The data explored here represents, to our knowledge, the largest wild-caught caterpillar rearing program ever conducted. To date, the project has accumulated more than 883,000 caterpillar–food plant records (Janzen, 1988; Janzen et al., 2009; Quicke et al., 2024). Records of larval food plants for the butterfly species (i.e., Papilionoidea) collected before 2007 were included in the catalogue by Beccaloni et al. (2008), which also categorized some of these recorded trophic associations as plausible versus dubious.

The scientific objective of the ACG caterpillar-rearing program has always been purely documentary in that it was to obtain a detailed snapshot of which plants are consumed by caterpillars in a diverse and successional tropical wildland. Thus, since there has been no standardization of sampling effort, the number of caterpillar rearings corresponds to the frequency with which their food plant was encountered. In addition, the intensity of survey effort varied throughout the course of the project, and sampling efficiency increased commensurate with the increased experience of the parataxonomists involved.

These features mean that it is not possible to compare directly all features of the present study with those of a number of other large field studies that have been conducted in recent years (e.g., Novotny et al., 2002a, 2004, 2010; Dyer et al., 2007), which usually restricted sampling to a particular group of plant species and/or standardized searching effort in relation to leaf area. Furthermore, most other studies have been restricted to relatively short time windows. In contrast, the data we make available may allow the extraction of subsets from which valid comparisons with other studies can be made.

Using a data set of 29,215 different caterpillar–food plant associations, we have compiled species accumulation curves, compared species richness among three different ecosystems in ACG, explored trophic interactions, quantified faunal turnover over time (largely due to successional changes in food plants), examined diet breadth and plant chemical defenses, and briefly reviewed a variety of other parameters among exophytic caterpillars revealed by the immense ACG database. Our overarching purpose is to investigate the factors mediating the diet breadth of the reared Lepidoptera species and families, and in this respect, we take two approaches. First, we examine the utilization of food plant families with well-known protection either physical or chemical against folivores. These families were selected based on scientific literature and investigated separately. Second, we investigate the food plant diversity of individual caterpillar species on each plant family as well as their diets across all other plant families. The latter approach revealed a number of rather unexpected trends which we discuss further.

Study site: The Área de Conservación (ACG) in northwestern Costa Rica is a 169,000-ha tract of land within the national system of protected areas. The site has been described in detail by Janzen (2004); Janzen et al. (2016), and Janzen and Hallwachs (2017). In general, ACG supports three primary ecosystems: dry forest (ca. 70% of the land area), rain forest (ca. 25%), and cloud forest (5%–10%). The last of these has diminished during the course of the study, most likely as a result of warming due to climate change (Smith, 2023). In ACG, dry forests occur predominantly at lower elevations and along the Pacific coast and are dominated by deciduous trees that often lose their leaves during the dry season. It is represented by a range of successional ages from very recent to centuries old rain forest is moister and characterized by dense canopies of evergreen trees, high humidity, and abundant rainfall, with many fewer and less abundant deciduous trees. Cloud forest occurs at the highest elevation in ACG (above 1,000 m elevation) and is characterized by a persistent cloud cover but exhibits considerable botanical overlap with the rain forest.

Field collecting and most caterpillar and plant identification were carried out by resident parataxonomists (“gusaneros” in the local language) (Janzen and Hallwachs, 2011). For those unfamiliar with this occupation, here is a brief description of their roles. A parataxonomist is broadly analogous to a paramedic or paralegal person but in the arena of field taxonomy/biology. They are resident men or women selected from the rural workforce, without a university degree. They are trained on the job to conduct an inventory by finding insects, knowing their names, knowing the plant species they are feeding on, rearing the caterpillars and their parasites, collecting and preserving the individual insects that emerge, and entering all the relevant data into a database. Most such parataxonomists began their career in the 1980–1990s by taking an intensive 6-month introductory course and then being continually coached and updated over the next 20–35 years while on the job. New parataxonomists are trained through an apprenticeship with the more experienced ones. Each is based at a particular field station where they have spent decades learning the taxonomy of their focal insects and plants.

