Contrasting pollination syndromes in closely related species suggest that floral trait divergence is associated with differences in pollination system, but empirical observations are required to confirm syndrome-based predictions. We present a comparative study of two closely related Erica species with contrasting pollination syndromes from the Cape Floristic Region of South Africa. Erica cylindrica has narrowly tubular pale and strongly scented flowers and is known to be hawkmoth-pollinated. The closely related Erica infundibuliformis has bright flower colours and appears to lack scent, traits that are suggestive of pollination by long-tongued nemestrinid flies (rhinomyiophily). Floral trait measurements revealed that both species exhibit predominantly upright flower orientation and elongated floral tubes, although tube length of E. infundibuliformis is consistently greater than that of E. cylindrica. For both species, petals are brighter than floral tube surfaces, but flowers of E. cylindrica lack the strong UV reflectance found in E. infundibuliformis. Nectar of E. infundibuliformis is more concentrated and produced in larger volumes. Scent composition, but not evening scent emission rates, differed between the species: scent of E. cylindrica is dominated by aromatic compounds, whereas scent of E. infundibuliformis is dominated by (E)-ocimene and other terpenoid compounds and is emitted at higher rates during the day than the evening. Pollinator observations contradicted trait-based predictions: although a single nemestrinid fly captured in the vicinity of E. infundibuliformis did carry Erica pollen, almost all other diurnal flower visitors were nectar-robbing Hymenoptera which did not carry Erica pollen. Contrary to predictions, at two sites and over two flowering seasons, flowers were consistently visited in the evenings by several species of settling moths and hawkmoths which carried pollen, almost exclusively of Erica, on their proboscides. Our findings thus suggest that, despite objective differences in key floral traits between the closely related hawkmoth-pollinated E. cylindrica and E. infundibuliformis, moths are also important pollinators of E. infundibuliformis. A bimodal pollination system involving predominant pollination by moths and occasional visits by long-proboscid flies could partially reconcile findings with predictions. Our study further suggests that hawkmoth pollination may be more widespread in both Erica and the broader Cape flora than has hitherto been assumed and emphasises the importance of nocturnal pollinator observations.

Colour, Erica cylindrica, Erica infundibuliformis, flower orientation, hawkmoth, long-proboscid fly, moth-pollination, scent

There is strong evidence that pollinators have been important drivers of the radiation of angiosperms, especially in lineages in which interspecific variation in suites of floral traits is associated with variation in functional pollinator groups (e.g. Grant and Grant (1965); Johnson et al. (1998); Goldblatt and Manning (2006); Whittall and Hodges (2007)). Such covariation often leads to recognition of particular suites of floral traits as ‘pollination syndromes’ (Faegri and van der Pijl 1979; Fenster et al. 2004). Pollination syndromes can be used in a predictive framework: based on a subset of species for which both floral traits and pollination systems have been documented, pollinators can be predicted for those species lacking pollinator observations. This approach has been utilised successfully for pollination systems characterised by high degrees of specialisation (Johnson and Steiner 2000; Rosas-Guerrero et al. 2014; Dellinger 2020), both for particular pollination guilds (Pauw 2006, 2022; Valverde-Espinoza et al. 2021;) and across an entire flora (Johnson and Wester 2017). Confirmation of syndrome-based predictions strengthens the evidence for associations between particular floral traits and specific pollinator groups. This information, in turn, provides better understanding of the ecology and sensory perception of functional pollinator groups (Schiestl and Johnson 2013). However, instances in which empirical evidence contradicts predictions also advance our understanding of evolutionary ecology. Firstly, falsification of syndrome-based predictions has revealed that pollination systems in some plant species may not be as specialised as suggested by pollination syndrome theory (Ollerton 1996; Waser et al. 1996). Secondly, observations that contradict syndrome-based predictions have generated useful insight into the ecological context in which specialised interactions implied by syndromes break down (e.g. de Merxem et al. (2009)). Finally, mismatches between observations and syndrome-based predictions shed light on differences between human and pollinator sensory perception (Cozien et al. 2019; Castañeda-Zárate et al. 2021), emphasising the importance of objective approaches for quantification and interpretation of floral traits, especially those such as floral scent and colour for which human perception is particularly biased (Leonardos et al. 1969; Kevan et al. 1996). Therefore, empirical tests of predictions derived from pollination syndromes can provide useful insights into plant-pollinator interactions and function of floral traits.

