Species Diversity in a Wet Tropical Forest Food Web

Complete table of contents for layout elements

Keywords

Abstract

I. Intro to Heliconius, Passiflora and flea beetles
II. Flea Beetle survey
III. Comparison of Heliconius and Flea Beetle Biology and Community Structure
IV. Ant Predation
V. Cyanogenesis Survey
VI. Seasonality
VII. Aposematic Coloration and Mimicry
VIII. Species Analogs
IV. Species Diversity
X. Partitioning and Populating the Passiflora/herbivore Community
XI. Summary

Acknowledgements

Literature Cited

Table 1. Passiflora x Heliconius Interaction Matrix
Table 2. Flea Beetle x Passiflora Intraction Matrix
Table 3. Flea beetle adult feeding experimental feeding preferences
Table 4. Flea Beetle Larvae x Passiflora Interaction Matrix
Table 5. Cyanogens and other Allelochemicals of La Selva Passiflora
Table 6. Cyanogenesis in adult flea beetles
Table 7. Aposematic coloration and mimicry in Heliconius and Flea beetles
Table 8. Cross-community Analogs
Table 9. Niche analysis based on larval feeding mode
Table 10. Partitioning and populating the Passiflora community: determinants of species diversity
Table 11. Passiflora/herbivore community: summary of niche dimensional elements

Figure 1. Passiflora taxonomic subgroups represented at La Selva
Figure 2. Flea beetle larval morphology
Figure 3. Circle diagram illustrating flea beetle species barcode sequence groupings.
Figure 4. Ants vs flea beetles on Passiflora
Figure 5a-b. Total HCN Release from Crushed Passiflora species
Figure 6a-d. Within-species variation in HCN content for three variable species
Figure 7, Cyanogenesis in root and stem tissue of Passiflora
Figure 8. Kinetics of HCN gas release
Figure 9. HCN kinetics by P. megacoriacea showing extremely slow release
Figure 10. HCN release kinetics by P. quadrangularis showing two peaks
Figure 11. Changes in P. quadrangularis HCN release kinetics as leaves mature
Figure 12. P. auriculata branch HCN profile
Figure 13. P. biflora branch with unexpected complexities
Figure 14. Passiflora leaf thickness

Appendix 1. Procedure for measuring HCN production
Appendix 2. Gallery of HCN profiles for different Passiflora species
Appendix 3. Natural history observations and photographs (Passiflora, Heliconius and flea beetles)

Species Diversity in a Wet Tropical Forest Food Web

John Smiley, Research Affiliate, Office of Research Affairs, University of California, San Diego

I. Introduction and Background Wet tropical forests such as those found in the lowlands of northeastern Costa Rica have an exceptionally species-rich fauna and flora. Some of this richness is associated with a high degree of specialization, in which many species share habitats in specific, slightly different ways. Species richness can also be attributed to geographic differentiation, with many species replacements as one moves across the landscape. The Heliconius butterfly-Passiflora vine food web is a classic example where we know a great deal about both types of diversity, including multiple niche dimensions for both insects and plants. Much less-known are the Passiflora-feeding flea beetles (Coleoptera: Chrysomelidae: Galerucinae: Alticini) that parallel Heliconius in many ways, including species diversity and host plant relationships. The study described here is an attempt to describe this community, and compare it with the Heliconius community in its relationship to Passiflora. This comparison presents a tool for understanding plant-feeding insects. Since the primary commonality between the two insect communities is their association with Passiflora, I assume that community properties shared by both types of insect (such as species diversity) must derive from their relationship to the host plants. Other niche dimensions, such as adult floral resources for the butterflies, should be less important. My goal is to use this assumption to discover ecological processes that are important in structuring wet tropical forest insect communities. I chose to simplify investigation by restricting analysis to one field site: the La Selva Biological Station in the lowlands of northeastern Costa Rica (10°26'N/-83°59' W/elevation 50m). By doing this, I can build on my previous work at La Selva (Smiley 1978a, 1978b, 1982a, Gilbert and Smiley 1978) and draw on results from nearby sites in southwestern Costa Rica (Smiley 1985a, 1986). Work at La Selva began October 2012 and was essentially completed December 2015 after about 15 months of field and lab work

Heliconius butterflies

Heliconius are found in the neotropics from Southern North America through Central America and across South America as far as Bolivia, Paraguay and Brazil. Brown (1981) summarizes what is known of their distribution including subspecies extent as well as pockets of endemism. The Heliconius from La Selva are part of a suite of 16 "Transandean" species and subspecies. This area includes Ecuador and Colombia west of the Andes; Colombia and Venezuela west and north of the Orinoco plains; and Central America. Five of these species (and several other subspecies and color morphs) are restricted to small areas of endemism (in Guatemala and the Chiriquí region), or at high altitude. Only one widespread (i.e. non-endemic) Transandean low altitude species is not found at La Selva. H. ethilla does not reach Costa Rica from Panama. Thus it is reasonable to describe the La Selva Heliconius community as "complete" in the sense that nearly all the species in the Transandean bio-region are represented at the field site. Notably, three mimicry pairs are present including the black-red-and-yellow H. melpomene (Linnaeus 1758) and H. erato (Linnaeus 1758), the black-and-white H. cydno (Doubleday 1847) and H. sapho (Drury 1782), and the orange-black-and-white H. hecale (Fabricius 1776) and H. hecalesia Bates 1866. An additional mimicry pair, H. sara (Fabricius 1793) and blue form H. doris (=Laparus doris Linnaeus 1771) is potentially present, but the blue form of H. doris seems to be rare or absent from La Selva. The 10 Heliconius species from La Selva range across the Transandean region; i.e. none are locally endemic. A recent phylogenetic analysis (Kozak et. al. 2015) confirms Brown's 1981 classification of Heliconius into two principal lineages which I will call group I and group II. These may also be named by their principal host plant associations: the Decaloba/Astrophaea feeders and the subgenus Passiflora feeders. Kozak et. al. also nest the Brown (1981) Laparus doris within the Heliconius group II lineage and in this summary I will refer to this species as Heliconius doris. Table 1. Passiflora x Heliconius Interaction Matrix reflects these groupings as well as the Kozak et. al. ordering of the species.

Smiley (1978a) described the Heliconius community at La Selva, including estimates of population density, habitat preference and host plant relationships. The La Selva habitats include primary forest, forest edge, and second growth areas. Using these habitat designations Smiley found that the co-mimic species flew side by side in the same habitat, and that Passiflora abundance was reduced 100-fold in forest habitats as compared with early successional sites. He also found that Passiflora species were somewhat habitat-specific, with forest, forest edge and second growth species. The host plant interaction matrix, drawn from from Smiley (1978) and DeVries (1987), may be seen in Table 1. The final count of Heliconius species at La Selva included 10 species, 11 if you include the heliconiine Dryas iulia, which also uses new leaves of many of the same Passiflora as Heliconius. Smiley (1978) also found that Heliconius population densities were low at La Selva as compared with many other neotropical sites, probably because the relatively cloudy conditions are detrimental to butterfly flight.

Passiflora vines

Neotropical lowland rainforests typically support a Passiflora community of 10-15 species (Gilbert and Smiley 1978). What determines this number, given that the genus includes over 500 species? Passiflora biogeography has not been analysed as thoroughly as Heliconius, yet it is clear that the 12 low elevation species found at La Selva (not including two montane species that barely enter La Selva: P. nitida and P. lancearia) represent a fairly complete sample of the species found in the lowland wet forest of the "Transandean" region described by Brown (1981) for Heliconius. The assembled community is not random with respect to Passiflora taxonomic diversity. Passiflora has been classiflied into subgenera, supersections and sections (Feuillet and MacDougal 2003), and in only one case is more than one species from a section represented at La Selva. P. oerstedii and P. menispermifolia are classed together in the section Granadillastrum, yet these are widely different in foliage, habit and cyanogen chemistry. The other 10 species are each drawn from a different taxonomic subgroup (many of which are very species rich, totalling over 300 species; see Appendix 3). Although there are 34 other Passiflora subgenera, supersections and sections not represented in the 11 groups above, the great majority are from different geographic regions including the old world tropics, Amazonia, the Andean region, the drier parts of Mesoamerica and higher elevations nearby. Figure 1. Passiflora taxonomic subgroups represented at La Selva illustrates the diversity in habit and leaf shape among the taxa represented at La Selva.

Every one of the 12 La Selva Passiflora species is widely different in multiple characters, including habitat preference, woodiness, leaf shape, cyanogen chemistry, extrafloral nectary placement, leaf pelage, pollination syndrome and many other traits. One might think of them as separate genera or even separate families in terms of their appearance and ecological interactions. Eight of the twelve species range geographically across much of the Transandean region, with only four, arbelaezii, costaricensis, megacoriacea and lobata having a substantially lesser range (Nicaragua to Panama or western Colombia). Only P. quadrangularis, a feral cultivated species, ranges all the way across South America. Occasional human transport is possible for some of the other species, many of which are very attractive plants. All are currently maintaining wild populations at La Selva.

As mentioned above, the genus Passiflora includes over 500 species and various subgenera, supersections and sections. Eleven of the La Selva species belong to three of the largest subgenera: the shrubby Astrophaea, the small-flowered Decaloba, and the large-flowered subgenus Passiflora. The 12th, P. arbelaezii, has been assigned to the subgenus Tryphostemmatoides, but shares many characteristics with Decaloba such as flower size and herbivore susceptibility. In this work, I will usually treat P. arbelaezii as a member of the Decaloba group. The Heliconius group I lineage is associated with subgenera Astrophaea and Decaloba (and Tryphostemmatoides), while the La Selva group II lineage, including H. doris, is primarily associated with host plants in subgenus Passiflora. The symmetry of these subgeneric relationships , along with other evidence, have suggested that Heliconius coevolved with Passiflora as the plant genus diversified during evolutionary time. The arrangement of species in Table 1 reflects this symmetry. A phylogenetic analysis by Muschner et al. (2012) suggested that the main Passiflora subgenera are each monophyletic, arising 32-38 million years ago in Central America or adjoining South America. These authors also found evidence that subgenus Passiflora underwent its major species radiation more recently than Decaloba (10-15 million years as opposed to 30 million years). These divergence times differ substantially from the more recent divergence times reported for Heliconius in Kozak et. al. (2015), possibly negating the hypothesis of concurrent diversification of butterfly and host plant.

