• In nursery pollination systems, pollinator offspring usually feed on pollinated fruits or seeds. Costs and benefits of the interaction for plant and pollinator, and hence its local outcome (antagonism–mutualism), can be affected by the presence of ‘third-party’ species. Infection of Silene latifolia plants by the fungus Microbotryum violaceum halts the development of fruits that provide shelter and food for larvae of the pollinating moth Hadena bicruris. We investigated whether the moth secures its benefit by selective oviposition on uninfected flowers.
• Oviposition was recorded in eight natural populations as a function of plant infection status, local neighbourhood, plant and flower characteristics.
• Oviposition was six times lower on flowers from infected than on those from uninfected plants. Oviposition decreased with decreasing flower and ovary size. Moths could use the latter to discriminate against diseased flowers.
• Although moths show an adaptive oviposition response, they reduce the future potential of healthy hosts because they still visit infected plants for nectar, vectoring the disease, and they reduce any fitness advantage gained by disease-resistant plants through selective predation of those plants.
Nursery pollination mutualisms are mutually beneficial interactions between pollinators and plants, in which the plant provides oviposition sites and shelter or food for the offspring of the pollinator. At least 13 nursery pollination mutualisms have been identified, that have evolved independently and convergently (Dufaÿ & Anstett, 2003). Nursery pollination mutualisms, like any mutualism, should be considered as one of the outcomes of a reciprocal exploitation between interacting species (Bronstein, 1994; Anstett et al., 1997; Dufaÿ & Anstett, 2003), in which both plant and pollinator are selected to increase their own benefit and reduce the cost of the interaction. Depending on the net result of these conflicts of interest, a gradient from antagonistic to mutualistic associations can occur that can vary locally, spatially, and among species combinations (Gomulkiewicz et al., 2003; Palmer et al., 2003).
A nursery pollinator system that is considered to be predominantly at the parasitic end of the spectrum (Brantjes, 1976b, 1976c; Bopp, 2003) is the interaction between moths of the genus Hadena and their host plants in the genus Silene, and a few other genera in the pink family (Caryophyllaceae) (Kephart et al., unpublished). Hadena moths pollinate their host plants and at the same time oviposit on the ovaries inside the flowers. The moth larva develops inside the maturing fruit at the expense of developing seeds. This is considered a primitive example of a nursery pollination mutualism, compared with the more specialized fig/fig wasp and yucca/yucca moth systems (Dufaÿ & Anstett, 2003). First, the pollinators have no active pollination, and do not appear to minimize pollination to ensure fruit set for their own offspring. Second, plants do not appear to reduce the cost of maturing infected fruit by selective abortion of infected fruits, hence plants appear to have no control over pollinator population size. In such systems, the persistence of mutualism depends more on external factors than is seen in more specialized systems (Stanton, 2003).
Many external factors can affect the costs and benefits of the interaction for plants and pollinators in nursery pollination mutualisms. Among these are the presence of ‘cheaters’ that do not pay the costs but do gain the benefits; alternative pollinators that do not predate on seeds; and suites of abiotic or biotic factors that affect the quality of the interacting partners for each other. In the Hadena–Silene nursery pollinator system, one such biotic factor is the specialized plant-parasitic anther smut fungus Microbotryum violaceum, which takes advantage of the mutualism. It sterilizes its Caryophyllaceous host plants and produces its spores in the anthers of infected hosts (Thrall et al., 1993). The spores are then transmitted by the pollinator that becomes a vector of the disease (Jennersten, 1983).
