Associations between sex-organ deployment and morph bias in related heterostylous taxa with diff erent stylar polymorphisms 1

Heterostyly is a genetic polymorphism in which populations are composed of two (distyly) or three (tristyly) fl oral morphs that differ in the reciprocal placement of stigmas and anthers within fl owers ( Darwin, 1877 ; Barrett, 1992 ). Th e genetic control of the polymorphism in distylous plants usually involves a single Mendelian diallelic locus in which the long-styled morph is of genotype ss and the short-styled morph is of genotype Ss , although in several species the dominance relations at the S -locus are reversed ( Lewis and Jones, 1992 ). Th e fl oral morphs are maintained in populations by negative frequency-dependent selection resulting from intermorph (disassortative) mating. With this genetic system and disassortative mating, a 1:1 morph ratio (isoplethy; Finney, 1953 ) is expected in equilibrium populations. Th e classic textbook depiction of heterostyly as a balanced polymorphism (e.g., Roughgarden, 1979 ; Silvertown and Charlesworth, 2009 ; Charlesworth and Charlesworth, 2010 ) is largely based on knowledge of distyly in Primula and emphasizes how the reciprocal positioning of sex organs (reciprocal herkogamy) associated with a self and intramorph incompatibility system (heteromorphic incompatibility) promotes outcrossing in populations. Th e heterostylous 1 Manuscript received 28 September 2016; revision accepted 29 November 2016. 2 Department of Plant Biology and Soil Sciences, Faculty of Biology, University of Vigo, As Lagoas-Marcosende 36200 Vigo, Spain; 3 CFE, Centre for Functional Ecology and Department of Life Sciences, University of Coimbra, Calçada Martim de Freitas, 3000-456 Coimbra, Portugal; 4 Department of Ecology and Evolutionary Biology, University of Toronto, 25 Willcocks Street, Toronto, Ontario, Canada M5S 3B2; 5 Department of Biology and Centre for Environmental and Marine Studies, University of Aveiro, 3810-193 Aveiro, Portugal; 6 Department of Ecology and Evolution, Stony Brook University, 650 Life Sciences Building Stony Brook, New York 11794 USA; and 7 Departamento de Biología Vegetal y Ecología, Universidad de Sevilla, Apartado 1095 41080 Sevilla, Spain 8 Author for correspondence (e-mail: victoferrero@gmail.com) doi:10.3732/ajb.1600345 Associations between sex-organ deployment and morph bias in related heterostylous taxa with diff erent stylar polymorphisms 1