Heliconius x Passiflora relationships at La Selva

Other broad patterns may be seen as we examine the Heliconius-Passiflora relationship in Table 1. It appears that three members of the group II lineage, while mainly associated with subgenus Passiflora, also use Astrophaea/Decaloba as alternative, less preferred host plants. A fourth, H. melpomene, is specialized to use only P. oerstedii (and occasionally P. menispermifolia) in subgenus Passiflora, but its larvae are capable of feeding on many Decaloba species if placed there artificially. Unlike the other group II lineage Heliconius, H. doris is obligately monophagous on P. ambigua. In contrast, the five group I species (including D. iulia) were never seen to feed on subgenus Passiflora at La Selva, and their larvae die if placed on those species. Thus a basic asymmetry exists between the La Selva Heliconius lineages. Smiley (1985a) discusses this in more detail.

The La Selva Passiflora/herbivore community is embedded in a diverse, high biomass matrix of plant and and animal species. La Selva contains a mix of habitat types that include riverine forest, montane "foothill" forest and second growth forest. Crossing these forest types are roads, wide trails, narrow trails, mowed clearings and artificially maintained successional plots. The latter include five adjacent parcels approximately 0.5 ha in area, that are clear cut to the ground on a staggered five year rotation. Sampling at ground level in these habitats, Smiley (1978a) was able to classify Passiflora and Heliconius according to their stage of succession. Here, I simplify this quantitative measure by dividing species into "forest" and "second growth" categories, with old second growth transitioning into forest at about 20 years of succession.

At La Selva, most Passiflora vines are relatively small, and compete for light and soil nutrients among ~2100 other plant species. Passiflora-specialist herbivores have the problem of locating their relatively small host plants hidden in the "sea" of other plants. Once the plant is located they have to deal with the plants' defense mechanisms as well as predators or parasites in the plant's vicinity. Some of these parasites and predators have very strong impacts on fitness for the species under consideration.The most important known predators (e.g. ants and predatory wasps; Smiley 1985b, 1986b) are often spatially limited by territorial interactions or nest siting limitations, creating a mosaic in the forest matrix. The mosaic includes "bad" and "good" habitat patches as defined by the presence/absence of the predators specifically relevant to the consumer being considered. This creates a situation in which herbivores can be thrown into intense competition within mosaic patches where the predators are absent, and yet where the host plant population is seemingly untouched in the remaining protected patches.

Smiley (MS in preparation) has modelled such a habitat mosaic for Heliconius ismenius (Latreille 1817) based on measured values at Corcovado National Park in Southeast Costa Rica (Smiley 1986). In the absence of consumers, patches of this mosaic range in suitability for caterpillar growth and survival from very low to very high fitness values. However, in nature when consumers are present, the majority of mosaic patches will be unsuitable (low fitness) because resources in the more suitable patches will have been consumed. In this case, consumer survival arises from the relatively few mosaic patches which are marginally suitable. In these patches, resources persist and consumer growth/survival occurs. Modelling revealed that if fitness varies between patches by a factor of 10 to 100, then population fluctuations stabilize . The findings also suggested that two species can share the same resource yet experience reduced competition if they have different predators or other spatially patchy limiting factors. In such cases the mosaic patches in which their resources are found will be different and they will grow/survive on different individual plants or plant parts.

The dependence of H. melpomene on P. oerstedii may be another example where predation by a novel set of predators (microhymenopteran egg parasitoids) brings about host plant specialization (Smiley 1978b). Many group II Heliconius have egg laying behaviors that seemingly minimize ant predation, such as searching for small isolated ant-free plants and placing eggs on tendrils and leaf tips where they are less likely to be found. However, the tiny stalked petiolar nectaries of P. oerstedii are not very attractive to ants. Instead they are well-suited for visitation by microhymenopteran parasitoids (Hymenoptera: Encyrtidae and Scelionidae). At La Selva, H. melpomene has adapted its behavior to placing eggs out of sight in the cluster of tiny leaves at the shoot apex.

The leaf shapes in Passiflora have long been of interest to ecologists wishing to understand the selective forces that could bring about such enormous diversity (Gilbert 1982, Dell'Aglio 2016)). Does a novel leaf shape confer an advantage, perhaps hiding the plant from searching Heliconius or other herbivores? Do the searchers have a "search image" which can be countered with a novel leaf shape? Do plants hide in the background by mimicking the foliage of other, non-related plants, or does a novel leaf shape pre-adapt plants to occupy different habitats away from searching herbivores? It seems possible that searching Heliconius could create selective pressure for novel leaf shapes and plant habits in Passiflora, especially the more generalist species. Also, there are other non-specific herbivores in the habitat such as leaf-cutter ants (Atta cephaloides). Passiflora foliage is sometimes strongly preferred by foraging mammals such as collared peccaries (Pecari tajacu observed in the Passiflora garden at La Selva), and dried Passiflora foliage samples have been preferentially raided by Neotoma woodrats in California (JTS personal observation). The mammals may be attracted to the harmane alkaloids, perhaps seeking a narcotic effect. Flea beetles with their reduced mobility and smaller eyes seem less likely to respond to leaf shape differences.

Josiine moths

In addition to Heliconius, the notodontid moth tribe Josiini (Lepidoptera: Notodontidae: Dioptinae: Josiini) includes at least 102 species, most of which specialize on Passiflora (Miller 2009). Josiini have brightly colored adults and larvae. Miller has proposed that three of the larger genera are each associated with the major groupings of Passiflora as follows: Getta (7 species) on Astrophaea, Josia (20 species) on Decaloba, and Lyces (19 species) on subgenus Passiflora. About half of the Josiine diversity has no host plant information and many species are extremely rare. In the case of Josia, 6 of the 10 species with host plant information feed on a single Decaloba subgroup Xerogona. One of these species (Josia frigida) is found at La Selva, with larvae specializing on P. costaricensis, a Xerogona species (See Appendix 3; P. costaricensis for photos of the moth and larva). According to Miller (2009) Costa Rica hosts at least two other Josiini: Getta tica with larvae on Passiflora tica in the subgenus Astrophaea, and Lyces cruciata with larvae on P. menispermifolia. The foliage of both P. costaricensis and P. menispermifolia is covered in long hairs, both possess few attending ants, both are strongly cyanogenic, and neither are preferred hosts for Heliconius at La Selva (see below for more on these characteristics). The Josiini thus offer another example, parallel to Heliconius, of a clade of herbivores adapting to Passiflora and specializing on the principal subgenera. Miller (2009) suggested that the Josiini may have evolved concurrently with the evolutionary radiation of Passiflora.

Flea beetles

In addition to Heliconius, Passiflora also host an equally diverse assemblage of flea beetles (Coleoptera: Chrysomelidae: Galerucinae: Alticini). Unlike Heliconius and Passiflora, almost nothing is known about neotropical flea beetles. Basic natural history, including host plant relationships, community ecology, immature stages and species level taxonomy are unknown for most genera. Also, dried specimens quickly lose color in museum collections, obscuring vibrant adult coloration. Smiley (1982a) presented evidence that, at La Selva, the flea beetle community exhibits many parallels to the Heliconius community. Parallels included similar species diversity and a similar but not identical pattern of feeding preferences across Passiflora subgenera. Since the only common factor is the sharing of the host plant community and ambient environment, Smiley (1982) proposed that species richness in these communities is controlled by host plant relationships. Pollen availability, capacity for speciation, selection for aposematic coloration, searching efficiency, and other factors which might be considered to influence Heliconius community structure (Gilbert and Smiley 1978), must be relatively less important. Whether or not Smiley (1982) is correct, it seems likely that knowledge of the flea beetles can can help us understand the community ecology of Heliconius, and the other way around.

The biogeography of the flea beetles has not been looked at, and range data has not been compiled for most species. Furth et al (2003) surveyed Alticini at La Selva extensively using sweep sampling, fogging and other unselective sampling methods, and found 68 genera and 247 species. They also estimated the true total for Costa Rican Alticini to be 1000 species and for the world, probably 500 genera and 8000 species. A biogeographic survey of the Passiflora-feeding species could be revealing. It would be very interesting if, for example, the 10 Passiflora feeding flea beetle species shared Brown's Transandean region with the 10 La Selva Heliconius. However, the only comparison for lowland rainforest Passiflora-feeding flea beetles is between La Selva and Corcovado National Park, in the in the Chiriquí area of endemism in Southwestern Costa Rica and extreme Western Panama. As is the case with Heliconius, many of the Chiriquí flea beetles are quite different from those at La Selva. Catherine Duckett worked in Corcovado National Park and produced a key to the alticine species found on Passiflora pittieri. Comparing that key to the flea beetles from La Selva reinforces the conclusion that the flea beetle beetle community is quite different.

Even with incomplete knowledge of the host plant relationships of most flea beetles, two kinds of evidence suggest that the Passiflora-feeding flea beetles beetles are not, like Heliconius, a monophyletic lineage specifically associated with Passiflora. First, some of the genera including Disonycha have published records of host plants in other families besides Passifloraceae. Second, at La Selva I have observed profound differences in the morphology of the juvenile stages of flea beetles represented among the Passiflora feeders. The genera Monomacra and Parchicola have cylindrical root- and stem-feeding larvae with tiny spherical "balloon" organs on the dorsal and lateral surfaces of the body. These appear to be modified setae, the function of which is unknown. Their eggs are linear and cylindrical, and are laid near the base of the plant in clutches of about 10-20. Disonycha larvae have differently shaped balloon organs and eat leaves (I haven't seen their eggs). Pedilia lack balloon organs but they eat stem and leaf tissue and lay clutches of linear cylindrical eggs end-to-end on the new leaves. The genus Ptocadica have leaf-eating, tapered larvae with no "balloon" organs and lay single ovate eggs on leaves. I have not yet observed the juvenile stages of Monomacra chontalensis, Yellow Ptocadica or Black-tibia Parchicola but I assume they are not too different from their congeners. Photos of these previously unknown juvenile stages may be found in Figure 2. Flea beetle larval morphology, below, or by selecting the flea beetles in Table 2. Flea Beetle x Passiflora Interaction Matrix and following the links. Based on egg and larval morphology I would group Parchicola, Monomacra, Disonycha and possibly Pedilia together in one or more possibly related groups, with Ptocadica in a separate, unrelated group.