For the plant, pollination in diseased populations carries a severe risk of becoming infected and sterilized. Presence of M. violaceum can also reduce seed set on healthy plants because of increased spore-to-pollen ratios and inhibition of pollen germination (Marr, 1997, 1998). For the pollinator, infection destroys two important rewards of the host plant. First, ovaries of infected flowers are aborted at an early stage and hence do not provide developing seeds as a food source for the larvae of the pollinator. Surprisingly, studies in the Hadena bicruris–Silene latifolia system have shown that H. bicruris larvae are very plastic in their food choice (Biere et al., 2002). On diseased plants that do not offer maturing fruits, larvae consume flowers and leaves and reach pupation rates similar to those on uninfected host plants. However, these pupae suffer a 40% lower weight and a > 25% reduced probability of survival to the adult stage. Second, diseased plants lack shelter against predator and parasitoid attack. In natural populations of S. latifolia, parasitoids are a main source of mortality of H. bicruris larvae (Elzinga et al., 2005). On healthy plants, a first-instar caterpillar eats its way into the ovary and develops inside the maturing seed capsule until its fourth or fifth instar, providing shelter against predators and parasitoids. On diseased plants, such shelter is lacking. Accordingly, experimental studies have shown that, because of the higher exposure of H. bicruris larvae to parasitoids on diseased S. latifolia plants, these larvae suffer twofold higher rates of parasitism than do larvae on healthy plants (Biere et al., 2002). Hadena bicruris still visits infected host plants of S. latifolia, probably because infection of female hosts reduces nectar rewards only to a limited extent (Shykoff & Bucheli, 1995; Biere & Honders, 1996a; Shykoff & Kaltz, 1998), but it is not known whether the moth discriminates against diseased plants for oviposition, which would be an adaptive response.
In this paper we investigate the following questions. (1) Does H. bicruris show an adaptive response to infection of its host plants by M. violaceum by avoiding oviposition on such plants? (2) What mechanisms could be involved in an adaptive oviposition response? Hadena bicruris uses both olfactory and visual (morphological) cues in host selection. We restrict ourselves to nonolfactory cues, and study (i) which population (host density, nearest neighbour distance), plant (floral display) and flower (corolla, ovary size) characteristics are associated with high oviposition rates; and (ii) which of these characteristics differ between healthy and diseased plants and hence could potentially be used by the moth to discriminate between rewarding and less rewarding hosts.
Materials and Methods
Silene latifolia Poiret (= S. alba [Miller] Krause, Caryophyllaceae), the white campion, is a dioecious, short-lived perennial plant from disturbed sandy habitats including river banks, agricultural field margins and road verges. The anther smut fungus Microbotryum violaceum [Pers. Pers.] Deml & Oberw. (=Ustilago violacea [Pers.] Fuckel, Ustilaginaceae) is a plant-parasitic fungus that sterilizes its host plants in the Caryophyllaceae (Thrall et al., 1993). Infection of female S. latifolia plants results in abortion of the ovary and development of anthers that – as in diseased males – only contain fungal teliospores. Infection also induces a number of alterations in the plant phenotype, including the production of more but smaller inflorescences and flowers (Day & Garber, 1988; Biere & Honders, 1996a; Shykoff & Kaltz, 1998). Hadena bicruris [Hufn.] (Noctuidae), the Lychnis, is a moth specialized on S. latifolia and a few other Caryophyllaceous host species. In western Europe it is a major pollinator of S. latifolia (Jürgens et al., 1996). It visits both male and female plants for nectar, and usually oviposits a single egg (Brantjes, 1976a) on the ovary of flowers on female host plants at dusk. The first-instar caterpillar chews a hole in the ovary wall, enters the fruit and stays inside the capsule, feeding on the developing seeds, until its fourth or fifth instar. The caterpillar then leaves the capsule and consumes the contents of a few uninfested capsules before pupating in the soil (Biere & Honders, 1996b).