In plant groups for which phylogenetic relationships have been reconstructed, syndrome-based predictions of pollination systems are particularly useful for investigating potential pollinator-driven divergence between closely-related species with contrasting pollination syndromes. Differences in floral syndromes suggest that divergence is potentially driven by adaptation to different pollinators (Johnson 2006; Van der Niet et al. 2014a). Such pollinator shifts often occur along ‘lines of least resistance’ (Stebbins 1970; Johnson et al. 1998), potentially involving only minor divergence in few key traits. Investigation of phenotypic differences between closely-related taxa that differ in pollination system can thus be used to identify which key traits underpin pollinator shifts (Goldblatt et al. 2004; Shuttleworth and Johnson 2009; Castañeda-Zárate et al. 2021). However, floral divergence may not necessarily be accompanied by pollinator shifts (e.g. Ellis and Johnson (2009)) and empirical pollination studies are, therefore, required to evaluate predictions.

The flora of the Cape Floristic Region is characterised by a high incidence of specialised pollination systems (Johnson and Steiner 2003; Johnson 2010) and strong predictability of pollination systems based on floral syndromes (Vogel 2012; Johnson and Wester 2017). The monophyletic “Cape clade” of the genus Erica is characterised by tremendous floral diversity (Baker and Oliver 1967; Schumann and Kirsten 1992; Vogel 2012). Rebelo et al. (1985) used floral traits to categorise flowers of 426 Erica species as consistent with four broadly defined pollination syndromes, distinguishing anemophilous, ornithophilous and entomophilous syndromes and coining the term “rhinomyiophily” to identify the suite of traits indicative of pollination by long-proboscid flies (Rebelo et al. 1985). Although some empirical studies have shown that the syndrome-based categorisations by Rebelo et al. (1985) are not consistently reliable (summarised in Van der Niet (2021)), predictions of pollination by long-proboscid flies (LPF) have been confirmed for several Erica species with a characteristic suite of traits including unscented, brightly coloured tubular flowers with a narrow orifice (Lombardi et al. 2021; Newman and Johnson 2021; Pauw 2022; McCarren et al. 2023).

The LPF syndrome of Erica flowers is consistent with that of LPF-pollinated species from other plant families (Goldblatt and Manning 2000). Further, narrowly tubular rhinomyiophilous Erica flowers restrict access to floral nectar for most other potential pollinators. It is, therefore, likely that predictions based on the LPF syndrome are reliable in Erica and that LPF syndrome traits in species which are closely related to species with different floral traits indicate a pollinator shift. Erica cylindrica, with narrow floral tubes and upward-facing, pale-coloured and strongly scented flowers, was recently discovered to be pollinated by hawkmoths, a pollination system that is hitherto unique in Erica and rare in the CFR (Van der Niet and Cozien 2022). The hawkmoth-pollinated E. cylindrica is part of a small clade of species that are all characterised by tubular, upward-facing flowers (Baker and Oliver 1967; Schumann and Kirsten 1992; Pirie et al. 2016); within this clade, Erica infundibuliformis is closely related to E. cylindrica (Pirie et al. 2016). Floral traits of the two species show strong similarities, with two key exceptions in colour and scent: flowers of E. infundibuliformis are bright pink and not reported to be scented. The combination of apparently unscented, brightly coloured upward-facing flowers and a narrow flower opening conforms to the rhinomyiophilous pollination syndrome and is considered indicative of pollination by LPF in Erica (Rebelo et al. 1985). Within LPF-pollinated ericas, flower orientation may also be an important trait for distinguishing pollination by two main groups of LPF pollinators, tabanid and nemestrinid flies. Since tabanid flies are unlikely to feed on upward-facing flowers, plants with such flower orientation are likely pollinated by nemestrinid flies (McCarren et al. 2022; McCarren et al. 2023). Thus, based on differences in floral syndromes, E. infundibuliformis was predicted to be pollinated by long-proboscid nemestrinid flies in contrast to pollination by hawkmoths in E. cylindrica.

