Selection and inertia in the evolution of holocentric chromosomes in sedges (Carex, Cyperaceae)

Authors

  • Marcial Escudero,

    1. Department of Molecular Biology and Biochemical Engineering, Pablo de Olavide University, Ctra. Utrera km 1, 41013-Seville, Spain
    2. The Morton Arboretum, 4100 Illinois Route 53, Lisle, IL 60532-1293, USA
    Search for more papers by this author
  • Andrew L. Hipp,

    1. The Morton Arboretum, 4100 Illinois Route 53, Lisle, IL 60532-1293, USA
    2. Department of Botany, The Field Museum, 1400 S Lake Shore Dr, Chicago, IL 60605, USA
    Search for more papers by this author
  • Thomas F. Hansen,

    1. Centre for Ecological and Evolutionary Synthesis, Department of Biology, University of Oslo, PB 1066, Blindern, 0316, Oslo, Norway
    Search for more papers by this author
  • Kjetil L. Voje,

    1. Centre for Ecological and Evolutionary Synthesis, Department of Biology, University of Oslo, PB 1066, Blindern, 0316, Oslo, Norway
    Search for more papers by this author
  • Modesto Luceño

    1. Department of Molecular Biology and Biochemical Engineering, Pablo de Olavide University, Ctra. Utrera km 1, 41013-Seville, Spain
    Search for more papers by this author

Author for correspondence:
Marcial Escudero
Tel: +1 630 373 1979
Email: amesclir@gmail.com

Summary

  • Changes in chromosome number as a result of fission and fusion in holocentrics have direct and immediate effects on the recombination rate. We investigate the support for the classic hypothesis that environmental stability selects for increased recombination rates.
  • We employed a phylogenetic and cytogenetic data set from one of the most diverse angiosperm genera in the world, which has the largest nonpolyploid chromosome radiation (Carex, Cyperaceae; 2= 12–124; 2100 spp.). We evaluated alternative Ornstein–Uhlenbeck models of chromosome number adaptation to the environment in an information-theoretic framework.
  • We found moderate support for a positive influence of lateral inflorescence unit size on chromosome number, which may be selected in a stable environment in which resources for reproductive investment are larger. We found weak support for a positive influence on chromosome number of water-saturated soils and among-month temperature constancy, which would be expected to be negatively select for pioneering species. Chromosome number showed a strong phylogenetic signal.
  • We argue that our finding of small but significant effects of life history and ecology is compatible with our original hypothesis regarding selection of optima in recombination rates: low recombination rate is optimal when inmediate fitness is required. By contrast, high recombination rate is optimal when stable environments allow for evolutionary innovation.

Introduction

Stebbins (1958) and Grant (1958) argued that the function of genetic recombination is to bring about a workable compromise between the contradictory demands for immediate fitness and evolutionary flexibility. In stable communities, already-established species have little to lose through recombination, because rare allelic combinations may have extreme fitness and the loss of gametes is outweighed by those with increased fitness. By contrast, unstable environments may select for lower recombination rates, as those individuals that survive necessarily carry successful allelic combinations and combinations and selection favours high reproductive potential to build up a population as quickly as possible (Mather, 1943; Stebbins, 1958; Grant, 1958; Barton, 1995; Burt, 2000). While this theory has not been studied in a large number of taxa, studies in a handful of plant genera have found correlations between recombination systems and ecological strategies (e.g. Australian Senecio, Lawrence, 1985; Erythrina, Forni-Martins & da Cruz, 1996; and Carex, Bell, 1982).

The sedges, Carex (Cyperaceae), with c. 2100 species, comprise one of the largest angiosperm genera in the world and the largest in the northern temperate regions. The genus exhibits a remarkable chromosome number radiation, ranging from 2= 12 to 124 chromosomes (Roalson, 2008), and intraspecies variation can cover a range of > 10 chromosomes (Luceño & Castroviejo, 1991; Roalson, 2008). Carex species have holocentric (or holokinetic) chromosomes, which are characterized by the absence of localized centromeres. Holocentric chromosomes are found in all studied species of Cyperaceae and its sister family Juncaceae (Greilhuber, 1995) as well as several unrelated angiosperm genera, green algae, and mosses from Bryopsida. In addition, holocentric chromosomes may be found in Rhizaria, velvet worms, nematodes, and several orders of arthropods (reviewed in Mola & Papeschi, 2006). Fragments from fissions of holocentric chromosomes segregate normally during meiosis (reviewed in Cope & Stace, 1985), and single-chromosome fission or fusion events appear not to be underdominant (Faulkner, 1972; Luceño, 1993; Hipp et al., 2009). As a consequence, holocentry allows rapid evolution of chromosome number, mainly from agmatoploid (chromosome fission; Davies, 1956) and symploid (chromosome fusion; Luceño & Guerra, 1996) events.

The drivers of among-species differences in chromosome number distribution within the genus are unknown. Previous works at the section level in Carex have produced contradictory results. Chromosome number exhibits clade-specific shifts in equilibrium values, evolutionary rate, and phylogenetic heritability within a c. 5 million-yr-old clade (Carex section Ovales; Hipp, 2007), suggesting that the dynamics of chromosome rearrangement are nonuniform across the genus. The older (c. 13 million-yr-old) section Spirostachyae exhibits uniformly high phylogenetic heritability and no clade-specific shifts in rate or equilibrium value (Escudero et al., 2010). It remains to be seen whether there are detectable ecological or life-history correlates of cladogenetic shifts in the dynamics of chromosome evolution. Chromosome divergence has a demonstrated effect on the rate of hybridization and gene flow within species (Escudero et al., 2010; Hipp et al., 2010), suggesting that karyotype evolution plays a role in species diversification within the genus.

