Repurposing genome skimming data for non-model plant functional phylogenetics: A case study in Rhodiola.

BACKGROUND: Investigating the evolution of functional genes in non model plants is often hindered by the lack of reference genomes and transcriptomic resources, especially for taxa inhabiting extreme environments. Here, focusing on the salidroside biosynthesis pathway in the medicinal alpine genus Rhodiola, we asked whether genome skimming data could be used to test three a priori predictions: predominant purifying selection across most pathway genes, lineage specific shifts in selective constraint under heterogeneous environments, and corresponding differences in predicted protein binding properties. METHODS: We integrated genome skimming with codon based selection analyses, environmental variable analysis, and two deep learning based tools, Evo2 for nucleotide level conservation scoring and AlphaFold 3 for protein structure prediction, to reconstruct phylogenies, detect selection signals, and evaluate relative binding patterns through molecular docking. Functional genes were mined using GeneMiner2, and phylogenetic signal analyses were performed with RASP to examine associations between gene evolutionary patterns and climatic or edaphic factors across 18 Rhodiola species. KEY RESULTS: A total of 37 target genes, including 4HPAAS, 4HPAR1, 4HPAR2, and 34 UGT family members, were retrieved with a mean recovery rate of 96.7%. Six genes (4HPAAS, 4HPAR2, UGT3, UGT9, UGT20, and UGT21) showed strong purifying selection, high structural conservation, and significant phylogenetic signals correlated with diurnal temperature range and precipitation gradients. Divergence time estimation placed functional gene diversification in the late Pliocene-early Quaternary, coinciding with major uplift events of the Qinghai-Tibet Plateau. Comparative phylogenetic regressions (PIC and PGLS), together with PAML tests, further highlighted three candidate FGs (4HPAR2, UGT10 and UGT26) showing lineage-specific shifts in selective constraint associated with environmental gradients. CONCLUSIONS: This study illustrates that genome skimming data, combined with codon based and AI based analyses, can be used to test biologically grounded predictions about the evolution of functional genes in non model plants. Our results remain preliminary, but they identify a small set of candidate genes for future functional and ecological validation.