The aim of this study was threefold: firstly, to quantify floral traits to objectively characterise differences between E. cylindrica and E. infundibuliformis; secondly, to verify predictions of LPF pollination in E. infundibuliformis with empirical observations and, finally, to use these combined data to test whether a shift in colour and scent mediates a shift between hawkmoth and nemestrinid fly pollination between the two species.

Corolla length differed significantly amongst all three populations (Figs 1, 2) and was shortest in E. cylindrica and longest in E. infundibuliformis from Agtertafelberg. In both species, more than 75% of flowers of an inflorescence faced upright and this proportion did not differ amongst the populations (Fig. 2).

Figure 2. 

Comparisons of floral and inflorescence characteristics between Erica cylindrica and populations of E. infundibuliformis: flower orientation (A), corolla tube length (B) and evening scent emission rates (C). Letters indicate significant pairwise comparisons at P < 0.05 (B).

Floral spectral reflectance patterns were largely similar for flowers of both populations of E. infundibuliformis, despite some variation in brightness (Fig. 3). Petals of flowers from both populations of E. infundibuliformis had greater overall brightness than flowers of E. cylindrica and also were consistently characterised by a steep increase to maximum reflectance in the UV region, around 350 nm (Fig. 3), which was almost absent in flowers of E. cylindrica. Overall brightness varied somewhat amongst flowers within both E. infundibuliformis populations, but was usually consistently close to maximum between 350 and 700 nm, with an average of approximately 50% for petals and 10% for floral tube surfaces (Fig. 3), which is approximately double that observed for flowers of E. cylindrica. The variation in flower colour within E. infundibuliformis populations was mostly in the degree of reflectance between 500 and 600 nm (Figs 1, 3).

Figure 3. 

Reflectance of floral petal surfaces (grey) and outer surfaces of corolla tubes (black) of Erica infundibuliformis from Agtertafelberg and Stettynsberg. Solid lines represent reflectance measurements recorded from individual flowers, dashed lines show means for all spectra recorded for the respective floral parts in each population.

A total of 82 compounds were detected in the scent samples of E. infundibuliformis (Table 1). Just over half of these (n = 42) could not be identified, but these unknown compounds on average only made up between 0.49% and 2.87% of the entire blend. The largest number of identified compounds were monoterpenes and these were also the most dominant in terms of scent emission, with (E)-ocimene being the most important compound at both sites and during both day and evening (Table 1). Of the 48 compounds produced by E. cylindrica, only ten were in common with E. infundibuliformis and the bouquets of these two species were consequently significantly different in composition (Fig. 4). The main difference between the species involved the unique presence of a large number of dominant aromatic compounds in E. cylindrica, contrasting with the large number of dominant monoterpenes in E. infundibuliformis (Table 2). Evening emission rates did not differ significantly amongst the populations (Fig. 2), whereas there was an effect of both population and period (but no significant interaction) in the comparison of daytime and evening scent emission rates of the two E. infundibuliformis populations: emission rates were higher at Agtertafelberg and during daytime (Fig. 5).

Figure 4. 

Non-metric multidimensional scaling of daytime and evening scent bouquets of Erica cylindrica and E. infundibuliformis populations.

Figure 5. 

Comparison of daytime and evening scent emission rates in the two Erica infundibuliformis study populations. Different letters indicate significant pairwise differences in emission rates at P < 0.05. Asterisks refer to levels of significance: ** P < 0.01, * P < 0.05.

Flowers of E. infundibuliformis produced a mean ± SD of 0.49 ± 0.42 μl of nectar (n = 24 flowers), with a sugar concentration of 35.8 ± 10.9% (n = 19 flowers), whereas flowers of E. cylindrica produced a mean ± SD of 0.19 ± 0.17 μl of nectar (n = 10 flowers), with a sugar concentration of 24.1 ± 5.3% (n = 6 flowers). Both nectar volume and sugar concentration were higher for E. infundibuliformis compared to E. cylindrica (nectar volume: z = 2.052, P < 0.05; sugar concentration: z = 2.39, P < 0.05).