II. Flea beetle survey at La Selva. In 1975, while sampling Passiflora for Heliconius eggs and larvae, I discovered small numbers of colorful flea beetles on the foliage of some of the Passiflora. I began recording these observations and collecting specimens while walking Heliconius transects. Over time this resulted in a collection that, after examination by staff of the US National Museum, revealed the presence of five species in four genera as well as 3-4 morphospecies without generic names. The results of this collection formed the basis of the Smiley (1982) article referred to above. Then, in 1991 I returned to La Selva and re-discovered the flea beetles, finding several of the common species in just two days of searching. I made another short visit in 2009, and spent several days searching, again finding most of the 1975 survey species. Encouraged by the apparent stability of the flea-beetle populations, I returned for slightly longer surveys in 2010 and 2011. The 1991-2011 surveys augmented the original 1975 flea beetle x Passiflora interaction matrix without adding any new species. In 2010 I brought my voucher collection to David Furth of the United States National Museum, and he brought the nomenclature up to date, assigning modern generic names and splitting one morphospecies "Yellow Monomacra" into two Parchicola species, "DF1" and "DF2". The resulting list included 9 Passiflora-feeding species.

Genetic barcode analysis brings the species list to 10

After collecting more beetles in 2012-2014, I collaborated with Carlos Garcia-Robledo, then at the US National Musuem. We conducted a genetic "barcode" analysis of the cytochrome oxidase 1 "Fulmor" region on 110 La Selva flea beetle specimens. Collected beetles were preserved in 95% EtOH. One leg of each voucher specimen was removed and DNA was extracted. Amplification of the mitochondrial gene cytochrome oxidase COI was conducted in 96-well plate formats using the COI Folmer primer. PCR was followed by ExoSap purification of amplified products and then subjected to standard sequencing using BigDye Di-Deoxy terminator sequencing. Sequences of beetles were aligned using multiple sequence alignment. To determine the accuracy of using DNA barcodes to identify the beetle species included in this study, we calculated the similarity of each COI sequence to each of the other sequences measured as the percentage of bases/residuals that are identical. We then calculated the average inter and intraspecific similarities among sequences and estimated the DNA barcode gap, i.e., the similarity threshold at which a COI sample included in this study can be identified as a conspecific (similarity>95%) or as a heterospecific(similarity<85%). We generated a neighbor-joining tree and estimated bootstrap support after 100 replicates using Geneous Pro V 5.6.5. (Biomatters-development-team, 2012). Figure 3. Circle diagram illustrating flea beetle species barcode sequence groupings illustrates the results.The analyses supported all of David Furth's species designations except that Parchicola "DF1" included two "cryptic" species which I designated "yellow-tibia" and "black-tibia" after finding a characteristic color difference in the hind legs. The analysis also supported the existence of two species within David's Ptocadica "straminea" which I had previously designated Yellow Ptocadica and Red Ptocadica based on the color of the elytra, colors seen only seen in life but obscured in museum collections. Four individuals of Parchicola DF2 included more variation than the remaining nine specimens, and it may be interesting to further analyze this species.

With these corrections and changes, the "completed" 2016 La Selva collection included 10 flea beetle species known to feed on Passiflora, 4 with binomial species names and 6 with generic names but no species name. I then compiled my field notes and created a flea beetle x Passiflora interaction matrix, with over 1600 observations (see Table 2). In this matrix I ordered the flea beetle species with Astrophaea/Decaloba-feeders on top and subgenus Passiflora-feeders on the bottom, to facilitate comparison with the Heliconius matrix. It is important to note that, except for Red Pedilia, the species in Table 2 were all present in 1975 during the initial collection, suggesting a steady, stable ecological community over a 40 year interval. Red Pedilia was undoubtably present in 1975 as well, but was not sampled owing to the rarity of its host plant, P. pittieri.

Although I have not attempted to quantify the abundance of flea beetles on Passiflora at La Selva, their ground-level numbers are within an order of magnitude of those of Heliconius (1-70 adults per hectare, depending on habitat) as determined by Smiley 1978. Their population densities are small, and many hundreds of hours of observation were needed to produce the data in Table 2. Perhaps the dense forest matrix and the lack of an extended dry season contribute to the persistence of low density populations of flea beetles, Heliconius butterflies and Passiflora plants. In all my hours of survey I have never seen any of the flea beetles in Table 2 feeding on plants other than Passiflora, in spite of the fact that they often rest on any adjacent plant species. Like Heliconius, they appear to be obligately specific to Passiflora as food plants.

Methods for working with flea beetles

I captured adult beetles on plants using two ounce polypropylene condiment cups with snap-on lids. Being careful not to touch or shake the plant, I positioned the open cup over the beetle. I then positioned the lid opposite the cup, under the leaf. The leaf (and beetle) were then trapped by gently bringing the lid and cup together. When contact is made, the beetle hops into the cup and the cup plus opposing lid are gently slid off the leaf leaving the beetle trapped in the cup. Delicate larval beetles were captured by removing their substrate leaf and placing the leaf in a cup. Cups with beetles were brought into the laboratory at La Selva.

I employed several techniques to investigate the feeding ecology of the flea beetles. I confined adult beetles in "Glad" brand plastic 15x20x10 cm containers. I would add small cuttings of Passiflora vines, with the cut end immersed in a 10 ml vial with cotton to prevent water leakage. Each day I would record the quantity of leaf eaten by visually estimating the area removed. I tested the seven common flea beetle species against the most common Passiflora species. The results may be seen in Table 3. Flea beetle adult feeding experimental feeding preferences. I developed a method for manipulating the beetles by drilling holes in the side of the containers and introducing CO2 through a tube. This would immobilize the beetles for 30-60 seconds, enough time to move them from one container to another. Containers were kept at ambient temperature and humidity in the La Selva "Ambient Laboratory."

I also confined flea beetle larvae in 15x20x10 cm containers with Passiflora cuttings and measured leaf consumption. Measurements were limited to the leaf-feeding larvae of Red Pedilia and Red Ptocadica, the only species with sufficient numbers of individuals for experimentation. To obtain larvae and observe life history information on other species, I confined 5-10 adult beetles with potted Passiflora vines in 60 x 60 x 90 cm "Bioquip" mesh cages. Only the most common Passiflora x flea beetle combinations were tested. These were kept in an outdoor "shade" house and checked once or twice weekly for sign of mating behavior, eggs, larvae or new adult recruitment. In this way I was able to determine larval feeding mode (root/stem or leaf) as well as obtain specimens and photographs of these previously unknown juvenile forms. Eggs and larvae of root/stem-feeding species were kept in glass petrie dishes with moistened filter paper and Passiflora rootlets extracted from leaf litter at the base of live plants, and were examined and photographed through a stereo microscope. Larva/host plant interaction relults are summarized in Table 4. Flea Beetle Larvae x Passiflora Interaction Matrix and natural history observations and photographs may be found in Appendix 3.

III. Community Parallels and Similarities Comparison of the Interaction Matrices in Tables 1 and 2 reveals strong parallels between flea beetle and Heliconius host plant relationships, as well as some important differences. Species richness is about the same (10-11 species). Both herbivore communities include 2-3 "generalist" species that are capable of eating Passiflora from both major subgenera. Both include 3-4 species which specialize on subgenus Passiflora, both include one species that specializes on subgenus Astrophaea, and both include 5-6 species that specialize on subgenus Decaloba. The biggest difference is in the degree of host specificity; three Heliconius are obligately monophagous with 2 others facultatively so, while only one flea beetle (Red Pedilia) is monophagous. However, it is important to note that 95% of the records in Table 2 are from adult flea beetles, and larval flea beetles are much more difficult to find and identify. Some species appear to be more host-specific as larvae (e.g. M. violacea; see below). Nevertheless, several examples suggest that most flea beetle larvae are usually found on the plants preferred by the adults. For example, adult Ptocadica bifasciata are most commonly found on auriculata and biflora, and larvae found on those plants were verified by genetic barcode as belonging to that species. Red Ptocadica larvae were always found on P. lobata, the strongly preferred host for adults of that species. Caged Parchicola species (DF2 and Yellow-tibia) and Disonycha quinquelineata were all found to lay eggs and reproduce on plant species preferred by the adults (Table 4. Flea Beetle Larvae x Passiflora Interaction Matrix). Another similarity between the two communities is that Passiflora species favored by Heliconius (P. pittieri, P. lobata, P. auriculata, P. biflora, P. vitifolia, P. ambigua and P. oerstedii) are also favored by flea beetles. The remaining species are seldom fed upon by either group.

Like Heliconius, some species of flea beetles may live long lives as active adults. One Monomacra chontalensis was caught on a wild plant on 12/2/13 and lived in captivity on a P. biflora for over 3 months, until 3/8/14. Another lived for 2 months in the same cage. Other caged flea beetles, for example, Ptocadica species, lived only 2-4 weeks. Heliconius are known to live for several months as reproductively active adults, not in diapause. Such longevity may enable the insects to locate widely scattered resources and endure periods of unfavorable conditions.

The Heliconius and the Passiflora-feeding flea beetle communities appear to be structured in parallel, with similar species diversity across Passiflora subgenera. This suggests a common set of Passiflora traits and attributes that both orders of herbivores independently exploit (and compete for) without interference from the other. In other words, flea beetles and Heliconius do not compete with or exclude each other from the La Selva community, but flea beetles do compete with (and exclude) other flea beetles and Heliconius compete with (and exclude) other Heliconius. The observation that the two communities saturate with species in parallel, on the same host plant species, suggests that the host plant resource has a powerful structuring effect. What are the resources in this complex competitive framework, and how are they divided?

