Discussion
Here, we use community modelling to uncover determinants of wild plant virus diversity, a vastly unexplored component of biodiversity (Roossinck 2011). We show that virus communities exhibit a clear nested structure with abundant co-occurrences, both pairwise and higher-order. The observed coexistence patterns are mediated by host plant characteristics, abiotic environment and spatial structure, as well as associations among the viruses. Many present-day threats posed by infectious diseases involve interactions that are manifested across nested scales of biological organisation (Johnson et al. 2015), and our findings shed light on how plant virus coexistence is maintained at different spatial scales.
The simplest within-host viral communities detected in our data consisted of single infections, but nearly half (46%) of the infected plants hosted multiple infections. Single infections were typically observed only for the few most common viruses, and half of all the viruses never occurred as single infections. Our findings are in line with other studies that have found high levels of coinfection (Al Rwahnih et al. 2009; Rey et al. 2012; Tugume et al.2016). However, our approach enabled the detection of virus communities in a substantial number of host individuals, with capacity to detect viromes consisting of up to 24 distinct taxa. This suggests that virus communities are highly variable within and among hosts.
From the difference between species-area and coexistence curves, we see that a larger number of hosts is needed to maintain the coexistence patterns that could be derived from the overall virus richness. Indeed, increasing environmental heterogeneity and varying responses of viruses to this heterogeneity promote coexistence by reducing competition, as predicted by the classic species sorting paradigm, which has been shown to be influential for microbial communities (see e.g. Székely & Langenheder 2014 and refs. within). Combined with the weakening effects of demographic stochasticity and the homogenising effects of dispersal, these mechanisms lead to more stable coexistence at increasing spatial scales (Hart et al. 2017; Levine & Hart 2020).
We found that the conditional network models (CRFs), outperformed the model incorporating only associations between viruses (MRF) in explaining virus community structure. The differences between the CRF model variants were less pronounced, as seen from their equal model fit and performance. Hence, we conclude that there is environmental variation and processes operating at different spatial scales that influence the wild virus community coexistence structure both directly and indirectly. The environmental characteristics and the spatial variables explain community structure almost interchangeably. Previously, e.g. within-host diversity of pathogens has been shown to increase with latitude, while pathogen turnover follows an opposite trajectory, suggesting limited transmission in lower latitudes (Seabloomet al. 2010).
Our analysis revealed spatial variables to influence associations between viruses, resulting in indirect effects on their co-occurrence. These viruses are vector-dispersed, and most likely dispersal-limited at larger spatial scales (Pleydell et al. 2018), leading to less homogenisation at these scales. However, at the scale of a host populations, variation among host genotypes (Sallinen et al.2020) and demographic stochasticity (through e.g. herbivory, as seen from the indirect effects of herbivore damage on the associations between viruses) are expected to be important, as well as abiotic interactions mediating coexistence (Kozanitas et al. 2017).
After accounting for the effects of the host and habitat characteristics, as well as spatial structure on virus distributions, non-random, significant positive associations between viruses remain. Although these associations are based on purely observational co-occurrence patterns, given the spatial scale of our sampling and our method for detection that targets the host’s defence response (Kreuzeet al. 2009), we consider these indicative of biotic interactions between the viruses within a host plant (Wintermantel et al.2008). The interactions among coinfecting viruses may involve positive effects on replication (Pruss et al. 1997; Taiwo et al.2007), or even obligate dependencies as some viruses require their specific helper virus in order to complete their lifecycle (DaPalmaet al. 2010). Viruses may also suppress host immunity allowing subsequent infections to escape recognition by host immunity (González-Jara et al. 2005). As expected, the inclusion of explanatory variables reduced the number of direct associations between viruses identified by the models, suggesting that shared environmental responses play an important role in the assembly of the communities (see e.g. Leathwick et al. 2006; Ovaskainen & Soininen 2011). After accounting for host- or habitat-related or spatial variables, support for several direct associations (e.g. Tombusviridae-Alphaflexiviridae, Betaflexiviridae- Potyviridae and Tombusviridae-Betaflexiviridae) was no longer detected.
Importantly, we found that the inclusion of the additional conditioning variables changed the association networks especially through their indirect effects. For example, we found a significant indirect effect of the coarsest-scale spatial variable (MEM1) on the association between Fimoviridae and Alphaflexiviridae, while the direct effect of this spatial variable was not significant. Hence, the occurrence probability of Alphaflexiviridae is jointly affected by the occurrence of Fimoviridae as well as spatial structure. Due to symmetry, Fimoviridae is similarly affected jointly by the occurrence of Alphaflexiviridae and spatial structure. These indirect significant effects were more frequent (in total 32 effects) in comparison to direct effects (in total 10). The indirect links can be indicative of either biotic interactions between viruses only manifesting under certain environmental conditions, or these environmental conditions having an effect only in interaction with the other virus (Kozanitas et al. 2017). Such indirect effects have traditionally been challenging to detect, yet not accounting for them can lead to over- or underestimation of signals of biotic interactions in co-occurrence data (Blanchet et al. 2020).
Although our virus community data set along with its environmental explanatory variables is extensive and of high quality, our results are limited by our sample size and its effects on the parameter estimation of our modelling method. We use regularisation to avoid overfitting, but we note that the estimated parameters could in theory change with more data included. However, as seen from the species-area and coexistence-area curves, our sampling effort captures the detected virus diversity already before all the samples have been included, indicating a promising sample size.
Our results demonstrate that natural plant virus communities are characterised by single infections of few, dominating virus taxa as well as diverse, non-random coinfections. Virus diversity can be explained by coexistence-promoting mechanisms, some of which we could tease apart with our modelling. We show that host and habitat characteristics, as well as spatial structure, resolve some of the observed co-occurrence patterns, to some degree interchangeably. Importantly, we find that some virus-virus associations are mediated by either host or habitat characteristics, or the spatial structure of the host populations. However, a substantial part of the explained virus co-occurrence patterns can be attributed to positive, direct associations among the viruses. Moreover, we show that additional conditioning variables changed virus association networks especially through their indirect effects. Thus, our study contributes to increasing understanding on how plant virus coexistence and thus biodiversity is maintained at different spatial scales. Our results highlight a previously underestimated mechanism of how human-driven environmental change can influence disease risks by changing biotic associations between viruses that are conditional on their environment.