Many dipteran larvae form social foraging clusters which can dig deep into food and might, in part, be a way to escape parasitic wasps. Clustering larvae generate a tunnel into the food substrate (Dombrovski et al. 2017). Clusters might protect against wasps in two ways: by drowning predators trapped in deep cavities, or by preventing wasps from entering these structures. These hypotheses were tested in the following experiments. While most fly larvae cluster, in this paper we focus on Drosophila melanogaster CantonS (Dmel) prey as it is the most robustly tested larvae population for clustering parameters. Two parasitic wasps were used, Leptopilina boulardi (Lb) and Leptopilina heterotoma (Lh), which both prey on CantonS (CS) larvae but have differing features—Lb using a passive immune evasion that prevents immediate immune cell death in the host, while Lh injects an active immune suppression venom that directly kills host immune cells (Schlenke et al. 2007). However, both wasps inject a complex cocktail of immune- and neuro- affecting factors that alter many aspects of host function (Schlenke et al. 2007).

Observations of wasps around clusters

CantonS larval clusters were established as described (Liao et al. 2024). About 100 L2 larvae were added to a pre-used 25 mm vial and allowed to form clusters after 3 days. About 50 wasps were added, a mixture of male and female adults. Wasp behavior was video recorded from the side and from above at 6 frames/minute. Wasps of both species are seen to regularly enter clusters (Fig. 1). In addition, while many dead wasps were seen buried in the food, no cases of these getting caught specifically by clusters were documented. High-resolution video recording was also conducted and, aside from some retreat movements to the ovipositor, larvae and clusters remained unperturbed by wasp presence (Movie S1a–d). The conclusion from these observations is that while wasp death does occur with larvae, it was not observed specifically in clusters and that wasps enter these structures quite freely. Therefore, wasp death was measured specifically in the context of differing amounts of clustering.

Fig. 1

Observations of wasp behavior. A–B. Lb (A) and Lh (B) wasp death is seen as a layer of carcasses (arrow). C–D. Magnified view of clusters in vials showing Lb (C) and Lh (D) wasps in clusters. E. Hi-resolution still images from videos of wasps in clusters, Lb (E) and Lh (F)

Do clusters specifically kill wasps?

To measure wasp death, 5 male and 5 female wasps were placed in a vial with 100 L2 larvae which begin clustering the next day (Fig. 2). The rationale for using male and female is that females might be more susceptible to drowning as they spend more time on the food stinging. At least 4 vials were used for each measure and wasps counted every day. The number of days needed to go from 3 to 2 surviving wasps of each gender was determined and plotted (Fig. 2). Varied larval conditions were used: 0 and 10 larvae will have no clustering while 40 and 100 larvae do cluster (Liao et al. 2024). Placing 40 or 100 larvae in the dark greatly reduces clustering as it has been previously shown that vision is required for clustering (Dombrovski et al. 2019). Larvae with genetically induced blindness is observed to enter clusters and interfere with the required coordination, thus dismantling the structures (Dombrovski et al. 2019, 2020). No male/female differences were seen across both species and in all clustering conditions. This indicates that stinging behavior specifically does not potentiate drowning. Both Lb and Lh wasps show increased mortality with increasing numbers of larvae, but this is not offset by maintaining vials in the dark. Therefore, while larvae in high numbers do kill wasps, this is not likely based on clustering behavior.

Fig. 2

Adult wasp death. A Number of days for over 2 of 5 added male (blue) or 5 female (pink) Lb wasps to end up drowning in food. Number of vials used for each measure indicated. Statistics were compared using ANOVA and significance or probabilities are indicated when non-significant. B Same as 2A. except Lh wasps

Does clustering affect wasp infectivity?

While larval clusters do not seem to potentiate the capture and drowning of wasps, they could reduce infectivity by making it more difficult to sting. To examine infectivity, 10 wasps (5 male and 5 female of both species) were placed in vials along with various numbers of larvae. Upon complete pupation, all wasps were removed and incubation completed to measure the number of wasps. Infectivity was calculated as the number of wasps produced per larva in each vial. For Lb, infectivity was about 90% for vials of clustering larvae independent of population size above 40 (Fig. 3a) which is similar to previous reports (Jones and Hurst 2020). However, upon reduction of clustering either via incubation in the dark or addition of clustering-spoiler blind larvae, infectivity is reduced significantly. These data indicate that clustering, in fact, favors infectivity. In contrast to Lb however, Lh shows a steady 80% infectivity across all larval states. These data indicate clustering enhances infection success for Lb but not for Lh, revealing a species-specific interaction between parasitoid strategy and larval social behavior. This raises the question, if clustering favors infectivity, how might parasitic infection affect larval clustering?

Fig. 3

Wasp infectivity. A Infectivity was calculated as percentage larvae successfully infected with wasp eggs. Statistics were compared using ANOVA and significance or probabilities are indicated when non-significant. B Same as 3B except using Lh wasps

How does infection affect clustering?

