Prediction of candidate NLRs

To predict the candidate NLR (nucleotide binding leucine rich repeat) resistance genes in our reference transcriptome, the NLR-Parser pipeline was applied (Steuernagel, Jupe, Witek, Jones, & Wulff, 2015). We picked the highest scoring domain found per reading frame per transcript and manually screened the ORFs containing that domain. The transcripts were left out if the ORF was too short or if they were missing the appropriate start and stop codons and BLAST queries did not identify hits from NCBI nr database. The NLR transcripts were considered to be expressed in a genotype level assembly if the count was greater than one in at least two replicates. The Venn diagram of expressed and differentially expressed NLR genes across all genotypes was made using venn package (https://CRAN.R-project.org/package=venn) in R.
Evolutionary analysis of the NLR transcripts was carried out usingAntirrhinum majus L. (snapdragon) proteins and coding sequences (Li, Zhang, et al., 2019) as outgroup, since it is evolutionarily the most recently diverged plant from P. lanceolata where full genome assembly is available. Multiple sequence alignment of all of the NLRs in reference assembly was carried out using MAFFT, and the phylogenetic tree was estimated using FastTree (Price, Dehal, & Arkin, 2010). ClusterPicker (Ragonnet-Cronin et al., 2013) script with 90 percent similarity and genetic distance of 0.03 with gap option was used to cut the phylogenetic tree into clusters. The longest sequence in each cluster was used as BLAST queries against snapdragon proteins and the closest hit was selected to represent the ancestral state and added to the cluster. The phylogenetic tree was first visualized using the newick format output tree produced by ClusterPicker and the final visualization was carried out with ggtree package (Yu et al., 2017) in R.