They were dried in a forced-air drying oven at 55°C for seven days, reweighed and used to calculate dry matter percentage. All yields were recorded on a dry matter basis , adjusting plot weights by the average dry matter percentage.An experiment was established in April 2018 consisting of four replications laid out in a randomized complete block design. This trial was a large dormancy evaluation but included all populations in the yield trial described above except for the cycle one phenotypic selection population – NY1221. The trial was located in Davis, CA on a Yolo silty clay loam . Plants were sown in trays in the greenhouse 2 months prior to transplanting. Plots consisted of a single row of 25 plants spaced 30 cm apart with a 90cm gap between plots within rows and 60cm spacing between rows. Plants were harvested throughout the season when they reached the target maturity of bud to early flowering stage using a self-propelled forage harvester. Fertilizer was applied to maintain P and K at appropriate levels for high yielding perennial forages, with weeds, insects and other pests monitored and standard control measures applied if necessary. Plant height was measured 25 days after the final harvest in October in 2018 and 2019. Plant height was considered the distance from the soil surface to the tallest point of the plant at its natural height . A single measurement was taken from each plant and an was average height across all plants in the plot was determined.All analyses were performed using R statistical software .
For the yield trial, analysis of variance was conducted to estimate the effects of population, location, flower display buckets and harvest on forage yield. Locations were analyzed independently followed by a multi-environment analysis. In addition to an overall annual yield analysis, the yield of individual harvests within location was conducted and across locations, harvests were analyzed by season, with spring , summer , and fall groups included. For the single location models, population, block, harvest, and population × block interaction were treated as fixed effects . For the multi-environment models, population, location, harvest within location, block within location, population × location interaction, and population × harvest within location were treated as fixed effects. Means and standard errors were calculated on a per harvest basis for each population using the emmeans package . Pairwise comparisons were conducted using the multicomp package . The significance threshold was set at 0.05 using the Tukey method for multiple comparisons.Forage yield differed among populations at each location and in the overall analysis . Forage yield also differed among populations for the different seasonal harvests at each location. The base population , phenotypic selection cycle one , genomic selection cycle one – high , and genomic selection cycle one – random were the top performing populations overall across both locations; the genomic selection cycle one – low and genomic selection cycle two populations were the poorest performing. These trends were also observed in the seasonal breakdown of harvests, with the exception of GSC1-L which performed well in the autumn harvests. In Ithaca, NY0847, GSC1-H, and GSC1-R had the highest yield per harvest over the four-year duration of the trial. GSC1-L had low yield in the spring harvests and overall. GSC2-H and GSC2-L were consistently the lowest yielding populations. Although NY1221 was not significantly different to top performing varieties in the seasonal breakdown of harvests, the GSC1-H population had a significantly higher forage yield than NY1221 overall. In Tulelake, there was less separation between populations. NY0847 and GSC1-H yielded significantly more biomass than GSC1-L in spring, summer and overall; however GSC1-L outperformed GSC1-H in the fall. There was no significant difference between genomic selection and phenotypic selection in Tulelake.Dry matter yield is the most important trait for profitable alfalfa production, yet somewhat inexplicably, over the past 30 years, there has been no improvement in on-farm alfalfa yields in the USA .
Genomic selection has been shown to increase the rate of genetic gain in many of the major crops grown in the United States, including alternatives to alfalfa, such as maize , in livestock rations . In this experiment, we evaluated and compared populations developed through traditional recurrent phenotypic selection and genomic selection to investigate whether genomic selection could be a viable option to address the lack of yield improvement in alfalfa. In this experiment heritability estimates for DMY are higher than have previously been reported , probably due to the use of ten replications within each location of the trial to obtain reliable estimates of forage yield in a densely sown sward. H2 estimates for Tulelake were lower than Ithaca, due in part to the inclusion of establishment year harvests . Considering only the populations developed using the genomic prediction model, the GSC1- H population had higher yield than the GSC1-L population, with the population whose parents were chosen randomly falling intermediate between the others. Thus, the model had the abilityto shift populations in the expected directions for biomass yield. Across all entries, the GSC1-H population was among the top yielding populations, and GSC1-L was among the lowest yielding. However, the genomic prediction model appeared to break down on the second cycle of selection, with both the high and low GSC2 populations performing poorly. This is concerning, as one of the major benefits of genomic selection is the potential ability to conduct multiple cycles of GS in the span of a phenotypic selection cycle. However, if the model breaks down after a single cycle, this benefit cannot be realized. Nevertheless, conducting a single cycle of GS in the space of a year is still considerably faster than a PS cycle. Approximately 9000 markers were used for the first cycle og genomic selection and fewer were used for the second cycle, which may explain some of the poor performance of the C2 populations; alternatively, the relative value of marker loci could have shifted following the first cycle, so that the model is simply not useful. In addition, there may have been an inadvertent shift in the dormancy of the C2 populations, which could contribute to lower total DMY.