To study oviposition on healthy and M. violaceum-infected S. latifolia plants, we selected eight natural populations of S. latifolia in the Netherlands (Table 1), in which previous studies had shown that plant population size exceeded 200 individuals and that both H. bicruris and M. violaceum were present. Populations censuses were carried out four times during a single year: in early June, late June, late July and late August 1996. Criteria for days on which censuses were performed were absence of rainfall during the preceding 24 h, and low-to-moderate wind speed (max. 4 Beaufort). At each census we sampled, on average, 60 flowering female plants per population. In small populations these were nearly all the flowering female plants available (Table 1); in large populations individuals were sampled along a linear transect. One population was mown after the third census so no fourth census could be carried out. On average, samples consisted of 40 healthy and 20 M. violaceum-infected female plants. From each plant sampled, we randomly selected one flower that had opened the previous evening (healthy plants: stigmas freshly unfolded; diseased plants: spore-filled anthers not yet or just dehisced) and examined the presence and number of H. bicuris eggs in the flower. All eggs observed were white, confirming they had been laid the previous night (older eggs turn darker; Brantjes, 1976b). To identify factors affecting oviposition choice we took measurements at four hierarchical levels: flowers, plants, local neighbourhood and population. At the flower level we measured petal lobe length (a proxy of flower diameter), ovary length and maximum ovary width, using a digital caliper. At the plant level we measured the number of open flowers and the total number of flowers on the plant. As local neighbourhood characteristics, we measured the number of healthy and M. violaceum-infected flowering male and female S. latifolia plants within a 4 m radius around the focal plant, using a measuring tape with one end fixed next to the focal plant, and the distance to the nearest neighbouring S. latifolia plant. The 4 m radius was based on studies of the transmission of M. violaceum in S. latifolia, indicating that this is a representative scale at which pollinators respond to the frequency of diseased plants (Biere & Honders, 1998). At the population level, we counted the number of healthy and M. violaceum-infected flowering female S. latifolia plants in the population. At the third and fourth censuses, we also counted the number of female S. latifolia that had ceased flowering to obtain an estimate of the total number of reproductive female plants per population. For each population and census day, data for average day temperature at 150 cm height were obtained from the nearest weather station of the Royal Netherlands Meteorological Institute (KNMI, De Bilt, the Netherlands). On plants carrying more than one open flower that had opened the night before, oviposition and flower traits were measured on a second flower to indicate the repeatability of oviposition estimates on a single plant. During the season, eggs were regularly collected and reared out in the laboratory to confirm that they were H. bicruris eggs.
Table 1. Locations (north and east coordinates) and characteristics of eight Silene latifolia populations used to study oviposition by Hadena bicruris
Generalized linear models (GLM) were employed to analyse oviposition data (presence/absence of H. bicruris eggs on a flower), using a binomial error distribution and a logit link function (procedure GENMOD, sas ver. 8.2, SAS Institute, Cary, NC, USA). A scale parameter was estimated to correct for overdispersion. First we analysed the effects of population, census day and flower disease status on oviposition to assess whether oviposition differed between healthy and diseased flowers, and whether the magnitude of the difference varied among populations or census days. Second, we analysed effects of traits measured at the four hierarchical levels on oviposition in a single analysis. To deal with the hierarchical structure in the data, a type I (sequential) analysis was run in which effects of traits were estimated in the following order: weather, population traits, local neighbourhood traits, plant traits and flower traits. Nonsignificant effects were removed to obtain the final model. Measurements of ovary length (L) and width (W) were combined into a single index of ovary size (S) proportional to the volume of the ovary, S = L × (W)2. Ovary size measurements were missing from populations WH, EL and M1 during the first census, so these censuses were omitted from analyses that include ovary size. Third, to detect traits that could possibly be used by female moths to discriminate between healthy and diseased flowers within a population, we used GLM with a normal error distribution and an identity link function to analyse the effects of population and disease status for each of the measured flower, plant and local neighbourhood parameters separately. Nearest-neighbour distance, density of conspecifics, total number of flowers and number of open flowers were ln-transformed before analysis to meet assumptions of normality and homogeneity of variances.
Plants with more than one open flower were used to test the consistency of oviposition choice. We calculated the number of concordant observations (both first and second flowers had received either eggs or no eggs) and discordant observations (one had received an egg, the other not), and used a χ2 test to compare the observed numbers with those expected if oviposition on the second flower was random with regard to whether there was an egg on the first sampled flower.
Effect of fungal infection on oviposition rate
Oviposition by H. bicruris was significantly higher on flowers of healthy S. latifolia plants than on flowers of diseased plants (Fig. 1; Table 2). On average, 52.0% of flowers sampled from healthy plants and only 8.1% of flowers from diseased plants received at least one H. bicruris egg during the first night of opening. The magnitude of the difference varied among censuses (Table 2, interaction disease status × population × time; Fig. 1).
Table 2. Generalized linear model of the effects of population, census day and disease status of Silene latifolia flowers (healthy vs. Microbotryum violaceum-infected) on oviposition by Hadena bicruris
Quasi-F values and residual mean deviance are indicated.
Consistency of differences in oviposition rate between plants
For plants with more than one open flower, cases where both flowers had either received eggs or had received no eggs were significantly overrepresented ( = 339.0, P < 0.001), suggesting consistent differences in oviposition between plants. This was true for both healthy ( = 149.7, P < 0.001) and diseased plants ( = 9.9, P < 0.01).