In addition to affecting rates of gene flow among populations or lineages, chromosome number variation in Carex affects the arrangement and constitution of linkage groups (Faulkner, 1972) and recombination rates within species (Bell, 1982). Recombination in organisms with holocentric chromosomes is likely to be dictated by chromosome number; holocentric species display one to two chiasmata per chromosome, irrespective of chromosome size (Nokkala et al., 2004). Bell (1982) used a comparative data set of chromosome numbers in Carex to test for correlations between recombination rate and habitat, predicting that lower chromosome number would be associated with open, xeric, montane, novel and northerly habitat and a higher rate of production of small propagules. He utilized categorical assignments of species to habitats in a nonphylogenetic comparative study to demonstrate a strong correlation between recombination rate and potential reproductive output as well as significant correlations between chromosome number and a few of the habitat variables. However, his results were ‘at best highly equivocal’ (pp. 429) with respect to his predictions.

The goal of this study was to tease apart the effects of phylogenetic inertia (i.e. resistance to adaptation) and adaptation in the evolution of among-species differences in chromosome number in sedges (Carex), a cytogenetically hypervariable group. We evaluate the hypothesis that environmental stability selects for high recombination rates, and that unstable habitats select for low recombination rates. We used for this purpose a phylogenetic comparative method (the stochastic linear Ornstein–Uhlenbeck model; Hansen et al., 2008) that simultaneously estimates the optimal or equilibrium relationship between a trait and its environment and the rate of evolution or adaptation towards the selective optimum or equilibrium; phylogenetic inertia can be understood as the inverse of this rate. In doing so, we provide evidence for a link between the evolution of chromosome number and the ecological diversification of the temperate zone’s largest angiosperm genus.

Description

Phylogenetic data and analysis

The phylogenetic data analysed for this project are the most comprehensive Carex-wide phylogeny published to date (Waterway et al., 2009; Supporting Information Notes S1). Molecular sampling for that study included 139 accessions (one of two populations of Cymophyllus fraserianus was excluded) sequenced for two nuclear ribosomal DNA regions (the internal transcribed spacer (ITS), excluding the 5.8S gene, which lies between ITS1 and ITS2; and the external transcribed spacer (ETS)) and two adjacent chloroplast regions (the trnL intron and the trnT-trnL spacer) (Notes S1). Genus Carex is paraphyletic with respect to the other genera of Cyperaceae tribe Cariceae (Cymophyllus, which is monospecific, Kobresia with c. 50 spp., Schoenoxiphium with c. 20 spp. and Uncinia with c. 60 spp.). Consequently, all five Cariceae genera are represented in the present study. Sequences were downloaded from NCBI GenBank (http://www.ncbi.nlm.nih.gov/genbank/).

Phylogenetic trees with branches proportional to time were estimated using the uncorrelated log-normal relaxed clock model (Drummond et al., 2006) as implemented in beast v. 1.5.4 (Drummond & Rambaut, 2007). The GTR + I + G substitution model was used as it showed the best fit for all three data sets (ITS (excluding 5,8S), ETS and trnL intron –trnL-F) based on the Akaike Information Criterion in MrModeltest 1.1b (http://www.abc.se/~nylander/). The analysis was conducted on the combined data set using several independent MCMC (Markov chain Monte Carlo) runs of 10 000 000 generations each to assess convergence, assuming the Yule tree prior with the mean substitution rate set at 1.0. A consensus of 9000 trees (1 tree per 1000 generations was retrieved) from one of the runs was obtained with maximum clade credibility and mean heights options of TreeAnnotator v. 1.4.5 implemented in beast with a posterior probability of 0.95 and a burn-in of 1000 trees (1 000 000 generations).

For phylogenetic comparative analyses, we pruned 39 of 139 tips (the five outgroups and 34 ingroup species without cytogenetic counts; Supporting Information Table S1) for a total of 100 species with at least one cytogenetic sample each. Four extra tips were pruned (accordingly our final data set included 96 species): two species without georeferenced specimens to estimate the climatic envelope, and the two species of the Siderosticta group, which is sister to the remainder of the genus and evolves primarily by polyploidy (Yan-Cheng & Qiu-Yun, 1989) rather than chromosome fission and fusion, as is typical in the rest of the Carex genus (Luceño & Castroviejo, 1991). The crown of this large clade (Carex excluding the Siderostictae group) dates to the Late Eocene and Oligocene 30.8 million yr ago (95% highest posterior density = 21.8–41.07; Escudero et al., 2012).

Chromosome data and predictors

Cytogenetic data for the 96 species were taken from a recent compendium of chromosome numbers for the genus (Roalson, 2008). Chromosome means were estimated for each species. Standard errors to be used in the phylogenetic analysis were not estimated separately for each species, as sampling intensity varied widely among species. Instead, we took the sample-size weighted average of sample variances across all species as the global estimate of average within-species variance, then we estimated the squared standard error of the mean separately for each species by dividing the global variance estimate by the sample size of the individual species (as recommended in Labra et al., 2009 and Hansen & Bartoszek, 2012); the sample size for each species ranged from one to 39 counts (Table S1). Initial analyses were performed on log-transformed and raw data. These trials demonstrated that data transformation had negligible effects on our results and conclusions, and consequently only the results from analysis of raw (untransformed) data are reported.