During each collecting foray, parataxonomists would collectxmostly in only one of the aforementioned three ecosystems. However, sometimes they ventured into two (e.g., parataxonomists working at Estación Cacao might, for example, sample plants and caterpillars from dry forest/rain forest intergrade, rain forest, and cloud forest). Hence, the three ecosystems do intergrade, and some of the trails in each ecosystem have similar plants that are generally associated with recently disturbed habitats (ruderal species).

Caterpillar collecting: Methods of collecting, rearing, and vouchering external foliage-feeding caterpillars in the ACG have been described previously in detail elsewhere (e.g., Janzen et al., 2016; Janzen and Hallwachs, 2017; Quicke et al., 2024). Members of the experienced team of 10–30 resident parataxonomists physically searched for external feeding caterpillars, including those in rolled leaves or silk-leaf nests up to approximately 2 m above the ground level, primarily along trails (Janzen, 2004; Janzen and Hallwachs, 2011). Surveys were conducted both by day and night and all year around. Although this method is fairly effective for nearly all Macroheterocera (macro-Lepidoptera) as well as for those hidden in leaf rolls or nests, which reveal their presence, it proved less effective for discovering caterpillars that are concealed feeders. Despite the considerable experience accrued by the parataxonomists, sampling was inevitably biased toward more apparent or distinctive species and against least apparent, more cryptic ones or those that hide elsewhere during daylight hours. The parataxonomists did not collect leaf-miners, stem or gall borers, or any species feeding on rhizomes.

When a parataxonomists encountered multiple individuals of a caterpillar species on a given plant on the same occasion, they sometimes collected and reared up to around 20 of them to increase the chance of a successful rearing as well as the probability of rearing parasitoids. For the analyses presented here, we consider this as a single rearing event and refer to this as a unique or independent rearing.

Caterpillars discovered in the field were taken to “rearing barns” and singly placed in plastic bags with cuttings of the food plant and hung from clothes lines in the shade (photographs showing these are provided in Quicke et al. (2024)). New foliage was provided at 2–4-day intervals. Each caterpillar was labeled with a unique voucher number in the form of YY-SRNP-X…… (e.g., 09-SRNP-15328), where the prefix is the last two digits of the year (e.g., 2009), “SRNP” refers to the project “call letters” assigned in 1977 (when the initial project site was referred to as Santa Rosa National Park), and the suffix is a unique number assigned within the year. As adults emerged, they were frozen, pinned, and labeled. These voucher codes are provided in Supplementary Table S1 (also available at https://doi.org/10.5683/SP3/NX043G), which includes several additional fields to the ones used in these analyses, i.e., barcode BIN, parasitoids when reared, including latitude, longitude voucher fate, and taxonomic identifier information. Most specimens have been transferred to the United States National Entomological Collection (USNM), Smithsonian Institution, Washington DC, where they now reside, but due to capacity restrictions, species-level duplicates and some paratypes are also deposited in the Museo Nacional de Costa Rica in Santo Domingo de Heredia in the suburbs of San Jose, Costa Rica.

Species identification: The identification of ACG Lepidoptera has been dynamic, involving the recruitment of taxonomic experts worldwide. Many of the early “species concepts” based mainly or solely on morphology have been refined through more integrative methods relying on a combination of morphological and molecular data. Although some of the undescribed and cryptic species have been formally named (e.g., Burns and Janzen, 2005; Burns et al., 2008; Metz et al., 2017, 2020; Solis and Styer, 2003; Solis et al., 2005, 2020; Metz, 2024), most are now recognized based on morphology and/or food plant associations following revisiting them in the light of barcoding results (e.g., Hebert et al., 2004). Even so, members of many putative species complexes are still discriminated only by their barcode index number (BIN) in the Barcode of Life Data System (BOLD) (https://boldsystems.org) (Ratnasingham and Hebert, 2013). In the present dataset, the undescribed putative species are indicated by interim alphanumeric place-holder names usually with a researcher’s or project’s initials and a number. Sometimes, further evidence reveals that even certain of these comprise one or more distinct species, or that despite barcode differences, some putative species should be recombined with members of another BIN. In general, the result of barcoding and associated integrative taxonomy has revealed far higher levels of dietary specialization than earlier work suggested for Lepidoptera (Hebert et al., 2004; Janzen et al., 2011), their dipterous and hymenopterous parasitoids (Smith et al., 2006, 2007, 2008), and other insects (e.g. Dolson et al., 2020, 2021; Underwood et al., 2024). Hence, for many Neotropical Lepidoptera families, it should be noted that the current state of their taxonomy and generic classification is far from established, and in the future, many of the generic assignments of species will probably change.

