Introduction
Viruses, like all organisms, occur as communities with other species (Díaz-Muñoz 2017). Importantly, the diversity of co-occurring viruses may have profound impact on virus evolution and epidemiology (Alcaideet al. 2020). Hence, understanding the conditions and scales at which virus communities vary is critical when predicting how virus communities respond to environmental change or to disease control measures (Gilman et al. 2010; Malmstrom et al. 2011; Massart et al. 2017; Halliday et al. 2020). Over the past decade metagenomic surveys exploring virus diversity in wild hosts ranging from plants to insects and mammals (Wren et al. 2006; Roossinck 2010, 2012; Ng et al. 2011; Letko et al. 2020) have revealed a tremendous, largely undescribed, virus diversity across environments. It is becoming clear that the diversity of virus taxa even within the same host can be highly variable, and that viral co-occurrences are common in nature (Roux et al. 2015; Díaz-Muñoz 2017; Munson-Mcgee et al. 2018; Alcaide et al. 2020).
Predicting how pathogen communities respond to environmental change requires a principled community ecology framework to disentangle possible drivers of coexistence patterns (Johnson et al. 2015; Seabloom et al. 2015). As Popovic et al. (2019) recently outlined, there are three main drivers of observed species co‐occurrence patterns (Figure 1). First, two species might share similar responses to environmental variables such as temperature (Suzuki et al. 2014; Obrępalska-Stęplowska et al. 2015; Alcaide et al. 2021). Second, two species may exhibit similar responses to the occurrence of a third species: for example, some viruses require another virus species to mediate their replication within a host (Pirone & Blanc 1996; Syller 2012), resulting in an indirect association between the dependent species. Third, two species might exhibit a direct, biotic association: a virus can for example facilitate the establishment of another (Six & Klug 1973; Waterhouse & Murant 1983). Hence, the sequence of arrival (Fukami 2015), can have far reaching implications for the structure of pathogen communities (Karvonen et al. 2019).
Spatial structure and abiotic habitat conditions are considered important for structuring pathogen communities (Bergner et al.2020) via their effect on both hosts and pathogens (Makiola et al. 2019) (Figure 1). While demographic stochasticity is important at smaller spatial scales (Tilman 2004), abiotic environmental heterogeneity increases with increasing spatial scale, promoting greater coexistence through species-specific environmental responses (Chase & Leibold 2003). Dispersal presumably decreases coexistence by forcing species to interact and by homogenising intra- and interspecific interactions (Snyder & Chesson 2003). Most plant viruses are vector-dispersed, and thus the distribution of vectors can also influence the structure of plant pathogen communities (Schröder et al. 2017). A special characteristic of pathogen communities is that their immediate environment is a living organism in itself, and hence the influence of the abiotic environment (e.g. weather) on pathogens can be either direct or mediated by the host and/or vectors (Figure 1). Moreover, the distinction between abiotic and biotic effects becomes blurred, as the interaction between a host and a pathogen can be considered both an environmental effect as well as a biotic association (Figure 1). Pathogens can only occur where they have susceptible hosts and thus spatial variation in resistance is expected to have direct impacts on pathogen (co-)occurrence patterns (Jousimo et al.2014; Carlsson-Granér & Thrall 2015).
Ecological knowledge often relies on observational methods, but inferring signals of biotic interactions from co-occurrence data is challenging (Blanchet et al. 2020). However, studying species associations through the perspective of conditional probabilities helps to overcome some of these challenges. Conditional probability refers to the probability that two species will be found together aftercontrolling for the other species in the network. Implementations for ecological applications have been recently developed (Harris 2016; Clarket al. 2018; Popovic et al. 2019). Markov random fields (MRFs) are a group of graphical network models which enable the estimation of conditional dependencies from networks of interacting variables (Sutton & McCallum 2011; Clark et al. 2018). In addition to analysing conditionally dependent species co-occurrence patterns after accounting for the occurrences of all other species in the community, also additional covariates can be included, resulting inconditional random fields (CRFs) (Azaele et al. 2010; Harris 2016; Clark et al. 2018; Popovic et al. 2019). MRFs and CRFs allow us to discover how species are associated with each other and their environment, and importantly also how the environment influences these associations (Clark et al. 2020).
Here, we aim to determine the relative roles of spatial environmental heterogeneity and both direct and indirect virus-virus associations in determining the coexistence structure of within-host virus communities of naturally occurring Plantago lanceolata populations, identified via deep sequencing of small RNAs from field collected plant samples (Kreuze et al. 2009). To overcome the challenges of inferring signals of biotic interactions from co-occurrence data (Blanchet et al. 2020), we incorporate relevant environmental covariates at the scale of both viruses and the hosts, analyse co-occurrence patterns through conditional probabilities, and use a reasonably-sized data set of 400 plants, sampled hierarchically at biologically relevant scales (across a population network and within populations), covering a wide environmental range. Specifically, we ask:
Q1. What are the relevant spatial scales of virus diversity and co-occurrences, and can we detect signals of coexistence mechanisms influencing the relationship between viral diversity and co-occurrence at different spatial scales? Q2. Do host and habitat characteristics, and spatial structure of the host populations influence virus community structure, or can we explain the structure solely using the direct and indirect associations between the viruses? Q3. After accounting for the effects of the host and habitat characteristics, and spatial patterns on virus distributions, are there any remaining non-random negative or positive direct associations between viruses? Q4. Do associations between viruses change when comparing conditional network models (CRFs) with only host- and/or habitat-related and/or spatial explanatory variables included in the model, demonstrating how these different sources of environmental heterogeneity explain the structure of the virus community?
Our results demonstrate that there are non-random co-occurrence patterns between viruses, which are only partly resolved by the host and habitat characteristics and spatial structure. We find that the majority of the explained virus co-occurrence patterns can be attributed to direct and indirect associations among the viruses.