Most efforts to understand the population biology of pathogens have focussed on specialist pathogens, and population biologists have successfully developed a formal understanding of the dynamics and evolution of single-host pathogens. However, most pathogens of humans, animals and plants are multi-host pathogens 
. As stated by Woolhouse et al. 
“understanding the more complex population biology of multi-host pathogens will be one major challenge in the 21st century “. There is evidence that within an ecosystem the prevalence of multi-host pathogens may differ largely for the different species of their host range [e.g., 
]. Similarly, there is evidence of large differences in the prevalence on a host species of the various pathogens that are able to infect it [e.g., 
]. However, no attempt has been made, to our knowledge, to analyse if differences in the distribution of multi-host pathogens over their hosts are random or if there are associations between hosts and pathogens. The uncovering of associations between hosts and pathogens would be highly relevant to understand and model the population biology of multi-host pathogens, and for understanding the phenomenon of generalism itself.
We present here indices and tests to analyse if there is association between multi-host pathogens and their hosts. The proposed indices of selectivity for the pathogen and for the host measure the degree of association between hosts and pathogens. The tests analyse the homogeneity of distribution of a pathogen over different host species or populations, and of different pathogens on a host, and analyse how significantly the values of the indices departs from zero (i.e. no association). The literature on pathogen ecology does not abound with data on the prevalence of various pathogens on various hosts. Hence, we have applied these indices to our unpublished data on the prevalence of five insect-borne plant viruses on 21 species of wild plants in an agroecosystem in central Spain over a three year period.
The analysis of the prevalence of the different viruses in each host species by the homogeneity test that we propose, shows that half of the analysed plant species showed an index of selectivity of the host (ISH) significantly different from zero. The distribution of the host species showing virus selectivity was not related to taxonomy, habitat (fallow fields, edges or wastelands), seasonality or vegetative cycle (annual vs. perennial) (not shown). Interestingly, there was a positive correlation between the ISH and the average virus prevalence for these 21 host plant species, showing that the more selective hosts are more prone to be virus-infected, obviously by the virus(es) that better infects them. This phenomenon suggests that in spite that each host encounters a wide array of pathogens, mechanisms of escape and/or resistance 
to some of them would operate, which could explain their selectivity. In fact, contingency analysis of counts of infected hosts by different viruses, suggest that different viruses specialise on different hosts.
The analysis of the homogeneity of prevalence of a virus over its host species showed that for three of the five analysed viruses there was a significant host association, i.e., the value of the index of selectivity for the pathogen (ISP) significantly departed form zero. One major and unexpected finding of the analysis was that there was a positive and highly significant correlation between the value of the ISP and the prevalence of the viruses. The value of the ISP was not conditioned by the number of host plant species infected by each virus, as there was no correlation (r
0.173 in a Spearman rank correlation test) between ISP and the number of plant species that each virus infected in the analysed system i.e., the more selective viruses were not those infecting a smaller number of plant species. Thus, the more host-selective viruses were those that did best in the analysed ecosystem. This result could be highly relevant for understanding the evolution of generalism in pathogens. Although most described pathogens are generalists, the advantages of generalism are poorly understood. A generalist strategy provides the pathogen with more opportunities for transmission and survival, but it is predicted that evolution would favour specialism, because pathogen-host co-evolution could result in functional trade-offs that would limit the generalist fitness in any one host 
. Our results are compatible with the hypothesis that specialism is advantageous for pathogens, as host selectivity is the rule for the analysed set of generalist viruses, and the more host selective is the virus, the more successful its strategy. Hence, our results could suggest that for generalist pathogens a degree of host specialisation, i.e. host-selectivity as defined here, is a successful strategy. Host specialisation in generalist pathogens would also be relevant for important issues of host and pathogen biology, as host specialisation will affect host-pathogen co-evolution and co-speciation, would reduce the opportunities for host switches and jumps, thus constraining the evolution of host expansion, and may result in spatial heterogeneity of hosts, thus favouring the stable maintenance of pathogen and host diversity 
. In addition, host specialisation may affect the opportunity for different pathogens of sharing a host and, thus, the consequences of multiple infection for pathogen and host evolution, as discussed below.
We propose here also a simple procedure to estimate association among pathogens, which enables to compute an association index whose significance can be tested against the null assumption of independence of infections that follow a binomial distribution. The test was applied to the same data set as above, and the second major contribution of our analysis is the finding that co-infection was mostly non-random and that associations among the five analysed viruses were mostly positive. This result is relevant because co-infection of different pathogens may have important consequences for the pathogens, the infected hosts, and for host-pathogen co-evolution 
. For viruses, co-infection of a host may result in the generation of new genotypes by recombination or by reassortment of genomic segments between different viral species or strains, often with dramatic changes in host range or pathogenicity. The classical example is the reassortment of avian and human strains of influenza A resulting in novel viruses with pandemic potential 
, but examples abound for both animal and plant viruses [e.g., 
]. In the individual host, co-infection may lead to aggravated disease, often resulting from extracellular cooperativity of independently replicating viruses, by which one virus modulates the host response to infection to the benefit of the other 
. In addition, direct interactions of different viruses in co-infected cells may result in complementation of highly pathogenic defective genotypes, in increased virus replication or in modified cell and tissue tropisms [e.g., 
]. Alternatively, there is also evidence that mixed infections of pathogens result in reduced pathogenicity and less severe disease 
. Examples from viruses include mixed infection with satellite or with defective interfering nucleic acids 
. In our data set, association between viruses depended on each particular virus-host system. Hence, data suggest that in some hosts, but not in all, co-infection would be advantageous for some viruses, though the underlying mechanism remains to be analysed.
The analysis here reported of plant virus infection on weeds has uncovered two major features that should be relevant to understand the population biology of viruses: i) the more host-selective viruses do better on the analysed ecosystem, ii) viruses tend to associate positively in co-infected hosts. It would be of high interest to know how general are these features and in which types of pathogens would they occur. The indices and tests that we propose here could be of general use in the analysis of the ecology of pathogens, and we hope that our results would prompt research on the ecology of pathogen-host and pathogen-pathogen associations, as these analyses might uncover pathogen properties relevant to the formal understanding of the population biology of multi-host pathogens.