We characterized climate envelopes for the 96 species in our data set based on climatic conditions at localities of georeferenced specimens reported in the GBIF network (The Global Biodiversity Information Facility; http://www.gbif.org). Duplicated and imprecise localities (a minimum precision of 0.0001 latitude and longitude degrees was required) were excluded using R software (R Development Core Team, 2010). In addition, geographic/climatic outliers (localities outside of reported distribution/habitat in the literature; Kükenthal, 1909; Chater, 1980; Ball et al., 2002) were excluded by visual inspection of the data set. Georeferenced specimen records (= 1892 samples per species on average, SD = 3187; range (2–) 30–19 000 samples per species; Table S1) were then used to extract estimated bioclimatic variables for each locality from the WorldClim climate database (at 2.5 min scale; http://www.diva-gis.org/climate). Five of the 19 BIOCLIM variables were used (measurement units and BIOCLIM abbreviations in parentheses): annual mean temperature (°C, BIO1); temperature seasonality (SD × 100, BIO4); temperature annual range or continentality (°C, BIO7); annual precipitation (mm, BIO12); and precipitation seasonality (coefficient of variation, BIO15). These five variables were chosen among the 19 available BIOCLIM variables because they are ecologically meaningful in terms of stable vs unstable habitats. For example, we expect stable habitats to be correlated positively with annual mean temperature (BIO1) and annual precipitation (BIO12), and negatively with seasonality of temperature (BIO4) and precipitation (BIO15), and temperature continentality (BIO7) (for the relationship between BIO1 and BIO12 and the different habitats on Earth, see Sadava et al., 2007). Means and standard deviations over georeferenced specimens were estimated for each species in R (R Development Core Team, 2010) using the package dismo (Hijmans et al., 2010).

Morphological data were taken from (in priority order): Flora of North America (Ball et al., 2002), Flora Europaea (Chater, 1980), and the only world-wide monograph of tribe Cariceae (Kükenthal, 1909). Four measurements were used to characterize the relevant aspects of vegetative and reproductive morphology in Carex: culm length (cm), leaf width (mm), lateral inflorescence unit length (the single peduncle unit of an inflorescence (mm), following Reznicek, 1990), and utricle length (mm). For unispicate species, we report the length of the single inflorescence unit. A proxy of the number of utricles per spike was estimated as length of the lateral inflorescence unit length/utricle length, where the length of the lateral inflorescence unit length was divided by the utricle length. Because observations in floras are given as ranges of variation, following Bell (1982), midpoints of the frequent ranges of variation (excluding low and high outliers) instead of means values were taken. Culm length and leaf width characterize the studied species vegetatively; lateral inflorescence unit length, utricle length and number of utricles per lateral inflorescence unit characterize them reproductively. We interpret lateral inflorescence unit length as a surrogate for the total investment of plants in each inflorescence unit (Bogdanowicz et al., 2011), and utricle length and number of utricles per inflorescence unit as estimates of the rate of production of propagules per inflorescence unit and their sizes (see r/K strategies in Pianka, 1970).

Soil moisture and light habitat were coded following Waterway et al. (2009) as follows: soil moisture: (1) water-saturated soil, (2) moist to dry upland soils, or (3) either intermediate positions on the moisture gradient or variation in habitat with respect to moisture; light habitat: (1) shaded conditions in forest habitats, (2) open habitats, or (3) intermediate positions on the insolation gradient or variation in habitat with respect to the insolation. Data for the core subgenus Carex clade were directly taken from Waterway et al. (2009) and for the remaining species data were inferred from floras (Chater, 1980; Ball et al., 2002). These two categorical variables are also ecologically meaningful in terms of stability vs instability of the habitats. For example, extreme and unstable habitats such as arctic tundra (Sadava et al., 2007) will be characterized by dry soil and high insolation categories. In contrast, more stable conditions such as temperate rainforest (Sadava et al., 2007) will be characterized by water-saturated soil and low insolation categories.

In this study, all analyses were conducted at the level of species. Accordingly, mean diploid chromosome numbers (response variable) and bioclimatic parameters were estimated at species level based on data from individuals. We did, however, include uncertainty attributable to within-species variation as measurement error in the comparative analyses, as described in Hansen & Bartoszek (2012). Separate microevolutionary studies (e.g. at the population or individual level; cf. Felsenstein, 2008; Stone et al., 2011) may complement the results presented here, but would address different research questions and need a different study approach and methodology.

