001

0.05. At the left part of the slash the P-value corresponds to the TIPs test and at the right to the PICs test. "--" indicates an absence of statistical test. Standard error values for each category is available in the additional file 3.

Figure 2

Distribution of .

Mean values of For each category the P-value associated with the statistical test is indicated by *** P < 0.001, ** 0.001