Network link prediction: Community structure shaped by ecological factors
Despite the uncertainty in the host range estimation, network link prediction (NLP) models confirmed the influence of the host evolutionary history on the structure of fish-monogenean communities. The host phylogeny contributed considerably to the acceptable performance of theplug-and-play algorithm (AUROC: 0.72), which outperformed the more complex Poisson N-mixture model. However, host-parasite links appear to be mostly predicted by ecological parameters as the ecosystem variable (Table 1) contributed the most to model performance (Fig. 5c). Therefore, ecological opportunity might play a major role in the assembly of cichlid-Cichlidogyruscommunities similar to neotropical fish-monogenean communities (Bragaet al. 2014), and these opportunities are likely created by host geographical and habitat distribution.
The uncovered significance of opportunity is highly relevant for aquaculture and fish conservation efforts. Introductions of infectious diseases can have devastating effects on native ecosystems (Thompson 2013). For instance, economic consequences for tilapia aquaculture were felt in some countries in association with the co-introduction of the tilapia-lake virus (Eyngor et al. 2014; Fathi et al.2017). Moreover, introductions of Nile tilapia (Oreochromis niloticus L.) and other large cichlids have led to co-introductions of their monogenean parasites, e.g. in continental Africa (Jorissenet al. 2020), Madagascar (Šimková et al. 2019), Asia (Paperna 1960; Duncan 1973; Wu et al. 2006), Australia (Wilsonet al. 2019), and the Americas (Jiménez-Garcia et al.2001; de Azevedo et al. 2006), and to occasional host switches to native fishes (Jiménez-Garcia et al. 2001; Šimková et al.2019). Our results suggest that more of these host range expansions might occur through anthropogenic introductions. Therefore, introduced populations and their surrounding environments should continue to be monitored.
Our results show that NLP can be a useful tool to verify traditional statistical analyses and to gain further insight into ecological and evolutionary mechanisms shaping host-parasite interactions. For instance, we inferred that the trophic level of the host is one of the more informative predictors of cichlid-Cichlidogyrusinteractions. Host size, life style, and parasite phylogenetic and attachment organ morphological parameters also improved model performance (Fig. 5b). In contrast, previous studies on fish parasites have delivered inconclusive results for the role of host and parasite traits on parasite community composition. Parasite community composition correlated with the host trophic level in some cases, e.g. for shelf fish off Buenos Aires (Timi et al. 2011), but not in others, e.g. for freshwater fish in Canada (Locke et al. 2013) and marine fish in Finland (Locke et al. 2014). No studies investigated the effects of life style as coded here (Table 1) but other studies suggest that host habitat preference can affect parasite communities (Locke et al. 2013). Host size was suggested as important predictor for the community composition of ectoparasitic monogeneans (Sasal & Morand 1998; Sasal et al.1999; Šimková et al. 2001; Desdevises et al. 2002; Morandet al. 2002). However, these correlations might reflect phylogenetic patterns of host size (Poulin 2002) explaining the variable importance of host size here. Lastly, no correlation ofattachment or reproductive organ morphology with community composition was found for species of Cichlidogyrus unlike for other monogeneans, e.g. Dactylogyrus (Šimková et al. 2001; Jarkovský et al. 2004). Instead, the morphology mostly reflects phylogenetic relationships of the parasites (Vignon et al. 2011; Cruz-Laufer et al. 2021b). The results of these studies highlight the challenge of linking host and parasite traits with community composition parameters and generalising observed patterns as sampling biases (Fründ et al. 2016) (Fig. 5a) and character coding (Pavoine et al. 2009) can influence the results. NLP can complement these analyses by indicating possibly undetected interactions (Fig. 5b) and assessing the predictive power of the ecological, evolutionary, and morphological parameters (Fig. 5c).