Host repertoire
We estimated the realised host repertoire as the structural and phylogenetic specificity of species of Cichlidogyrus (Esseret al. 2016). Structural specificity was quantified through the specialisation index di’ with significance levels inferred from NM1 and NM2. This index measures the deviation from a perfectly nested network and is derived from Shannon’s entropy, which takes diversity, abundance, and evenness of interactions into account (Blüthgen et al. 2006, 2008). Phylogenetic specificity was quantified as mean pairwise distance (MPD) and mean nearest taxon distance (MNTD) of host species to investigate ancient and recent relationships respectively (Clark & Clegg 2017). For the null model (NM3), we randomly redistributed the species labels of the host phylogenetic distance matrix to test if phylogenetic distances differed significantly from a random distribution. We calculated MPD, MNTD, and 1000 null estimates through the functions mpd , mntd , and taxaShuffle in theR package picante v1.8.2 (Kembel et al. 2010) for each parasite tree. We assessed the MPD and MNTD through the Z-scores provided by these functions. Negative Z-scores indicate a greater phylogenetic distance than expected under NM3, positive scores a smaller distance. Significance was assessed as proportion of MPD and NTD values smaller (negative Z-scores) or greater (positive Z-scores) than for the observed network.
Next, we investigated the effects of host environment and phylogeny on the host repertoire of the parasites. We calculate MPD and MNTD from functional-phylogenetic distance (FPDist) matrices. These matrices are derived from functional (FDist) and phylogenetic (PDist) distance matrices of the host species infected by each parasite species. We inferred the FDist matrices from the host niche dendrograms and the PDist matrices from the host phylogenetic trees and scaled the matrices by dividing the values through the respective maximum distance. We accounted for uncertainty in the host niche/tree topology by drawing random samples of dendrograms/trees every time we calculated FPDist = (aPDistp + (1 - a)FDistp)1/p (Cadotte et al.2013) with p = 2 to calculate Euclidean distances (see Cadotte et al. 2013; Burbrink et al. 2017; Clark & Clegg 2017). Finally, we applied 100 incremental increases from 0 to 1 to the weighting parameter a . We calculated FPDist through the functionFPDist in the R package funphylocom v1.1 (Walker 2014). We generated null distributions by randomly redistributing the labels of the FPDist matrices resulting from 1000 random draws from FDist and PDist matrices. We applied the same redistribution algorithm as for NM3. For interpretation of the FPDist plots, we followed Cadotte et al. (2013).