To measure the effect of infection on clustering, infected larvae were generated by placing 40 L2’s in the vial with at least 20 wasps of both species. Larvae were removed 4 days later, at the third day of the L3 stage, at which time maximal clustering is most often achieved. Under these conditions, over 90% of larvae are infected (Fig. 3). Infected larvae were placed in a 2D clustering assay and monitored for behavioral changes. Clusters were observed for Lb-infected larvae but not for those of Lh (Fig. 4a). Lb clusters, however, appeared more disorganized than those of uninfected CS (Fig. 4b). Specifically, larvae posterior spiracles are not as aligned as is generally observed (Dombrovski et al. 2017). Clustering was measured as described by averaging the number of larvae in clusters over 3 time points per assay. Compared to uninfected CS, Lb-infected larvae cluster more and Lh do not. Because inclusion in clusters requires coordination between larvae, it is possible that Lb-infected larvae might be better at this synchronization. However, measuring the coordination between larvae using high-resolution video recording (Fig. 4d) indicates the opposite. Uninfected larvae coordinate with a 0.6″ delay: when one moves, its closest neighbor moves about a half second later. Infected larvae delay on average 1.2″, which is nearly statistically random for a 2″ locomotion cycle, or no coordination (Dombrovski et al. 2017). Therefore, some means, other than coordination, keeps these larvae in clusters, which has been observed for salivary-gland-depleted larvae (Liao et al. 2024). To test whether Lb-infected larvae stay in uninfected clusters longer, food-coloring-labeled infected animals were added to uninfected clusters and the time residing in the cluster measured (Dombrovski et al. 2017; Liao et al. 2024). Lb-infected larvae were observed to spend about twice as long in a cluster over uninfected (Fig. 4f). Lh-infected larvae do not enter clusters (Fig. 4e). In conclusion, Lb infection appears to increase clustering by increasing the average time a larva resides in a cluster.

Fig. 4

Infected larval clustering. A Larvae infected with Lb (upper panel) or Lh (lower panel) were placed in a 2D cluster assay. Robust clusters form for Lb infection but not for Lh. B Close up of a Lb-infected larval cluster. Larval posteriors are not as well aligned as generally observed. C Clustering is higher in Lb-infected than for control. Only one brief cluster was seen for 11 Lh infected larval assays. Control is from a previous study(Dombrovski et al. 2017). D Infected larval (arrow) inclusion into an uninfected cluster. E Coordination of larvae was calculated from high resolution videos. F Cluster inclusion time for Lb and Lh infected larvae. 15 Lh-infected larvae were examined and only 3 entered clusters briefly. Control is from a previous study (Dombrovski et al. 2017). Statistics were compared using ANOVA and significance or probabilities are indicated

Do uninfected larvae affect the survival of the infected?

Previous studies have indicated that stung larvae can be subjected to cannibalism from those uninfected. Such behavior may confound the below measures. In the transplant experiments shown in (Fig. 4e), no Lb-infected larvae were observed being attacked by their uninfected neighbors. Lh larvae are regularly seen to sink (Movie_S4) below clusters, and this might be just due to lack of locomotor ability. To measure infected larval mortality, 10 early third-instar Lb- or Lh-infected larvae were placed in a vial with various numbers of uninfected cohorts. Mortality was measured by counting how many wasps came out of each vial. Lb-infected larvae show increased loss when there are increasing numbers of uninfected cohort-mates. In addition, incubation in the dark, which reduces clustering behavior (Dombrovski et al. 2017, 2019, 2020), decreases the Lb death rate as measured for 100 × uninfected larvae. Lh-infected larvae (Fig. 5b) have about a 50% survival rate uninfluenced by other uninfected cohorts residing within their vial. While no Lb-infected larvae were seen to be specifically attacked in clusters, it is possible that they are predated as pupae. To test this, 10 Lb-infected larvae were placed in a vial and allowed to pupate. These were marked in the vial. Various numbers of uninfected cohorts were added, and the pupae were counted every day for 5 days. Pupae that were removed and pulled into the food by uninfected larvae were counted. On average, about 2 out of 10 are lost off the vial side, regardless of the conditions (Fig. 5c). This does not account for the high mortality of Lb-infected larvae when with a high number of clustering cohorts.

Fig. 5

Survival of infected larvae. A 10 Lb infected early third instar larvae were added to a vial and survival measured as final wasps produced. Uninfected age-matched sibs were added at indicated numbers. B Same as 5A except that Lh infected larvae were used. C Lb infected pupal death was measured by allowing 10 larvae to pupate and then monitor them until wasp hatching. Various number of uninfected larvae were added which pupated around their infected sib. Pupae that were lost off the side of the vial were scored as lost. For A, B and C, the number of vials used for each measure indicated. Statistics were compared using ANOVA and significance or probabilities are indicated when non-significant

Correlations of findings with clustering

To summarize the findings in Figs. 1, 2, 3, 4, 5, cluster frequency was estimated from previous studies and compared to adult wasp death, infected larval death, and infectivity (Fig. 6). There is little correlation between the amount of clustering and wasp death (Fig. 6a). While wasps die more with more larvae, this is more likely a result of increased food fluidity than actual clusters. However, there is a correlation between clustering and infected larval death when there are uninfected cohorts. Countering previous hypotheses, this may not be due to cannibalism, as no such behavior was observed in any of the cluster transplants in 2D assays. However, these observational studies did not look outside of vial clusters for cannibalism or other signs of larval death from the many other changes in altered food. This might instead be due to infection through the sting site. Finally, there is a correlation between infectivity and clustering (Fig. 6c). The more larvae cluster, the more wasps can infect them.

Fig. 6

Summary of correlations with clustering. A–C. Clustering frequencies measured in previous studies(Dombrovski et al. 2017, 2019, 2020; Williamson et al. 2021; Liao et al. 2024) were compared to adult wasp death (A), infected larval survival (B) and infectivity (C). The correlations were scored by R2 calculation. While adult wasp death is not related to clustering, infected larval survival and infectivity is. D. Model for the role of clustering in the parasitic wasp life cycle. Two phases might exist for Lb infected larvae: at low frequency these infected larvae are at high risk for being killed by the uninfected. At higher frequency of infection, larvae are drawn more into clusters which further increase the degree of infection