R E S E A R C H A R T I C L E
Heterostyly is a genetic polymorphism in which populations are composed of two (distyly) or three (tristyly) fl oral morphs that differ in the reciprocal placement of stigmas and anthers within fl owers ( Darwin, 1877 ;Barrett, 1992 ). Th e genetic control of the polymorphism in distylous plants usually involves a single Mendelian diallelic locus in which the long-styled morph is of genotype ss and the short-styled morph is of genotype Ss , although in several species the dominance relations at the S -locus are reversed ( Lewis and Jones, 1992 ). Th e fl oral morphs are maintained in populations by negative frequency-dependent selection resulting from intermorph (disassortative) mating.With this genetic system and disassortative mating, a 1:1 morph ratio (isoplethy; Finney, 1953 ) is expected in equilibrium populations.
Th e classic textbook depiction of heterostyly as a balanced polymorphism (e.g., Roughgarden, 1979 ;Silvertown and Charlesworth, 2009 ;Charlesworth and Charlesworth, 2010 ) is largely based on knowledge of distyly in Primula and emphasizes how the reciprocal positioning of sex organs (reciprocal herkogamy) associated with a self and intramorph incompatibility system (heteromorphic incompatibility) promotes outcrossing in populations.Th e heterostylous syndrome has evolved on numerous occasions in unrelated animalpollinated families of fl owering plants ( Ganders, 1979 ;Lloyd and Webb, 1992a ;Barrett et al., 2000 ), and is perhaps the most wellstudied discrete fl oral polymorphism.
Investigations of heterostyly have broadened since Darwin's seminal work on Primula and Lythrum ( Darwin, 1877 ) to include many more families.It is now evident that reciprocal herkogamy may vary considerably in expression and can be associated with diverse compatibility systems, while still functioning to promote varying degrees of disassortative pollen transfer ( Barrett and Richards, 1990 ;Dulberger, 1992 ;Lloyd and Webb, 1992a , b ;Barrett and Cruzan, 1994 ;Ferrero et al., 2012 ;Zhou et al., 2015 ).Moreover, although the frequencies of style morphs in populations are governed by the aggregate patterns of mating in preceding generations, a variety of stochastic and deterministic processes can result in morph ratios that deviate signifi cantly from equality.Founder events and genetic drift commonly result in biased morph ratios (anisoplethy), especially in species in which features of the life history (e.g., clonality and episodic sexual recruitment) slow progress to the isoplethic equilibrium ( Ornduff , 1972 ;Morgan and Barrett, 1988 ;Eckert and Barrett, 1995 ).Although less commonly documented, morph-specifi c diff erences in reproductive fi tness can also cause biased morph ratios in heterostylous populations ( Barrett et al., 1983 ;2004 ;Brys et al., 2008a ;Weber et al., 2013 ).Th us, determining the causes of anisoplethic morph ratios in heterostylous populations is a complex problem that usually commences with a study of the reproductive correlates of morph-ratio variation, an approach we use here.Haldane (1936) fi rst recognized that distylous populations should proceed more rapidly than tristylous populations to an isoplethic equilibrium.He pointed out that in the absence of "illegitimate unions" (self and intramorph mating) in distylous populations, the frequencies of the L-and S-morphs should be fully restored to equality in one generation aft er any particular perturbation ( Haldane, 1936 , p. 396).Th is inference assumed a tight association between the stamen and style polymorphism and heteromorphic incompatibility.However, not all species with stylar polymorphism possess strong heteromorphic incompatibility, and in species in which self and intramorph mating are permitted, other morph ratio dynamics and equilibria are possible.For example, if style morphs in a population diff er in rates of disassortative and assortative mating, various L-morph and S-morph biased anisoplethic equilibria are possible.Indeed, there is empirical evidence of biased morph ratios in species in which enforced disassortative mating does not occur because of the absence of heteromorphic incompatibility ( Ganders, 1975 ;Ray and Chisaki, 1957 ;Barrett et al., 1996Barrett et al., , 2004 ; ;Baker et al., 2000a , b ;Pérez-Barrales and Arroyo, 2010 ;Simón-Porcar et al., 2015a ).In such cases, fl oral morphology plays a more signifi cant role in governing patterns of pollen transfer and mating than in the species envisioned by Haldane, in which illegitimate unions were prevented by physiological incompatibility.