Effects of weather, neighbourhood and flower characteristics on oviposition rates
Oviposition rates varied among censuses, ranging from 15 to 90% on healthy plants, and from 0 to 33% on diseased plants (Fig. 1). A large part of the variation in oviposition rates on healthy plants among censuses (32%) could be explained by differences in average day temperature at 150 cm during the census (Fig 2; F1,29 = 15.9, P < 0.001). Oviposition rate increased, on average, by almost 3% for every °C temperature increase. Differences in oviposition rate on diseased plants among censuses could not be explained by differences in temperature (F1,28 = 0.38, P = 0.54). Mean oviposition rates varied among populations (Table 2), ranging from 33.4 to 64.4% on healthy plants, and from 2.7 to 12.1% on diseased plants in populations Drijber and M2, respectively. Differences in oviposition rate among censuses could not be explained by differences in population size, or expressed as total number of female flowering plants in the population over the whole season, or as number of female flowering plants with open flowers during the census (healthy plants, F1,29 = 0.75, P = 0.39; diseased plants, F1,28 = 0.99, P = 0.33). Within populations, oviposition rate was affected by local neighbourhood characteristics. Oviposition rate on healthy plants increased with decreasing density of conspecifics in a 4 m radius (Table 3). Likewise, oviposition rates on diseased plants increased with distance to the nearest neighbouring female plant. At the plant level, oviposition was affected by floral display (number of open flowers on a plant). Interestingly, oviposition rates increased with larger floral display on healthy plants, whereas they decreased with larger floral display on diseased plants (Table 3; Fig. 3a). At flower level, oviposition was higher on flowers with a larger diameter (petal lobe length) (Table 3; Fig. 3b) and especially on flowers with larger ovary size (Table 3; Fig. 3c), on both healthy and diseased plants.
Table 3. Generalized linear models of effects of local neighbourhood, plant and flower characteristics on oviposition by Hadena bicruris on flowers of healthy and diseased (Microbotryum violaceum-infected) plants of Silene latifolia
Healthy plants F1,1122
Diseased plants F1,549
To adjust effects for differences in weather conditions during censuses and for overall differences in oviposition between populations and census dates, effects of average day temperature, population and census date were fitted before all other variables. Quasi-F values, direction of effect (in parentheses), residual mean deviance and significance of effects are indicated.
Differences in neighbourhood and flower characteristics between healthy and infected plants
Several neighbourhood, plant and flower characteristics showed significant differences between healthy and diseased plants, offering nonolfactory cues that could be used by H. bicruris to adjust oviposition to the disease status of plants (Table 4). At the neighbourhood level, diseased plants occurred in local areas with a higher density of conspecifics and a shorter distance to nearest neighbours. Further, the fraction of conspecifics in a 4 m radius that were infected with M. violaceum was twice as high around diseased plants than around healthy plants, indicating that disease was clustered within populations. The total numbers of flowers and of open flowers on diseased plants were two and more than three times higher, respectively, than those on healthy plants. Diameter of flowers (petal lobe length) was not significantly smaller (P = 0.07) on diseased plants, but ovary size was, on average, about one quarter that of those on healthy plants (Table 4). All traits differed among populations, and for several traits the magnitude of the difference between healthy and diseased plants varied among populations (Table 4, interaction population × disease status).
Table 4. Differences in local neighbourhood, plant and flower characteristics between healthy and diseased (Microbotryum violaceum-infected) plants of Silene latifolia
Sample sizes (N) and means (±1 SE) for healthy and diseased plants are indicated. Quasi-F values from generalized linear models and the significance of effects of disease status, population and their interaction are given for each trait.