At both study sites, moths were the most frequently observed visitors that fed legitimately from E. infundibuliformis flowers (Table 3, Fig. 6). At Agtertafelberg, a total of 12 hawkmoths were observed feeding on flowers of E. infundibuliformis over all three evenings of observations. At Stettynsberg, five hawkmoths were observed on a single evening, in addition to 15 settling moths which were observed feeding during daytime and evening hours (Table 3). Both hawkmoths and settling moths carried up to 1000 pollen grains per individual (overall mean ± SD 540.7 ± 362.9 grains, n = 11), of which on average 97% constituted Erica tetrads (SD = 7.4%, n = 11); 90% of Erica pollen carried by moths was located on moths’ proboscides (mean ± SD 90.1 ± 18.6%, n = 11) (Table 3).

Figure 6. 

Interactions between Erica infundibuliformis and flower visitors. Legitimate visits with potential for effective pollen transfer are shown on the left: Hippotion osiris with white pollen grains visible along the proboscis (A), Temnora sp. (B), both at Agtertafelberg; and Geometrid moth (possibly Acrasia sp.) visiting a flower at Stettynsberg (C). Illegitimate robbing visits, all photographed at Agtertafelberg, are shown on the right: Xylocopa sp. (D), Apis mellifera capensis (E) and a Sphecidae wasp sp. (F) feed on nectar through a puncture in the base of the floral tube without contacting reproductive parts of flowers. Hopliini sp. visiting flowers, possibly feeding on floral tissue (G).

No dipteran visitors to flowers of E. infundibuliformis were observed at Stettynsberg, despite the presence of several individuals of Philoliche (Tabanidae) which visited Iridaceae (Tritoniopsis cooperi, Geissorhiza confusa) and Proteaceae (Serruria) species in close proximity to Erica at this study site. Pollen on captured individuals of Philoliche also did not include Erica pollen (Table 3). At Agtertafelberg, one Bombyliid (Exoprosopa sp.) was observed foraging on flowers of E. infundibuliformis and one nemestrinid, Moegistorhynchus sp. nov., was captured in the population and carried approximately 250 grains of exclusively Erica pollen.

Floral larceny was observed almost constantly during diurnal observations at Agtertafelberg, but was not observed at Stettynsberg. At Agtertafelberg, 136 incidences of insects feeding through slits in the side of floral tubes, without potential for contact with reproductive parts, were observed, of which 90% were by carpenter bees and the balance by other hymenopterans including wasps and honeybees (Table 3). Monkey beetles were present on flowers, but did not feed on any flower parts (Table 3). Of 18 captured representatives of insects considered to exhibit robbing behaviour, only one individual carried more than ten grains of Erica pollen (Table 3). Carpenter bees captured on E. infundibuliformis flowers carried an average of 188 pollen grains (SD = 392.38, n = 10), of which less than 1% were Erica tetrads (mean ± SD 0.44 ± 1.26%, n = 9).

Rates of nectar robbing (as assessed from the proportion of flowers with evidence of puncturing in the tissue of the floral tube) varied between 30% (Stettynsberg) and 35% (Agtertafelberg), whereas anther ring disruption varied between 41% (Agtertafelberg) and 58% (Stettynsberg) of assessed flowers, but neither differed significantly between sites (Fig. 7).

Figure 7. 

Frequencies of nectar robbing and anther ring disruption (as a proxy for pollination rate) in the two Erica infundibuliformis study populations.

Results from this study revealed divergence in several floral traits, including scent, colour, corolla tube length and nectar, between E. cylindrica and E. infundibuliformis. Despite these floral differences, pollinator observations revealed that both species are pollinated by moths, contrary to the idea that the differences in floral traits indicate a difference in pollination system.

