The analyses performed in the present study, coupled with previous and complementary results (Pérez et al., 2004), provide significant support for the Darwinian hypothesis of the determinant role played by pollinators in flower evolution. In particular, a quantitative model, as reformulated by Lloyd & Webb (1992b), is supported for Narcissus considering the three main components of the model in a phylogenetic context: style polymorphism, breeding systems and pollinator efficiency. Previous studies which focused on style polymorphism and perianth morphology to infer pollinator type suggested that approach herkogamous monomorphism precedes style dimorphism in the course of evolution in Narcissus, followed by distyly and/or tristyly (Graham & Barrett, 2004; Pérez et al., 2004). Our results are congruent with this sequence of style evolution, providing further support for the idea that style dimorphism is ancestral to heterostyly, and that the evolutionary transition is more closely related to changes in pollinators than breeding system evolution.
Self-incompatibility in Narcissus
Barriers to selfing as revealed by our experiments in Narcissus sect. Apodanthi agree, to a great extent, with the incompatibility system reported for other Narcissus species (Bateman, 1954; Dulberger, 1964; Barrett et al., 1997; Sage et al., 1999; Baker et al., 2000b; Arroyo et al., 2002). Narcissus albimarginatus, N. cuatrecasasii, N. calcicola, N. scaberulus and N. marvieri are primarily self-sterile, in congruence with no link between the incompatibility system and style polymorphism (Table 4); crosses within and between morphs set equivalent number of seeds. All self-incompatible species had values of self-compatibility index much lower than 0.75, whereas self-compatible species had higher values, as expected according to this cut-off rule proposed by Lloyd & Schoen (1992) to discriminate between self-compatible and self-incompatible species. However, different seed-set success was retrieved in some intra- and intermorph crossings without any particular direction (Fig. 1). Morph-inconstant skewness is probably unrelated to morph-specific differences in incompatibility (but see Ornduff, 1988) and probably reflects some uncontrolled effects during sampling or experimental pollination, for example biparental inbreeding as a result of genetic substructuring of source populations, especially those that are small (cf. Hodgings & Barrett, 2006), or differential dichogamy between morphs, as reported for N. assoanus (Cesaro et al., 2004). Self-sterility in these apparently self-incompatible species could also be achieved by the effects of inbreeding depression. Although the results reported in this study agree with those for a few self-incompatible Narcissus species investigated in detail (particularly N. tazetta by Dulberger, 1964, and N. triandrus by Sage et al., 1999), it would be desirable to carry out an appropriate crossing programme and a developmental study to confirm these findings.
Narcissus rupicola and N. watieri are predominantly self-compatible, as reported in a few other Narcissus species (Baker et al., 2000b; Medrano et al., 2005, but see also reviews in Bateman, 1954, and Barrett et al., 1996). Both species are Mediterranean mountain species distributed at the highest elevations (up to 2500 m a.s.l.) and occur in either the most northern (N. rupicola) or southern (N. watieri) parts of the geographic range of sect. Apodanthi, where weather conditions are harsh during their early blooming season and pollinator reliability may be low. Under such conditions, the loss of incompatibility may have been selected via reproductive assurance (Baker, 1955). By contrast, the closely related N. marvieri is mostly self-incompatible, at least for the L-morph. Further searches for populations large enough to provide a significant number of S-morph plants are needed to test whether both morphs are self-incompatible in this species, as in all other self-incompatible dimorphic Narcissus species tested to date (Barrett & Harder, 2005).
The results of insect observation satisfied the predictions of the theoretical model of Lloyd & Webb (1992a,b). In sect. Apodanthi, species with different flower morphology and stylar polymorphism (see Pérez et al., 2004) have different pollinator fauna (Table 4). Males of Anthophora sp. were the main flower visitors in the distylous N. albimarginatus (68% of all visits) and in the style-dimorphic N. cuatrecasasii (60% of visits), in which there is a relatively high degree of reciprocity of stigma and anther heights (Fig. 2), as hypothesized in previous work (Arroyo & Barrett, 2000; Graham & Barrett, 2004; Pérez et al., 2004).
Medium-proboscis bees play an important role in one more species. Data on pollinator fauna visiting N. marvieri (over 95% of anthophorid visits) are partial and refer to a peculiar, isolated population in Morocco (Jebel Tazekka), which has a mean flower-tube length (14.1 ± 0.2 mm) much shorter than that of the remaining populations in the Atlas Mountains (Djebel Zerekten, 24.0 ± 0.3 mm; Tizi-n-Ifar, 24.9 ± 0.2 mm). The flower morphology of typical populations of N. marvieri from the Atlas Mountains is very similar to that of long-tubed N. rupicola (see next paragraph); in fact, some authors considered the former as a mere subspecies of the latter (Maire, 1959). We do not expect a pollinator fauna dominated by bees in long-tubed populations of N. marvieri, a prediction to be tested in further research on the possible effect of change of pollinators on flower morphology at the within-species level (e.g. Arroyo & Dafni, 1995).