Because of the functional link between fl oral morphology, pollen transfer, and mating in heteromorphic taxa, the degree of reciprocity of stigmas and anthers between the style morphs and their spatial separation within a fl ower (herkogamy) should play a key role in governing rates of disassortative mating with consequences for morph ratios ( Barrett et al., 2004 ).In species with heteromorphic incompatibility, weak sex-organ reciprocity will be costly because pollen transferred to incompatible stigmas by animal pollinators involves gamete wastage.Th is loss in male outcrossed siring opportunities through illegitimate pollen transfer should select for greater sex-organ reciprocity and greater herkogamy.In contrast, in species in which intramorph mating is permitted, because of the absence of heteromorphic incompatibility, pollen wastage does not occur and we might expect less stringent selection for reciprocal herkogamy and perhaps smaller herkogamy distances.
Th eoretical models of the evolution of distyly include stigmaheight dimorphism as an intermediate stage in the transition from stylar monomorphism to distyly ( Charlesworth and Charlesworth, 1979 ;Lloyd and Webb, 1992b ).Comparative evidence in Narcissus and Lithodora involving phylogenetic reconstructions of the evolutionary history of stylar polymorphisms generally support these models ( Graham and Barrett, 2004 ;Pérez-Barrales et al., 2006 ;Ferrero et al., 2009 ).However, stigma-height dimorphism is clearly a stable polymorphism in each of these taxa because it is reported from a signifi cant number of species.Moreover, at least in Narcissus , there is experimental evidence that despite incomplete sexorgan reciprocity and intramorph compatibility, disassortative mating is promoted by stigma-height dimorphism ( Simón-Porcar et al., 2014, 2015a ), and the stylar morphs are subject to frequencydependent selection ( Th ompson et al., 2003 ) in the same manner as in distylous populations.
Here, we investigate the reproductive correlates of variation in style morph frequencies in populations of 11 dimorphic taxa formerly belonging to Lithodora (recently split into Lithodora and Glandora ; Th omas et al., 2008 ;Ferrero et al., 2009 ).We chose this group for several reasons.First, our earlier investigations revealed considerable variation in both sex-organ reciprocity and compatibility systems, including species with classical distyly with heteromorphic incompatibility, distyly with self and intramorph compatibility and, fi nally, stigma-height dimorphism with self-incompatibility and intramorph compatibility ( Ferrero et al., 2009( Ferrero et al., , 2011a( Ferrero et al., , 2012 ) ).To our knowledge, this range of reproductive variation surpasses that found in any dimorphic group of closely related species.Second, our preliminary observations of natural populations indicated a wide range of style-morph ratios raising the possibility that their contrasting reproductive systems may play a role in causing this variation.
Our study addressed three main questions: (1) Are the patterns of morph frequency variation in populations of Lithodora and Glandora associated with the type of stylar polymorphism of each species?We predicted greater deviation from isoplethy in taxa with stigma-height dimorphism compared to those that were distylous.
(2) Can quantitative measures of the degree of sex-organ reciprocity predict variation in morph ratios?We predicted that greater reciprocity between the style morphs should increase rates of disassortative pollen transfer and mating leading to more even style morph ratios in populations.(3) Are diff erences between the style morphs in herkogamy (spatial separation of stigmas and anthers) associated with biased morph ratios?Because morph-specifi c differences in herkogamy have the potential to infl uence selfi ng and assortative mating in stylar dimorphic populations (see Baker et al., 2000a ), we predicted increasing morph-ratio bias where diff erences in herkogamy between the morphs were most evident.To address these questions, we surveyed morph frequencies in 66 populations of seven taxa of Glandora and 39 populations of four taxa Lithodora over much of their geographical range.In each population, we sampled fl owers for morphological characterization of each style morph.We tested our predictions using phylogenetic methods to take into account the nonindependence of data and we also analyzed each species independently.