The moth H. bicruris showed a strong, adaptive oviposition response to infection of its host plant S. latifolia by the anther smut fungus M. violaceum. Oviposition on flowers of the less rewarding, infected host plants, that do not provide a high-quality food source or shelter against parasitoid attack (Biere et al., 2002), was more than six times lower than that on flowers of uninfected female plants. In the short run, this will secure the nursery pollination benefit that the moth obtains from the host plant in the presence of the fungus as a ‘third party’ that competes for control over flowers of the host plant. A question that requires further study is whether the low level of oviposition on infected flowers reflects a lower visitation rate of infected flowers, or an active decision not to oviposit on infected flowers once they are visited. No studies so far have specifically addressed visitation patterns of healthy and infected S. latifolia flowers by H. bicruris. However, studies of visitation patterns by the joint guild of night visitors suggest that diseased S. latifolia plants receive similar numbers of visits to healthy females (Shykoff & Bucheli, 1995). As infected plants produce more flowers than uninfected plants, visitation rates per flower are proportionally smaller. This is in accordance with the patterns in nectar rewards, the only reward for H. bicruris that is shared between healthy and diseased females. Nectar volume and nectar sugar production do not differ between healthy females and diseased plants (Shykoff & Bucheli, 1995), but tend to be lower in diseased plants on a per flower basis (Biere & Honders, 1996a), although the latter does appear to be the case for infected males rather than for infected females (Biere & Honders, 1996a; Shykoff & Kaltz, 1998). Similar patterns have been found for day visitors and for other Caryophyllaceous species. For instance, healthy and infected Viscaria vulgaris produce similar amounts of nectar, but as infected plants produce more flowers, nectar production per flower is lower in infected plants and they receive proportionately fewer visits (Jennersten, 1988). If this pattern holds for the H. bicruris–S. latifolia interaction, we would expect moth visitation rates per flower to be maximally about twofold lower on diseased females, which produce around twice as many flowers as healthy females. The sixfold lower oviposition rates on infected flowers suggests that, in addition to visitation differences, moths make an active decision not to oviposit on diseased flowers when they visit them.
The cues that H. bicruris moths use for oviposition discrimination against diseased plants in S. latifolia are unclear. It is not known whether moths are able to detect and respond to the presence of fungal spores in the flower. Floral scent is known to trigger landing on the flowers, the number of landings being proportional to the intensity of the scent (Brantjes, 1976c). Recently, several of the floral scent compounds responsible for this response have been identified (Dötterl et al., 2005). It would be interesting to see whether there are any infection-induced olfactory cues to which moths can respond with respect to oviposition behaviour. Part of the lower oviposition on diseased plants may not be caused by active discrimination, but may be a consequence of a general density dependence in the oviposition response. For both healthy and diseased plants, we observed a ‘saturation effect’ that has previously been observed in both natural and experimental patches (Elzinga et al., 2005): oviposition decreased with increasing density of conspecifics or decreasing distance to the nearest conspecific. Because of the observed higher incidence of diseased plants in denser patches, part of the lower oviposition on diseased plants might be explained by this general response of moths to host density. This is certainly not the case for responses to variation in floral display. Based on the observed increase in oviposition rate with an increasing number of open flowers on healthy female plants, we would expect higher oviposition rates on diseased plants, which produced larger numbers of open flowers. However, on diseased plants the opposite pattern was found: oviposition per flower decreased with increasing number of open flowers. One possible interpretation of this pattern is that moths are attracted to plants with larger floral displays, but whereas they visit and select a large number of the flowers on such plants for oviposition when they are healthy, they visit or accept only a limited number of flowers for oviposition when they are diseased, resulting in a lower fraction of oviposited flowers on diseased plants as the number of flowers on these plants increase.
