Background: Strong selection can occur during switchgrass sward establishment.  Differences in establishment selection due to environment or management could provide information on genotype‐by‐environment variation and could influence strategies for breeding perennial grasses.

Methods: Leaf samples were collected before sward establishment and from 3‐year‐old swards for two breeding groups (lowland and hybrid) at three locations. Within two locations, samples were collected from paired fertilized (110 kg ha^-1 N) and unfertilized plots. Allele frequencies from pooled DNA samples were studied through multivariate analysis of variance, genomewide trait predictions (heading date and winter survivorship), and genomically estimated breeding values (GEBVs) for individual sward survival within an independent data set.

Results: This study found only minor variations in selection due to location or management. Predicted heading dates of the hybrid population had significant changes due to fertilization and location. There were strong correlations among sward establishment survival GEBVs between growing environments (hybrid r = 0.77; gulf r = 0.97). Interestingly, this study found a small number of genotypes that were over‐represented in established swards across all growing environments.

Conclusions: This study reinforces a prior report of selection during sward establishment and indicates that only a small degree of establishment selection is location‐specific within these diverse growing conditions.

This dataset includes breeding populations of switchgrass half-sib families derived from the Liberty cultivar which were planted at three locations (Arlington, WI, Prairie du Sac, WI, and Hancock, WI).  Collectively, these will be referred to as the hybrid breeding population.  These families were collected from two generations of genomic selection, which will be referred to as G1 and G2.  A small number of diverse families from the gulf ecotype were also sampled with fewer replicates and will be referred to as the gulf breeding population. 

The Arlington half-sib families are from swards within a previously published study evaluating the impact of plant spacing and will be referred to as SPACE (https://doi.org/10.1002/glr2.12007). The Prairie du Sac and Hancock families are from a separate study evaluating the impact of nitrogen fertilization and marginal land sites, referred to as SLOW (Casler, in progress). The SPACE study was planted with 19 hybrid half-sib families from G1 and G2 while the SLOW study contained 69 hybrid half-sib families from G1 and G2.  The SPACE study included 18 gulf half-sib families and the SLOW study included 4 gulf half-sib families.  No overlapping gulf families occurred between SPACE and SLOW.

Field Locations, Designs, and Measurements

The primary experimental and sampling unit in both studies are seeded swards (0.9 m by 2.4 m) which were seeded at a rate of 600 PLS m^-2 with five drilled rows, 15 cm apart.   The SPACE Arlington, WI site was established in 2019. The Prairie du Sac, WI and Hancock, WI field sites were established as part of an experiment focused on nitrogen use efficiency and marginal land productivity.  Briefly, the Prairie du Sac location is in a field where the topsoil was effectively eliminated during construction of an ammunition manufacturing facility. The Hancock, WI field site is in the central sands region of Wisconsin, with poor nutrient and moisture holding capacity.  The experimental design was a paired-plot augmented split block.  It assigned paired plots with common genotypes in adjacent high nitrogen (110 kg ha^-1) and no nitrogen (0 kg ha^-1) swards, with buffer strips of the cultivar Cave-In-Rock between to reduce nitrogen contamination.

DNA Collection and Sequencing

Leaf samples were collected during the first three growing seasons in SPACE. These greenhouse samples were aggregated for the current study to calculate initial allele frequencies for all half-sib families within the gulf breeding group and the G1 and G2 generations of the hybrid breeding population.

For the SLOW study, the Prairie du Sac and Hancock site was established in 2018 and leaf samples were collected in 2021. All leaf samples consisted of leaves from 100 distinct individuals. For leaf sampling of the hybrid breeding population, each location was divided into four spatially clustered blocks. Within each of the four blocks at each site and each hybrid breeding generation (G1 and G2), a random set of 10 paired (high and low N) hybrid sward plots were selected for sampling and 10 leaves were collected within each sward. This resulted in four samples from each block for the hybrid breeding group: high N G1, low N G1, high N G2 and low N G2.  

Overall, this resulted in 34 sets of 100 leaf samples. These leaf samples were freeze-dried prior to pooling and DNA extraction. Evenly-sized sub-samples of each leaf were cut prior to DNA extraction since pool construction error variance is reduced within numerically large pooled DNA samples (Craig et al., 2009). Technical replicates for sixteen randomly selected families were created from leaf subsamples prior to DNA extraction and sequencing.

Sequencing and bioinformatics were carried out on the pooled DNA samples of SPACE and SLOW simultaneously using methods previously presented in SPACE. The family SPACE pools were combined into breeding population samples by taking the mean allele frequencies of all applicable family samples. Specifically, the initial breeding population frequencies were obtained from SPACE samples by taking the mean allele frequency from pooled leaves collected from pots in a greenhouse for G1 (n=12), G2 (n=9), and gulf families (n=18). The same method was carried out for the samples SPACE families collected after establishment in Arlington. Specifically, the mean allele frequencies were calculated for the DNA pools collected during the third year of study (G1 n=9, G2 n=9, gulf n=14). After this filtering and sample aggregation, sites containing missing values (6.5% across all loci and samples) were imputed using the expectation-maximization method of the rrBLUP package (Endelman, 2011). After filtering and imputation, 236,262 high-confidence markers were used during further analysis.