MATERIALS AND METHODS
Study group -Lithodora and Glandora are primarily composed of small, insect-pollinated shrubs, distributed around the Mediterranean basin ( Th omas et al., 2008 ).Glandora consists of six species, fi ve occurring in the western Mediterranean region, which is the main center of diversity.Lithodora consists of three species; one distributed in the western Mediterranean ( L. fruticosa ) and the remaining two equally distributed between the central and eastern Mediterranean.Several subspecies are recognized: two in G. prostrata and three in L. hispidula .Some taxa are narrow endemics: G. moroccana in central Rif (N.Africa), G. nitida in the southeastern part of the Iberian Peninsula ( Blanca et al., 2003 ), G. oleifolia in the eastern Pyrenees, and L. zahnii in southern Peloponnese ( Fig. 1 ).We were not able to sample L. hispidula subsp.cyrenaica, which is endemic to Libya.Most species fl ower in April-May, although some have longer fl owering periods from January-June (e.g., G. prostrata ).Th e two genera exhibit a range of compatibility systems including fully self-compatible taxa to those that are self-incompatible, including some with typical dimorphic incompatibility and others that are intramorph compatible ( Ferrero et al., 2012 ).Associated with this variation are diff erent types of fl oral polymorphism: distyly, stigma-height dimorphism, and "relaxed stigma-height dimorphism" (see Ferrero et al., 2011a , b ).In the latter condition, each anther within a fl ower occurs at a diff erent height.Insect visitors to fl owers of both genera include solitary bees (mainly Anthophora ), species of Lepidoptera (principally Macroglossum stellatarum ), and Diptera (mostly Bombylius ), and these visitors have been shown to diff er in pollination effi ciency and the range of species they visit ( Ferrero et al., 2011b ).