Although E. infundibuliformis was visited by a large number of insect species during both daytime and the evening, legitimate visits were almost exclusively limited to settling moths and hawkmoths, which carried large amounts of Erica tetrads. These observations strongly contradict the expectation that the pollination system of E. infundibuliformis differs from that of the hawkmoth-pollinated E. cylindrica. Moth visits were observed consistently on all evenings, over multiple years and at two different sites with somewhat different flowering phenology; the two study sites have non-overlapping flowering periods, such that by the time of peak flowering in December-January at Agtertafelberg, flowering at Stettynsberg – which peaks in November – is completely over. Consistent observations of moth pollination, in combination with the fact that pollen loads on moths consisted largely of Erica pollen, therefore suggests an established plant-pollinator interaction, rather than opportunistic foraging by (hawk)moths from plants that are adapted for pollination by other insects (see Haber and Frankie 1989; Martins and Johnson 2013). This study, hence, adds another record of moth pollination in Erica. The observed visits were associated with disrupted anther rings (a proxy for pollination in Erica) in roughly half the flowers examined, suggesting that moth visitation results in reasonably high pollination rates. Similar to the mechanism of pollen transfer in E. cylindrica (Van der Niet and Cozien 2022), pollen was found on the moths’ proboscises. Visits by Helicoverpa armigera at Stettynsberg were somewhat unexpected, as the proboscis of this moth species is much shorter than the corolla tube of E. infundibuliformis. It is possible that this moth species, which is also the main pollinator of the moth-pollinated E. plukenetii subsp. breviflora (Van der Niet et al. 2014b), is nevertheless able to access nectar that accumulates as droplets along most of the length of the corolla tube. A single bombyliid fly briefly visited E. infundibuliformis flowers during extensive daytime observation hours. Additionally, one individual of a Moegistorhynchus sp. was captured in the E. infundibuliformis population at Agtertafelberg. This fly not only carried Erica tetrads, but also had a tongue that closely matched the corolla tube of the local E. infundibuliformis plants in length. The pollen load and close morphological match with flower dimensions together suggest that Moegistorhynchus flies do occasionally visit the species, albeit at a lower rate than moths during the time that we observed plants. Low visitation rates of LPF to Erica species that are specialised for this pollinator group, with complete absence in some years, appears to be the norm (e.g. Newman and Johnson 2021; McCarren et al. 2023), raising the possibility that, in some years, flies may be more common visitors than observed during our study. However, regardless of whether and to what extent LPF contribute to pollination of E. infundibuliformis, evidence from this study shows that moths unambiguously contribute to pollination in this Erica species.

Moth visitation to E. infundibuliformis flowers is surprising because strong nocturnal floral scent is considered a key characteristic of moth-pollinated flowers (Faegri and van der Pijl 1979; Knudsen and Tollsten 1993; but see Martins and Johnson (2013)), whereas E. infundibuliformis flowers have not been described as scented to the human nose. Nevertheless, a weak scent was detectable when flowers were assessed in the field at the Stettynsberg study site, possibly owing to the elevated presence of linalool at this site in particular, and GC-MS analysis of E. infundibuliformis scent confirmed the presence of a scent bouquet that is richer in compounds than that of the closely related E. cylindrica, which is perceived as strongly scented by humans (Rebelo et al. 1985). Further, comparison of rates of scent emission showed that, at the time when moth pollinators are most active, during the evening hours, rates of scent emission are similar for both E. infundibuliformis and E. cylindrica. The main difference between the scent of the two Erica species is that E. cylindrica flowers emit aromatic compounds, whereas the scent of E. infundibuliformis flowers is dominated by monoterpenes, such as ocimene. In analyses of human perception of volatiles, based on gas chromatography coupled with olfactometry, (E)-ocimene is often reported as an active compound (Lee et al. 2011; Zhao et al. 2022), but it is sometimes described by humans as grass-like and perhaps, therefore, not considered a typical floral volatile. Together, these results reinforce the need to objectively quantify floral traits when assigning pollination syndromes to plant species.