The base population and all populations developed through genomic selection were included in a separate trial investigating the autumn height of various alfalfa populations, a proxy for autumn dormancy. The GSC2-H population was significantly shorter than the GSC1-L population indicating selection for plants containing alleles for less fall growth. Future applications of genomic selection should include selection criteria to ensure fall dormancy remains unchanged during the selection process. This also shows a potential risk of genomic selection in a breeding program – the possibility for undesired shifts of non-target traits. Breeders should be aware of this when making selection decisions, and this result highlights the need to measure other traits of importance during the breeding process. The GS model was developed based solely on phenotypic information from plants grown in Ithaca, and not surprisingly, the selected populations performed better relative to other entries in Ithaca than in Tulelake. This suggests that any potential gains derived from genomic selection require the inclusion of phenotypic information from the target environment. This observation has potentially significant implications for the viability of incorporating genomic selection into alfalfa breeding. Already suffering from a paucity of breeders and breeding resources, flower bucket expanding breeding trials to include more environments may not be possible for many breeding programs. Further investigation is required to determine the requisite number of environments that need to be evaluated in order for GS to work robustly. The lack of separation between the base population and populations selected for high yield, either by PS or GS, in the overall analysis provides some insight into what breeders have experienced over the past 30 years of alfalfa improvement. The selected populations have not increased yield in an experiment designed to replicate a commercial production environment. Notably, however, in Ithaca the GSC1-H population performed better than the PSC1 population that was selected through phenotypic evaluation, even though both relied on yield information from Ithaca in making selections or in developing the GS model. Regardless, the lack of DMY gain from the base population using either phenotypic or genomic selection remains a significant concern. Further improvements to the predictive model are possible and may yet result in real gains at the commercial production level. The GS populations evaluated in this experiment derived from a predictive model developed using clonally replicated space-plant yield. A poor correlation between individual space-plant yield and DMY of a densely planted sward is often obtained , so an alternative approach evaluating the DMY of families in densely planted small plots might be a better approach .
These families can be bulk genotyped to obtain allele frequency marker data rather than individual genotyping calls . This method better captures commercial yield in the model so more accurate predictions can be made, aligns well with current the current breeding methods in alfalfa, and can be implemented alongside family-based recurrent selection. Genomic selection is still in its infancy in alfalfa; however, our data indicate there is the potential for greater genetic gain with GS than has been obtained with the use of phenotypic selection alone. Significantly more research is required to investigate alternative models and selection strategies across the wide range of environments in which alfalfa is grown. With the cost of genotyping decreasing, new high throughput technologies being developed, and a greater understanding of the alfalfa genome, the potential for GS to improve yield is quite high. The results of this work will be beneficial not only to alfalfa production but also will help guide decision making for breeding of other outcrossing perennial forages.Alfalfa is one of the most important perennial forage crops in the world. It is the third most valuable field crop in the United States in which California leads the nation for hay and seed production, generating in excess of $1B in 2022 . Its high yield and nutritional value are key drivers for California’s dairy industry, the state’s top valued agricultural commodity . In addition to its economic value and importance as a forage, alfalfa provides a host of beneficial ecosystem services. Its nitrogen fixing capabilities and perennial nature promote sustainable cropping systems and contribute to nutrient cycling . Alfalfa also plays a role as an important insectary and habitat for native fauna . Cultivated alfalfa is predominantly derived from the subsp. sativa, an allogamous autotetraploid . It is commercialized as synthetic populations consisting of highly variable and heterozygous plants . Alfalfa breeding programs are based on recurrent phenotypic selection, with or without progeny testing . They are designed to increase the frequency of desirable alleles in a population while maintaining genetic variability for continued genetic improvement . Breeding goals in alfalfa are characteristic of those in other crops: increasing yield, enhancing forage quality, and improving tolerance to biotic and abiotic stresses . Simply inherited traits with high heritability have been greatly improved through traditional breeding methods; however, improvement in complex, quantitatively inherited traits have been less successful , most notably yield for which there has been little to no improvement over the last 30 years . Long breeding cycles , multiple harvests per year, limited breeding resources, inability to make gains in the harvest index, significant genotype by environment interaction , and selection based on vigor of spaced plants or short family rows are all factors contributing to the low rate of yield progress . Yield improvements in alfalfa in the past can be mainly attributed to improvement of ‘defensive’ traits i.e., improvements in pest and disease resistance . This helps alfalfa reach its yield potential, but it does not result in an increase in yield per se. To select on yield per se, selections could be based on yield data from the first full year of production before plant mortality becomes an influencing factor impacting yield, while persistence could be evaluated at the end of a multi-year trial. Marker-assisted selection is a modern tool that has great potential in addressing the lack of genetic gain in alfalfa yield . The availability of a large number of single nucleotide polymorphism markers, cost effective genotyping assays, and the recent availability of chromosome-scale, haplotype-phased genome assemblies facilitate the dissection of complex traits and provide a pathway for genetic improvement .