Surveys of style-morph ratios -
We surveyed 2-31 populations from four Lithodora taxa and 4-22 populations of seven Glandora taxa throughout the Mediterranean basin and the Atlantic coast of southwestern Europe (see Table 1 for taxa and sample sizes).Our sampling covered most of the distributional range of the taxa included in our survey ( Fig. 1 ).In each population, we randomly sampled one fl ower per plant from 100 individuals (where possible), and in smaller populations we sampled all individuals.Because taxa of Lithodora and Glandora do not appear to propagate by clonal reproduction, the sampling of genets was relatively unambiguous.We preserved fl owers in 70% ethanol for later classification on the basis of style length and measurements of sex-organ reciprocity.In each population, we calculated an index of style morph bias as the absolute value of the diff erence in the number of individuals of the L-and S-morphs, divided by the total number of fl owering individuals sampled (see Endels et al., 2002 ;Brys et al., 2008a ).Th e index identifi es deviations from isoplethy, with values ranging from 0, when both style morphs are present at equal frequency, to 1, when there is only one style morph present in the population.
We used G -tests to determine whether the style-morph ratios of individual populations diff ered signifi cantly from isoplethy and used pooled goodness-of-fi t G -tests to determine whether pooled morph ratios for individual taxa diff ered signifi cantly from isoplethy.G-heterogeneity statistics were calculated to test for heterogeneous morph ratios among populations of taxa.

Sex-organ reciprocity and herkogamy -
In the laboratory, we measured stigma and anther heights of the preserved fl owers using digital photography and the image analysis soft ware analySIS (version 5.0.), following procedures detailed in Ferrero et al. (2009Ferrero et al. ( , 2011a ) ).We then applied the method of Sánchez et al. (2008Sánchez et al. ( , 2013 ) ) to calculate an index of reciprocity ( R ) between complementary sex organs in each population.Th is index is based on comparing the position of all anthers of each fl ower with the stigmas of all fl owers of the opposite morph in a population.When reciprocity between anthers and stigmas is highest, the value of the index approaches one.Values that depart from one toward zero occur when reciprocity is low, and are modulated by the average standard deviation of all height measurements for each sex-organ level, so the greater the dispersion of values the greater departure from one.Th e computational soft ware is licensed by Creative Commons Attribution 3.0 (see Sánchez et al., 2008Sánchez et al., , 2013)).Herkogamy was calculated as the difference between the means of stigma and anther heights for each morph in all populations ( Fig. 2 ).We averaged anther heights within a fl ower in these calculations.We used two-way ANOVA to compare the degree of herkogamy between morphs and species.
Inference of the time-calibrated phylogenetic tree -We estimated a time-calibrated phylogeny of taxa of Boraginales from DNA sequences downloaded from GenBank (Appendix S1, see the online Supplementary Data tab with this article).Th is was undertaken to test our hypotheses at the species level while accounting for the fact that lineages are not independent ( Felsenstein, 1985 ).Most taxa in our trees belonged to Boraginaceae.Outgroup taxa consisted of members of Hydrophyllaceae, Heliotropiaceae, Cordiaceae, and Ehretiaceae ( Mansion et al., 2009 ).Th e six DNA regions used to infer the trees were: trn L intron, trn L -trn F intergenic spacer, ndh F gene, the rbc L gene, part of the trn K intron, and the mat K gene (both coded as the mat K region).Th ese analyses are described in detail in Appendix S2 and information on node calibration is specifi ed in Appendix S3. 2 ) and the type of stylar polymorphism ( Table 1 ) at the species (or subspecies) level.For this, we characterized the taxa into two groups based on their degree of reciprocity: distylous populations, which generally display greater levels of reciprocity, and nondistylous populations, including those with stigma-height dimorphism and relaxed stigmaheight dimorphism, which exhibit weaker reciprocity ( Ferrero et al., 2011a ).Th en, we calculated a mean morph bias per taxon.We fi tted a phylogenetic-corrected Linear Model using Generalized Least Squares: function 'gls' in the R package 'nlme' ( R Development Core Team, 2014 ).We used the index of style morph bias as the dependent variable and the type of polymorphism (distylous vs. nondistylous) as the independent variable.Th e correlation structure of the data was given by the expected covariance of taxon traits, given the phylogenetic tree and evolutionary model.Th e phylogeny of Boraginales containing the taxa of Glandora and Lithodora was pruned accordingly.We performed three analyses, one in which the phylogenetic signal (Pagel's λ ) was estimated by maximum like-lihood, and another two in which λ was forced to 0 (i.e., there is no infl uence of the evolutionary history of the species on the relationship between traits) and 1 (i.e., the phylogenetic signal is maximum, and the relation between variables has evolved according to a Brownian motion model of evolution).Th e fi t of the models was assessed using the Akaike weights (AICw) ( Burnham and Anderson, 2002 ).

