Animals have evolved the capacity to learn, and the conventional view is that learning allows individuals to improve foraging decisions. We describe a first case of maladaptive learning where a parasitoid learns to associate chemical cues from an unsuitable host, thereby re-enforcing a reproductive cul-de-sac (evolutionary trap). Telenomus podisi parasitizes eggs of the exotic stink bug Halyomorpha halys at the same rate as eggs of its coevolved host, Podisus maculiventris, but the parasitoid cannot complete its development in the exotic species. We hypothesized that T. podisi learns to exploit cues from this non-coevolved species, thereby increasing unsuccessful parasitism rates. We conducted bioassays to compare the responses of naïve vs. experienced parasitoids on chemical footprints left by one of the two host species. Both naïve and experienced females showed a higher response to footprints of P. maculiventris than of H. halys. Furthermore, parasitoids that gained an experience on H. halys significantly increased their residence time within the arena and the frequency of re-encounter with the area contaminated by chemical cues. Maladaptive learning in the T. podisi - H. halys association is expected to further decrease parasitoid reproductive success and have consequences for population dynamics of sympatric native and exotic host species.

Generalized linear models (GLMs with Gaussian distribution for time data or Poisson distribution for count data) were fitted to test the effects of host herbivore species (H. halys or P. maculiventris) and parasitoid previous experience (experienced or naïve) on total parasitoid residence time and number of re-encounters with the contaminated area. Residence time data were subjected to Box-Cox transformation for normalization before the analyses. Significance of the model terms was evaluated by means of F test or Likelihood Ratio Test. Significance of the different variable levels was assessed using the Tukey HSD multiple comparisons procedure, adopting a significance level α = 0.05. Analyses were conducted under R statistical environment (R Core Team, 2020).