Phylogenetic comparative analysis

In our analyses, the evolution of the response trait (mean diploid chromosome number) was modelled as an Ornstein–Uhlenbeck process, which is often used as a model of adaptive evolution that explains the evolution of a character as a result of the action of natural selection or other deterministic forces (Hansen, 1997; Butler & King, 2004; Hansen et al., 2008). This model has previously been used in comparative studies of chromosome number evolution by Hipp (2007) and Escudero et al. (2010), and evolution of recombination rates by Dumond & Payseur (2008). The Ornstein–Uhlenbeck model includes two components: a deterministic component, a tendency to evolve towards a ‘primary’ optimal state that can be fixed or varying on the tree; and a stochastic component that represents unknown secondary factors affecting the trait. This model is described by the equation inline image, where dY(t) is the change of our trait (chromosome number) over an infinitesimal time interval dt, α determines the rate of change towards a primary optimum or selection equlibrium, θ, Y(t) is our trait at time t, σ is the standard deviation of the secondary stochastic changes and dB(t) are independent random variables that are normally distributed with mean zero and unit variance (i.e. white noise). The model parameters can be translated into parameters more easily interpreted in terms of trait evolution. The phylogenetic half-life, t1/2 = loge (2)/α, is an estimate of the time it takes for a species to evolve halfway from its ancestral state towards its primary optimum in a new niche, and is thus a quantification of phylogenetic inertia (Hansen, 1997). Phylogenetic inertia is the resistance to adaptive change (i.e. it comes about when related species inherit an inert trait from a common ancestor), which then resists adaptive change to new optima (Hansen et al., 2008). The second estimated parameter, the stationary variance, of the Ornstein–Uhlenbeck process (νy = σ2/2α) is an estimate of the residual trait variance expected among species that have fully adapted to their niches, and is directly proportional to the variance of the stochastic change and inversely proportional to the rate of change towards the primary optimum.

The phylogenetic Ornstein–Uhlenbeck model as originally formulated assumed categorical optima for different niches mapped onto the phylogeny (Hansen, 1997; Butler & King, 2004); recently it was extended to continuous predictor variables that do not need to be mapped on the phylogeny (Hansen et al., 2008; see also Labra et al., 2009 for some further developments). This extension retains the Ornstein–Uhlenbeck process as the model of adaptation, but assumes that the primary optimum, θ, is now a linear function of randomly changing predictor variables. The method returns an estimate of the regression of the primary optimum on the predictor variables. This optimal regression describes the optimal adaptive relationship free of ancestral influence or phylogenetic inertia, that is, the predicted relationship we would see if the species were given enough time to complete their adaptation to their current environments. The method assumes that the predictor variables evolve as a Brownian-motion process and that a linear model appropriately describes the relationship between the optimum or equilibrium and the predictor variables. Nevertheless, there are no specific assumptions about the state of any variable along the phylogeny farther than the tip values (i.e. the approach does not rest on any particular internal node reconstructions, although the model does imply a distribution for each internal node).

For our study (evolution of chromosome number in holocentric organisms), it is more correct to refer to the evolution towards an optimal state as an approach to a selective equilibrium state in chromosome number. This karyotypic equilibrium would be determined by the rates of fixed fusion and fission, and we used the model to test whether this equilibrium, and hence the rates of (fixed) fusion and fission, are dependent on the environment (i.e. on the environmental predictor variables we test). Accordingly, we will refer to chromosome-number or karyotypic equilibria rather than optima (see also Hipp, 2007).

In the current study, we utilized a combination of the linear modelling approaches described above for continuous predictors and an ANOVA and ANCOVA extension of the method implemented in slouch (Hansen et al., 2008; Labra et al., 2009) to evaluate the effect of two categorical ecological predictors, soil moisture and insolation, on the evolution of chromosome number as well as new methods for the incorporation of measurement error and computation of the intercept. For this approach, categorical states were mapped onto the phylogeny using several reconstruction methods (when the predictor variable is categorical, there are assumptions about the state along the phylogeny). We report results from parsimony reconstruction, but as results may depend on ancestral character estimations, we also evaluated several other hypotheses for ancestral characters (e.g. using ancestral character estimations based on the all-rates-different likelihood model using the ace function as implemented in the ape R package (Paradis et al., 2004; R Development Core Team, 2010)). All analyses yielded results that are qualitatively the same; thus, only parsimony analyses are reported in this study. We used Akaike’s Information Criterion (AIC) and AIC weights (AICw) to compare models to the ‘no-specific-adaptation model’, in which chromosome evolution evolves towards a single karyotypic equilibrium (this is a regression of form Y = b + ε, where Y is a vector of the trait of interest (in this case, diploid chromosome number), b is the intercept, and ε is the residuals vector). Use of AICw is a way to standardize between 0 and 1 the weight of evidence in favour of each alternative model (Burnham & Anderson, 2002). We also report AIC weights for each parameter relative to all models tested, as an assessment of the total evidential support for that parameter relative to the complete set of models evaluated. Because the plausible set of models considered in our analysis was selected a posteriori, AIC weights for multiple-predictor models are included as a means of evaluating whether combined systems of environment, habitat, or morphological variables are significantly better at predicting chromosome number than single predictors, not as absolute estimates of the posterior support for each model.