Morphological traits associated with style-morph ratios -We fi rst examined the relation between style morph bias ( Table
Th e AICw of any particular model varies from zero (no support) to unity (complete support) relative to the entire set of models ( Johnson and Omland, 2004 ).Second, we tested whether the variability in morphological traits among population was related to deviations from isoplethy.If reciprocity and the degree of herkogamy play an important functional role in governing disassortative mating ( Fig. 2 ) and thus promoting isoplethy, we would predict greater variation in morphological traits among populations associated with greater deviation from isoplethy.For this reason, we used the interquartile range (IQR) to standardize measurements of data dispersion allowing comparisons among species.For each taxon, we calculated the IQR of the morph-bias index, the IQR of reciprocity and the herkogamy distance, and we tested for the eff ect of the morphological measurements on the IQR of the morph-bias index.We used the function 'gls' in the R package 'nlme', following the same procedure as for the previous analysis.
Finally, we analyzed the relation between the morph-bias index and the reciprocity and herkogamy distances independently for each taxon.We carried out this analysis only in those taxa with more than four populations sampled and used General Linear Models.We constructed models for diff erent subsets of factors and used the Akaike weights (AICw) to determine which candidate model best Reproductive traits of Glandora and Lithodora species.N p and N f refer to the number of populations and fl owers sampled, respectively, SC: selfcompatible, SI: self-incompatible, IMC: intramorph compatible, IMI: intramorph incompatible.Mean and standard deviations are provided for corolla size, reciprocity index ( R ), and herkogamy distance for the L-and S-morph.?= absence of information.Measurements are in millimeters.

Species
N p N f

Range of distribution
Corolla size R  explained the data.In all analyses, we used 0.05 as the signifi cance level.
Sex-organ deployment -A total of 105 populations of seven taxa of Glandora and four of Lithodora were sampled.For each taxon, mean values for corolla length, herkogamy, and reciprocity are summarized in Table 1 .In these taxa, style length ranged between 7.3-14.6mm in the L-morph and 3.4-8.2mm in the S-morph; whereas anther height ranged between 5.2-10.7 mm in the L-morph and 5.6-13.6 mm in the S-morph.Mean values for corolla length and sex-organ position are summarized in Fig. 5 , and detailed information on measurements for each population is presented in Appendix S5.Reciprocity values varied from high reciprocity in populations of G. moroccana (0.77 ± 0.02) to low reciprocity in L. fruticosa (0.25 ± 0.22; Table 1 ).Th ere were signifi cant diff erences in the degree of herkogamy between morphs ( F 1, 187 = 14.68,P < 0.001), species ( F 10, 187 = 52.39,P < 0.001), and their interaction ( F 10, 187 = 5.14, P < 0.001).Herkogamy distance in the L-morph was signifi cantly greater than the S-morph in four out of the eight taxa we investigated (see Tables 1, 2 and Fig. 5 ).Herkogamy in the L-morph varied from 2.55 ± 0.46 mm in populations of L. fruticosa , to 6.29 ± 0.81 mm in G. oleifolia , whereas herkogamy in the S-morph varied from 2.32 ± 0.34 mm in popula-tions of L. hispidula subsp.versicolor , to 6.42 ± 0.35 mm in G. oleifolia ( Table 1 ).

Relations between morph ratios and fl oral traits -
We compared the mean morph-bias index per taxon between distylous and nondistylous taxa (stigma-height dimorphic and relaxed style dimorphic).As predicted, distylous taxa were signifi cantly more isoplethic than nondistylous species, which oft en had biased style-morph ratios ( t = −4.6,df = 10, P < 0.001).Th ere was evidence that phylogenetic relationships among taxa contributed toward morph-ratio variation, because the value of λ in our analysis was 1, indicating that the mean values for morph-bias index of related taxa were more similar to each other than random values drawn from the same tree.With regard to the relations between variation in deviations from isoplethy and sex-organ deployment at the species level, we found no signifi cant associations between the IQR of morphbias index (IQR b ) and the IQR of reciprocity (IQR r ), L-morph herkogamy (IQR L ), or S-morph herkogamy (IQR S ) (Appendix S6).Note that because we found diff erences in the herkogamy distance between morphs, we considered the herkogamy distance in the L-morph (IQR L ) and the S-morph (IQR S ) as independent factors in these analyses.Th e phylogenetic signal in the analysis had a value of λ = 0, indicating that values of IQR b among closely related taxa were no more similar than among less-related taxa (Appendix S6).