Although the function of scent for moth attraction in E. infundibuliformis was not established experimentally, some evidence supports the idea. Despite the scent of E. infundibuliformis not conforming to a typical bouquet associated with moth pollination, hawkmoth (including the species visiting E. infundibuliformis) antennae respond to (E)-ocimene (Shuttleworth and Johnson 2022) and high emission rates of the two stereoisomers of ocimene in an orchid were found to coincide with the peak activity of hawkmoths (Steen et al. 2019), suggesting a functional role for monoterpenes in moth attraction. Additionally, the objective absence of scent recorded by GC-MS analysis of flowers of two Erica species pollinated exclusively during the day by LPF (McCarren et al. 2023) further supports a functional role of scent emission for pollinator attraction in E. infundibuliformis. Although lower scent emission during the evening than during daylight hours contradicts a typical moth syndrome (Jürgens et al. 2014; Balducci et al. 2020; Powers et al. 2020), this pattern has also been found in other moth-pollinated plant species (Van der Niet et al. 2015). An interesting analogous case of interspecific variation in floral scent composition, as found in this study between E. infundibuliformis and E. cylindrica, occurs in two closely-related Zaluzianskya species, although these species do differ in the pollination system: Z. natalensis, with a bouquet dominated by aromatics, is pollinated by hawkmoths, whereas Z. microsiphon, dominated by monoterpenes, is pollinated by long-proboscid nemestrinid flies (Campbell et al. 2016). However, unlike in Erica, in this Zaluzianskya species pair, the flowers of the fly-pollinated species close during the evening and experimental manipulations and bioassays showed that flower visitation by hawkmoths was determined by flower orientation rather than by scent composition.

Flowers of both moth-pollinated Erica species studied here were found to be predominantly facing upwards, which is highly unusual in the genus (Van der Niet and Cozien 2022). In other systems, the upward-facing orientation of flowers has been shown to be important for pollination by hawkmoths (Fulton and Hodges 1999; Campbell et al. 2016). Upward-facing flowers should, therefore, not be considered exclusively associated with nemestrinid fly pollination in Erica (see McCarren et al. 2022; McCarren et al. 2023), as this study further confirms its association with pollination by moths (and hawkmoths in particular) in Erica (see also Van der Niet and Cozien (2022)).

Variation in corolla tube length is often associated with covariation with pollinator morphology as the match may be important for effective pollen transfer (e.g. Van der Niet et al. 2014b; Newman and Johnson 2021). In this study, corolla tube length was found to differ amongst the three Erica populations, with the longest corolla tubes found for E. infundibuliformis at Agtertafelberg and the shortest in E. cylindrica, albeit only marginally shorter than E. infundibuliformis at Stettynsberg. However, there was no consistent covariation or trait matching between pollinator proboscis length and floral tube length amongst populations or species; many legitimate flower visitors had proboscides that exceeded the floral tube in length (Fig. 6). This general absence of covariation is not unexpected for interactions in which pollination precision is low, as is the case for both the study species of Erica, when anthers of the flowers do not protrude beyond the floral tube and pollen is mostly placed quite imprecisely along the moth proboscis (also see Johnson 2024). It is possible that shorter floral tubes at Stettynsberg might reflect higher rates of visitation by relatively short-tongued settling moths versus predominant visitation by longer-tongued hawkmoths at Agtertafelberg, but more pollinator observations are required to firmly establish any difference in moth species composition between the two sites.

The two studied Erica species differ in colour as perceived by humans and differences were confirmed by objectively measured reflectance spectra of the two species. In both species, brightness of the corolla tube is lower than for the petals, but in both Agtertafelberg and Stettynsberg, reflectance of petal lobes of E. infundibuliformis is approximately twice that recorded for E. cylindrica (see Van der Niet and Cozien (2022)). Flowers of E. cylindrica have maximum reflectance between the 500 and 700 nm region, mainly in the human-visible part of the spectrum, whereas flowers of E. infundibuliformis also strongly reflect light in the UV range of the spectrum, with a steep increase in the UV region around 350 nm and consistently high maximum reflectance from 400 to 700 nm. The flower colour is perhaps the most puzzling trait in association with moth-pollination, as moth-pollinated flowers usually do not exhibit this steep increase in reflectance in the UV range (e.g. Johnson and Raguso (2005)), which is, instead, strongly characteristic for other fly-pollinated Erica species (Lombardi et al. 2021; Newman and Johnson 2021; Pauw 2022; McCarren et al. 2023). The colour of the floral tube, which does not reflect light in the UV region may be particularly visible to many Hymenoptera and, therefore, explain their attraction to this flower part for nectar consumption (Chittka et al. 1994). However, they may also be attracted by the scent or their presence may be driven by local distribution, since they were not observed at Stettynsberg.