Results

Phylogeny and environmental predictors

The topology of the consensus tree (Fig. 1; Notes S2 for parenthetical format; matrix in Notes S3) is congruent with the phylogeny published in Waterway et al. (2009); the few minor topological differences can be explained by methodological differences in phylogeny reconstruction between our study and Waterway et al.’s. The tree is scaled to 1.0 total length (from the root to the tip of any single leaf in the ultrametric tree) to facilitate interpretation of parameter estimates. The response and predictor variables are summarized in Table S1. The total variance in climatic variables explained by among-species differences ranges from 30% (BIO12 and BIO15) to 60% (BIO1), and accordingly, 40–70% by within-species differences. Among the five climatic predictors, one pair exhibits correlations of |r | > 0.70 (BIO4:BIO7, R2 = 0.94). Among the four morphological predictors, no pair exhibits correlations of |r | > 0.70 (maximum R2 = 0.18, utricle length:lateral-inflorescence unit length). Correlations are also low between morphological and climatic predictor pairs (maximum R2 = 0.18, BIO1:leaf width). Climatic and morphological variables exhibit a strong phylogenetic signal; only in one case was t1/2 < 0.5 (0.30 for BIO15). A t1/2 = 0.5 in units of tree height means that a species entering a new niche would need a time span equal to half the tree length before it has lost half the influence of its ancestral state. Regarding habitat characterization (see Fig. 1 for ancestral character estimation of soil moisture), Waterway et al. (2009) suggested clustered distribution on the tree for these categorical variables, which may reflect niche conservatism. A priori, diploid chromosome number may track all the examined predictors (bioclimatic, morphological and categorical habitat), as all of them display strong phylogenetic signals. In addition, our analyses do not strongly violate the assumption of SLOUCH that predictor variables follow Brownian motion (all include t1/2 = ∞ in their support set).

Figure 1.

Ultrametric phylogenetic tree from the consensus of 9000 trees from beast analysis of 139 accessions of Cyperaceae. Forty-three accessions were pruned after analysis. Ancestral character estimation (based on Fitch parsimony) of categorical variables of soil moisture are shown on the branch tree. Blue, water-saturated soils; green, intermediate; red, dry uplands. Diploid chromosome number range is indicated.

Phylogenetic effects in chromosome number for the ‘single equilibrium O–U model’

In a no-predictor (single equilibrium) O–U model, an estimate of t1/2 = 0 (α = ∞, instantaneous adaptation) implies that there is no influence of the past on trait value (no phylogenetic effect), and all species represent independent draws from the trait distribution. By contrast, if t1/2 = ∞ (α = 0, no adaptation or Brownian-motion model), phylogeny is a strong predictor of trait value. In our case, the point estimate of t1/2 for the no-predictor O–U model suggests a strong phylogenetic effect (t1/2 = 0.60 in units of tree length; Fig. 2a), with supported values (values with a log-likelihood until two units lower than the maximum log-likelihood; Edwards, 1992) ranging from a moderate to very strong phylogenetic effect and the hypothesis of species independence strongly rejected (support interval over t1/2 = 0.26–∞; Fig. 2a).

Figure 2.

The support (log-likelihood) difference from the best model is shown on the ordinate relative to the different parameter combinations. (a) Phylogenetic effects on chromosome number in Cyperaceae based on a model including only an intercept (2~ 1; best estimate: t1/2 = 0.60 (0.26–∞) in units of tree height; vy = 216.5 chromosome number squared). (b) Phylogenetic effects and inertia in chromosome number based on a model including temperature seasonality as predictor (best estimate: 0.52 (0.22–7.5); vy = 176). (c) Phylogenetic inertia in chromosome number based on a model with lateral inflorescence unit length as predictor (best estimate: t1/2 = 0.50 (0.22–6.75); vy = 171.5). (d) Phylogenetic inertia in chromosome number based on a model including lateral inflorescence unit length, soil moisture and temperature seasonality (best estimate: t1/2 = 0.38 (0.14–1.85); vy = 140).

Adaptation and inertia in chromosome number

None of the predictor variables explained a lot of the variation in chromosome number, but we still found clear evidence for weak effects from several variables. As we will argue in the Discussion, we judge these effects to be biologically important. Mean chromosome number is more strongly predicted by inflorescence unit length than by any other predictor (R2 = 0.063, AICw = 0.945 relative to a no-predictor O–U model) (Table 1). We interpret inflorescence unit length as a proxy for total resource investment in each inflorescence unit. Chromosome number is negatively correlated with temperature seasonality (BIO4) but with marginal support (R2 = 0.038, AICw = 0.778; Table 1; Figs 2b, 3a). The correlations between chromosome number and the remaining continuous predictors are weak (R2 = 0.000–0.038; Table 1). For categorical habitat predictors, the correlation is overall weak (R2 = 0.035–0.054; Table 1). The best-supported model with only categorical habitat predictors is a model with soil moisture as the sole predictor (R2 = 0.054, AICw = 0.660; Table 1; Fig. 3c), in which chromosome number is positively correlated with soil moisture (dry soil = 51.5 ± 4.7 chromosomes, intermediate = 63.5 ± 6.2 chromosomes, and water-saturated soil = 73.7 ± 7.8 chromosomes). The slouch confidence intervals for the regression parameters and primary optima are conditional on the alpha and sigma parameters of the Ornstein–Uhlenbeck process and they are local confidence intervals. For BIO4, lateral spike unit length and soil moisture predictor, we have taken some alternative values of alpha and sigma (at the edges of the support intervals, and a few internal points) to estimate global confidence intervals. Our conclusions are also supported by the global confidence intervals for the regression parameters and primary optima (results not shown).

Table 1.   Models, phylogenetic half life (t1/2) scaled in tree-length units with 2-unit support interval in parentheses, stationary variance (vy) in units of chromosome number squared, intercept (± SE) in units of chromosome number and slope (± SE) from phylogenetic regression, R-squared (R2 in %), Akaike Information Criterion (AIC) and Akaike Information Criterion weights (AICw) in comparison with the no-specific-adaptation model (single-equilibrium O–U model: 2~ 1) are shown. In all models the response variable is 2n, where 2n is diploid chromosome number. Predictor variables are given to the left of the tilde, and a ‘1’ means that the model has only an intercept.
Model [K]t1/2vyIntercept (± SE)Slope (± SE)R2AICSimple AICw**Global AICw*
  1. Significant and marginally significant AICw are in bold. D, dry soil; W, water-saturated soil; I, intermediate; O, open environment; F, forest; TS, temperature seasonality; LIUL, lateral inflorescence unit length; #chr, chromosome number.

