Transcriptome data were cleaned with Trimmomatic version 0.39 (Bolger et al., 2014) using a sliding window of four base pairs with a Phred quality score cutoff of 20, a minimum length of 20, and a leading and trailing minimum of three. The software packages Bowtie2 (Langmead and Salzberg, 2013) and Tophat2 (Kim et al., 2013) were used to make indexes of the reference S. caninervis genome (Silva et al., 2020) for mapping and assembly. Htseq-countversion 0.9.1 (Anders et al., 2015) was used to estimate read counts per sample per gene and final analyses were performed in R (R Core Team, 2019) using DESeq2 (Love et al., 2014) to test for differential transcript abundance in UV-filtered and UV-transmitted samples. To test for candidate genes associated with UV-reduction and UV tolerance, transcript abundance was assessed with the DESeq2 formula: ~pair+window_treatment. In these comparisons transcript counts were normalized with DESeq2’s default model and significance was a adjusted with the BH correction. For each comparison, transcripts were considered candidates for that effect if they had an absolute logarithmic (base 2) fold change (LFC) of at least 1 and an adjusted P-value (-adj) of 0.005 or less. Normalized transcript counts were log2-transformed and LFCs were shrunken with the Approximate Posterior Estimation for generalized linear model for plotting and ranking genes (Zhu et al., 2019). Variation in transcript abundances was reduced to two dimensions with principal components analysis.