The apparent contradiction between observed pollinators and floral traits of E. infundibuliformis raises the question for what kind of pollinator the species is adapted. Although the possibility that LPF pollination was underestimated cannot be excluded, effective pollination by moths is unambiguous. Some of the quantified traits, such as the presence of floral scent, suggest a functional role in moth pollination, but not fly pollination. Flower colour, on the other hand, was more typical for LPF-pollinated ericas (McCarren et al. 2021). One possibility is that, for these two traits (scent and colour) in this system, there are not strong trade-offs associated with attraction of moths versus flies, as has been suggested for other specialised pollination systems (Aigner 2001) and shown for particular floral traits in other plant species (Muchhala 2007; Peakall et al. 2010). Cases where trade-offs have been demonstrated involved roles of highly specific sex pheromones for pollinator attraction or morphological mismatches. It is possible that the presence of certain scent compounds may not compromise fly pollination (Campbell et al. 2016), whereas flower colour may be less critical for nocturnal moth pollination. The pollination system of E. infundibuliformis may, thus, represent a combination of traits in which an absence of trade-offs facilitates a bimodal pollination system, which has been observed in several species of Tritoniopsis (Manning and Goldblatt 2005) and in Protea punctata (Johnson et al. 2012). Indeed, visitation by hawkmoths and LPF has been demonstrated in other long-tubed plants in South Africa that are seemingly unscented to the human nose, such as Satyrium hallackii (Johnson 1997). Such bimodal pollination systems may be particularly important if visitation by one of the pollinator groups is unpredictable, as in this case may apply to LPF. Interpretation of traits of E. infundibuliformis in the context of bimodal pollination rather than as exclusive adaptations for moth pollination is also consistent with the observed trait divergence relative to E. cylindrica. If both species were specialised for moth pollination, the observed differences in floral traits, most strikingly flower colour, which has shifted between quadrants in a model of fly vision (Fig. 8), would be particularly puzzling. Further studies, including temporal selective pollinator exclusion experiments to assess the potential role of diurnal versus nocturnal pollinators (e.g. Wenzell et al. 2024) and phylogenetic analyses that can be used to determine the evolutionary direction of floral trait divergence, are needed to obtain a better understanding of the evolution of pollination systems (Van der Niet et al. 2014b).

Figure 8. 

Loci of colours of petals and corolla tubes for flowers of Erica infundibuliformis from Aftertafelberg and Stettynsberg and for E. cylindrica, plotted in the fly vision colour space of Troje (1993). Loci of most, but not all spectra of both petals and floral tubes of flowers from both populations of E. infundibuliformis fell within a single quadrant in the fly vision colour space; spectra of both petals and floral tube surfaces of E. cylindrica fell into a different quadrant, suggesting that neither species has within-flower contrast for flies, but that flies may perceive flowers of the two species differently.

This study adds to a number of cases in which syndrome-based hypotheses were contradicted by empirical observations (e.g. de Merxem et al. 2009; Cozien et al. 2019; Castañeda-Zárate et al. 2021). In the study system examined here, the combination of rarity of hawkmoth pollination in the CFR, in general and in Erica, in particular, in combination with the inadequacy of human perception for identification of floral scent, were probably the main reasons why moth pollination was not correctly predicted. It is also possible that the trait combination in E. infundibuliformis represents a syndrome indicative of bimodal pollination by flies and moths, but data are currently inadequate to confirm or refute this idea. Our findings again underline the importance of verifying pollination systems predicted by syndromes with empirical pollinator observations (Van der Niet 2021). Results also indicate that hawkmoth pollination, which has now been documented at multiple sites and years and during different seasons, may be more common in the CFR than previously suggested (Johnson 1997) and emphasise that, despite the associated challenges, pollinator observations during evening hours should be considered a critical component in any pollination study of plant species with long-tubed flowers and to characterise the pollination systems of any region (cf. Cai et al. 2024).

We thank Denis Brothers and Øystein Opedal for help with insect identification. James Rodger is thanked for help with fieldwork. We thank Johannes Botha of Stettynsberg and the Mountain Club of South Africa for access to their land for fieldwork. We thank Sam McCarren and two anonymous reviewers for constructive comments on an earlier draft of this manuscript.