For these communities there are many candidate traits that might affect community diversity and structure. These include habitat specificity, host plant size preference, host plant extrafloral nectaries, ant/wasp predation, egg placement, pollen competition, mimicry space, leaf shape diversity, root ecology, response to wet/dry seasons, degree of monophagy, plant taxonomic diversity, plant allelochemicals and probably other factors as well (Gilbert et al 1975). Allelochemicals include harmane alkaloids, tannins, flavonoids, and cyanogenic glycosides. Some of these traits would seem to to be irrelevant to the flea beetle community; for example, pollen competition and leaf shape. Likewise, flea beetles may respond to host plant root ecology (some of the flea beetles have root-feeding larvae), something perhaps irrelevant to the butterflies. In contrast, other factors such as predation, plant chemistry, mimicry selection (aposematic coloration) and habitat (stages of succession and seasonality) will likely have strong effects on both types of herbivores. Next I will examine some of these factors in the light of new life history information revealed by the flea beetle survey, and will suggest possible roles in community structure for the butterflies as well as the flea beetles.

IV. Ant Predation and Passiflora Extrafloral Nectar Glands Most Passiflora have extrafloral nectar glands on their leaves and petioles that secrete sugars and amino acids, attracting ants and wasps to the plant (Apple and Feener 2001; Durkee 1983). Ant/wasp predation is severe on fast-growing, soft-bodied Heliconius larvae, and most larvae die if their host plant is attended by ants (Smiley 1985b and 1986). Larger plants are more likely to recruit and retain ant attendants (Smiley 1978), and the oviposition behavior of many Heliconius includes a preference for small isolated plants. In contrast, flea beetles are most easily found on large patches of host plant with multiple growing shoot tips. Adult flea beetles possess a powerful spring mechanism in their hind legs that employs over 100g of acceleration to escape predators. They allow potential predators to approach closely, springing away only when they are actually touched. After their spring, which takes them 20-30 cm almost instantly, they open their wings and fly to a soft landing, usually less than a meter away. This adaptation makes adult flea beetles relatively "ant-proof," and it is common to find flea beetles with what appear to be ant mandible scars on their soft elytra.

Analyzing over 1200 independent field observations over a five year period, I found that Passiflora with ants in attendance also have more adult flea beetles (Figure 4). Furthermore, I found that plants with Ectatomma tuberculatum ant have more flea beetles than plants with "other" ant species. The data in Figure 4 are conservative in that many plants in the "no ants seen" category may well be tended by ants at other times. It would be interesting to study the Passiflora/Ectatomma relationship in more detail. How many plants are colonized, and how large do they have to be? Are nests located at the base of the plants? How far will they forage? Do Ectatomma repel other predators?

Larval flea beetles also seem to be well-protected from ant predation, although this has not been demonstrated experimentally. For Ptocadica, Pedilia and Disonycha stem- and foliage-feeding larvae, the body shape is slug-like, and covered with rounded "hill and valley" protuberances, very much like the ant-tended larvae of lycaenid and riodenid butterflies. The dorsal surface is expanded over the head and legs, hiding them from view. Most ants ignore these larvae, even when they crawl along a stem covered with worker ants moving up and down the plant. Parchicola and Monomacra larvae may also be ant-resistent, with a similar slug-like appearance. They feed on stems and exposed rootlets at the base of the plants. Both Disonycha, Parchicola and Monomacra larvae have microscopic balloon-like organs that may also function in antipredator defense, perhaps dispensing a chemical of some kind (see Figure 2, flea beetle larvae). Probably as a result of their effective predator defenses, flea beetles are able to "infest" larger plants on a long-term basis, plants likely to be attended by ants. They are also slow-growing, making them more vulnerable to discovery (see Appendix 3 descriptions of flea beetle larvae).

V. Cyanogenesis Survey. Passiflora Allelochemicals are diverse, and include tannins, alkaloids, cyanogenic glycosides, flavonoids and other compounds. Table 5. Cyanogens and other Allelochemicals of La Selva Passiflora includes a partial summary of compounds discovered in the La Selva species. It appears that many species contain small amounts of harmane alkaloids while only a few have tannins (Smiley and Wisdom 1985). Nearly all species are known to contain cyanogenic glycosides, and many species have unique forms of these compounds. C-glycosylflavonoids have been found in some species including P. menispermifolia at La Selva (Ulubelen et al 1981).

Cyanogenic glycosides are relatively stable compounds that, if mixed with a specific β-glucosidase enzyme, release toxic hydrogen cyanide gas. They are diverse and widespread in Passifloraceae and related plant families, and most Passiflora synthesize one or more types. Compounds are reputedly stored in vacuoles in the leaf and stem tissues, and the corresponding β-glucosidase enzymes kept in the cytoplasm, physically separated from the vacuole contents (Spencer 1988). Mixing occurs when plant tissues are damaged or crushed, releasing toxic cyanide gas that poisons mitochondrial respiration in eukaryotic cells. Most plants and animals have a protective enzymatic system against cyanide poisoning, but if the system is overwhelmed then death occurs rapidly.

There are at least five distinct biosynthetic pathways for creating cyanogenic glycosides, all found in Passiflora. Table 5 lists the known types for the La Selva Passiflora, along with some other allelochemicals found in these species. The diversity is remarkable, and each species has a distinct cyanogen profile. If herbivores are able to detect these compounds and exploit them for mating success or predator deterrence, as has been shown for Heliconius sara and H. hewitsoni, there is great scope for host specialization and diversification.

Method for measuring cyanogenesis

Using a "Gasalert Extreme" portable cyanide gas meter (made by BW Technologies/Honeywell) developed for emergency personnel entering hazardous situations, I developed a method for measuring parts per million (ppm) of HCN gas produced by crushing plant and animal tissues. This procedure employs a "Glad" brand 150 ml plastic container with snap-on lid to hold and concentrate the HCN released from the sample, and a BW Technology "Sampler" 5ml/s air pump to deliver the gas from the cup to the meter. The technique is moderately sensitive, enabling quantification of just a few hundredths of a micromole of HCN gas per gram of tissue and is reasonably precise, yielding values with less than 10% variation around the mean. Unfortunately, about half of the HCN is lost during measurement, so that the measured parts per million are about 50% of the true absolute amount. The percentage loss is consistent, however, and as I discovered, is insignificant in comparison with the measured variability in the plants. To make more accurate measurements, I also developed a "closed glass container" procedure. This technique gives readings without gas losses that are as accurate as the instrument allows. However, given the equipment available, this procedure was much less sensitive and much more time-consuming than the plastic cup technique. Unless otherwise specified, all values below were determined using the "plastic cup" procedure, and are thus about 50% of the actual amount of HCN present. See Appendix 1 for more detail on measurement technique.

Cyanogenesis in Passiflora and flea beetles

Figure 5a. Amount of HCN gas released from crushed Passiflora tissues reveals five orders of magnitude variation within and between Passiflora species. Figure 5b, showing results from new leaves only, also reveals extensive variation, including 2 orders of magnitude within the more variable species. All species exhibited HCN gas release, although in the case of P. lobata it was mainly detected in the roots, in very small amounts. Two other species, P. oerstedii and P. vitifolia, also released very small, barely detectable amounts. Engler (1998) found that P. oerstedii actually contains moderate amounts of the cyanogenic glycoside gynocardin, but that the plant lacks the correct β-glucosidase enzyme to release HCN. In addition, P. vitifolia has been found to contain up to 15% dry weight of tannins, an amount that might interfere with hydrolysis and HCN release. These findings bring up the possibility that some Passiflora evolve altered HCN profiles including loss of HCN capacity. Four other Passiflora produced consistent amounts of HCN >1 μM/g (microMole HCN gas per gram fresh weight plant tissue), including P. menispermifolia, P. pittieri, P. arbelaezii, and P. costaricensis. The rest, including P. ambigua, P. biflora, P. megacoriacea and P. auriculata produce moderate but highly variable amounts of HCN.

I investigated HCN variation in more detail, analyzing the three most variable species, P. auriculata, P. ambigua and P. biflora. I located 11-16 individual plants of each species, each with multiple branches. I collected three leaves from each branch, including a new leaf not yet fully expanded, a full-sized new leaf, and an older leaf. In about 50% of all cases I could verify the common origin of each branch by tracing the stems downward and locating the connection, but in the other 50% the stems ran under logs and debris and could not be followed. In those cases I reasoned that the spatial separation between plants was sufficient that I could assume a common connection. I did not use leaf appearance, maturity, or degree of shading as criteria for identifying connected branches. The results of over 450 measurements expressed on a log-10 scale revealed the main sources of variation for each species: within branch, between branch, and between plant (see Figure 6 a-d. Within-species variation in HCN content). The species differed, with more between-plant variation in P. auriculata, and greater within-plant variation for P. biflora and, to a lesser extent, P. ambigua. In all three species variation was huge; well over an order of magnitude for all plants and approaching two orders for others. Within-branch variation was usually much smaller but still large, with 3 to 5-fold variation. This amount of variation is surprising and has not been previously reported for Passiflora, although experimental studies with Passiflora edulis have shown that branches from the same plant may often be decoupled in their response to stress signals (Izaguirre, et al 2013).

Like leaves, the roots and stems of Passiflora produce variable amounts of HCN when crushed. For most species the amounts are roughly correlated to the amounts in the leaves, but some are different although sample sizes are small. For example, P. biflora roots produced low amounts of HCN as compared with the leaves, while P. auriculata roots produced higher amounts. Also, P. ambigua produced relatively high amounts as compared to its foliage, but P. pittieri produced lower amounts. Figure 7. Cyanogenesis in root and stem tissue of Passiflora summarizes the results. These differences may turn out to be important, as half the flea beetle species depend upon root and stem tissue for larval development.

