Breeding objectives always involve consideration of multiple traits, even in situations where output of a single feature is dominant. While animal breeders have formalized the procedure of multiple trait selection, commercial plant breeders use non-formal ways of combining selection pressure on various traits, which are not published (Solkner et al. 2008). The longer generation intervals and, more importantly, the lower reproductive rates of animals emphasize procedures of simultaneous selection whereas crops are developed in multiple stages, emphasizing different traits at different generations. Thus, selection for traits with high heritability is usually performed at early generations. If a kind of pre-selection based on QTL linked markers might be carried for one or more traits at seedling stage, fewer generations would be needed. Long juvenility species would greatly benefit from this. However, to ensure the success of MAS, QTL mapping results, particularly QTL estimates, must be further investigated by cross validation and biometrical methods (Melchinger et al. 2004), even including Bayesian approaches (Meuwissen et al. 2001). Even so, there will always be QTLs of minor effects (and epistatically masked QTLs) that will be missed, making selection of favorable QTL alleles alone unable to reach the full genetic potential and the ultimate limits to selection (Kearsy et al. 2003).
There are two important genetic factors, epistasis and genotype by environment interaction (G x E) that limits our understanding of agronomic traits. In this sense, the use of marker-assisted selection (MAS) can be inefficient if the effects of G x E and epistasis cannot be anticipated (Openshaw and Frascaroli 1997; Moreau et al. 2004). Moreover, Bernardo and Charcosset (2006) showed that poor estimates of allele effects within the selected population, even when genes are known, reduced MAS efficiency. Therefore, when G x E and epistasis are important we should regularly re-estimate QTL effects within the breeding program (Podlich et al. 2004).
Significant G x E makes statistically impossible to interpret G and E main effects and to predict genotypic performance across changing environments. In its simplest form, G x E may rise from heterogeneous genetic variances among environments. Thus, genetic variance at one QTL may be sufficiently large in one environment, in comparison to non-genetic effects, to allow its detection here, but not in other environment (salinity, for instance). If it happens in one or more QTLs for the same environment, it will cause just a scaling effect but it will not alter the ranking of genotypes. But, if the variance heterogeneity affects different QTLs depending on the environments, a re-ranking of genotypes will happen which is a major concern to breeders, particularly those devoted to increase crop adaptability to abiotic stresses.
To determine genetic factors responsible for G x E, agronomic data must be collected on a mapping population in multiple environment trials. Then, comparison of QTL detection across environments is carried out by analysis of variance to test marker locus x environment interactions (Sari-Gorla et al. 1997; Villalta et al. 2007). Quantitative trait loci by environment interaction can also be evaluated by the regression of marker genotype mean on an environmental index to discern if the linear regression coefficients are significantly different (Campbell et al. 2003). QTL mapping results (e.g. Patterson et al. 1991; Stuber et al. 1992; Lu et al. 1996; Monforte et al. 1997b; Villalta et al. 2007) have shown that some QTLs can be detected in all tested environments while others are detected in some of them. However, in the absence of G x E, a QTL may be detected in one environment but not in others due to sampling or experimental error. On the other hand, G x E may exist even when QTL are detected in multiple environments (Yan et al. 1999). Noteworthy, Villalta et al. (2008) found that the position of maximum LOD for a leaf area QTL (and also for dried leaf weight) on Solanum chromosome 5, changed few cM when control and salinity growing conditions were compared. Only QTLs detected under control condition showed significant G x E interaction. Is it a matter of random errors or the presence of two QTLs in tandem, one conditioning the trait just under control and the other under both control and salinity conditions? Unfortunately, random errors cannot be discarded given the overlapping of confidence intervals.
Due to the cost of utilizing several QTLs, only markers that are tightly linked to no more than three QTLs are typically used for MAS (Ribaut and Betran 1999), although there have been reports of up to five QTLs being introgressed via MAS (Lecomte et al. 2004). The cost of using MAS compared to conventional plant breeding varies considerably between studies and need to be considered in a case-by-case basis. And, as Collard et al. (2005) have pointed out these studies did not include the large initial cost in their development. An estimate for the cost to develop a single marker was 51,140 € (Langridge et al. 2001). Model crops, where genomic tools and most QTL analysis are being developed have a clear advantage with regard to applications of QTL mapping. Fortunately, comparative mapping could be used to infer QTL position between related species what could be particularly important in "orphan" (or neglected) crops.
Markers tightly linked to loci controlling difficult, laborious and expensive but breeding-targeted traits are valuable tools to assist selection at seedling stage during the breeding program of a long-lasting juvenility crop or forest species. Thus, MAS have been most frequently used (or reported) to discard putative disease susceptible plants in the early generations and to introgress disease resistance genes into high-quality well adapted elite cultivars. The greatest efficiency of MAS is in early generations due to the increasing probability of recombination between the marker and QTL; the major disadvantage is the cost of genotyping a large number of plants given that it is applied at the beginning of the program (Collard and Mackill 2008). The continuously growing marker technology and availability of high-throughput equipment for DNA extraction and genotyping (for a review see Gupta et al. 2008) is expected to make MAS less expensive in the future.
A successful strategy to minimize the failure of marker-trait association is to maximize the efficiency of QTL mapping in tandem with the MAS process itself (Young 1999). Additionally, to cover a wider range of allelic diversity, multiparen-tal populations (like those commonly used by breeders) have shown to be better than biparental populations. Thus, Blanc et al. (2006) have recently shown by simulation that addressing a broader diversity, multiparental designs increase the power of QTL detection, which reinforces their superiority over biparental designs for MAS. As Melchinger et al. (2004) concluded, for routine applications of QTL mapping in plant breeding programs, it is mandatory that we move away from the analysis of large segregating populations of biparental crosses and develop new QTL methods and software, which are directly applicable to the genetic materials routinely tested by plant breeders.
Was this article helpful?