Moths and butterflies were the main flower visitors in N. rupicola and N. watieri. These long-tongued insects are nectar-feeders and contributed in both cases to > 60% of the visits. In fact, the maximum flower tube lengths of these species (N. rupicola: 23.0 mm; N. watieri: 27.2 mm) (see Pérez et al., 2004) fit well with the mean proboscis length of one of their most frequent visitors (Macroglossum stellatarum: 27.6 mm; V. González, pers. comm.). Short-proboscis insects (mostly syrphids) contributed to the remaining insect visits, especially in N. watieri. These insects only collect pollen from the upper-anther whorl as the floral tube is too long and narrow for them to access nectar at the base of the tube.
Our data on nocturnal insects are partial and not directly comparable with diurnal censuses. The number of insects captured was very low and only a small fraction of insects carried pollen, which means that they are probably much lesser important visitors than diurnal insects (but see Vogel & Muller-Doblies, 1975, and Pérez-Barrales, 2005 for other Narcissus species). They were moths (Noctuidae, Geometridae and Depressaridae) and Trichoptera. It is worth noting the presence of Trichoptera carrying Narcissus pollen in one Moroccan (N. marvieri) and one Spanish (N. cuatrecasasii) species. This has rarely been documented (Petterson & Hasselrot, 1994; D. Inouye, pers. comm.) and their relative proficiency as pollinators remains an open question.
In general, bees are more efficient than lepidopterans in terms of amount of pollen grains transferred (Herrera, 1987; Hiei & Suzuki, 2001). The bees we captured usually transported more than 50 pollen grains, while lepidopterans and syrphids carried fewer than 50 grains. However, it is also necessary to know where an insect carries pollen and its accuracy when touching sexual whorls, to assess its role as an actual pollinator (Armbruster et al., 1994, 2006). Most captured bees were males transporting pollen on all their body parts, whereas the other pollinators transported pollen on a narrower range of their body parts. Lepidopterans only introduce the proboscis to seek nectar and syrphids only feed on pollen from the upper anthers. Heterostylous species need pollinators to carry pollen of different morphs on different body parts in order to promote disassortative pollination (see Lloyd & Webb, 1992a, and references therein). Among all pollinators reported in this study, bees have the highest potential to transfer pollen between morphs. However, the lack of pollen dimorphism in Narcissus precludes direct estimation of the pattern of pollen transfer. It would be beneficial to carry out experimental studies, for example using different fluorescent dyes to track pollen movement, emasculations in controlled arrays (Cesaro & Thompson, 2004) or molecular markers. We hypothesize that pollen carried by bees on different body parts comes from stamens at different heights in the distylous N. albimarginatus and the close-to-reciprocal N. cuatrecasasii (Pérez et al., 2004). Whereas intermorph pollen transfer cannot by achieved by syrphids, the role of Lepidoptera remains more complicated (Stone, 1996).
Testing the Darwinian hypothesis of heterostyly
The present study brings further evidence to bear on Lloyd & Webb's (1992a,b) predictions of the importance of pollinators, as Darwin (1877) first suggested, for the evolution of heterostyly. Differences in the pollinator composition of species with different floral polymorphisms and the lack of association between style polymorphisms and incompatibility systems are the key elements of our results here. Style polymorphism is present in both self-compatible and self-incompatible species (Table 4). Accordingly, the incompatibility systems appear to be unlinked to genes responsible for style polymorphism (reviewed in Barrett & Harder, 2005). Taking the most reliable phylogeny into consideration (Fig. 4), it is plausible that self-incompatibility is the ancestral state, a situation that has been maintained in the three lineages of Narcissus sect. Apodanthi, irrespective of evolution of style polymorphism. Lack of correspondence between the evolutionary histories of the incompatibility system and style polymorphism (Fig. 5a) supports the view that the incompatibility is not critical for evolution of stylar polymorphisms, both for heterostyly (N. albimarginatus) and for style dimorphism (N. cuatrecasasii, N. scaberulus and N. calcicola). Joint occurrence of stylar dimorphism and extensive self-compatibility (N. rupicola) provides additional and strong support that incompatibility is not necessary to maintain style polymorphism. Loss of self-incompatibility is clearly derived from ancestral self-incompatible lineages in N. rupicola.