My HCN measurements on insect tissue were consistent with information in the literature: Heliconius of all stages are themselves cyanogenic when damaged or crushed. However, measurements of Heliconius sara and H. doris larvae actively feeding and growing in closed chambers reveal that biting, chewing, feeding larvae do not emit HCN gas in any measureable amount. Apparently these caterpillars have some mechanism to prevent HCN release while feeding, perhaps an enzyme inhibitor in the saliva. For example, no measureable HCN was released over 2-3 days in a closed 10 liter chamber, for 10 late instar H. sara larvae feeding in tandem. Similarly, H. doris larvae actively feeding on host P. ambigua did not release any measureable amount of HCN gas, even after consumption of 2 full-sized leaves. Crushing these leaves mechanically would have released about 5 parts per million HCN in the closed 10 liter bell jar. The limit of detection for the meter was 0.3 ppm, so the reduction in HCN emission for those feeding larvae was at least 94% if not more.

In contrast, Passiflora-feeding flea beetles present a very different result. Although their small size makes measurement difficult, no HCN has been detected in their tissues (limit of detection between 0.1 and 0.01 μM/g fresh weight) after crushing adults of several species and larvae of one species, Red Pedilia (Table 6. Cyanogenesis in flea beetles and Heliconius). In addition, examination of Table 2 and comparison with Figure 5 reveals that as a group flea beetles tend to avoid the most cyanogenic plants. Only Red Pedilia uses a highly cyanogenic host (Passiflora pittieri); all the others are commonly found on species that frequently produce reduced amounts of HCN from their tissues. This suggests that, as a group, flea beetles tend to be HCN avoiders, in marked contrast to Heliconiines.

When feeding on Passiflora, flea beetles eat many small, separate "meals", both adults and larvae. The result is that host leaves become "peppered" with 1-2 mm diameter feeding holes and scrape marks where their small jaws removed one layer of the leaf without biting through. Is it possible that flea beetles avoid cyanide poisoning by feeding slowly and allowing HCN to diffuse away before building up toxic concentrations? Also, second and third instar Red Pedilia larvae feed primarily on stem epidermis of P. pittieri, tissues with reduced HCN amounts. Parchicola and Monomacra larvae feed on stems and surface roots, tissues with low HCN amounts for some species, and Ptocadica often eat leaf veins, stipules, tendrils, stem tissues and single leaf layers, some of which may be low in HCN. It would be valuable to test flea beetles to see if they release HCN while feeding, but this is beyond the sensitivity of my equipment. Passiflora-feeding flea beetles (like many other beetles) can survive for many hours inside a killing jar with HCN gas.

Even though the Passiflora feeding flea beetles, unlike Heliconius, do not produce quantities of HCN when damaged, they are still brightly colored and highly visible. Are they chemically defended? Perhaps they are able to sequester the toxic cyanogenic glycosides or their aglycones. The bright colors might also advertise escape capability.

HCN kinetics and cyanogenic glycoside mobilization within Passiflora

The HCN measurements have revealed several kinds of variation in HCN production that might affect growth and survival of herbivores. Recording at 20-second intervals, Passiflora release HCN slowly at first, presumably the time required for cytoplasmic β-glucosidase enzymes to mix with their substrate cyanogenic glycosides. For most species, gas release peaks between 100 and 200 seconds, depending on the species. Release then tapers off in an exponential decay characteristic of enzymatic hydrolysis. In most cases, 90% or more of the total HCN is released in the first 5 minutes after crushing. P. ambigua, P. menispermifolia, P. biflora, P. arbelaezii and P. costaricensis peak at about 100 seconds. Two other species, P. pittieri and P. auriculata peak at about 200 seconds but the release curve looks similar otherwise (see Figure 8).

At the other extreme is P. megacoriacea with extremely slow release of HCN (Figure 9). The rate at which HCN is released from crushed leaves is about 10 times slower that of other La Selva Passiflora, taking over 2 hours instead of 15 minutes to release 99% of the gas. This could conceivably deter herbivory by reducing the HCN "footprint" for searching female Heliconius while retaining toxicity in the herbivore gut. The close relative P. coriacea contains the simple monoglycoside cyclopentyl cyanogens Volkenin and Epivolkenin, the complex diglycoside cyclopentyl Passisuberosin and Epipassisuberosin, and one of the aliphatic cyanogens (Engler 1998; Spencer 1988). Perhaps the slow release results from interference between the β-glucosidase enzymes specific to these various cyanogenic glycosides, as proposed in Spencer (1988). Alternatively, slow release could result from the presence of only tiny amounts of β-glucosidase enzymes.

Finally, P. quadrangularis appears to have 2 distinct cyanogen release systems, one of which is "fast release" and one "slow release". The fast release system is much like that of most other Passiflora species in that most of the cyanide gets released into the air within 5-10 minutes after crushing the leaf. The slow release system is quite different, starting right away but releasing a constant amount over a 30-40 minute period of time (Figure 10). The measurements suggest that the slow release system is primarily found in the new leaves, and declines to low levels (10-15% of the total) in full sized young leaves and mature leaves that are beginning to harden (Figure 11).

HCN Branch Profiles

HCN measurements along Passiflora branches typically show 1-2 peaks, sometimes in the young leaves, sometimes as leaves mature and sometimes both (Figure 12). Amounts usually range over a factor of 2 to 5. However it is quite common to see very different patterns, even within the same species. At times the plants seem to be withdrawing cyanogenic glycosides, with newer foliage having no measureable amounts and older leaves with moderate amounts (Figure 13). Damaging leaves, either by caterpillar feeding or incidentally by cutting leaves for sampling, sometimes results in changing amounts of cyanogenic glycoside ("induction"). Changes may include a doubling after 24h within the same leaf and no change to adjacent leaves, and other damage effects have been documented, including no change at all. Appendix 2. HCN profiles includes a "gallery" of HCN production profiles for Passiflora, highlighting the great diversity. The overall picture is that HCN release, as caused by changing concentrations of cyanogenic glycosides, seems to be under active control by many Passiflora species. For some species this control includes radical reduction for reasons still unknown. Is there between-plant variation in average amount? Is there genetic control? Do different kinds of herbivory induce different reactions in the foliage? These questions remain unanswered.

Since Heliconius are not harmed or deterred by HCN production and may benefit with enhanced growth rates or improved predator deterrence, could searching female Heliconius be attracted to HCN leaking from plants? There is some evidence that this is the case for H. erato searching for P. biflora (Burkholder 2008). In contrast flea beetles taken as a group seem deterred by HCN amounts greater than 1 μM/g. Thus Passiflora may be "trapped" between opposing selective agents; reducing HCN to avoid Heliconius and increasing HCN to avoid flea beetles. This opposing selection may be responsible for some of the variation discussed above. Among the cyanogenic Passiflora in Table 2, the species with the most variable HCN concentrations are those attacked most heavily by Heliconius and flea beetles.

Probably the best plant to investigate is P. auriculata. Heavily attacked by Heliconius sara and adults of three flea beetle species, P. auriculata is perhaps the most variable Passiflora in terms of foliar HCN quantity. Flea beetles tend to attack P. auriculata with lower amounts of HCN (See Appendix 3; P. auriculata). Does H. sara prefer plants with higher amounts of volkenin, the principal cyanogenic glycoside in P. auriculata? Does P. auriculata respond differently to butterfly herbivory, as opposed to flea beetle damage? Other highly variable species that would be good to look at are P. ambigua and P. biflora. P. oerstedii would also be interesting if we could find a way to measure the content of gynocardin. To do this we would need a source of the appropriate β-glucosidase enzyme.

If this hypothesis of opposing selection stands the test of further investigation, it suggests an interesting kind of mutualism between Heliconius and flea beetles: the presence of one type of herbivore may alter the plant community in a way that is beneficial to the other type. In this way Heliconius might compete for Passiflora (with competition structuring the butterfly community), while at the same time inducing or selecting for reduced cyanogen content. Flea beetles would then compete for the reduced cyanogen plants (with competition structuring the flea beetle community), inducing or selecting for increased cyanogens. The presence of one type of herbivore in the habitat would thus facilitate herbivory by the other type. This could lead to a situation where some Passiflora species are heavily used by butterflies and flea beetles, while other Passiflora such as P. megacoriacea, P. arbelaezii, P. costaricensis and P. menispermifolia, are relatively free from both kinds of herbivory (see Tables 1 and 2).

Passiflora Leaf Thickness

Smiley (1978a) observed a positive correlation between Passiflora leaf thickness and caterpillar groups size but did not offer actual leaf thickness measurements. To remedy this I located individuals of 11 species of Passiflora along roads and trails at La Selva, and one individual of seven species in the Passiflora garden, and measured leaf thickness. 1-3 representative branches of each Passifora species were chosen from each plant. For each branch, I selected one new leaf, about 50% of full size, one mature leaf, full-sized with mature-leaf coloration, and one old leaf showing signs of senescence such as epiphytic growths or discoloration. The thickness of each leaf was measured using a 0-150mm caliper with digital readout ("Mr. Toolz IP54"), care being taken to avoid major leaf veins. Moderate pressure was applied to the caliper roller until the 2.5 mm-wide flat surfaces of the caliper jaws were snugly closed on the leaf, as determined visually, and further tightening would result in damage to the leaf. The values were then recorded and the graphic comparison shown in Figure 14.

VI. Seasonality La Selva is located in a relatively wet climatic region: the lowland forest (elev. 35m) right at the base of the Central American cordillera, 55km from the Caribbean sea. Smiley (1978) found that Heliconus were relatively rare in this habitat as compared with other locations such as the Chiriqui area of endemism in southwestern Costa Rica. For example, during August 2015, after three months of unusually persistent clouds and rain, I found Heliconius populations greatly depressed. Normally common species such as H. cydno and H. hecale were not seen flying, nor did I see their eggs on Passiflora as would normally occur. I did see H. melpomene and H. doris. The only common species, even in the successional areas, was H. sara. In contrast, during that same period all seven of the common flea beetle species were found, although in reduced numbers. Mating was observed in several cases, as were a few aggregations of 5 or more beetles, suggesting successful recruitment of new adults and participation in normal social behavior. The only flea beetles not found were three rare species, difficult to find in any season: Monomacra chontalensis, Disonycha quinquelinata and Yellow Ptocadica. Rainy or cloudy weather, a dominant feature of the habitat at La Selva, appears to affect Heliconius more than flea beetles. The butterflies probably suffer greatly reduced flying time, while both types of herbivores are affected by reduced growth and availability of host plant.

