Network structure
We recorded 163 interactions between 62 snake species and 26 food resources (Figure 1 and Supplementary material Appendix 1 Figure A6) that were heterogeneously distributed among snake species, where most of them had few interactions (56.45% snake species interacted with one or two resource categories) and few species had many interactions (6.45% interacted with more than five resources; Supplementary material Appendix 1 Figure A1). The network structure show moderate connectance (C = 0.101) (Table 1), indicating that, from the variety of food items consumed by snakes, the species analysed use, on average, 2-3 resources. The snake-resource network also show significant nestedness (N = 33.14, p < 0.01), indicating that 1/3 of the interactions of the less connected species represent a subset of the interactions of the most connected species. Finally, the network also show significant modularity (M = 0.51, p = 0.03), indicating that the number of interactions within each modules is 51% larger than what is expected for a network with the same number of modules, the same number of interactions per species, but with random interactions between species.
Some snake species showed extreme specialisation, such as Dipsasspp., which feed exclusively on mollusks, and Drepanoides anomalus that rely upon eggs of squamate reptiles. Other species, such as Atractus spp., although specialist in the consumption of earthworms, may also feed on insects. Similarly, Micrurus surinamensis primarily consume fish but secondarily consume lizards and snakes. On the other hand, we found very generalist species, such asBoa constrictor and Epicrates cenchria , which interacted with six resource types, Corallus hortulanus , which interacted with eight resources and Eunectes murinus , the largest species of the network, which interacted with 11 resources. Among the food resources consumed by many snake species were lizards (24% of all interactions), anurans (16%), and small mammals (9%), comprising rodents and marsupials. Among the least consumed resources were large mammals (such as cervids), turtles and alligators, only consumed byEunectes murinus , onychophorans only consumed by Micrurus hemprichii , gymnophiona only consumed by M. lemniscatus , and salamanders, which were only consumed by Chironius fuscus(Supplementary material Appendix 2 Table A1).
To assess whether there was a difference in network structure based only on the presence of primary resources in the snake diet, we removed all non-primary resources and reanalysed the network. Even after removing the secondary resources, network average degree and connectance remained within the same range values (Table 1). The results also indicated that there was no significant difference between the network nestedness values with and without the presence of secondary resources (p = 0.147). Even after the removal of non-primary items, the network remained significantly nested (N = 29.46, p < 0.01). In contrast, the modular structure was nonsignificant after removal of non-primary resources (M = 0.47, p = 0.44). Similarity, to check if the type of food resource categorization could affect the network patterns, we used the same metrics to analyse the more specific and the less specific networks. Our results hold even with different levels of detail on resource description and all networks remained significantly nested and modular (Supplementary material Appendix 1 Table A2).
Our results supported the prediction that there is a positive association between the number of resources consumed and average body size (slope = 1.41, R2 = 0.46, p < 0.01, Figure 2), indicating that in general the largest species of snakes showed a greater number of food interactions. Exceptions to this pattern were Corallus caninus (k = 3) and Lachesis muta (k = 1), both specialists in the consumption of mammals. Among the seven largest snake species, five of them (Eunectes murinus, Boa constrictor, Epicrates cenchria, Corallus hortulanus, and Corallus caninus ) belong to the family Boidae. Thus, this family is over-represented among the set of heavier snakes in the network and our analysis may be biased by the confounding factors generated by all other traits shared by boid species. To circumvent this problem we explored if the correlation between average body mass and degree holds within speciose snake families. We performed correlation analyses between degree and average body mass for species of the family Colubridae and for those of the family Dipsadidae, the two largest snake families in the network. The results indicated that a positive correlation between average body mass and the number of resources consumed hold even for non-boid snakes and partially controlling for phylogenetic effects, there was (see Supplementary material Appendix 1 Figure A4 and A5).