The number of SNPs that are sufficient for GWAS, the cost and the genome coverage are the main determinants when selecting an appropriate genotyping method. Three most commonly used methods are Whole-Genome Sequencing (WGS), Genotyping by sequencing (GBS) and SNP Arrays. WGS aims to capture whole genome variation and is more expensive than other two methods. This method can capture common and rare variants which allows fine-resolution of variant mapping (Höglund et al., 2003). GBS uses restriction enzymes before sequencing in order to reduce genome complexity and price which leads to low coverage (Pavan et al., 2020). It is particularly useful for populations with slowly decaying LD where lower number of genotyped SNPs and their uneven genomic distribution do not affect the questions of the GWAS. Imputation can compensate for low coverage of GBS though it works best in homozygous populations. SNP arrays are the most popular of the genotyping method used in GWAS due to their generally low cost, high SNP density and ability to sequence large number of individuals (Pavan et al., 2020). One major drawback is dependence on the markers present in a population, that is, used as a reference for the development of the array, meaning that SNPs not present in that population will be missed when sequencing new populations. In general, the sequencing costs are decreasing and methods that generate longer reads are becoming more popular (Pacific Biosciences (PacBio) and Oxford Nanopore Technologies (ONT)) because they can discover hidden variation in the genome (De Coster et al., 2021). The decision on which method to use will largely depend on the cost but in general WGS, SNP arrays and long read methods are proving a good choice for viral infectious disease GWAS.

GWAS can be used to study virus-host interactions in any host species. Although model organisms in general have more advantages than non-model organisms (for example humans) when performing GWAS because different individuals: i) can be maintained through inbreeding, as in Arabidopsis thaliana or Caenorhabditis elegans, ii) can be grown in controlled environments under careful observation and iii) can be inoculated with a known viral dose that could be genetically and phenotypically well defined. All these points also highlight the main disadvantages of using non-model host and pathogen organisms with uncontrolled genetic and environmental background. An important advantage of A. thaliana, and many other plants, is self-fertilization where the same genotyped individuals can be maintained through time, thus allowing repeated phenotyping (Korte et al., 2012). Main disadvantages are caused by different breeding strategies that lead to extensive population structure and spurious hits later on in the association analysis (Mozzi et al., 2018). Fortunately, this can be corrected by proper use of statistical models, such as linear mixed models that take into account the population structure thus minimizing spurious hits and leading to recovery of more meaningful associations (Lippert et al., 2011; Zhou and Stephens, 2012; Lippert et al., 2014). In conclusion, the choice of an organism for GWAS should depend on the species of interest and the available information.

When talking about the number of individuals to use in a GWAS of viral infections, the general guideline should be the more the better. However, this is not always feasible due to cost or inability to recruit a large number of individuals. This can be represented as an inverse relationship between sample size and effect size. Where large sample sized studies have more power to detect smaller associations than small sample sized studies (Witte, 2010). Of course, the numbers vary from one organism to another and the studied trait, due to genetic diversity, recombination and ploidy (Genissel et al., 2017). For example, for some traits pretty good sensitivity and meaningful results can be achieved with ∼ 100 lines of A. thaliana; however, thousands of individuals might be needed in the case of human studies (The Wellcome Trust Case Control Consortium, 2007; Manolio et al., 2009; Korte and Farlow, 2013; Butković et al., 2021). This difference can be explained by the number of loci that explain the majority of the phenotypic diversity. In some A. thaliana traits, few loci explain the trait of interest and the diversity can be captured by genotyping less individuals. While in human studies, many small effect loci explain a trait and require genotyping of thousands of individuals in order to detect them (The Wellcome Trust Case Control Consortium, 2007; Korte and Farlow, 2013; Genissel et al., 2017). So far, in infectious disease GWAS an increase in sample sizes identified more meaningful associations (McLaren et al., 2015; Butković et al., 2021; Butković et al., 2022). Since viral infectious diseases are highly prevalent and are governed by common variants, a moderate increase in sample size should lead to enough statistical power needed to identify meaningful associations. However, if there are rare alleles responsible for a studied trait they will not be identified in moderate sample sized studies and would require larger number of samples (The Wellcome Trust Case Control Consortium, 2007; Hong and Park, 2012). Adequate sample sizes can be estimated with the help of Genetic Power or GAS Power calculators (readily available online), mostly intended for case-control studies (Purcell et al., 2003; Johnson and Abecasis, 2017). In summary, sample size should be directly correlated with the target species, the particular trait being studied, and any previous studies on similar topics that might provide even a raw estimate of appropriate sample sizes. Usually, significance thresholds need to be adjusted via correction methods, such as Bonferroni correction. With large sample sizes small effect loci become easier to detect and stricter significance thresholds are recommended to reduce the number of false positives (Fadista et al., 2016; Pulit et al., 2017; Wu et al., 2017).