VII. Aposematic Coloration and Mimicry Heliconius species belong to several distinct mimicry groups. In fact every species at La Selva with the exception of the zebra-striped H. charithonius, has a strikingly similar co-mimic flying in the same habitat (Table 7. Aposematic coloration and mimicry in Heliconius and Flea beetles, rows 1-8). This has been studied at La Selva and elsewhere, and has been found to be evolutionarily dynamic. Genetic analysis has revealed that coloration evolves rapidly and changes with the presence or absence of suitable co-mimics (Mallet, 2009; Dasmahapatra, et al. 2012). This explains how the nearly perfect mimicry pair at La Selva (H. cydno and H. sapho; large white patch on black wings) can cross over the mountains to the Chiriqui area of endemism and evolve a totally different color pattern (H. pachinus and H. hewitsoni; two yellow stripes on black wings). In spite of known gene flow and hybridization, selection for mimicry is strong enough to result in a nearly perfect match between the latter two species.

Do brightly colored, presumably aposematic flea beetles exhibit a similar degree of color convergence among species found together on the same plant? Table 7 (rows 9-13) shows that the Astrophea-Decaloba feeders do not converge. Each of those species has a strikingly different color pattern. For example, it is quite typical for a P. auriculata plant to simultaneouly host yellow Parchicola DF2 and red-brown-white Ptocadica bifasciata.

In contrast to the above, the subgenus Passiflora-feeding flea beetles suggest the possibility of mimicry. The two yellow Parchicola species are nearly identical, and Yellow Ptocadica is also similar, being distinguishable by its larger size and rounded body shape (Table 7, rows 14-15). Yellow Ptocadica is probably very closely related to Red Ptocadica and both may be represented in museum collections under the name Ptocadica straminea (David Furth, pesonal communication). Has Red Ptocadica evolved its red coloration to mimic Red Pedilia, and has Yellow Ptocadica evolved to mimic the yellow Parchicola species? It may be no coincidence that these species share the forest habitat, feeding on P. pittieri, P. ambigua, P. vitifolia and P. lobata. They would not be found together on the same host plant, but they do share the forest habitat.

It is likely that Passiflora-feeding flea beetles, like Heliconius, are dynamic in evolution of color forms. The community of Passiflora-feeding flea beetles at Corcovado National Park in Costa Rica in the Chiriqui area of endemism, is quite different-looking from that at La Selva. Selection by visual hunting predators is one possible agent influencing color evolution, although I do not know what type of predator this would be. Also, flea beetles aggregate on host plants and are somewhat social; color may play a role in social interactions. The subject of flea beetle color pattern evolution seems wide open to investigation.

VIII. Cross-community analogs What is the result if we try to compare Heliconius and flea beetles found on each Passiflora species? The nearly identical species diversity suggests a possible one to one correspondence, with each Heliconius having an analogous flea beetle sharing its host plant. To do this I will list the La Selva Passiflora species and associated herbivores one at a time, as shown in Table 8. Cross-community Analogs.

Four Passiflora do not appear to host many Heliconius or flea beetles. These include P. menispermifolia, P. megacoriacea, P. arbelaezii, and P. costaricensis. It is interesting that three of the four species have uniformly high amounts of cyanogenic glycosides, and the fourth, P. megacoriacea, has a unique "slow" HCN release system. P. menispermifolia and P. costaricensis are not only highly cyanogenic, but are densely covered with fine hairs. P. arbelaezii is a little known plant with a limited distribution: in fact P. costaricensis and P. megacoriacea also have limited ranges. Perhaps the narrow range of cyanogenic glycoside concentrations reflects a lack of disruptive selection caused by the lack of Heliconius/flea beetle herbivory. As mentioned above, P. costaricensis and P. menispermifolia are host plants for brightly colored moths in the tribe Josiini (Lepidoptera: Notodontidae: Dioptinae). H. erato is known to infrequently lay her eggs on P. megacoriacea and P. costaricensis.

Two species, P. pittieri and P. lobata, each host a pair of monophagous species, with H. sapho and Red Pedilia on P. pittieri and H. charithonia (Linnaeus 1757) and Red Ptocadica on P. lobata. A chemical basis for monophagy has been discovered for H. sapho, namely sequestration of the cyanogenic glycoside volkenin. The large, brightly colored Red Pedilia flea beetle may also sequester this compound, but this is not known. P. lobata leaves are covered with sharp hooked trichomes that kill most Heliconius larvae except for H. charithonia. They also kill Red Pedilia larvae when artificially placed on the plant, but do not harm Red Ptocadica (Table 4. Flea Beetle Larvae x Passiflora Interaction Matrix). If hooked trichomes are indeed the basis for monophagy in Red Ptocadica, I would predict that other flea beetles will be killed by the hooks as well. Although M. violacea adults are generalists and can be found feeding on most Passiflora species, their larvae appear to be successful primarily on P. lobata. The hooks should not be a deterrent in this case because the larvae feed on roots and/or basal stems, and do not come in contact with the hooks.

In subgenus Decaloba we find P. auriculata and P. biflora. These species often grow together at La Selva, and often share flea beetles. The Heliconius distinguish them, perhaps based on their very different cyanogen chemistry. H. sara is monophagous on P. auriculata, which contains the cyanogenic glycoside volkenin, and P. biflora contains the complex cyanogenic glycoside passibiflorin. The Heliconini using P. biflora as primary host include H. erato, H. hecalesia, and Dryas iulia (Less commonly, D. iulia also uses P. megacoriacea and P. arbelaezii). Three "generalist" species H. cydno, H. hecale and H. ismenius also lay eggs on P. biflora. P. biflora is probably the most widespread and abundant Passiflora in rainforest second growth, as well as being nearly a "universal host plant" edible to all but monophagous Heliconius. Among flea beetles, Ptocadica bifasciata, with leaf eating larvae, seems to prefer P. auriculata but also uses P. biflora to a lesser extent. In contrast Parchicola DF2 may prefer biflora but also uses P. auriculata as a larval and adult host. Monomacra chontalensis, (a primary/secondary forest species) also uses P. biflora, along with P. lobata, but does not use P. auriculata. The disturbed habitat species Disonycha quinquelineata also uses P. biflora. Adults of the "generalist" species M. violacea also use P. biflora and P. auriculata.

Among subgenus Passiflora, P. ambigua hosts the monophagous H. doris and P. oerstedii hosts the facultatively monophagous H. melpomene. Preliminary field and lab observations suggest that adults of three flea beetles are able to share both of these host species: Yellow-tibia and Black-tibia Parchicola, with stem/root feeding larvae, and Yellow Ptocadica, with leaf-eating larvae. However, during the wet season Pa. "yellow-tibia" was only found on P. oerstedii, indicating dependence on that species alone. In contrast, Pa. "black-tibia" and Pt. "yellow" are more common on P. ambigua and may be more dependent on that species. The other three subgenus Passiflora species do not host monophagous herbivores, but share three "generalist" Heliconius, H. cydno, H. hecale and H. ismenius. They (not including P. menispermifolia) also host Black-tibia Parchicola. There is also a parallel between H. cydno, the forest generalist Heliconius, and adult M. violacea, the "generalist" flea beetle. They are even the same color (blue-black)! However, they appear to have different "core" (most often used) plants: H. cydno on P. vitifolia and M. violacea on P. lobata. P. menispermifolia is seldom used by H. melpomene or H. cydno at La Selva, and no flea beetles have been seen feeding on this species.

Even though the three generalist Heliconius share many of the subgenus Passiflora species, there is some evidence for differing adaptations. H. hecale cannot feed successfully on P. menispermifolia while H. cydno and H. melpomene feed and grow well. The presence of H. ismenius in the community is not well-understood, though there is some evidence that it prefers the rare (at La Selva) but widely planted P. quadrangularis (Smiley 1978). The host plant preferences of Black-tibia and Yellow-tibia Parchicola species also suggests some differentiation among subgenus Passiflora, perhaps chemical in nature. Does P. vitifolia differ enough from P. oerstedii to present an opportunity for specialization? Tannins found in P. vitifolia might make a difference as could the habitat difference.

These comparisons suggest that some Passiflora species (P. vitifolia, P. ambigua, P. oerstedii, P. pittieri, P. lobata, P. auriculata, and P. biflora) are more likely to support herbivores than others (P. quadrangularis, P. menispermifolia, P. megacoriacea, P. arbelaezii, and P. costaricensis), and that when they do, they support both Heliconius and flea beetles. These findings support the prediction that there is little or no competitive exclusion between Heliconius and flea beetles.

It appears that some species are more likely to recruit herbivores than others. Does P. biflora, the most widely acceptable host, have an unusually extensive range? It grows in a wider variety of habitats than other rainforest Passiflora in Costa Rica (0 to 2000m above sea level), and extends all the way to Ecuador and Venezuela. Among La Selva Passiflora, eight of the species range geographically across the Transandean region, with only four, arbelaezii, costaricensis, megacoriacea and lobata having a substantially lesser range (Nicaragua to western Colombia). Overall, the cross-community analog comparision (Table 8) reinforces the suggestion made above, that species either successfully host both Heliconius and flea beetles or they host neither. However, only three analog pairs could be identified: H. sapho/Red Pedilia, H. charithonia/Red Ptocadica. Superficially M. violacea is analagous to H. cydno, but unlike that species, M. violacea larvae seem to be monophagous on P. lobata. The remaining species share many similaries with their counterparts, but the lack of host specificity prevents further 1:1 pairing. Below I look at the question of host plant specificity in more detail.