For the purposes of this study, we describe diet breadth as a count of taxa (i.e., species as delimited by morphology and/or additional evidence as noted above classified in families) and do not infer phylogenetics other than assumed monophyly at the family level. Summarization at the family level is a matter of convenience and not an implication that families are comparable among the included taxa.

For DNA barcoding, DNA was extracted from tissue from the leg of an oven-dried individual adult lepidopteran (Hajibabaei et al., 2006; Janzen et al., 2009; Janzen and Hallwachs, 2011, 2016). Sequencing of the barcoding region of the mitochondrial cytochrome oxidase c subunit 1 (COI) gene was performed following standard methods at the Centre for Biodiversity Genomics, University of Guelph (Craft et al., 2010; Wilson, 2012).

With regard to the dataset (available in Supplementary Table S1), this comprised not only caterpillar–host associations for adults successfully reared from larvae discovered in the field but also for associations in which caterpillars succumbed to parasitoids. In the latter case, the identity of caterpillars was reliably determined either through morphological or behavioral distinctiveness or because other individuals of the same species collected from the same plant and on the same occasion yielded an adult butterfly or moth and/or was identified through comparative DNA barcoding (154,833 Lepidoptera DNA barcodes were generated in the course of the study). The data are more thoroughly vetted than those of our previous paper (Quicke et al., 2024) with only caterpillars that could be identified to species or assigned a confident interim name (or BIN) included.

Data analyses were performed using the R environment for statistical computing (version 4.4.1) (R Development Core Team, 2024). In addition to functions provided in base R, we also employed functions from the following packages: circlize (Gu et al., 2014), e1071 (Meyer et al., 2019), bipartite (Dormann et al., 2008), RColorBrewer (Neuwirth, 2014), pals (Wright, 2025), and VennDiagram (Chen and Boutros (2011).

Species accumulation curves were calculated by the standard method, plotting cumulative number of species (y-axis) against sequential number of independent rearings (x-axis). Four curves were compiled, one for the entire data set and one for each of the three ecosystems.

Shannon’s diversity index H (Shannon, 1948) was used to quantify species diversity within a sample (Stireman et al., 2017) as it is less strongly influenced by a small number of dominant species than is Simpson’s index. To estimate the overall degree of specialization in a network, we used the standardized two-dimensional Shannon entropy, H₂’ (Blüthgen et al., 2006), which has values near zero in cases of extreme specialization and 1 when there is extreme generalism. Similarity among foodplant, caterpillars, and trophic interactions for the three ecosystems was calculated using Jaccard’s coefficient with the following formula: J(A, B) = |A∩B|/|A∪B|, which can be summarized as “the intersection of two sets/the union of two sets.”

Comparisons of species richness among lepidopteran families were illustrated using simple histograms with families on the x-axis and number of species on the y-axis. Histograms were compiled for the total data set and for each of the three individual ecosystems. The sequence of families along the x-axis was determined by species richness based on the total data set and retained for each of the ecosystems. In an attempt to quantify changes in similarity among caterpillars and foodplants over time, the data set was divided into six 5-year periods of comparable sampling effort (i.e., 1990–2020). Similarity among caterpillars and foodplants was calculated for each of the three ecosystems using Jaccard’s coefficient. Similarities were then divided into seven increments of 0.1, ranging from 0–0.1 to 0.6–0.7. For several analyses concerning unpalatable food plants, we treated the Arctiinae (tiger moths and allies) separately from other members of the Erebidae because their ability to feed on a wide range of toxic plants is well documented (Weller et al., 1999; Mason et al., 2014; Zaspel et al., 2014).