  2. *Global AICw is the AIC weight for each model relative to all models tested. As 20 models were evaluated, the prior expectation for AICw is 0.05.

  3. **Simple AICw is the AIC weight for each model relative to the no-predictor O–U model.

2~ 1 (single-equilibrium O–U)0.60 (0.26–∞)21758.042 ± 3.8760.000752.0990.0113
2~ 1 (Brownian motion) [= 2]57.165 ± 6.7940.000755.6120.14730.0020
2~ mean temperature (BIO1)0.76 (0.28–∞)24455.535 ± 4.7760.311 ± 0.275 #chr./°C0.013752.7050.57520.0084
2~ temperature seasonality (BIO4)0.52 (0.22–7.5)17669.917 ± 6.094− 1.646 ± 0.707 #chr./°K0.038749.5930.77780.0396
2~ temperature range (BIO7)0.52 (0.22–8.9)19066.171 ± 6.416− 0.272 ± 0.181 #chr./°C0.018752.2000.51260.0108
2~ mean precipitation (BIO12)0.62 (0.26–∞)22158.381 ± 5.331− 0.000 ± 0.003 #chr./mm0.000754.0820.27070.0042
2n ~ precipitation seasonality (BIO15)0.80 (0.28–∞)23953.549 ± 5.7040.140 ± 0.127 #chr./mm0.015751.3110.59730.0168
2~ culm length0.62 (0.26–∞)21757.031 ± 4.3970.027 ± 0.055 #chr./cm0.003753.8430.29490.0047
2~ leaf width0.64 (0.26–∞)21759.708 ± 4.103− 0.476 ± 0.303 #chr./mm0.021751.9920.51340.0119
2~ LIUL0.50 (0.22–6.75)17254.858 ± 3.5260.169 ± 0.065 #chr./mm0.063746.4130.94500.1944
2~ utricle length0.62 (0.26–∞)22158.422 ± 4.486− 0.092 ± 0.499 #chr./mm0.000754.0680.27200.0042
2n ~ utricles per LIUL0.54 (0.22–∞)19456.731 ± 3.7260.294 ± 0.178 #chr./utricles0.027751.4210.58400.0159
2~ soil moisture (3 categories)0.52 (0.20–∞)185D = 51.493 ± 4.725
I = 63.496 ± 6.212
S = 73.658 ± 7.842
0.054750.7800.65970.0219
2~ light (3 categories)0.50 (0.22–∞)190I = 59.807 ± 5.892
F = 51.348 ± 5.828
O = 69.075 ± 6.904
0.035752.7640.47120.0081
2n ~ light and soil moisture (9 categories)0.32 (0.16–1.02)136DF = 43.008 ± 5.117
DI = 55.527 ± 6.896
DO = 60.602 ± 8.113
IF = 59.9467 ± 9.070
II = 59.192 ± 6.013
IO = 79.800 ± 9.117
SF = 69.070 ± 12.501
SI = 78.909 ± 9.422
SO = 68.283 ± 5.235
0.163752.9330.39730.0075
2~ LIUL + temperature seasonality0.46 (0.20–3.1)15864.264 ± 6.44LIUL = 0.138 ± 0.064 #chr./mm
TS = − 1.237 ± 0.724 #chr./°K
0.084746.2760.94840.2082
2~ temperature seasonality + soil moisture0.40 (0.16–2.45)149D = 61.778 ± 6.629
I = 69.736 ± 6.433
S = 78.739 ± 7.340
− 2.432 ± 1.274 #chr./°K0.093749.1550.81340.0494
2~ LIUL + soil moisture0.40 (0.16–2.45)149D = 50.061 ± 4.109
I = 58.877 ± 4.927
S = 65.884 ± 6.271
0.251 ± 0.103 #chr./mm0.111746.4860.94300.1875
2~ LIUL + soil moisture + temperature seasonality0.38 (0.14–1.85)140D = 57.650 ± 6.806
I = 65.000 ± 6.655
S = 72.268 ± 7.639
TS = − 1.734 ± 1.267 #chr./°K
LIUL = 0.204 ± 0.101 #chr./mm
0.132746.7050.93690.1680
2~ LIUL + soil moisture and light + temperature seasonality0.26 (0.12–0.72)113DF = 43.530 ± 8.292
DI = 55.570 ± 8.035
DO = 58.969 ± 7.911
IF = 57.8418 ± 9.356
II = 55.951 ± 7.215
IO = 75.263 ± 8.340
SF = 65.781 ± 11.701
SI = 65.645 ± 10.522
SO = 67.005 ± 6.719
TS = − 0.446 ± 1.103#chr./°K
LIUL = 0.204 ± 0.085#chr./mm
0.232750.5040.68950.0251
Figure 3.

(a) Optimal (in grey) and phylogenetic (in black) regressions of diploid mean chromosome number on temperature seasonality in °K. (b) Optimal (in grey) and phylogenetic (in black) regressions of diploid mean chromosome number on lateral inflorescence unit length in mm. (c) Mean chromosome number and the standard deviation (wide boxes) of the data as well as mean ± SE of primary equilibria (thin boxes) for the three soil moisture categories (D, dry uplands; I, intermediate; W, water-saturated). For more details, see Table 1.

