Figure and table legends
Figure 1. A schematic representation of direct and indirect associations between viruses in the context of their hosts and environment. Within plant populations, individual plants host viruses. The viruses have associations with each other, either through direct mutualistic or antagonistic interactions (bold font), or indirect associations mediated by another virus, shared vector, and/or host immunological responses (small font). The environment can influence the (co-)occurrences either directly or through host effects at the level of host populations or individuals, or via the vectors (capital font). The solid bold arrow represents the direct association between two viruses, whereas the dashed arrows show the confounding indirect associations. The additional conditioning explanatory variables included in our constrained models are illustrated. The indirect effects (mediator species, changes and variation in host immune response) are included in the models implicitly through spatial variables and the model structure.
Figure 2 . Virus communities by population. In A, each pie chart represents the virus community of a host plant population, overlayed on the map of Ă…land. Infection load is shown with greyscale, as indicated in the legend. Sliced portions of same shade of grey show plants with equal number of viruses, but in different combinations. B) Nestedness of the virus communities when aggregated to host population level. Two most contrasting communities are shown in more detail in C and D. In these barplots, the within-host virus community composition of a virus-rich plant host population 861 (D) and virus-poor population 3222 (C) is shown with different colours indicating different virus taxa, as shown in legend.
Figure 3. Viral co-occurrence structure. A) Virus occurrences (diagonal) and co-occurrences (off-diagonals), organised according to the first principal component of the incidence matrix. The colouring of the font intensifies with the value. The diagonal elements show the overall number of occurrences of the focal virus, whereas the off-diagonal elements show the numbers of host plants where the two viruses co-occur. The numbers inside brackets next to virus names indicate the number of single infections. B) Species-area curve and coexistence curves. The uppermost black dotted line indicates the species-area curve, the grey dashed line indicates the maximum amount of possibly coexisting pairs (calculated from the species-area-curve), and the black solid line shows the actual average coexistence curve.
Figure 4. Virus networks. A) Markov random field co-occurrence network, representing direct pairwise associations between viruses, after accounting for the rest of the community (MRF). B) Constrained Markov random field co-occurrence network, representing direct pairwise associations between viruses, after accounting for the rest of the community as well as all environmental and spatial effects (CRFfull). C) Direct associations explained by host-related variables (CRFhost); D) direct associations explained by habitat-related variables (CRFhabitat); and E) direct associations explained by spatial variables (CRFspat). The viruses closer in space have stronger positive associations with each other and edge thickness is scaled by the strength of association. In summary, when the associations in panels C-D are subtracted from all the associations in panel A, the associations in panel B are left. Hence, the dashed lines in C-D represent associations that disappear from panel A after adding the explanatory variable group in question, resulting in the permanent associations illustrated with solid lines in panel B.
Table 1. Explanatory conditioning variables. See Supplementary Material and Methods for more thorough description and justification of the explanatory variables.
Table 2. The direct effects of the additional explanatory variables on the virus occurrence probabilities and the indirect effects of the additional explanatory variables through their influence on the association between a focal pair of viruses. Effects\(\geq|\)0.25| are highlighted with bold font. All mean coefficient estimates are significant (based on bootstrapping and a 90% confidence interval).