Parasitoid larvae consume their hosts to obtain the nutritional resources required for their development. Parasitoid wasps can optimally select the size of their hosts by laying unfertilised and fertilised eggs according to the amount of biomass available for consumption by the larvae. However, parasitoids may eventually experience low host availability within the optimal range of body sizes, affecting the survival of their offspring. In this study, we identified a situation in which all available hosts (Leucauge volupis) were smaller than those previously observed to be parasitised by Hymenoepimecis pinheirensis at the same study site. Therefore, we investigated how these parasitoids can bypass the scarcity of ideal hosts. Female wasps biased their oviposition toward the largest L. volupis females available. In this suboptimal scenario, they did not oviposit only unfertilised eggs, which developed into relatively small offspring (males). In this situation, they lay fertilised eggs mainly on larger spiders. Larval mortality was high, but the larvae attached to the larger spiders were more likely to complete their development. In general, H. pinheirensis females managed to target the best hosts available, but could not delay the oviposition of fertilised eggs or avoid offspring mortality. Here, we discuss the potential causes of asynchronies in the life cycles of parasitoid wasps and their hosts, the availability of optimal hosts, and how these factors may affect their populations.

To assess the influence of the body length (mm) of the hosts (predictor variable) on the sexes of the adult wasps that emerged from the pupae (response variable), we used a generalised linear model (GLM) with binomial error distribution and logit link function. We tested the relationship between the body lengths (mm) of the host (predictor variable) and the emerged adult wasp (response variable) using a general linear model (LM). We then performed a GLM with a binomial error distribution and logit link function to assess how the probability of the larva completing its development (response variable) was related to the body length (mm) of the host spiders (predictor variable). In this test, we only considered the larvae that were in the second half of the second instar or in the third instar at the time of collection. In this species, larvae spend approximately nine days from egg hatching to pupation, as seems to occur in other species with similar size and hosts. Therefore, it is probable that these selected larvae have undergone at least 77% of their development in the field since they took less than 48 hours in the laboratory to pupate. In both previous analyses, we opted to use the spiders' body length as the predictor variable since it represents the overall size of the hosts. Complementarily, we repeated these analyses using carapace width (mm) as the predictor variable (see Supplementary Material I). The carapace has a rigid cuticle and does not decrease if the larva has consumed a significant amount of haemolymph from the spider. Thus, this variable represents the relative sizes among the spiders at the time of the wasp attack.

To assess whether wasps selected host spiders for oviposition according to their body length, we categorised all L. volupis individuals (parasitised or not) according to their body length distribution. We followed the procedure of Gonzaga et al. (2015) and separated the spiders into intervals of 1 mm body length: class I (0-1.99), class II (2-2.99), class III (3-3.99), class IV (4-4.99), and class V (5-5.99). We performed a chi-square test to compare larval incidence among spider-size classes. Because none of the male spiders carried parasitoid larvae they were not considered potential hosts and were excluded from all analyses.

Finally, to verify whether the spider population of this study really consisted of smaller individuals than the spiders in Gonzaga et al. (2015), we obtained the original data and compared the body length of spiders between both studies. For this, we conducted a GLM with Gamma error distribution and identity link function. In this model, the body length of spiders (mm) was the response variable and the study and spider parasitism status composed the predictor variable (four levels: parasitised and non-parasitised spiders in this study and in Gonzaga et al. (2015)). We also performed a contrast analysis to compare each pair of levels of the predictor. In this test, we built a contrast matrix expecting that non-parasitised spiders were smaller than parasitised spiders in their respective study; and expecting that spiders in the present study were smaller than those in Gonzaga et al. (2015).

All analyses were conducted using R version 4.3.1 (R Core Team 2023). We graphically analysed the normality and homoscedasticity assumptions of the model residuals. We additionally evaluated residual overdispersion in the GLMs using the ‘DHARMa’ package (Hartig 2022). We used the ‘dplyr’ package (Wickham et al. 2022) to organise the data set, the ‘lme4’ (Bates et al. 2015) and ‘car’ (Fox and Weisberg 2019) packages to perform the statistical models and estimate P-values, respectively, the ‘pscl’ (Jackman 2020) and ‘rsq’ package (Zhang 2022) to calculate the coefficients of determination of the models, the ‘multcomp’ package (Hothorn et al. 2008) to conduct the contrast analysis, and the ‘ggplot2’ (Wickham 2016, Wickham et al. 2023), ‘ggdist’ (Kay 2024), ‘tidyquant’ (Dancho & Vaughan 2023), and ‘patchwork’ (Pedersen 2024) packages to build the graphics.