At flower level, we found that oviposition rates increased with flower size. As experimental inoculation studies have generally shown that M. violaceum infection results in reduced flower size in S. latifolia and other Caryophylaceous species (Biere & Honders, 1996a; Shykoff & Kaltz, 1998), this trait could be involved in reduced oviposition rates on diseased flowers. However, in our field study we found no evidence for reduced size of diseased flowers. We speculate that this is because visitation and spore deposition rates are higher on plants with larger flowers (cf. Shykoff et al., 1997), hence plants that became diseased had larger flowers to start with, infection reducing their flower size to the average size of flowers on uninfected plants. The most striking difference that we observed between healthy and infected flowers was the reduced size of the aborted ovary on infected plants. In H. bicruris, nectar feeding and oviposition behaviour are strongly related. The former obligately precedes the latter, and experimentally satiating moths results in loss of selectivity with respect to oviposition site. ‘Measuring’ the perimeter of the ovary during nectar feeding to discriminate against diseased flowers during subsequent oviposition therefore appears to be a feasible scenario. On both healthy and on diseased plants, oviposition strongly increased with ovary size. This result should be interpreted with some caution. In principle, larger ovary size could be a result, rather than a cause, of higher oviposition rates; ovaries on which oviposition is recorded are also pollinated and may have grown more quickly than ovaries of unvisited, unfertilized flowers during the hours between pollination and our recording of ovary size and oviposition. However, we argue that this could explain only part of the association. First, increases in oviposition rate with ovary size were observed for both healthy and diseased flowers; as the latter cannot be fertilized, increased growth of fertilized ovaries can be excluded in that case. Second, the period between pollination and recording is probably < 12 h, during which fertilization could result in a c. 7% increase in ovary size (Oostenik, 1976), much smaller than the average size difference between ovaries with and without eggs. If we then interpret the positive association between oviposition rate and ovary size as an oviposition preference for larger ovaries, we could hypothesize that the lower oviposition rates on diseased flowers are caused partly by their smaller ovaries. Surprisingly, a similar hypothesis proposed for the role of the ovary in the almost perfect oviposition discrimination of H. bicruris against healthy staminate (male) compared with healthy pistillate (female) flowers could not be supported by the data (Brantjes, 1976c). Like flowers on diseased plants, flowers on male plants are deceit flowers from the viewpoint of nursery pollination, providing neither shelter nor food, but they do produce nectar, usually in lower quantity but with higher sugar concentration per flower (Shykoff & Bucheli, 1995; Biere & Honders, 1996a; Shykoff, 1997; Shykoff & Kaltz, 1998). They are visited for nectar, but under normal conditions H. bicruris does not oviposit on staminate flowers. However, on the basis of manipulation studies with experimental addition/removal of flower parts, the presence of neither calyx, stigmas, ovary or amount of nectar appeared to be decisive in oviposition discrimination between pistillate and staminate flowers (Brantjes, 1976c). This means either that the H. bicruris response to detached organs differs from that in a natural flower setting, or that the presence and size of ovaries are not an important cue in oviposition decisions.
Pathogen impact on nursery pollinator mutualism
The pathogen M. violaceum effectively exploits the nursery pollinator system. It converts the nursery sites for Hadena offspring (flowers producing fruit) into nursery sites for its own offspring (flowers producing anthers) through masculinization of the flower (Scutt et al., 1997). It further induces the plant to produce more flowers (more anthers and sporulation sites) without severely disrupting its transmission opportunities, as infection neither severely reduces the life span of (then only spore-producing) plants (Alexander & Antonovics, 1995), nor does it reduce flower size and nectar rewards to such a level that they lose attractiveness for pollinators vectoring the spores. In other systems, fungal infection may even lead to higher visitation rates on infected than on uninfected plants (Lara & Ornelas, 2003). From the plant's perspective, the presence of M. violaceum in a population can reduce the pollination efficiency of the nursery pollinator by the presence of spores in the deposited pollen load, and poses a severe risk of attracting pollinators because they might be carrying spores causing sterilization. Plants can respond either by evolving resistance (Alexander, 1989; Alexander et al., 1993; Biere & Antonovics, 1996) or shifting phenology to minimize spore-to-pollen ratios (Pettersson, 1991; Alexander et al., 1993; Biere & Honders, 1996b). From the nursery pollinator's perspective, selective oviposition by H. bicruris on healthy plants can be seen as an adaptation of the moth to secure its benefits in the nursery pollination system in the short run, but selective oviposition on healthy plants and the continued visitation of diseased plants by H. bicruris pose a severe threat for the nursery pollinator system in the long run. First, moths continue to act as vectors of the disease, reducing the future potential of healthy hosts for their offspring. Second, the selective advantage of disease-resistant plants in S. latifolia populations in which both M. violaceum and H. bicruris are present is diminished by increased oviposition on resistant, healthy individuals. This may slow down the evolution of resistance in such populations, reducing the potential of healthy nursery plants, and increasing the risk of local population extinction. We conclude that, despite the adaptive oviposition response of H. bicuris, the presence of M. violaceum as a third party in this nursery pollinator system may increase the costs of the interaction for both parties and will locally reduce opportunities in this system to shift from antagonism to mutualism.
We thank two anonymous reviewers for their constructive comments on an earlier version of the manuscript, and Staatsbosbeheer for their permission to perform field work in the nature conservation areas used in this study.