To examine the spectrum of plant defense against caterpillar herbivory, we created two lists of plant families, taxa that are conspicuously equipped with chemical and/or structural defenses, and one for families whose species generally lack such obvious protection. The well-protected list comprised those plant families that have latex, abundant silica, or notable toxins (e.g., aristolochic acid, alkaloids, cycasin, glucosinolates, and glycosides). The families considered well defended (with T = toxic, L = latex, S = silica) are Alismataceae (T), Amaryllidaceae (T, L), Annonaceae (S), Apocynaceae (T, L), Araceae (T, L), Arecaceae (S), Aristolochiaceae (S, T), Boraginaceae (S), Bromeliaceae (S), Burseraceae (S), Cactaceae (L, S), Cannabaceae (L, S), Cannaceae (S), Caricaceae (T, L), Chloranthaceae (S), Chrysobalanaceae (S), Costaceae (T, S), Celastraceae (T, L), Convolvulaceae (T, L), Cyperaceae (S), Dilleniaceae (S), Euphorbiaceae (T, L), Gesneriaceae (T), Heliconiaceae (S), Hernandiaceae (S), Icacinaceae (T, L), Loranthaceae (S), Magnoliaceae (S), Marantaceae (S), Menispermaceae (S), Moraceae (T, L), Nymphaeaceae (T), Musaceae (T, L, S), Olacaceae (T, L), Papaveraceae (T, L), Passifloraceae (T), Piperaceae (T, S), Poaceae (S), Sapotaceae (T, L, S), Solanaceae (T), Thymelaeaceae (S), Ulmaceae (S), Urticaceae (S), Zamiaceae (T), and Zingiberaceae (S). All remaining angiosperm families were classified as having largely palatable foliage.

The classification of angiosperm food plants follows APG IV (The Angiosperm Phylogeny Group et al., 2016). The classification of grasses follows GPWF II and GPWG III (GPWG II, 2012; GPWG III, 2025). The classification of ferns follows PPG I (Pteridphyte Phylogeny Group; Schuettpelz et al., 2016). Definitions of Lepidoptera families follow those of van Nieukerken et al. (2011).

We compiled 435,897 individual rearing records of which 197,366 represent unique, in other words independent, observations (i.e., caterpillar–plant associations that were collected on separate dates and from different plants). Within that data set, 2,489 plant species served as hosts for one or more caterpillars, and 7,160 species of Lepidoptera were reared. Caterpillar and food plant accumulation curves are shown for the whole data set (Figure 1) and for each of the three ecosystems (Figures 2A–C). For the total data set, trophic interactions showed the steepest curve, with accumulation curves for caterpillar species and plant species significantly more gradual. The accumulation curves for the three different ecosystems showed similar patterns, with caterpillar species far outnumbering plant species beyond the first few sampling bouts. In all cases, there was an initial steep accumulation of species with rearing effort, with a continuing nearly linear curve that never showed signs of reaching an asymptote, although rain forest Lepidoptera species accumulation did show a potential transition towards an asymptote. The rather relatively sharp increase in food plant species in the cloud forest data (Figure 2C) reflects, at least in part, changes that occurred during the period of COVID-19 restrictions (March 2020 to December 2021) as well as the transfer of a parataxonomist to and from another station.

