Discussion
There has been considerable interest in the molecular mechanisms of G x E across a diversity of phenotypes, species, and environments. G x E is common and is often driven by differential sensitivity of alleles and may play an important role in adaptive plasticity and local adaptation (Des Marais et al., 2013). With its large scale, our study evaluated the genetic basis and examined the QTL x E of panicle morphological traits in switchgrass grown at 10 field sites in the central United States (Figure 1). Overall, we detected moderate heritability (except for the field site Stillwater, OK) for panicle traits (Table 2) and positive phenotypic and genetic correlations between traits at each site and across sites (Table 3). These data suggest considerable standing genetic variation in inflorescence characteristics available for natural or artificial selection to act upon. We identified several QTL with significant QTL x E effects and the potential environmental factors underlying the QTL x E, indicating that panicle traits in switchgrass result from the combination of QTL and environment. We also detected pleiotropic effects between panicle traits and flowering time as well as tiller count and biomass, suggesting a possible shared genetic basis between different traits.
Our study identified genomic regions (QTL) that contribute to panicle trait variation across a broad latitudinal gradient. These QTL exhibited constant effects (i.e., no QTL x E), antagonistic pleiotropy, or condition-specific effects across the studied environmental gradients. QTLs with condition-specific effects are relatively easy to incorporate into breeding programs because the selected favorable alleles will confer an advantage in some environments, without a negative effect in other environments (El-Soda et al., 2014). Antagonistic pleiotropy is a genetic trade-off at an individual locus or QTL, that results in opposite effects (i.e., sign change) on a trait in different environments (Wadgymar et al., 2017). QTL with antagonistic pleiotropy can result in tradeoffs and challenges in breeding if the preferred allele depends strongly on the environment (El-Soda et al., 2014; Lowry et al., 2019). Studying the molecular genetic basis of specific QTL should greatly contribute to the mechanistic understanding of such QTL x E. In our study, most of the QTL are conditionally neutral. This is consistent with a recent meta-analyses which found that asymmetry of QTL effect is more often caused by conditional neutrality than it is by trades-offs (Wadgymar et al., 2017). Overall, our results show that panicle traits are controlled by a combination of QTL and the environment and, in a number of cases, their complex interaction with the environment.
Inflorescence architecture is influenced by the vegetative-to-reproductive phase transition, which also largely determines patterns of vegetative growth and resource allocation. In our study, 11 of 18 inflorescence QTLs co-localized with flowering time or vegetative growth genomic intervals, which supports the hypothesis that pleiotropy impacts the phenotypic integration of these vegetative and reproductive structures. An exciting opportunity lies in the search for the candidate genes that may underlie this integration. Fortunately, extensive genetic mapping efforts in crops and model systems have identified a number of candidate genes and a basic understanding of their role in the development of the inflorescence. For example, a locus on chromosome 9N (at 38.02 cM) was associated with the whole process of vegetative-to-reproductive transition (PL, PBN, SBN, FL50, TC and BIO). This QTL cluster is in the vicinity of homologs ofOsCOL10 and OsTB1 , which are known as the key regulators in flowering and branch development (Tan et al., 2016; Takeda et al., 2003). Specifically, OsCOL10functions as a flowering time repressor downstream of Ghd7 and the OsTB1 gene negatively regulates lateral branching in rice. Moreover, the locus on chromosome 3K (at 38 cM) was clustered with QTLs for PBN, FL50 and TC. Significantly, this QTL clustering region had large effects for PBN, suggesting a major QTL that coordinates vegetative and reproductive processes. We identified a homolog of GA2ox3 in this region, which is considered as a key factor in gibberellin catabolism and plays a central role in plant development (Sakamoto et al., 2004). These results imply that there may be a shared genetic basis between vegetative and reproductive divergence within switchgrass populations.
The low to moderate prediction accuracy (0.34-0.62) of the multienvironment mixed model (Eq. 1) is likely due to two factors. Our model only accounts for significant QTL, while there are likely many smaller QTL, that were below the threshold for detection, which contribute to variation in these traits. Unfortunately, our power to detect these small effects is likely low due to our modest sample sizes (380 progeny). Additionally, the QTL model does not consider epistatic effects or dominance effects between QTL. Epistasis is known to be an important factor that affects genetic variation and phenotypic expression in populations, especially for developmentally regulated traits like inflorescence architecture. Epistatic effects on panicle related traits have been identified in several studies (Leng et al. , 2017; Ye et al ., 2009). Further inclusion of epistasis into the multi-environment QTL model may help improve model prediction. However, our approach provides a way of predicting the performance of new genotypes under environments similar to the tested environments, and can potentially help with suitable genotype selection for traits of interest under a specific environment.
Temperature and photoperiod were the most significant predictors of QTL x E interactions. This is consistent with the pattern of additive effects of most of the QTL (Table 3), where QTL displayed conditional neutrality with effects either in the northern or the southern sites (Figure 5). Previous study also showed that temperature-based growing degree days and photoperiod affected switchgrass morphology (Mitchell et al., 1997). Solar radiation was also a significant driver of QTL x E for secondary branching number (SBN) QTL. This is consistent with a rice study in which SBN was more plastic in response to different light resources (Adriani et al., 2016). No environmental factors were detected for some of the QTL x E interactions, possibly because the appropriate environmental factors (i.e., growing degree days, soil moisture etc.) were not explored or were obscured by complex interactions between environmental factors. The relative low heritability for panicle traits at the field site Stillwater, OK (STIL, Table 2) also suggested that there may be other environmental factor affecting panicle traits but not being accounted for, such as effective soil moisture. As noted in Table 1, Stillwater, OK, Overton, TX (OVTN), and Manhattan, KS (MNHT) all have sandy loam soil which may impact water status and subsequently plant growth and organ expansion. In our study year, OVTN received ample rain (~1400mm), MNHT had slightly cooler temperature and received approximately 1000mm rain, while STIL only received around 700mm rain (Figure 1). This study could be expanded in the future to include more field sites, multiple years, and more environmental data collection such as soil composition and nutrient availability to better capture the environmental drivers underlying the QTL x E interactions and trait plasticity across large geographic regions and across multiple years.
In summary, our results suggest that variation of panicle traits in switchgrass is due to a combination of QTL and the environment, with QTL displaying different effects across geographic regions. Future work focusing on identifying the driver of QTL by environment interactions and understanding the mechanisms underlying them will facilitate the selection of suitable genotypes for specific environments in switchgrass breeding programs.