IX. Heliconius and flea beetle diet breadth and diversity on Passiflora subgenera and species Smiley (1985) reported that four of the five subgenus-Passiflora-feeding Heliconius have larvae capable of feeding on subgenus Decaloba, but the reverse is not true: The Decaloba-feeding species generally cannot feed and grow on subgenus Passiflora. If flea beetles respond to Passiflora subgenera in the same way it would suggest similar physiological adaptations to Passiflora biochemistry. Table 2. Flea Beetle x Passiflora Intraction Matrix reveals a conspicuous lack of Decaloba-feeders on subgenus Passiflora, very much like data from Heliconius seen in Table 1. Passiflora x Heliconius Interaction Matrix. Much more work is needed, but the limited data presented in Table 3. Flea beetle adult feeding experimental feeding preferences suggests a similar pattern to that reported in Smiley 1985a for Heliconius. For example, subgenus-Passiflora-feeding Parchicola (either the yellow-tibia or the black-tibia species, or both) were able to feed on P. biflora as well as the preferred P. oerstedii and P. vitifolia, while Parchicola DF2 and Ptocadica bifasciata (both Decaloba feeders) could not feed successfully on P. oerstedii. Also, Smiley 1982 reported that yellow Ptocadica could eat P. auriculata and P. biflora along with the preferred P. vitifolia, P. ambigua and P. oerstedii. These results suggest that flea beetles respond like Heliconius to chemical differences between the Passiflora subgenera, including their response to the "barrier" (Smiley 1985a) that protects subgenus Passiflora from Decaloba-feeding herbivores.

A second conclusion from the data in Table 2 is that flea beetles are less likely to be monophagous than Heliconius. Monophagous herbivores such as H. sara in Table 1 sequester and make use of specific dietary chemicals in their host plant. Table 5. Cyanogens and other Allelochemicals of La Selva Passiflora shows that each Passiflora species posseses a unique set of cyanogenic glycoside compounds. If Passiflora-feeding species experience increased survival and/or reproductive success by ingesting specific cyanogenic (or other) chemicals, then monophagy is more likely to evolve. This is probably the case for the obligately monophagous H. sara, H. doris, and H. sappho and to a lesser extent, the facultatively monophagous H. melpomene and H. charithonia (Gilbert and Smiley 1978). Among flea beetles, only Red Pedilia appears to be obligately monophagous (Table 3). Adults of all the other flea beetle species at La Selva have been observed feeding on at least two Passiflora species (Table 2).

A possible reason for the lack of monophagy in flea beetles may be that they avoid strongly cyanogenic Passiflora and choose plants with reduced amounts of chemicals. They may also avoid HCN toxicity behaviorally rather than by specifically evolved biochemical processes. As a result they do not have the biochemical opportunity to obtain benefits by sequestering specific chemicals. Without chemical sequestration and the resulting chemical dependence, strict obligate monophagy is less likely to evolve. In some Heliconius cyanogenic glycosides are known to be sequestered and probably used as a chemical defense, and plant-derived chemicals may also be sequestered for courtship. The exceptional case of Red Pedilia may actually be consistent with this hypothesis, as P. pittieri is strongly cyanogenic as compared with all other flea beetle host plants at La Selva (see Figure 5. Total HCN Release from Crushed Passiflora species). Red Pedilia is thus forced to adapt to unusually high levels of cyanogenic glycosides in its diet, perhaps leading to obligate monophagy. By the same reasoning, H. melpomene and H. charithonia may partially owe their reduced degree of monophagy to the low levels of cyanogenic glycosides in their host plants, P. oerstedii and P. lobata, respectively. Whether or not this theory is correct, it would be of great interest to know if the Passiflora-feeding flea beetles sequester cyanogenic glycosides from their diet.

X. Partitioning and populating the Passiflora/herbivore community. If obligate monophagy is rare or absent in flea beetles, then another question arises. Why is the species diversity of flea beetles equal to that of Heliconius? Monophagy should reduce species overlap within the herbivore community, and the number of species in the community should increase as more Passiflora species become host to monophagous herbivores. La Selva Heliconius diversity appears to include an approximately equal number of monophagous and non-monophagous "generalist" species, as if the two strategies complement each other and do not compete. All other things being equal, the lack of monophagy in flea beetles should reduce species diversity as compared with Heliconius. The solution may lie in the fact that the Passiflora-feeding flea beetle community appears to also include two larval feeding strategies: root/stem feeders (Monomacra, Parchicola) and leaf eaters (Ptocadica, Disonycha and Pedilia). If the two strategies avoid competition and structure their communities independently, this could increase species diversity substantially. Table 9. Niche analysis based on larval feeding mode summarizes this aspect of diversity.

Table 9 incorporates much of the information presented in this work. It shows that seven of the Passiflora species possess at least one herbivore species from each of the four feeding types, as if a vacancy there would represent an empty, available niche to be filled. The other five Passiflora species represent those which are lightly used by herbivores at La Selva, as discussed above. Table 9 includes rare species as well as common ones, and there is the possibility that some of the rare ones include "spillover" species from adjacent habitats, species that are peripheral to the core interactions of the La Selva community. It may therefore be valuable to revise Table 9, omitting the less common species. For Passiflora these include P. costaricensis and P. quadrangularis, for Heliconiini, Dryas iulia, H. hecalesia and H. ismenius, and for flea beetles, D. quinquelineata and M. chontalensis. This may have the added benefit of omitting species for which there is less information on habitat and host plant preference, strengthening confidence in the results.

The omitted species probably are more common in other areas surrounding La Selva. For example, D. quinquelineata is more common in nearby coastal habitats (JTS personal observation) and M. chontalensis is more common at higher elevations in the adjacent mountains as determined by examination of specimens in the David Furth collection in the U.S.National Museum. This holds for plants as well, with P. lancearia (probable favored host for H. hecalesia) and P. nitida growing nearby on the mountain slopes. P. costaricensis and P. quadrangularis (favored hosts for Josia frigida and H. ismenius, respectively) seem to be restricted to the 5 successional plots which are artificially cleared every 5 years on rotation. Thus, Table 10. Partitioning and populating the La Selva Passiflora/herbivore community is similar to Table 9 except we have omitted two Passiflora, two flea beetles, two Heliconius, Dryas iulia and Josia frigida. I have also explicitly divided Passiflora species and their herbivores into two broad categories of habitat, based on stage of succession.

The "core" community description revealed in Table 10 reveals a remarkable degree of symmetry in the Passiflora/herbivore community at La Selva. It incorporates much of the information presented in this work and allows us to address the fundamental question: Why is species diversity approximately the same for Passiflora, flea beetles and Heliconius on La Selva Passiflora, and why is the number 8-10 species? Josiini, for example, a taxon of moths that may have evolved concurrently with Passiflora, has only 3 known species in Costa Rica, one on each of the principal Passiflora subgenera (Table 9).

To create Table 10 I first set aside the three common Passiflora species not heavily used by Heliconius or flea beetles. I then divided the remaining seven species by subgenus. In the large subgenus Decaloba I further divided by habitat; P. lobata, the common forest/edge species was separated from the second growth species P. auriculata and P. biflora. Similarly, in subgenus Passiflora I also divided by habitat, with the common forest species P. vitifolia and P. ambigua separated from the second growth species P. oerstedii. Subgenus Astrophaea has no second growth representatives, so that group was not divided by habitat type. The end result of this partitioning is five discrete groups of host plants totalling seven species, all commonly used by Heliconius and flea beetles. To each Passiflora group I then used Table 9 to assign the common Heliconius and flea beetle species.

The only significant discrepancy between Table 10 and the data in Table 1 and Table 2 is that H. cydno, and to a lesser extent, H. hecale, sometimes lay eggs on additional species of Passiflora (acting as "generalists"), and M. violacea adults will feed on any Passiflora. Otherwise, the host plant species representation in Table 10 is a reasonable summary of the relationshops in Tables 1 and 2. Remarkably, the 20 adaptive combinations ("niche boxes") represented in the table are all filled with one species each, with no "empty" niches, the only exception being the two "generalist" group II Heliconius which probably separate along more inclusive elements of habitat (Smiley 1978a). Only a few species occupy more than one Table 10 niche box: H. charithonia is specialized to group-feed on P. lobata (facultative monophagy) but will sometimes lay eggs on other species in the laboratory (non-monophagy), Red Pedilia larvae eat leaves in the smaller stages and stem epidermis in the older ones, and Yellow Ptocadica prefers forest but is found in both habitats on all three species of subgenus Passiflora.

Table 10 suggests that Heliconius builds up diversity (8 species) through a combination of subgenus specialization, probably chemical in nature but mechanism unknown (e.g. Decaloba feeders vs Passiflora feeders; see Smiley 1985b), forest vs. second growth habitat specialization (e.g. cydno vs. melpomene), and degree of monophagy (e.g. erato vs. sara). The latter factor, monophagy, nearly doubles Heliconius diversity over what it would be if there were only generalists (4-5 additional species). In contrast, flea beetles, not evolving monophagy, occupy a similar set of five host plant niches, but double their species diversity by diversifying larval feeding behavior (e.g. rootlet/basal stem feeding vs leaf feeding; 4-5 additional species). This suggests that flea beetle adult feeding is less important in structuring the communities than larval feeding, since species overlap is much greater among adults than among larvae. Yet big questions remain. Why do species specialize on the three Passiflora subgenera and on the two habitat types? How do the four modes of larval feeding coexist on the same host plant species without competitve exclusion? Below, I address these and other questions relating to competion and coexistence in the Passiflora/herbivore community. Table 11. Community niche dimensional elements is a summary of the discussion about these questions.

Diversification across Passiflora subgenera

Apparently, neither Heliconius nor flea beetles successfully exploit the three Passiflora subgenera at the same time as primary host plants. What is the difference that the herbivores cannot breach? It is not simply the ability of larvae to survive, since the group II Heliconius along with a few other heliconiine species successfully feed and grow on all three subgenera (Smiley 1985a). As in many instances of host plant selection by insects, the butterflies act like plant "taxonomists," apparently responding to taxonomic categories rather than presence or absence of known chemical constituents. Concurrent evolution of herbivores and Passiflora over tens of millions of years would be a possible explanation, but this hypothesis is not supported (or strongly refuted?) by current evidence. Heliconius seems to have diversified more recently than Passiflora, and flea beetles seem to have colonized Passiflora multiple times. However, there are no estimates for the timing of flea beetle colonization in evolutionary time. Another hypothesis is that the three Passiflora subgenera are chemically distinct in some as yet unknown way that enforces specialization. Smiley (1985a) explores this question by measuring caterpillar growth rates in the different Heliconius species, but the fact is that we know very little about the chemical adaptations that enable Heliconius to feed on and sequester Passiflora nutrients and chemicals. We know much less about the flea beetles' adaptations. However, the presence of sulfated cyanogens in subgenus Passiflora and their apparent absence from subgenus Decaloba (Table 4) suggests one strategy for moving forward.