The steep and continuous nature of the curves suggests that considering all measured parameters (caterpillars, host plants, and interactions), significant numbers remain undersampled. However, we included associations that were detected only once which some authors have argued against, suggesting that such rearings may be abnormal or atypical (Marohasy, 1998; Novotny et al., 2002b). While it is true that a female herbivorous insect might make occasional “mistakes”, that does not mean these events are unimportant. A shortage of the normal plant species, various types of stress or old age, might cause females to perform some egg dumping on possibly unsuitable plants on which their caterpillars have lower but nevertheless non-zero chance of developing successfully. If the resulting caterpillar does complete its development successfully on that plant, this is presumably how host ranges evolve. Furthermore, because of the non-targeted and opportunist nature of the caterpillar collection, the dataset includes a substantial number of Lepidoptera species that were encountered and reared only on a few occasions—for example, 1,751 caterpillar species were only reared on one occasion, 772 twice, and 494 three times.

In this light, we suspect that the steep increase in trophic interactions results from the fact that that species were only encountered rarely, and since very few species collected on numerous occasions were monophagous, increased rearing of uncommon taxa with more trophic links kept being revealed. Thus, both host specificity of caterpillars and the herbivore ranges of food plants based on these data are likely to be marked underestimates.

The overlap in caterpillar, plant species, and trophic interactions among the three ecosystems is illustrated in Figures 3A–C. Jaccard coefficients of similarity are presented in Table 1. The overlap of food plant species among ecosystems was 13% (i.e., 324 shared species), and the overlap of lepidopteran species was 8.6% (i.e., 612 shared species). Only 1.5% of species–species interactions (i.e., 441) occurred in all three ecosystems.

Of the 439 most widely recovered trophic interactions, those linking only four combinations of caterpillar–food plant families were represented 10 or more times, and 39% of the family-level trophic links were found only once. The widespread caterpillar species were dominated by Hesperiidae, 29 of which fed on Fabaceae and 11 on Poaceae. Next were Sphingidae–Rubiaceae associations (n = 24) and the third were Notodontidae–Fabaceae links (n = 10), although the last is almost entirely due to the subfamily Hemiceratinae, specifically Hemiceras and its association with Inga species. Although Inga species are not ruderals, they are typically competitive and stress tolerant. While a total of 62 nymphalid species were reared from all three ecosystems, these were collectively reared from 22 different plant families.

Based on the total data set for ACG (Figure 4A), Erebidae (n = 1,212 species) was the most species-rich family, followed by Geometridae (n = 812), Hesperiidae (n = 666), Depressariidae (n = 659), and Crambidae (n = 497). The pattern of species richness in rain forest (Figure 4B) was similar to that of the total data set, with the same top 10 families (i.e., Erebidae, Geometridae, Depressariidae, Crambidae, Hesperiidae, Notodontidae, Nymphalidae, Noctuidae, Gelechiidae, and Tortricidae); however, there were a few slight differences in their relative positions—for example, Hesperiidae dropped from third to fourth most species-rich, Notodontidae dropped from sixth to seventh, and Noctuidae dropped from eighth to ninth.

In dry forest (Figure 4C), Erebidae again were the most species-rich family (n = 497 species), but in this ecosystem, Hesperiidae was the second (n = 354) and Crambidae the third (n = 254). Cloud forest deviated the most from the overall ACG pattern (Figure 4D), with Geometridae (n = 194 species) representing the most species-rich family, followed by Erebidae (n = 126) and Hesperiidae (n = 120). Overall, Depressariidae were less species-rich in both dry and cloud forest compared to the overall ACG and rain forest ecosystem. Crambidae and Hesperiidae were comparatively more species-rich in dry forests, whereas Hesperiidae, Notodontidae, Nymphalidae, and Sphingidae were more species-rich in cloud forest in comparison to other ecosystems. As would be expected, many microlepidoptera families that typically feed on detritus, fungus, roots, or other cryptic microhabitats (e.g., Tineidae, Hepialidae, Blastobasidaae, Oecophoridae) are poorly represented in the reared samples. In this light, it was surprising that Tortricidae and Gelechiidae, microlepidopteran families that include many concealed feeders, are among the top 10 families sampled. It is noteworthy that nearly all of the gelechiids reared belonged to Dichomeridinae and Anacampsinae, with only a few belonging to Gelechiinae.