DISCUSSION
In this study, we investigated the relationships between variation in the position of female and male sex organs and population-style  morph ratios in a lineage of heteromorphic plants, which vary in stylar polymorphisms and incompatibility-compatibility systems.
We examined this association independently in 11 taxa of Glandora and Lithodora , but also in a phylogenetic context because it has been shown that heteromorphic traits show numerous transitions during the evolutionary history of this lineage ( Ferrero et al., 2009 ).As predicted, our results demonstrated that across all species, the type of sex-organ deployment found in a particular taxon was associated with the amount of deviation from isoplethy.Distylous taxa collectively exhibited less deviation than those with stigma-height dimorphism and relaxed stigma-height dimorphism.However, within taxa there was little evidence for an association between intraspecifi c variation in sex-organ position and morph ratios.Only in L. fruticosa, a species with stigmaheight dimorphism and weak reciprocity between stigma and anther heights, was there evidence that variation in sex-organ position may infl uence morph-ratio variation.Our discussion considers the reproductive, ecological, and genetic factors that may help to explain the striking variation in style-morph ratios exhibited by this unusual lineage of heteromorphic plants.

Causes of the general association between type of stylar polymorphism and morph-ra-
tio bias -Diverse factors infl uence the relative frequency of the L-and S-morphs in dimorphic species of angiosperms.These include stochastic forces associated with finite population size, and deterministic forces resulting from diff erences among the style morphs in fertility and mating patterns.Although we did not estimate population size quantitatively in our survey, we grouped distylous and nondistylous populations into three classes (small, medium, and large, Appendix S4), but found no statistically significant relation between population size and morph-ratio variation (Generalized Linear model: F 2, 99 = 0.8, P = 0.465) or in the interaction: Size × Type of polymorphism ( F 2, 99 = 1.3, P = 0.271).However, there was a signifi cant relation between the type of polymorphism and morph-ratio variation ( F 1, 99 = 11.4,P = 0.001), confi rming our results at the species level.Population size undoubtedly infl uenced variation in morph ratios in some of the very small populations in our sample; however, it does not appear to have played a signifi cant role in causing the overall diff erence between distylous and nondistylous species in morphratio variation.Prolifi c clonal propagation or diff erences in the fertility or strength of incompatibility systems between morphs are known to aff ect their frequency in populations of some heteromorphic plants (e.g., Th ompson et al., 2003 ;Wang et al., 2005 ;Brys et al., 2008b ;Castro et al., 2013 ), but in the case of Lithodora and Glandora , the species appear to be nonclonal and we have no evidence for consistent fertility diff erences or variation in the type and strength of incompatibility between the style morphs within species.
We found equal proportions of the style morphs in 76% of the 105 populations that we surveyed in this study.In the more widespread taxa that were sampled more extensively, the opportunities for detecting bias increased because of the number of populations sampled.The most plausible hypothesis to explain the observed diff erence between the two major classes of stylar polymorphisms in the relation between sex-organ reciprocity and morph-ratio bias concerns the aggregate patterns of mating in populations.In those with a high level of disassortative mating, equal morph ratios are expected as a result of negative frequencydependent selection.In species with a dimorphic incompatibility, as in Glandora nitida, this mating pattern is guaranteed and all three populations sampled were isoplethic.However, for the remaining three distylous taxa in our survey for which the compatibility system has been determined experimentally ( Ferrero et al., 2012 ), none possess dimorphic incompatibility and all are self-and intramorph compatible.In these species sex-organ reciprocity is the primary mechanism promoting disassortative mating and indeed there is evidence that this may occur.Glandora moroccana and G. diff usa are fully self-and intramorph compatible with high sex-organ reciprocity ( Ferrero et al., 2011a( Ferrero et al., , 2012 ; see Fig. 5 ) and all populations sampled were isoplethic ( Fig. 3G ).Elsewhere among heteromorphic plants, high levels of disassortative mating resulting in equal morph ratios have been reported in self-compatible, tristylous Eichhornia paniculata ( Barrett et al., 1987 ;Kohn and Barrett, 1992 ) and in stylar dimorphic Narcissus papyraceus ( Simón-Porcar et al., 2014, 2015a ).We observed abundant pollinator visitation by long-tongued bees in distylous Glandora and Lithodora populations ( Ferrero et al., 2011b ) and these insects probably play a key role in promoting disassortative mating and causing isoplethic morph ratios.
In this study, we distinguished species of Glandora and Lithodora with stigma-height dimorphism and relaxed stigma-height dimorphism (sensu Ferrero et al., 2009 ) from those exhibiting distyly.We made this distinction because angiosperm species with these polymorphisms generally lack strong sex-organ reciprocity (reviewed in Barrett et al., 2000, and see Sánchez et al., 2008, 2013 ).In species with stigma-height dimorphism, anther levels are positioned similarly in the fl oral tubes of the L-and S-morphs, whereas with relaxed stigma-height dimorphism, they are scattered in their distribution within the corolla and show little reciprocity.Our measures of reciprocity ( R , see Sánchez et al., 2008 ) generally supported the expectation that stigma-height dimorphic taxa exhibit less reciprocity of sex-organ positioning compared to those that were distylous ( Table 1 ).Values of R ranged from 0.77-0.58for distylous species and from 0.58-0.25 for stigma-height dimorphic species, including the relaxed form.
In three of the four species with stigmaheight dimorphism, the style morphs, although self-incompatible, are intramorph compatible allowing assortative mating in populations.Th is type of incompatibility system has been reported in other taxa of fl owering plants with stigmaheight dimorphism and can cause asymmetrical mating and biased morph ratios ( Barrett and Cruzan, 1994 ;Baker et al., 2000a ;Arroyo et al., 2002 ;Barrett and Hodgins, 2006 ).However, the extent to which the style morphs diff er in levels of assortative mating depends on a variety of factors including the size of populations and their pollination biology, and the particular confi gurations of sex organs in the style morphs.For example, in stigma dimorphic Narcissus assoanus , both L-morph biased and isoplethic populations are reported in southern France and these differ in size and the quantity and quality of pollinator service ( Baker et al., 2000a , c ). Diff erences in the types of pollinators visiting populations of other stigma-height dimorphic Narcissus species have also been implicated in causing variation in morph ratios ( Arroyo and Dafni, 1995 ;Simón-Porcar et al., 2014 ).A theoretical model of pollination and mating in stigma-height dimorphic populations, based on empirical data on the fl oral biology of Narcissus species, demonstrated that greater assortative mating in the L-morph, because of the reduced herkogamy of this morph compared to the S-morph, could explain the L-biased populations that are most commonly encountered in this genus ( Baker et al., 2000a ).However, in Glandora and Lithodora , the degree of herkogamy does not differ between the style morphs in a consistent way ( Table 1 ).Th us, it is unclear the extent to which herkogamy generally infl uences mating patterns and morph ratios in a consistent way across species.
In our survey of stigma dimorphic species, a wide range of morph ratios were encountered but overall, as expected, they exhibited a greater tendency to depart from isoplethy compared to distylous species.We found some exceptions to the general association between weak reciprocity and biased morph ratios.For example, G. prostrata subsp.prostrata possesses relaxed stigma-height dimorphism, intramorph compatibility, and a low reciprocity value of R = 0.26.Nevertheless, of the 31 populations we surveyed, 26 exhibited equal morph ratios and the remaining 5 were L-biased.In closely related G. prostrata subsp.lusitanica , which also has the same characteristics, among 10 populations sampled, eight were isoplethic and the remaining two were S-biased.It is unclear what reproductive mechanisms operate in populations of this species to promote disassortative pollination and why the direction of bias diff ers between the two subspecies.

