Cross-Species association statistics for genome-wide studies of host and parasite polymorphism data

Abstract

1Uncovering the genes governing host-parasite coevolution is of importance for disease management 2 in agriculture and human medicine. The availability of increasing amounts of host and parasite 3 full genome-data in recent times allows to perform cross-species genome-wide association studies 4 based on sampling of genomic data of infected hosts and their associated parasites strains. We aim 5 to understand the statistical power of such approaches. We develop two indices, the cross species 6 association (CSA) and the cross species prevalence (CSP), the latter additionally incorporating 7 genomic data from uninfected hosts. For both indices, we derive genome-wide significance thresh-8 olds by computing their expected distribution over unlinked neutral loci, i.e. those not involved 9 in determining the outcome of interaction. Using a population genetics and an epidemiological 10 coevolutionary model, we demonstrate that the statistical power of these indices to pinpoint the 11 interacting loci in full genome data varies over time. This is due to the underlying GxG interactions 12 and the coevolutionary dynamics. Under trench-warfare dynamics, CSA and CSP are very accurate 13 in finding out the loci under coevolution, while under arms-race dynamics the power is limited 14 especially under a gene-for-gene interaction. Furthermore, we reveal that the combination of both 15 indices across time samples can be used to estimate the asymmetry of the underlying infection ma-16 trix. Our results provide novel insights into the power and biological interpretation of cross-species 17 association studies using samples from natural populations or controlled experiments. 18 Keywords 19 population genomics; linkage disequilibrium; single nucleotide polymorphism; host-parasite co-20 evolution 21

Novel genomic approaches to study antagonistic coevolution between hosts and parasites

“…parasites) and noninfected hosts (Märkle et al, 2021) (here randomly implies an unbiased sample of the coevolving populations). These indices measure the degree of association between allele frequencies in the host (infected, noninfected and both types) and allele frequencies in the parasite samples, mirroring measures of linkage disequilibrium in population genetics.…”

“…Thus, coevolving loci are more likely to be detected under (long-term) trench-warfare than under the arms-race dynamics. Furthermore, obtaining time samples improves the statistical power of these indices in particular, and most likely of all co-GWAS methods, as they capture the temporal fluctuations in allele frequencies at the coevolutionary loci (Märkle et al, 2021). The association indices do not replace co-GWAS to pinpoint loci under coevolution, but allow to link the results of co-GWAS with the theoretical models of coevolution and thus could be used for statistical inference of coevolutionary parameters at significant pairs of associated SNPs found in co-GWAS (Märkle et al, 2021).…”

“…However, Märkle et al, (2021) predict that few hundreds of samples at one-time point (on the order of sample sizes in Wang et al, 2018) might be enough to obtain adequate statistical power. It is also predicted that the statistical power of co-GWAS would be increased if samples are available at several time points.…”

“…As an addition to natural co‐GWAS, two cross‐species association indices have been developed based on pairwise comparison of host and parasite SNPs, using randomly chosen infected (along with their parasites) and noninfected hosts (Märkle et al, 2021) (here randomly implies an unbiased sample of the coevolving populations). These indices measure the degree of association between allele frequencies in the host (infected, noninfected and both types) and allele frequencies in the parasite samples, mirroring measures of linkage disequilibrium in population genetics.…”

“…Finally, we note that the sample sizes in the natural co‐GWAS performed on human‐parasite systems (Ansari, 2017; Bartha, 2013; Lees et al, 2019) are large (several thousand) and obtained at a single point in time. However, Märkle et al, (2021) predict that few hundreds of samples at one‐time point (on the order of sample sizes in Wang et al, 2018) might be enough to obtain adequate statistical power. It is also predicted that the statistical power of co‐GWAS would be increased if samples are available at several time points.…”