Another important selection criteria are the geographical distribution and kinship of the samples. More heterogeneous samples can increase the genetic variability, allow replication in independent populations and may detect SNPs associated with the trait of interest in mixed populations (Korte and Farlow, 2013; Li and Keating, 2014). There are downsides to making samples more heterogeneous since this can decrease the power to recover significant variants; causing a non-causative marker to become more significant than the causal one and lead to the problem of handling linkage disequilibrium across heterogeneous populations (Korte and Farlow, 2013; Li and Keating, 2014). For example, a naturally occurring plant population can be previously exposed to a virus and gain resistance alleles while other populations lack them. This difference can lead to overestimation of the impact these resistance alleles have in the global diversity panel since they will become significant when comparing different populations. To avoid all the problems that population heterogeneity can cause, one solution is to densely genotype a local population. Although this approach leads to underrepresentation or absence of some variants of interest that may be relevant to global phenotypic diversity (Korte and Farlow, 2013). Also, focusing on a local population can lead to insufficient statistical power if the size of the population is small. Again, there is no golden rule when selecting how geographically or kinship diverse the samples shall be, the answer mostly depends on the goal of the GWAS. For example, more geographically heterogeneous samples would allow mapping of variants relevant to a global response to infection while less heterogeneous samples can map variants specific to a certain population. But appropriate measures should be taken into account where false positives would be avoided. For example, performing principal component analysis will account for genetic differences between populations which can later on be included as a covariate in the association analysis (Price et al., 2006; The Wellcome Trust Case Control Consortium, 2007; Cook and Morris, 2016). Another good measure that can be used is admixture mapping, a method that tries to associate an allele with a disease based on ancestry information between different populations (Shriner, 2013).

Before starting a GWAS, it is essential to define the infection-related trait to be studied and the method of virus delivery. When measuring viral infection, it is typical to measure the development of symptoms during a specified time period, frequency of infected individuals, viral accumulation, or presence/absence of symptoms related to infection (Fellay et al., 2007; Butković et al., 2021; Choudhury et al., 2019; COVID-19 Host Genetics Initiative, 2021; Crosslin et al., 2015; Adebamowo et al., 2020; Foresman et al., 2016; Feng et al., 2019; Liu et al., 2021; Garcia-Etxebarria et al., 2015; Montes et al., 2021; Pimenta et al., 2020; Rubio et al., 2019; Tian et al., 2017; Xiao et al., 2019; Yang et al., 2019; Zhang et al., 2020; Zignego et al., 2014). All these traits can be divided into categorical (binary case-control studies where diseased individuals get compared to disease-free controls or categories of disease symptoms on a scale ranging from asymptomatic to death) or quantitative measures (frequency or percentage of infected individual, viral accumulation). Generally speaking, quantitative traits are preferred because they improve the power of GWAS and can be better interpreted while case-control studies can lead to large measurement errors (Bush and Moore, 2012).

Expertise is necessary to be able to discern symptoms of infection as opposed to other conditions similar in appearance. Viral accumulation is not a visual trait and therefore, is not dependent on the trained eye of the observer, making it easier to measure. Although viral accumulation is not always directly correlated with the strength of the symptoms and might not involve the same variants for the two traits. This is of special interest if the goal of the GWAS is to discover genetic regions related to development of detrimental symptoms in the host. A common problem when visually detecting infected individuals are asymptomatic infections, mostly caused by errors in the inoculation procedure as opposed to resistant individuals that have no phenotypic symptoms. Another problem in visual evaluation of symptoms is resistance (no symptoms and no virus multiplication), tolerance (prevention of symptom onset despite virus accumulation) or recovery of individuals from infection during the observation period, that may confuse the observer and lead to incorrect measurements. A possible way to deal with these problems is to detect the presence of a virus (RT-PCR, RT-qPCR, PCR … ) in order to confirm or reject the infection status of an individual.

An important factor contributing to development of symptoms is the inoculation procedure. There are many different ways to inoculate individuals, for example, in plants using mechanical inoculation (rubbing using abrasive material) and Agrobacterium-delivery, adding virus infected medium to a cell culture, and vector dispersal. All of the methods except vector dispersal, lead to a controlled administration of the virus amount and remove bias due to large differences in initial virus concentrations between individuals (Gokhale and Bald, 1987). Still, these inoculation methods might lead to different responses of the plant to the virus compared to natural modes of infection and should be taken into account when interpreting results.

A good predictor of how much a trait is explained by genetic factors is heritability (Zaitlen and Kraft, 2012). Due to the presence of many small effect or rare variants, epistasis and phenotypic plasticity, it is impossible to identify all of the components influencing a complex trait in any chosen method (Frazer et al., 2009; Aschard et al., 2012). One solution to this problem might be to study the trait of interest in the context of the environment which might explain some of the missing heritability.