Substantial parts of ACG are undergoing natural (i.e., benign neglect) succession from former agricultural land to forest (see Janzen, 1990), but climate change also contributes to changes in the vegetation through differences in rainfall patterns and increasing average temperatures (Herrera, 2016). Changes in floral and faunal composition over time are shown in Figure 5. It has now become apparent that for many years overall insect abundance, including Lepidoptera and the species that parasitize them, has shown a considerable decline since the 1980s (Janzen and Hallwachs, 2021). This follows an apparently global trend—for example, Hallmann et al. (2017) report greater than 75% decline in total flying insect biomass in just over 27 years in 63 protected areas in Germany. Furthermore, a recent metanalysis (Rumohr et al., 2023) indicates that the major contributing factors to this are anthropogenic in origin.

Overall, food plant and caterpillar communities in rain forest (Figure 5A) showed the most change over the first 10 years but have remained more stable and consistently higher since the year 2000. However, it is possible that some of the consistency is a result of collecting bias, as parataxonomists grew increasingly familiar with the fauna and where to search for caterpillars and the intensity of collecting effort could not be standardized (in effect, a biased sampling effect). The floral and faunal composition of dry forest was more consistent than in rain forest throughout the sampling period. The cloud forest biota was the most consistent between 1995 and 2010 but, as in dry forest, the caterpillar fauna changed considerably during 2015–2020. In all cases, plant species composition was more constant than caterpillar composition.

In general, the number of caterpillars and food plant species, respectively, were comparatively low in the first 5-year period but subsequently increased, allowing us to more meaningfully calculate several standard food web statistics (connectivity, rescaled connectivity, link density, and generality) (see Bersier et al., 2002; Lewis et al., 2002; Blüthgen et al., 2006; Dormann et al., 2009; Gilbert, 2009; Kerkig et al., 2025), and these are presented in Supplementary Table S2 separately for each ecosystem and 5-year sampling windows. Overall, both generality and linkage density were highest in the rain forest and lowest in cloud forest, However, it should be noted that the reliability of the values might be reduced because of the incompleteness of the samples with many species represented by only a few individuals (singletons and doubletons) (Stireman et al., 2017).

When examining family–family interactions, we included only those lepidopteran families with at least 40 distinct trophic interactions (Figure 6) and for food plant families with more than 100 distinct interactions. While the most species-rich lepidopteran families generally attack a similar range of food plant families, some distinct preferences are clear—for example, Pieridae were reared almost exclusively from Fabaceae due to the number of Dismorphinae rearings, with far smaller numbers from Capparaceae, Santalaceae (former Viscaceae clade thereof), and Bignoniaceae. Uraniids (all belonging to the Epipleminae) were predominantly reared from Bignoniaceae and as were many Sphingidae. Both of these plant families have species that produce iridoid glycosides (Nayar and Fraenkel, 1963; Quicke et al., 2024), suggesting that epiplemine uraniids and sphingids have evolved mechanisms to cope with these plants which usually involve suppressing enzymes in the gut that cleave them to release the unstable and toxic aglycone residue (Dobler et al., 2011). In this context, sphingids have been rather well-studied but not the uraniids.

Moths of the family Lasiocampidae (lappet moths) were predominantly reared from Lauraceae (870 unique rearings), Phiditiidae almost entirely from Bignoniaceae (n = 184) with a few from Lamiaceae, while more than 99.9% of Attevidae rearings were from Simaroubaceae (n = 265). The database confirmed that out of 13,627 unique Hesperiinae (grass skipper) rearings, not one was from a dicotyledonous plant family and predominantly from Poaceae and Arecaceae (Figure 6). Another conspicuous preference is the large number of trophic links between Sphingidae and Rubiaceae.

We explored the degree of specialization of those caterpillar species that were associated with the 40 plant families for which the data includes more than 1,000 independent rearings. For these taxa, we recorded the diet breadth of their herbivore species across all plant families to identify which plant families had the most specialist herbivores (Table 2). Here and throughout, we use the simple heuristic of the number of plant families (plant family “penetrance”), recognizing that the taxonomic range of any given group of herbivores does not track taxonomic rank.