Association between sex-organ positions and morph ratios within
species -Our study of Glandora and Lithodora taxa was designed to investigate general patterns of morph-ratio variation among dif-ferent classes of stylar polymorphism.As a result, the power to detect signifi cant associations between sex-organ position and morph ratios was weak in many of the taxa.Th is was particularly the case for rare species, because of the limited number of populations that could be sampled.Indeed, in seven of the 11 taxa in our survey, less than 10 populations were sampled.However, where larger samples were obtained, several patterns were revealed-the most interesting of which involved L. fruticosa .
Among the 22 populations of L. fruticosa , a negative relation between the degree of herkogamy of the S-morph and deviations from isoplethy was evident ( Fig. 6 ).As the distance between stigmas and anthers of the S-morph became smaller in populations, the frequency of this morph also declined.In contrast, there was no relation between the herkogamy of the L-morph and morph ratios (Appendix S7).Two potential mechanisms could explain this pattern.Th e reduction in frequency of the S-morph could arise if intramorph mating in this morph caused morph-specifi c inbreeding depression resulting from sheltered load at the S -locus ( Strobeck, 1980 ;Barrett et al., 1989 ).A second possibility is that the smaller distance separating female and male sex organs within a fl ower may result in self-interference ( Barrett, 2002 ;Cesaro et al., 2004 ;Valois-Cuesta et al., 2011 ).In self-incompatible species like L. fruticosa , self-interference may reduce female fi tness because stamens physically impede pollen deposition on stigmas ( Barrett and Glover,FIGURE 5 Variation in corolla length and sex-organ height (mean ± SD in millimeters) in fl owers of the L-and S-morph in Glandora (seven species) and Lithodora (four species).The number of populations per species that were used for calculations is summarized in Table 1 .
Integration of phylogenetic and species level approaches -In this study, we used a phylogenetic approach in our statistical analyses.Th is was necessary for our comparative study because the two taxa we investigated were previously considered a single genus, although they are clearly diff erentiated based on phylogenetic evidence ( Th omas et al., 2008 ;Ferrero et al., 2009 ).Th e primary rationale for the use of phylogenetically based statistical methods was the expected nonindependence among species trait values due to their phylogenetic relatedness ( Felsenstein, 1985 ). Th e branching pattern and estimated times of divergence in our Bayesian analysis based on six DNA regions was largely consistent with previous work, although here we obtained stronger support for most clades ( Ferrero et al., 2009 ;Mansion et al., 2009 ).We found a phylogenetic signal when analyzing the infl uence of the type of polymorphism on morph-ratio bias.However, our analyses at the species (and subspecies) level revealed that variation among populations in sex-organ deployment was largely unrelated to morph-ratio variation.Considerable eff ort involving a large number of populations sampled over a broad geographical range is oft en necessary to detect significant associations between functional traits and style-morph ratios (e.g., Barrett et al., 2004 ).However, where this is not possible, comparative studies conducted within a phylogenetic framework can expose novel associations between morphological characters and style-morph ratios providing opportunities for future hypothesis testing at the population level.

FIGURE 2
FIGURE 2Expected infl uence of sex-organ deployment on pollen transfer, mating patterns, and morph ratios in species with contrasting stylar polymorphism.(A) distyly; (B) stigma-height dimorphism.Equality in the proportion of morphs (isoplethy) is expected to be maintained when:(1) sex-organ reciprocity favors disassortative pollen transfer and; (2) decreased herkogamy increases assortative pollen transfer and selfi ng.

FIGURE 4
FIGURE 4 Geographical distribution of style-morph ratios in species of Lithodora throughout its range in the Mediterranean basin.The frequency of the L-morph (white) and S-morph (black) are indicated for each population.Image on the top shows the localization of the detailed areas.(A) L. fruticosa , N = 22 populations; (B) L. hispidula subsp.versicolor , N = 6 populations; (C) L. hispidula subsp.hispidula, N = 7 populations; and (D) L. zahnii , N = 4 populations.

FIGURE 6
FIGURE 6 Plot of response surface illustrating the eff ect of reciprocity, and herkogamy in the S-morph, on the morph ratio of populations of Lithodora fruticosa .The morph-bias index was calculated as the absolute value of the diff erence in the number of individuals of the L-and Smorph, divided by the total number of fl owering individuals sampled.Reciprocity index was calculated following Sánchez et al. (2013) .

TABLE 2 .
Morph frequency data collected from surveys of populations of Glandora and Lithodora .Average morph ratios per species and the number of isoplethic and anisoplethic populations (L-biased or S-biased) are shown for each species.Sample sizes for populations and fl owers are given in Table1; L = L-morph; S = S-morph.Signifi cant results for departure from isoplethy after pooled and heterogeneity G -tests at α = 0.05 are highlighted in bold.