Multiple-predictor models explain up to 23% of the variance in chromosome number (R2 = 0.084–0.232; Table 1). However, all strongly supported multiple-predictor models include lateral inflorescence unit length as a predictor, and the supports for these models as estimated using AIC weights do not exceed the AIC support of the model with only lateral inflorescence unit length as a predictor (AICw = 0.194 relative to all models considered; multiple-predictor models range from AICw = 0.168 to 0.208). These models differ, however, in their estimates of phylogenetic inertia. Phylogenetic half-life estimates vary from t1/2 = 0.26 to 0.80 in units of tree length (Table 1), and confidence intervals for many of the models reject the pure Brownian-motion model (i.e. the upper bound on t1/2 < ∞; Table 1). Many of these same models are favoured over the single-equilibrium (no-predictor) O–U model (Table 1; simple AICw > 0.5), although only inflorescence unit length strongly rejects the single-equilibrium model (simple AICw = 0.945). For the most strongly supported models, half-life estimates vary from t1/2 = 0.38 to 0.52, with confidence intervals rejecting both instantaneous adaptation (t1/2 = 0) and Brownian motion (t1/2 = ∞, or approx. t1/2  6) (Table 1).

Discussion

Adaptation, inertia, or unpredictable chromosome number variation?

In general, selection on recombination rates is a weak, secondary force (Barton, 1995; Otto & Barton, 2001). Moreover, the effects of small changes in chromosome number through fusion or fission will only lead to small changes in overall recombination rates. This means that we do not expect to find strong direct selection and rapid adaptation of chromosome number to external variables. Instead, evolution of chromosome number must be erratic and influenced by genetic drift and/or stochastic indirect selection as a result of coincidental associations of the chromosomal mutations to other aspects of the individual phenotype. Accordingly, Dumond & Payseur (2008) concluded that the genomic rate of recombination in mammals follows a neutral evolution (Brownian-motion) model. Hence, it would be unrealistic to predict a strong, dominant association between chromosome number and environmental predictors. Instead, selection for varying recombination rate can at most be one relatively weak force among many others that affect chromosome number variation (reviewed in Hipp et al. (2009) for Carex). To be able to detect such a systematic, but weak primary effect, we need a large number of species to see the pattern among strong ancestral effects and variation from the many secondary effects.

In this study of 96 species of sedges, we were able to detect systematic associations of chromosome number to life-history and habitat variables. Specifically, we found that sedges with higher chromosome numbers tend to have larger lateral inflorescence units, and, to a lesser extent, occur in habitats with water-saturated soils and low temperature seasonality. The total amount of variance in chromosome number explained by these predictors was low. Separately it ranged from 3.8 to 6.3%, and in combination these traits explained 13% of chromosome variation in Carex, leaving 87% unexplained. As explained above, this is to be expected, and in the light of strong phylogenetic inertia, explaining even a small component of total variation in chromosome number can be considered a strong effect. The study also demonstrates that phylogenetic inertia is significant component of every model evaluated (viz., t1/2 > 0). Hence, although the majority of variation in chromosome number remains unexplained by our analyses, we regard our results as consistent with the general hypothesis that chromosomal fusion and fission are influenced by selection to adapt recombination rates to the life history of the organism.

Phenotypic and genetic consequences of chromosome number variation in holocentric organisms

The physiological and ecological implications of polyploidy have been studied a great deal (reviewed in Otto & Whitton, 2000; Balao et al., 2011). Because ploidy changes entail changes in DNA content and gene number, expression differences associated with polyploidization can have dramatic effects on phenotype (Otto & Whitton, 2000). The phenotypic and physiological implications of small cytogenetic changes, however, where changes in DNA content are minimal (quantitative aneuploidy) or null (qualitative aneuploidy: agmatoploidy or symploidy), are not well understood. In Carex, where polyploidy is rare and chromosome evolution proceeds primarily by fission, fusion, and translocations, there is no expected adaptive significance of chromosome number per se (Hipp, 2007). However, chromosome evolution in the genus is expected to affect recombination rates (Bell, 1982) and constitution of linkage groups (Faulkner, 1972). As Bell (1982) observed, chromosome number is an appropriate proxy for recombination rate only if chiasma frequency is fairly constant per chromosome and chromosome rearrangements do not entail chromosome duplication or deletion. Recent research demonstrates that, in organisms with holocentric chromosomes, among-chromosome variation in chiasma frequency is very limited (one or two chiasmata per chromosome, irrespective of chromosome size; Nokkala et al., 2004; M. Luceño, pers. obs. in Rhynchospora, Cyperaceae), and recombination rates have been demonstrated to be dependent on the number of chromosome arms (proper homologue disjunction at meiosis requires at least one chiasma per chromosome arm; Pardo-Manuel de Villena & Sapienza, 2001). Moreover, chromosome number changes in Carex are not associated with correlated changes in DNA content (Chung et al., 2011, 2012).