Next we examined the overall recorded diet breadths, again as measured by overall taxonomic penetrance, exhibited by caterpillar species feeding on these families displayed across all of their range of food plant families (Figure 7). Plant families having the most specialist caterpillar species (lower left) are indicated in red, and ones with the most generalists are in green (towards right and top). The twelve families hosting the most specialist caterpillars were, in descending order, Apocynaceae, Poaceae, Acanthaceae, Arecaceae, Solanaceae, Piperaceae, Rubiaceae, Melastomataceae, Rutaceae, Costaceae, Moraceae and Marantaceae, in order of decreasing herbivore specialization.

From the standpoint of botanical understanding, Figure 7 suggests that several plant traits are associated with having trophic associations involving highly specialist caterpillars. Arecaceae, Costaceae, Piperaceae, and Poaceae have particularly tough leaves (sclerophyly) with particularly high silica content, Apocynaceae and Moraceae possess latex which contains various entomotoxic compounds, and Solanaceae and Piperaceae contain many toxic secondary metabolites, notably alkaloids. The picture is less clear as to why Melastomataceae and Rubiaceae are associated with particularly specialist caterpillar species, although both families have diverse secondary metabolites (some used in ethnopharmacology) (Ocampo Serna and Isaza Martínez, 2015; Martins and Nunez, 2015). Some Melastomataceae are aluminum accumulators (Timpone et al., 2025), including Miconia species which are represented in our data set by 42 species (Sittenfeld et al., 2002). Whether the presence of high concentrations of aluminum in their leaves affords them protection against caterpillar herbivory does not seem to have been investigated. However, in the cases of some other plant species, it appears to be a deterrent (Behmer et al., 2005; Ribeiro et al., 2017) although there is no overall consensus about the evolutionary selective value of metal hyperaccumulation in plants (Raskin et al., 1994; Jansen et al., 2002).

Several Lepidoptera are notorious for the broad range of plant families that serve as larval hosts. For example, Gelechiidae (twirler moths), a family with more than 4,000 described species, are documented from more than 80 plant families, with a proclivity for Asteraceae and Fabaceae (Powell, 1980). The 424 species of gelechiids reared in the ACG are recorded from 78 plant families but are represented in the data set only by species with exophytic caterpillars, whereas most gelechiids are leaf miners and stem borers. Although this family is under-sampled in our data set because the majority are leaf miners and stem borers, our data nevertheless reveals a remarkably broad range of host plant families.

We analyzed rearing records from three taxonomic cohorts outside Angiospermae: Lycopodiopsida (clubmosses, spikemosses, and quillworts), Polypodiopsida (ferns), and Gymnospermae, the immediate sister group of the angiosperms. Ferns originated at least 423 million years ago (Ma) (Nitta et al., 2022), and gymnosperms diverged from their common ancestor with angiosperms between 337 and 208 Mya (Morris et al., 2018). We treat the cycad family Zamiaceae separately because of its well-known extreme toxicity. Utilization of these plant groups in terms of unique rearings and distinct trophic links are summarized by family in Figure 8. Caterpillars of only two families, Erebidae and Nymphalidae, utilized lycopods at ACG. Ferns, by contrast, are more widely used, with moth larvae of the Erebidae, Noctuidae, Crambidae, and Geometridae as the top four Lepidoptera families feeding on these primitive plants. There were too few rearings from non-cycad gymnosperms to discuss (Cupressaceae (n = 6), Pinaceae (n = 2), Podocarpaceae (n = 2)).

The data included a substantial number of species belonging to well-known, chemically defended Lepidoptera, such as the nymphalid tribe Heliconiini (including the Heliconius butterflies), Ithomiini (the glass wings), and Danaiini (including the milkweed butterflies) and the subfamilies Melitaeinae (Nymphalidae), Pierinae (Pieridae), Arctiinae (Erebidae), and the family Zygaenidae. We summarize these below.

