Co-evolution between plants and herbivores is a major cause of both plant and insect diversity and adaptation (Ehrlich and Raven, 1964). The role of host shifts and key innovations as a driving force in herbivore diversification has been widely studied. Similarly, in recent years there has been considerable interest in the role of herbivores in promoting plant diversification, specifically through the Janzen-Connell effect (Janzen, 1970; Connell, 1971). This hypothesis states that herbivores could exert negative frequency dependent selection by adapting to exploit the commonest host plants in their local environment. This could in turn favor rare plant species and promote local plant species diversity.

The Janzen-Connell hypothesis has generally been discussed in the context of specialist herbivores preventing the local establishment of common plant species. However, an alternative mechanism is that more generalist herbivores might learn a “search image” for locally common plant species. This could similarly generate negative frequency dependence, but on a much shorter timescale (Sinervo and Calsbeek, 2006). In visually searching predators, this could be driven by learning of distinctive cues for finding host plants, such as leaf shape. Variation in the size and shape of leaves is often considered to be mainly a result of the physiological and biomechanical demands imposed by different habitats (Brown et al., 1991). The role of herbivores in influencing the evolution of leaf size and shape has mainly considered in the context of physical barriers to herbivory (Brown et al., 1991). The role of leaf shape as an adaptation against visual herbivores has been less well studied, but one example is leaf mimicry in the Boquila trifoliolata vine, which mimics the leaves of its supporting trees to avoid visual herbivores (Gianoli and Carrasco-Urra, 2014).

In order to test the idea that herbivores might use visual cues such as leaf shape in finding their host plants, we need to demonstrate that the relevant herbivores can indeed use shape cues. Shape perception in insects has primarily been studied from the perspective of foraging bees (Anderson, 1977; Zhang et al., 1995), which show a preference for radial patterns when searching for nectar (Lehrer et al., 1995). Monarch butterflies, Danaus plexippus, are capable of learning shape only in association with color, showing that both stimuli must appear together in the context of foraging for nectar (Cepero et al., 2015). Furthermore, leaf shape detection and learning has been demonstrated in oviposition preference in Battus philenor (Rausher, 1978; Rausher and Papaj, 1983; Papaj, 1986; Weiss and Papaj, 2003) and in Eurema, which landed more often on leaves that resemble their host (Mackay and Jones, 1989).

Perhaps the most promising system in which visually searching herbivores interact with a diverse community of leaf shapes is among Heliconius butterflies and their Passiflora host plants (Gilbert, 1982). Leaf morphology in the family Passifloraceae, both between and within species, is among the most variable observed in any plant group (Figure 1A). In any locality, Passiflora species exhibit a wide variety of leaf shapes even if they are closely related and inhabit similar physical conditions (Benson et al., 1975; Gilbert, 1975, 1982). Some species also show a huge range of intra-specific variation in shapes, especially between young and old leaves (Gilbert, 1982). For example, Passiflora suberosa shows a high degree of leaf plasticity when raised in different light intensities (Barp et al., 2006). In addition, some Passiflora are very similar in form and texture to other non-host plants, which might be a form of mimicry. Gilbert (1975) speculated that visual searching behavior by Heliconius butterflies acts as a diversifying evolutionary force on Passiflora vines. Heliconius larvae feed almost exclusively on the family Passifloraceae, and can cause severe foliage damage (Gilbert, 1982). This close insect-plant interaction has led to the evolution of various defense mechanisms in Passiflora plants in response to selective forces imposed by Heliconius caterpillars.

The Heliconius-Passiflora interaction is already well established as an example of insect-host co-evolution. Passiflora species possess a range of defensive traits, such as production of chemical compounds that provide feeding barriers (Smiley, 1985a; Engler et al., 2000) and mechanical protection such as hooked trichomes that are able to pierce larvae, resulting in death for the majority of Heliconius caterpillars on P. adenopoda (Gilbert, 1971). In turn Heliconius charithonia has evolved to overcome these trichomes and is the only species that can feed on this host. In addition, extra-floral nectaries on some Passiflora species are similar to Heliconius eggs (Gilbert, 1982). In Passiflora cyanea projections on the stipules resemble, in shape and color, eggs of Heliconius ethilla (Williams and Gilbert, 1981). Females avoid ovipositing in the presence of a conspecific egg on the host, as young larvae are often cannibalistic (Nardin and Araújo, 2011), and are therefore deterred by these egg mimics. Egg-mimicry of extra-floral nectaries provides strong evidence that ovipositing females use visual cues in the selection of suitable Passiflora vines. Another function of extra-floral nectaries against herbivores is the production of nectar that attracts ants to the plant, to collect this valuable food resource (Apple and Feener, 2001; Izaguirre et al., 2013). Ants in turn attack Heliconius larvae and eggs (Smiley, 1985b, 1986).

Based on field observations, females of Heliconius butterflies use visual cues while searching for host plants (Brown, 1981). Females may inspect objects that resemble a Passiflora structure, such as similarly shaped leaves or vines that look like tendrils (Benson et al., 1975; Gilbert, 1982). A female searching for a specific Passiflora plant typically flutters slowly just above the vegetation, periodically approaching, and landing on leaves (Figure 1B). Upon landing, she drums her forelegs and presumably stimulates tarsal chemoreceptors, allowing the female to “taste” the plant with her gustatory receptors (Briscoe et al., 2013). It has been hypothesized that Passiflora leaf shape variation might make it harder for Heliconius females to detect host plants.