For example, in our data set, chromosome number ranges from 2= 26 (= 13) in Carex pedunculata to 2= 96 (= 48) in Uncinia phleoides and, accordingly, one might expect 13–26 chiasmata in C. pedunculata and 48–96 in U. phleoides during meiosis. Such differences in chiasma number during meiosis certainly entail differences in recombination rates. Thus Bell’s (1982) argument that chromosome number is an appropriate proxy for recombination rates in Carex is supported by the last three decades of additional research in holocentric organisms. We do not expect additional consequences from changes in chromosome number in holocentric organisms. The exception may be positional effects (i.e. after fission or fusion, a gene that was in the middle of a chromosome may be at the extreme and vice versa, which may have consequences in gene expression). Nevertheless, while we expect a systematic effect in recombination rates from changes in chromosome number, we do not expect it in gene expression.

Bell (1982) tested whether recombination rates are lower in species occupying novel, disturbed, or marginal environments and/or species with larger numbers of small propagules. He formulated the habitat hypothesis by investigating whether lower chromosome numbers are characteristic of open, xeric, montane, novel and northerly habitats, and he concluded that habitat correlations were equivocal. In addition, he reported a significant positive correlation between any measure of potential reproductive output and chromosome number. Bell (1982) viewed this finding as a violation of his expectations. Our study, however, refines Bell’s finding and suggests a different interpretation. First, the approach we have taken provides insights into both the phylogenetic component of chromosome variability and rate of evolution towards an equilibrium influenced by inflorescence unit length (and, to a lesser extent, seasonality and soil moisture). Secondly, our results of a correlation between chromosome number and bioclimatic variables, although only marginally significant (Table 1), are congruent with our initial hypothesis, as low chromosome number (low recombination rates) are related to extreme and unstable habitat (dry soil and unstable temperature). Thirdly, we found that variables strongly related to the number of propagules produced per plant and propagule sizes (utricles per inflorescence unit and utricle length) explained less chromosome number variation than inflorescence unit length. This suggests that shorter inflorescences may correlate with lower dependence on local resources, which is characteristic of plants growing in extreme and unstable habitats (e.g. Sonesson & Callaghan, 1991). Thus, our results suggest in aggregate a support for the adaptive significance of recombination rates.

Implications for patterns of lineage diversification

Chromosome rearrangements have previously been argued to be adaptatively significant in Carex (Faulkner, 1972; Luceño & Castroviejo, 1991; Bell, 1982), as they have been for numerous other organisms (e.g. Alvarez-Castro & Alvarez, 2005; Butlin, 2005; Alvarez-Castro & Carlborg, 2008; Hoffmann & Rieseberg, 2008). The current study is the first to demonstrate that chromosome number itself – which may be a result of many independent and different rearrangements – may be in part a product of correlation with life-history traits or (more weakly supported) adaptation to the environment. While the direction of chromosome evolution has long been of interest in Carex (Heilborn, 1924; Davies, 1956; Luceño & Castroviejo, 1991), chromosome evolution in the genus is clearly not directional (Reznicek, 1990; Hipp et al., 2009; Escudero et al., 2010). The current study provides a nonstochastic explanation for this observation and for previous findings that pure phylogenetic explanations of chromosome variance are inadequate (Hipp, 2007).

Carex is the most diversified angiosperm genus in the temperate regions of the Northern Hemisphere (Reznicek, 1990) and exhibits one of the most exceptional chromosome radiations in angiosperms. The relationship among chromosome number, reproductive traits, and environment reported here may well have played a significant role in the diversification of the genus. In one of the largest sections of the genus, for example, origins of an eastern North American clade from western North American ancestry entailed a significant decrease in equilibrium chromosome number, compatible with a shift in selective regime but not with a pure phylogenetic (Brownian-motion) model (Carex section Ovales; Hipp, 2007). The direction of evolution in this clade – which was largely excluded from Waterway’s data set and consequently from analyses presented in this paper – is in broad terms compatible with the observations reported here: the shift to a novel environment (from western to eastern North America) is associated with a significant decrease in chromosome number (lower recombination rates). Similarly, the change in the Northern Hemisphere from a stable, warm and humid climate during the Miocene and beginning of the Pliocene to a more unstable, dryer, colder climate during the end of the Pliocene and Pleistocene was essential to the expansion of Carex and diversification in the northern temperate regions (Escudero et al., 2012). Congruently, our results suggest a correlation of low chromosome number (low recombination rate) with short lateral inflorescence unit length (presumably associated with lower dependence on local resources), and with dry soil and unstable temperature (Table 1). The capacity to evolve rapid changes in chromosome number may have facilitated the spread of Carex in the temperate zone, as increased potential for evolution of recombination rates may ameliorate the effects of drift in the highly selfing populations typical of Carex (Friedman & Barrett, 2009) or allow for more rapid rates of adaptation to novel environments (Burt, 2000; Betancourt et al., 2009). The diversity of sedges may thus depend on their chromosomal diversity.

Acknowledgements

We thank Dr Francisco Balao, Dr Kyong-Sook Chung, Enrique Maguilla, Dr Richard Abbott and three anonymous reviewers for valuable comments on the manuscript; Dr Jason Pienaar, Dr Ian Pearse and Dr Luis Villagarcía for help with some analyses; and Paco Fernández, Mónica Míguez, Bethany Brown and Marlene Hahn for technical support. This research was supported by the NILS mobility project (UCM-EEA-ABEL-02-2009 to M.E. and T.F.H.); the Spanish Government (CGL2009-09972 to M.L. (PI) and M.E.); and the USA National Science Foundation (NSF-DEB Award #0743157 to A.H.).

Ancillary