ii) Larval and host plant data
We collected data on the larval social behaviour, Passifloraceae host species and adult wingspan of the Heliconiini included in each phylogeny (all data with references can be viewed in online repository). Data on larval social behaviour, recorded as a binary variable (0 = solitary, 1 = gregarious) were based on previous classifications of egg clutch size range (Brown, 1981; Beltran et al., 2007), with gregariousness recorded for any species with a range maximum ≥ 10 (as per Beltran et al.’s 2007 categorisation). We additionally analysed social behaviour as a categorical variable (0/G1/G2/G3) as per records of four levels of increasing ‘sociability’ based on clutch size (Brown, 1981): Solitary species were recorded as zero. Gregarious level one (G1) species were any with a minimum of 10 eggs in their clutch size range, and those with median clutch sizes between 10.5 and 24. Finally, gregarious level two (G2) and three (G3) species were those with median clutch sizes between 25 and 30, and greater than 30 respectively. Use of Passifloraceae host species were mainly taken from published sources (Benson et al. 1975; Kozak, 2016; Young et al. 2023). For a small number of species, all in the ‘Neruda’ clade of the genus Heliconius , we only recorded the host at the genus level (Dilkea ) as we lacked species-level data, this meant we could not confirm how many separate species within this genus are used as hosts, despite there being 13 accepted species withinDilkea (POWO, 2022). The four species in the Neruda clade occur at lower densities than other Heliconius and lack some derived features of the genus. Re-running the analyses with these genus-level entries omitted had no meaningful effect on the results (Table A1).
Trait data on the Passifloraceae hosts were taken from two main sources (Benson et al., 1975; Ulmar and MacDougal, 2004). To test for the effect of host habitat on larval behaviour, we recorded four distinct, main habitats (forest interior, forest edge, open areas and humid glades), taken exclusively from Benson et al. (1975). To test if the presence of leaf trichomes predicts larval gregariousness, we recorded the vestiture (the presence or absence of leaf trichomes) of hosts as a binary variable (0 = glabrous (lacking trichomes), 1 = pubescent (possessing trichomes)), where leaves listed as ‘nearly glabrous’ were recorded as glabrous. A plant’s growth habit may give an indication of its size, for example a tree is likely to be bigger than a shrub, and it is possible that larger plants are better at supporting gregarious larvae. However, the Passifloraceae included in this study show a considerable lack of diversity in growth habit (only ~ 9% are not vines), making this trait an uninformative proxy for comparing overall host size. Instead, we focused on leaf size as an indicator of above ground biomass (e.g. Digrado et al., 2022). To determine whether larger leaves predict gregarious larvae and/or larger gregarious larvae, we recorded the mature leaf size of each Passifloraceae in our dataset, given by Ulmar and MacDougal as separate ranges (minimum and maximum) of leaf length and width (both cm). Here, we first calculated the median value of these ranges, then multiplied one median by the other to gain an estimate of the median leaf surface area, assuming an idealised leaf shape, which was used as a final value for ‘leaf size’(de Luna Souto et al., 2017). We acknowledge that, given the leaf shape diversity inPassilfora , leaf area is likely a crude estimate of overall leaf tissue available to larvae.
Finally, we obtained adult wingspan data from recordings of specimens collected in the field (Couto et al. 2022) and images from Butterflies of America (Warren et al., 2016). Where available, we measured images of five separate specimens, then calculated the mean wingspan (mm) for that species. Adult wingspan was used as a proxy for final instar larval size, as this has previously been found to be reliable (Nilsson and Forsman, 2003; Hebberecht et al., 2022; McLellan et al., 2023). This allowed us to test for a relationship between approximate larval size and host leaf size, to assess whether larger leaves are generally required to support larger larvae. Leaf size and wingspan data were log transformed for all analyses.