The first ever GWAS analyzing viral infection traits was reported for this virus (Fellay et al., 2007), wherein host genetic variants associated with HIV-1 RNA viral load, as a determinant of disease progression, were found. This study was performed on 486 European individuals and identified two significant polymorphisms for viral load and seven for disease progression. In the case of the viral load, the first one was SNP rs2395029 near the human leukocyte antigen (HLA) complex P5 (HCP5) gene that is localized within the major histocompatibility complex (MHC) class I region. This SNP is known to be in strong linkage disequilibrium with the HLA-B gene allele 57:01, previously described as associated with restriction of HIV-1 viral replication and low viral load (Migueles et al., 2000; Altfeld et al., 2003). The other significant SNP, rs9264942, is located 35 kb from the transcription initiation of the HLA-C gene and 156 kbp from the HCP5 and might be involved in the control of HIV-1 disease progression (Fellay et al., 2007; Thomas et al., 2009; Kulkarni et al., 2011). For disease progression, Fellay et al. (Fellay et al., 2007) identified seven significant SNPs near the ring finger protein 39 and zinc ribbon domain–containing one genes that can have an effect on HIV-1 disease progression. Following-up, Fellay et al. (Fellay et al., 2009) repeated a GWAS on more than 2,500 individuals. They confirmed the association with the HCP5 gene and the 35 kbp HLA-C region along with some other independent loci in the MHC region. Both studies by Fellay et al., however, failed to identify previously known HIV-1 resistance genes such as CC chemokine receptor type 5 (CCR5) (Lodowski and Palczewski, 2009) and transportin-3 (Rodríguez-Mora et al., 2019) which could result from the stringent statistical criteria in the GWAS, dozens of SNPs explaining only a small proportion of the variance, presence of rare variants or misclassification of cases and controls (Fellay et al., 2009; Sebastiani et al., 2009). Nonetheless, the CCR5 and HLA associations were found in a meta-analysis of 6,315 individuals (McLaren et al., 2015). Also, subsequent studies confirmed the previously identified associations (Fellay et al., 2007; Fellay et al., 2009) and discovered new ones associated with HIV-1 plasma levels in seroconverters (Dalmasso et al., 2008), rapid progression (Limou et al., 2009) and non-progression of AIDS (Le Clerc et al., 2009). The studies done on HIV-1 demonstrate the importance of sample size and correct labelling of cases and controls, where larger studies are able to find more meaningful associations.

TuMV causes significant loses of Brassicaceae crops around the world and ranks as one of the most damaging vegetable viruses worldwide (Tomlinson, 1987; Walsh and Jenner, 2002). It causes a variety of symptoms from chlorosis, stunting and necrosis (Butković et al., 2021).

So far, three GWAS have focused on TuMV, two studies used A. thaliana (Rubio et al., 2019; Butković et al., 2021) and one used Chinese cabbage (Brassica campestris ssp. Pekinensis) (Zhang et al., 2020) as hosts. The number of individuals greatly varied among these studies, Zhang et al. (Zhang et al., 2020) used 83 cabbage varieties in a controlled experimental setting, measuring virus load by RT-qPCR. Rubio et al. (Rubio et al., 2019) used 317 accessions and measured viral accumulation, frequency of infected plants and symptomatology across 2 years in a natural setting using the UK1 isolate. Butković et al. (Butković et al., 2021) used 450 accessions in a controlled experimental setting with generalist and specialist strains derived from the YC5 isolate, measuring disease progression, symptom severity and percentage of infected plants. There are overlaps between the genetic regions where the significant SNPs were mapped in all three studies but also many differences: e.g., significant SNPs mapped in DNA-J heat shock proteins for all three studies while SNPs mapping in nucleotide-binding site leucine-rich repeat (NBS-LRR) proteins were found by Butković et al. and Zhang et al. and SNPs in F-box family proteins were mapped by Zhang et al. and Rubio et al. The rest of the significant variants were mapped in genes with a wide variety of functions, most of which were previously described as related to disease resistance or susceptibility. Interestingly, Butković et al. described for the first time a factor related to symptom severity within gene AT2G14080, which is an NBS-LRR class protein. These proteins recognize specific pathogen domains and lead to resistance or a hypersensitive immune response that leads to cell death or necrosis (Marone et al., 2013). These three GWAS are a good example of how viral evolutionary history, selected traits, environment and host selection influence the genetic variants associated with the pathogen. Compared to HIV-1, smaller sample sizes in A. thaliana allow a pretty good recovery of meaningful associations. The influence of factors besides host and pathogen genetics should always be considered when planning or analyzing results on GWAS of infectious diseases.