Supplemental Material for Olatoye et al., 2020

Figure S1: Scree plots showing the proportion of variation explained (percentage, %; Y-axis) by each principal component (PC; X-axis) in (A) Miscanthus sinensis, and (B) Miscanthus sacchariflorus. These PCs are from a principal component analysis conducted on 5,140 genome-wide markers.

Figure S2: Phenotypic distribution of individuals in the study populations. Distributions of Miscanthus sinensis (blue), Miscanthus sacchariflorus (green), and the 09F2 population (orange) for traits Basal circumference (Bcirc; cm), Compressed circumference (Ccric; cm), Culm length (CmL; cm), Diameter of basal internode (DBI; mm), days to first heading (HD1; days), and Yield (Yld; g/plant). The median value of each population is represented in solid lines with colors corresponding to their respective populations. The trait values of the parental lines are represented in broken lines with blue corresponding to ‘Cosmopolitan Revert’ from M. sinensis, and green corresponding to ‘Robustus’ from M. sacchariflorus.

Figure S3: Barplots showing the narrow-sense heritability (Y-axis) for basal circumference (Bcirc; cm), compressed circumference (Ccirc; cm), culm length (CmL; cm), days to first heading (HD1; days), and yield (Yld; g/plant) (X-axis), color coded based on the three populations considered in this study Miscanthus sinensis (Msi), Miscanthus sacchariflorus (Msa), and F2 breeding population (09F2).

Figure S4: Principal component (PC) analysis of Miscanthus sacchariflorus and Miscanthus sinensis diversity panels. Open circles are individuals distributed along PC1 (X-axis) and PC2 (Y-axis) in (A) M. sacchariflorus and (B) M. sinensis. These PCs are from a principal component analysis conducted on 5,140 genome-wide markers. The diamond shapes represent the parents ‘Robustus’ from M. sacchariflorus and “Cosmopolitan Revert” from M. sinensis that were used to develop the interspecific F2 population (09F2). Color coding of the individuals was based on genetic clusters from previous analyses (Clark et al. 2014, 2018) conducted on these data.

Figure S5: Heatmap showing the genetic relatedness using 5,140 genome-wide markers between accessions in the Miscanthus sinensis and Miscanthus sacchariflorus diversity panels and 09F2 breeding population. This heatmap is presented for (A) all three populations, (B) Msi only, (C) Msa only, and (D) 09F2 breeding population only.

Figure S6: Distribution of individuals selected by CDmean on the Miscanthus sinensis (Msi) principal component axes for basal circumference (Bcirc), compressed circumference (Ccirc), culm length (CmL), days to first heading (HD1), and yield (Yld). The X-axis on each graph is principal component (PC) 1, while the Y-axis is PC2. Both PCs are from a principal component analysis of 5,140 genome-wide markers. The individuals selected by the CDmean procedure are colored, and the Msi parent of the 09F2 population, “Cosmopolitan Revert”, is indicated by a diamond.

Figure S7: Distribution of individuals selected by CDmean on the Miscanthus sacchariflorus (Msa) principal component axes for basal circumference (Bcirc), compressed circumference (Ccirc), culm length (CmL), days to first heading (HD1), and yield (Yld). The X-axis on each graph is principal component (PC) 1, while the Y-axis is PC2. Both PCs are from a principal component analysis of 5,140 genome-wide markers. The individuals selected by the CDmean procedure are colored, and the Msa parent of the 09F2 population, “Robustus”, is indicated by a diamond.

Figure S8: Linkage disequilibrium decay curves for (A) Miscanthus sinensis diversity panel and Miscanthus sacchariflorus within 50 kilobase (kb) window, (B) Miscanthus sinensis diversity panel, Miscanthus sacchariflorus, and 09F2 breeding population within 250 kb window, and (C) Miscanthus sinensis diversity panel, Miscanthus sacchariflorus, and 09F2 breeding population within a 2,000 kb window. On each graph, the X-axis is the physical distance between marker pairs and they Y-axis is the squared Pearson correlation between the markers.

Figure S9: Comparison of using diversity panels and F2 populations as GS training sets for making predictions in simulated F2 populations. For each prediction accuracy of a given F2 population, either the diversity panels, or a stratified random sample of the remaining 49 F2 populations were used as training set. Thus, each boxplot represents a distribution of prediction accuracies (Y-axis) across 50 simulated interspecific F2 populations for traits with contrasting genetic architectures (X-axis). Each boxplot was colored based on the approach used to train the genomic selection model which are: Msa (all 598 individuals in the Miscanthus sacchariflorus panel), Msi (all 538 individuals in the Miscanthus sinensis panel), F2.6H (600 randomly selected individuals from the 50 simulated F2 populations), Msi.Msa (sum of the genomic estimated breeding values, or GEBVs, estimated from Msi and Msa panels), F2.1K (1,200 randomly selected individuals from the 50 simulated F2 populations), MM.F6H (sum of the GEBVs estimated from Msi and Msa panels, and 600 randomly selected individuals from the 50 simulated F2 populations), MM.F1K (sum of the GEBVs estimated from Msi and Msa panels, and 1,000 randomly selected individuals from the 50 simulated F2 populations), F2.10K (all the individuals (n=10,800) in the 50 simulated F2 populations), and MM.F9K (sum of the GEBVs estimated from Msi and Msa panels, and 10,800 individuals from the 50 simulated F2 populations). Traits were simulated using five different scenarios, namely: D.QTN (traits simulated with completely different QTN in Msi and Msa but with the same effect sizes), D.QTN.Msa (traits simulated with different QTNs in each of Msi and Msa, with Msa QTNs having large effects while Msi QTNs had small effects), D.QTN.Msi (traits simulated with different QTNs in each of Msi and Msa, with Msi QTNs having large effects while Msa QTNs had small effects), P.QTN (traits where with 50% of the QTNs were the same across Msi and Msa, while 50% were different), and S.QTN (traits simulated in Msi and Msa based on the same QTNs and same effect sizes). All simulated traits had 20 additive QTN, 0 dominance QTN, and 0 epistatic QTN, while the heritabilities are as presented in Table 2. The white dots represent the mean value of each distribution.

Table S1: Summary statistics of phenotypic least squares means within Miscanthus sinensis and Miscanthus sacchariflorus diversity panels and the 09F2 breeding population.