One unexpected result was obtained from the evolutionary reconstruction of the breeding system. We detected one possible reversion event to achieve incompatibility in the course of speciation of N. marvieri, the sister species to the self-compatible N. watieri (suggested from maximizing reversal events with MacClade–acctran). Alternatively, we retrieved two reversals to compatibility as maximizing parallelism (deltran) (results not shown). This situation is puzzling, because it is fairly widely accepted that the evolution of selfing is quasi-irreversible (see Takebayashi & Morrell, 2001 for a review, and Igic et al., 2004, 2006 for a well-documented case, but see also Routley et al., 2004). Joint loss of heterostyly and self-incompatibility has been frequently reported (e.g. Kohn et al., 1996; Schoen et al., 1997; Truyens et al., 2005) as in all these cases style polymorphism and self-incompatibility are linked. If the loss of self-incompatibility is caused by modifiers unlinked to the S-locus, a reversion to self-incompatibility is possible, but is not believed to have occurred in either at population or phylogenetic levels (e.g. Solanaceae; Stone, 2002; Igic et al., 2006). An alternative explanation is that of population heterogeneity of breeding systems and style polymorphisms in the clade N. marvieri-N. watieri-N. rupicola, which could include both self-compatible and self-incompatible populations (see Busch, 2005). In this scenario, the ancestral condition to this clade would be self-incompatibility and style dimorphism, and loss of both conditions would have occurred only in some populations within each species. A more extensive population sample would be needed to investigate whether these three species do have self-incompatible and compatible populations, perhaps related to different degrees of herkogamy and dichogamy (Cesaro et al., 2004; Routley et al., 2004), and both style-dimorphic and monomorphic populations as reported in other Narcissus species (Arroyo & Dafni, 1995; Arroyo et al., 2002; Baker et al., 2000a).
If the evolution of incompatibility systems is not critical for maintaining style polymorphism, what is the evolutionary force underlying it? According to Lloyd & Webb (1992b), when the heteromorphic incompatibility system is not present, only higher levels of disassortative than assortative pollen transfer can maintain style dimorphism. Disassortative transfer is higher as the morph ratios approach equality (isoplethy). This hypothesis has positively been tested experimentally in style-dimorphic N. assoanus (Cesaro & Thompson, 2004). There are two main factors determining disassortative pollen transfer: reciprocity of anthers and stigma, and pollinator behaviour. Pérez et al. (2004) analysed the former in sect. Apodanthi, and we have focused in this paper on the latter. The most reciprocal species, distylous N. albimarginatus and style-dimorphic N. cuatrecasasii (Pérez et al., 2004), have the same pollinators, bees, as the other heterostylous group in the genus (N. triandrus complex; C. Gomez, unpublished data), and these are very different from other species in sect. Apodanthi. A particularly puzzling feature of our results is that both self-compatible species showed contrasting nearly isoplethic style dimorphism (N. rupicola; 38–53% of S-plants) and monomorphism (N. watieri), but had similar pollinators (long- and short-proboscis insects). Pollinator information for N. rupicola is reliable in terms of population number, time effort and nocturnal captures and diurnal observations, but that for N. watieri is limited. It has been argued that long-tongued insects (i.e. Lepidoptera) may be responsible for maintenance of style-height dimorphism (Cesaro & Thompson, 2004). In contrast, short-tongued, pollen-feeding insects (mostly syrphids) may favour intramorph pollination of the L-morph, and virtually no pollination of the S-morph, thus producing morph ratio bias and possible fixation of the former in morph-compatible species (Arroyo & Dafni, 1995; Pérez-Barrales, 2005). Perhaps the relative numbers of short- and long-tongued insects in N. watieri allow this process to operate by syrphid activity, aided by self-compatibility (a syrphid may allow easy self-pollination of L-flowers) and a flower that is attractive to syrphids: white flat tepals with a yellow central dot (upper stamens) (Dinkel & Lunau, 2001). Again, more research in different populations of these two critical species to ascertain breeding systems, pollinators and pollen transfer patterns throughout their ranges would provide critical information on the maintenance and dissolution of style dimorphism.
Lloyd & Webb (1992a, p. 174) specifically stated that ‘all attempts to date to uncover the evolutionary origins of heterostyly have been hindered by ignorance of the exact phylogenetic relationships among the species in taxa with heterostylous representatives’. Our results using both phylogenetic information and data on breeding systems and pollinators support the idea that the evolution of flower polymorphism in Narcissus is mediated by the pollinator fauna. Our finding that there is a disassociation between breeding systems and style polymorphism in Narcissus suggests that more work on the role of pollinator shifts for the evolution of flower polymorphisms is warranted. Additionally, the common joint presence of reciprocal herkogamy and heteromorphic self-incompatibility in most heterostylous groups remains an unresolved problem in the Lloyd & Webb (1992a,b) model, which requires further investigation in a phylogenetic context; for example, do ancestral taxa consistently suffer from some inbreeding which promotes self-incompatibility?
This study supports the idea of divergent natural selection in Narcissus mediated by the pollinator fauna. The consistency of the association between style polymorphism and pollinators in Narcissus and other taxa will test the generality of Darwin's predictions about the role played by pollinator variation in flower evolution.