Plant chemical defenses against insect herbivory include a broad range of toxic secondary metabolites such as alkaloids (Willaman and Schubert, 1961), glycosides (including cyanogenic glycosides which are then converted to cyanide) (Vetter, 2000; Park and Coats, 2002), terpenoids (Mumm et al., 2008; Kortbeek et al., 2019), glucosinolates and many others (Bennett and Wallsgrove, 1994). In addition to these smaller molecules, there are several entomotoxic proteins, e.g., sugar-binding lectins in both seeds and foliage production of that can be induced by herbivory (Van Damme et al., 2008). These molecules are attracting greater attention by virtue of our growing capacity to incorporate them into crops via genetic engineering (Carlini and Grossi-de-Sá, 2002; Grossi-de-Sá et al., 2015). Other secondary metabolites can still act as feeding deterrents even though they are less toxic (Simmonds, 2006) and some may have both insecticidal and antifeedant properties (e.g. Prota et al., 2014; War et al., 2020). The Apocynaceae, Rutaceae, Piperaceae, Solanaceae and many Rubiaceae are widely noted for their possession of toxic secondary plant compounds including various alkaloids, and in the case of the first of these, toxin-containing latex.

To examine the spectrum of plant defense against caterpillar herbivory, we created two lists of families, one comprising 43 well-defended species as suggested in the previously published literature (see the footnote to Table 6) and one with little obvious protection. The well-protected list includes plant families that have latex, those with abundant silica, and those with notable levels of toxins (e.g., aristolochic acid, alkaloids, cycasin, glucosinolates, and glycosides) as detailed in Section 2.2.

There are many ways of ranking herbivores on a generalist–specialist scale, although any linear scale is at best a simplification. Since we are most interested in the utilization of very diverse food plants rather than the simple number of plant species consumed (or the behavioral or morphological adaptations to the morphology or behavior of food plants), we chose to use the Shannon diversity index of the number of trophic links with different plant families. This rank order is depicted in the central column (Figure 12). We then depict quantitatively the trophic links these have with the most abundantly represented largely edible food plants (Figure 12; left-hand column) and with the most abundantly represented defended plant (Figure 12; right-hand column). The least generalized lepidopteran caterpillar families show many more trophic links with generally edible plants, particularly with Fabaceae, Rubiaceae, and Lauraceae, than they do with any of the well-defended food plant families. The most polyphagous caterpillars have relatively few trophic links with defended plants, although geometrids and non-arctiine erebids show quite a strong association with Piperaceae and Nymphalidae with Passifloraceae (the Heliconiina), Solanaceae (the Ithomiini), and Euphorbiaceae.

To test whether the difference in herbivore utilization pattern of the two plant palatability categories apparent in Figure 12 was statistically significant, we first compiled the lists of the caterpillar species that were recorded from each plant family on more than 40 separate occasions. Then, for each species in the list, we determined the number of food plant species and food plant families they were reared from as well as the Shannon diversity of the raw numbers (Figure 13). The mean values for the two palatability categories are given in Table 7. For both food plant species and food plant families, the mean values differed significantly between food plant groups (food plant species: t = 2.91, d.f. = 1,000, p < 0.005; food plant families: t = 6.21, d.f. = 1,041, p ≪ 0.001). As regards Shannon diversity, the food plant species ranges did not differ significantly (t = 1.3229, d.f. = 829, p = 0.19), but the diversity of food plant families was significantly greater for caterpillars feeding on the palatable category (t = 4.70, d.f. = 1029, p ≪ 0.001).

We interpret these results as showing that Lepidoptera species feeding on broadly palatable plant families are not particularly constrained in their food plant range by chemistry or sclerophyly and so, on average, utilize a wider range of food plant species and families than caterpillar species adapted to particular well-defended plant groups, which is not surprising. This is also indicated by the highly significant difference in Shannon food plant family diversity between the groups. However, the lack of statistical difference between groups in terms of Shannon food plant diversity may indicate that species adapted to cope with a particular class of chemical or physical defense mechanism can still utilize a similar range of food plant species within families with that defense mechanism.