Food deprivation and crowding shape stage‐dependent cannibalism‐associated mortality in caterpillars

Cannibalistic interactions in herbivorous insect larvae are strongly context dependent, but how food deprivation, crowding and developmental stage shape cannibalism‐associated larval loss, and whether loss patterns vary among thermal settings, remain poorly resolved. We conducted two complementary laboratory experiments using third‐ and fourth‐instar larvae of Satyrium formosanum (Matsumura) (Lepidoptera: Lycaenidae), a plant‐feeding butterfly associated with Sapindus mukorossi . The first experiment tested food‐deprivation duration, larval density and instar at 25°C. The second examined larval density, food availability and instar across five incubator temperature settings during a 24‐h assay. Because each temperature setting was represented by one incubator, differences among settings were treated as exploratory. Food‐deprivation duration strongly affected cannibalism‐associated larval loss, and density‐related patterns differed between larval instars. The highest model‐predicted loss occurred in third‐instar larvae after 48 h of group food deprivation at a density of 20 larvae per box, reaching 68.9% (95% CI: 56.9–78.8%). At the same endpoint, predicted loss at the highest density of 40 larvae per box was lower, indicating a non‐monotonic density response. In the 24‐h temperature–density–food experiment, food absence was associated with higher loss primarily at densities of 20 and 40 larvae per box, whereas model‐predicted loss under food‐present conditions was generally low. Larval loss also varied among incubator settings, but these differences cannot be interpreted as independently replicated causal temperature effects. Qualitative observations confirmed biting and attacks on living conspecifics, including simultaneous feeding by multiple larvae. Together with the complete survival of all single‐larva controls, these observations support conspecific attack and consumption as the principal source of grouped‐box losses. Because individual interactions were not continuously monitored, the quantitative endpoint was interpreted conservatively as cannibalism‐associated larval loss. These findings demonstrate that food deprivation, crowding and developmental stage should be controlled explicitly in group‐rearing experiments and larval bioassays.