However, shape detection has not yet been demonstrated in Heliconius butterflies. They can be trained to associate a color stimulus with a food reward, demonstrating a high precision of discrimination and learning (Swihart and Swihart, 1970; Swihart, 1971; Blackiston et al., 2011). Here we extend these experiments to show that Heliconius erato can be trained to associate a shape cue with a food reward, demonstrating the perceptual ability to detect and distinguish shapes. Next, we tested shape perception for leaf morphology using ovipositing females trained on artificial leaves. H. erato naturally feeds on three species with diverse leaf shapes in our study area, and our results show that learnt shape preference is therefore a plausible selective force on Passiflora leaf morphology in this community.

Here we have shown for the first time that Heliconius butterflies can use shape cues to search for both flowers and leaves during feeding and oviposition. One explanation for the observed spontaneous preference for three and five petal flower shapes is an innate preference for radial symmetry, which corresponds to the actinomorphic flowers used most commonly by Heliconius (Corrêa et al., 2001). After training, we conditioned individuals to shift their shape feeding preference to an artificial flower with two petals, the least preferred shape initially. Previous studies have shown that conditioned Heliconius butterflies can shift their preference to yellow and green flowers, against their innate preference for orange and red (Swihart and Swihart, 1970). Here we show a similar effect for shape cues. Color is perhaps a more reliable visual cue for finding flowers since it is not affected by the angle of approach, as observed in Monarch butterflies (Cepero et al., 2015). However, shape is a complementary cue and may be important to distinguish objects that are similar in color.

The pipevine swallowtail butterfly, B. philenor, has long been known to use leaf shape in oviposition, in experiments in which the butterflies were trained on both real plants (Papaj, 1986) and artificial leaf models (Rausher, 1978; Allard and Papaj, 1996; Weiss and Papaj, 2003). Here we have provided the first evidence that Heliconius also use shape for leaf detection. Our results therefore support field observations of Heliconius female butterflies visually discriminating different leaves while searching for host plants. The butterflies exposed to lanceolate leaves approach the lanceolate shape more than those exposed to P. biflora shape. It is worth noting that our artificial plants did not provide a negative stimulus against laying eggs during the training period, in contrast to the way that the flower shape experiment could provide a negative stimulus in terms of the absence of sugar water. The difference in response of the butterflies to the two leaf shape treatments is therefore the result of a learnt association between leaf shape cue and availability of an oviposition stimulus.

The lack of any significant difference between treatments in egg laying perhaps indicates that shape is not the only clue used to oviposit on host plants. Other cues such as leaf color, plant architecture and plant odor and taste are also likely to be extremely important (Rausher, 1978; Allard and Papaj, 1996). Specifically, in the confined insectary space female butterflies do not have trouble in eventually finding the shoots, irrespective of their associated leaf shape. Thus, once they have found both shoots the optimal strategy for a female is to distribute her available eggs evenly between the two shoots. Nonetheless, in the wild where long distance detection of host plants is likely to be more challenging than in a small insectary cage, it seems likely that leaf shape could play an important role in the location of host plants used by female butterflies.

We can therefore speculate about the potential for this learning behavior to influence the evolution of leaf shape in Passiflora. It has been previously suggested that diversification of leaf morphology might be a response to herbivore pressure (Gilbert, 1975; Rausher, 1978). Three elements of Passiflora leaf morphology may have evolved in response to Heliconius visual perception: mimicry, divergence in leaf shape between species and different adult and juvenile foliage (Gilbert, 1982). Negative frequency dependent selection could favor leaf polymorphism, as an unusual or rare leaf morphology would be more likely to escape the attentions of ovipositing butterflies using shape cues. Our results support this and we suggest that this might be an example of “enemy free space” competition (Jeffries and Lawton, 1984; Brown et al., 1991) between Passiflora plants for survival against Heliconius caterpillars.

Heliconius females can show strong host plant preferences that may not be perfectly aligned with larval food preference and survival (Copp and Davenport, 1978; Smiley, 1978; Kerpel and Moreira, 2005; Silva et al., 2014). The ability to learn to associate new leaf shapes with oviposition sites may allow females to tailor their search image to the local Passiflora community. Specifically, in the case of H. erato, there are three important host species in the Gamboa area which have dramatically different leaf shapes, P. biflora, P. auriculata, and P. coriaceae (Merrill et al., 2013). Furthermore, there is considerable variation in all three species both between individuals and between young and old leaves. It seems plausible that visual searching behavior by H. erato could play a role both in promoting the coexistence of these three species, and as a selective pressure favoring the evolution and maintenance of within species leaf shape diversity.

The Janzen-Connell hypothesis proposes that interactions between parasites and their host could be a driving force in maintaining plant species diversity (Wright, 2002) and even egg coloration polymorphism (Yang et al., 2010). Here we have demonstrated the potential for behavioral plasticity in animal responses to play a role in maintaining plant species diversity. If generalist herbivores commonly learn a “search image” for locally common plant species this could be an important source of negative frequency dependent selection favoring rare plant species. In the highly diverse and complex tropical rainforest environment, such an effect might play a role in maintaining species diversity and in particular in sustaining populations of rare species.

DD and CJ conceived, planned and designed the study. DD and ML conducted the experiments. DD analyzed the data. DD and CJ wrote the manuscript.

Cambridge Trust (UK) and CAPES (Brazil) to DD, and Smithsonian Tropical Research Institute (Panama) to ML supported this study.