Adaptation and specialization across successional stages

The second factor in Table 10 is habitat, doubling diversity by requiring both Passiflora and herbivore species to occupy either forest or second growth/disturbance but not both. This is most clear in the Passiflora and Heliconius communities, as described in Smiley (1978a). The flea beetle survey was not designed to sample habitats so the habitat designations are less quantitative. Nevertheless, as indicated in the table, there is a habitat difference between Yellow-tibia and Black-tibia Parchicola (sunny disturbed areas vs shady forest edge areas), and the same is true for the P. lobata-feeding Red Ptocadica vs the P. biflora/auriculata-feeding Parchicola "DF2" and Ptocadica bifasciata. The only flea beetle species found in both habitats seems to be Yellow Ptocadica and adult Monomacra violacea, yet both tend to be more common in forest.

In Table 10 the "habitat" axis (forest vs second growth) may be slightly artificial in the case of the Decaloba species, P. lobata. P. lobata is unique among La Selva Passiflora because of its unique hooked trichome defense. Among Heliconius only the adapted H. charithonia can feed, and among flea beetles the adapted Red Ptocadica can feed and the non-adapted Red Pedilia die on the hooks. But we don't know if other larvae such as Pt. bifasciata can survive or not. Thus the question: is P. lobata set aside as its own Decaloba-feeding group because it specializes in the forest habitat or because its trichomes kill non-adapted herbivores, or both? Perhaps the trichome defense enabled P. lobata to invade the forest habitat. Some non-adapted flea beetles successfully feed on P. lobata (Parchicola DF2 and Monomacra chontalensis).

The adaptive basis for habitat specialization is not known, but several factors are likely. At La Selva, Passiflora are about 100 times more abundant in second growth than in forest (Smiley 1978a). Forest adapted species (especially flea beetles) therefore have the additional challenge of locating host plants, and may alter their behavior and life-history accordingly. Second growth adapted species probably experience much greater turnover of host plant owing to the rapid changes in light regime and the possibility of desiccation. There are likely to be large differences in predation risk as well.

Herbivore life-history evolution and diversification: Larval feeding modes

Table 10 shows that four larval feeding modes coxist on the same host plant species. Three of the feeding modes involve feeding on new leaves of Passiflora, which would seem to bring them into competition. To resolve this I propose that coexistence is allowed because each mode has different requirements for growth and survival, and thus is successful on different individual plants or plant parts. Each of the 5 plant groups in the table has at least 1-2 flea beetles and 1-2 Heliconius herbivores. For the flea beetles it is clear that the root/stem feeding strategy can coexist with leaf feeding on the same host species. Although much more work is needed on the subject of larval feeding in flea beetles, I suggest that the conditions for successful reproduction may differ substantially between the two strategies. Larval root/stem feeders (and their eggs) need accessible rootlets, usually found in deep, decomposing leaf litter. They also need protection from soil predators and from dessiccation. In contrast, leaf feeding eggs and larvae need healthy leaf tissue and protection from predation. In addition, the leaf-eating Ptocadica tend to be solitary, and thus more likely to be successful on smaller plants.

Also in Table 10, I separated the Heliconius into two groups based on their degree of specialization. I reasoned that monophagous species can outcompete their generalist counterparts on their preferred plants, but that the generalists compete by using other plant species, different individual plants (smaller individuals?) or different plant parts, allowing coexistence. P. biflora and P. vitifolia lack monophagous Heliconius. Why? The answer may be in the relatively thin leaves possessed by those species (Smiley 1978a). Thick leaves (P. pittieri, P. auriculata, P. ambigua; see Figure 14) seem to be required for group feeding, obligately monophagous larvae. The hosts of the "ecologically monophagous" species (Smiley 1978b) have thinner leaves (P. lobata and P. oerstedii), and indeed do not support large groups with over 30 larvae. Perhaps thick leaves are necessary to physically support group feeding larvae feeding side by side. Alternatively, perhaps Heliconius evolved group feeding in order to more effiiciently consume the foliage. Group feeding has advantages and disadvantages in terms of greatly reduced growth rates and avoiding predation. In any case, the correlation is there, and opportunities for further diversification by evolving obligate monophagy may be limited by the lack of thick-leaved species.

XI. Summary. Table 10 and Table 11 are my best attempt to describe the defining community relationships in this system. They provide a non-trivial answer to the question "Why are there 8 species of Heliconius and 8 species of flea beetle on Passiflora at La Selva?" According to the table, the La Selva environment, with a fairly complete representation of the diversity of Brown's Transandean region, is home to three common, widespead Passiflora subgenera, each different enough chemically that Heliconius, Josiini and flea beetles must specialize in order to be successful. The two largest Passiflora subgenera include species growing in either forest or second growth, with some mixing in forest edge habitats. The herbivores adapt to these habitat differences by specializing on Passiflora in one habitat or the other. The result is five distinct sets of host plants: P. pittieri (subgenus Astrophaea; forest), P. auriculata/biflora (subgenus Decaloba, second growth), P. lobata (subgenus Decaloba, forest), P. oerstedii (subgenus Passiflora; second growth) and P. ambigua/vitifolia (subgenus Passiflora; forest).

Each of the five host plant groups is exploited by 3-4 herbivore species, one each from the four larval feeding modes: root/stem feeding, flea beetle leaf feeding, Heliconius solitary-generalist leaf feeding, and Heliconius monophagous-group leaf feeding. By this scheme, each "box" in the table represents a distinct ecological niche. Table 10 therefore presents a remarkable result (with one exception): no two species share the same niche, and all the niches are filled with exactly one species! Also, there are only 4 cases where a species fills more than one niche. This suggests that the Passiflora herbivore community is "nearly saturated" with species, and, barring serious disruption or disturbance, is probably closed to invasion by new species. This may explain why the flea beetle community has not changed over the 40 year interval since the beginning of the study.

The results of this study suggest that community alpha diversity (number of species at a given geographic location) has three main generators: (1) host plant evolutionary diversification at the subgeneric level; (2) adaptation to habitat type (stage of succession) for host plants and herbivores, and (3) evolutionary innovation of compatible life history strategies that allow coexistence on the same plant species (the four larval feeding modes in Table 11). In the case of the La Selva Passiflora/herbivore communities, these three factors act multiplicatively to define niches which are uniquely filled with very few exceptions.

Are there limits to this generation of diversity? The Passiflora genus arose over a 30 million year span, enough time for the evolution of over 500 species and 48 taxonomic sub-sections. However, only three principal subgenera appear to determine alpha diversity at La Selva, a very coarse partitioning of this intricately diversified plant genus. Is this limitation a consequence of millions of years of host plant switching within each subgenus? In other words, herbivore species that persist to the present day may have required a large set of alternative host plant species, a set corresponding to the diversity within a host plant subgenus. Obligate monophages such as H. sara, H. sapho and H. doris lack viable secondary host plants, but in this case there is also the possibility of multiple extinctions and recolonizations from related but geographically distant host plant species .

Adaptation to stage of succession ("habitat") appears to yield 1-2 sets of species for all three communities, including the host plants. Although the two speciose subgenera differ on average, with subgenus Passiflora being larger and more woody (forest adapted) and Decaloba smaller and more herbaceous (second growth adapted), each has evolved members that occupy the opposite habitat. Clearly there may be other habitats or environments, not well represented in the present study, that might contribute additional species.

In order to join the community, herbivore life history innovations must be different enough to reduce competition and allow coexistence. This is the case for the four feeding mode strategies, since they coexist on the same host plant species. How do the feeding mode differences allow for coexistence, when competition within a mode is strong enough to prevent two species from sharing the same niche in Table 10? Analysis not presented here (Smiley MS in preparation) sugggests that each of these life history strategies may be most successful on different individual plants in the La Selva environment, depending on factors listed in Table 11. There is also the possibility that some feeding modes may actually enhance growth and survival for other modes via opposing selection on plant quality and cyanogen content. Possession of multiple feeding modes on a plant species will likely foster variability within that species, perhaps leading to evolution of the remarkably complex cyanogen physiology/behavior described above for our seven species.

The results presented here are also notable in that species diversity is shown to be a consequence of competition and coexistence of larval forms, even in genera such as Heliconius that exhibit complex and intricate adult life histories. It would appear that generation of increasing species diversity is a slow process in this system, depending on macroevolutionary innovation of novel, compatible larval feeding syndromes and slowly evolving diversification of host plant subgenera. The overall picture is one of a well-structured community of species, stabilized by competition, predation and possibly spatial dispersion within the complex matrix of the tropical forest. Evolution of this structured community may be overlain by much more rapid microevolutionary adaptations such as rapid evolution of Heliconius wing coloration in response to presence/absense of co-mimics (Kozak, et. al. 2015, Jiggins 2017), evolution of egg-mimic structures in Passiflora (Gilbert 1982), and evolution of cyanogenic glycoside physiology/behavior in Passiflora foliage as suggested in this study. There may also be species replacement within the structured community. If, for example, one species went locally extinct it might be replaced by another, similar species from a nearby community that could adapt to the La Selva conditions. This type of microevolution could keep fine-tuning and/or re-populating the system without altering species diversity, larval feeding strategies, host plant subgenus affiliation and other basic properties.

This study shows the value of examining life histories of plant and animal species in relatively stable natural environments such as the wet forest at La Selva. Further study of some aspects of the system may be more easily conducted in environments where the organisms are more common and/or in laboratory facilities such as those at La Selva. It would be very interesting to pursue the possibility of chemical sequestration by the herbivores, including H. doris as well as several of the flea beetles. Ant/larval interactions would also be interesting to pursue, including discovering the function of the balloon organs in the flea beetle larvae. In addition, further investigation of cyanogenic glycosides and other chemicals would shed light on many aspects of the system, from plant responses to herbivory to the nature of variation within species. Finally, it should be possible to test the hypothesis that herbivores from different larval feeding modes are successful on different individual plants attended by different predators.

Acknowledgements