Transcriptome-Based Phylogenomics and Adaptive Divergence Across Environmental Gradients in Epimedium brevicornu

Abstract

Ecology and adaptive differentiation of Epimedium are central to understanding both its taxonomic complexity and medicinal value. In this study, we integrate transcriptomic and plastid data from four natural populations of E. brevicornu (HZ, QLH, TS, WD) to reconstruct their phylogenetic relationships, estimate divergence times, and identify candidate genes associated with local adaptation. Nuclear gene-based phylogenies provide higher resolution and greater topological consistency than plastid data, underscoring the utility of nuclear data in lineages affected by hybridization and incomplete lineage sorting. Molecular dating indicated that major intraspecific divergence occurred during the mid-Quaternary (0.61–0.45 Ma), coinciding with climatic oscillations and montane isolation. Population structure showed strong correlations with temperature and precipitation gradients, suggesting environmentally driven selection. Signatures of positive selection and accelerated evolutionary rates revealed population-specific enrichment of genes involved in stress response, protein modification, signaling, and carbohydrate metabolism—key pathways linked to high-elevation adaptation. Protein–protein interaction networks further indicated a two-tier adaptation mechanism: ancestral network rewiring combined with population co-evolution of interacting genes. Together, these findings advance our understanding of alpine plant adaptation and provide candidate genes for further functional and breeding studies in Epimedium.

1. Introduction

Epimedium L., the largest herbaceous genus within Berberidaceae, comprises approximately 60 species distributed from east Asia to northwestern Africa [1]. Several species have long been used in traditional medicine across China, Korea, and Japan for the treatment of sexual dysfunction, osteoporosis, and cardiovascular diseases [2,3]. Owing to its pharmacological and economic importance, this genus has attracted extensive research attention. China represents the main diversity center, harboring approximately 50 endemic species mainly within sect. Diphyllon [1,4,5,6]. These species occupy complex montane landscapes shaped by the uplift of the Qinghai–Tibet Plateau and Quaternary climatic oscillations, which promoted rapid radiation during the Pleistocene (~1.14 Million Years Ago, Ma) [7,8]. However, frequent hybridization, introgression, and incomplete lineage sorting have produced extensive genetic admixture, rendering taxonomy and species delimitation in sect. Diphyllon particularly challenging [4,6,9]. Morphological traits such as petal shape, spur structure, and indumentum have traditionally been used for classification, yet these often conflict with molecular evidence. The four morphological series within sect. Diphyllon are not monophyletic, and intraspecific morphological variation in traits, such as trichomes or floral coloration, frequently exceeds interspecific differences [4,6,9,10]. Molecular markers, including AFLP, ITS, and several plastid regions, provided preliminary phylogenetic frameworks [6,11,12], but their limited informative sites often yielded poorly resolved or polyphyletic patterns. Recent phylogenetic efforts have increasingly incorporated plastid genomes, which, due to their small size and conserved structure, provide valuable taxonomic markers [7,8,9,13,14]. Plastome data clarified certain controversies, such as the synonymy of E. dewuense with E. dolichostemon and the basal placement of E. koreanum. Nonetheless, plastid genomes offer limited resolution in rapidly radiating lineages because of low substitution rates, uniparental inheritance, and frequent plastome capture through hybridization, resulting in discordance with nuclear phylogenies [9].

Compared to traditional molecular markers such as ITS, AFLP, and plastid DNA sequences, genotyping-by-sequencing (GBS) provides genome-wide single-nucleotide polymorphism (SNP) data and has greatly improved clade resolution and species delimitation, particularly in taxonomically complex and rapidly radiating groups like sect. Diphyllon [4]. However, because GBS primarily samples anonymous non-coding regions and generally lacks functional annotations, its ability to infer the adaptive significance of genomic variation is limited. In contrast, nuclear genome and transcriptome data (e.g., RNA-seq) enable the recovery of high-confidence single-copy orthologous genes while also providing gene expression information. These datasets are especially valuable for both phylogenomic inference and tests of positive selection [15]. Functionally annotated, genome-scale markers substantially increase the power to disentangle population-level divergence and ecological adaptation, even in non-model medicinal plants where traditional markers are insufficient [16]. Against this backdrop, the advent of high-throughput sequencing, particularly multi-locus analytical frameworks based on transcriptomic or whole-genome datasets, has opened new opportunities to resolve the systematics and adaptive evolution of Epimedium across heterogeneous environmental gradients [17].

The pharmacological efficacy of Epimedium species is largely attributed to diverse secondary metabolites, especially flavonoids such as icariin and epimedins A/B/C, as well as lignans, alkaloids, and organic acids. These compounds are synthesized through phenylpropanoid and flavonoid pathways, with further modifications mediated by UDP-glycosyltransferases [18,19]. Considerable variation in metabolite accumulation profiles and transcriptomic expression has been observed among sect. Diphyllon species and populations [20,21], suggesting that ecological selection influences the biosynthesis of key medicinal compounds [22]. Most species in this section have narrow, fragmented distributions and strong habitat specialization, for example, E. chlorandrum is confined to limestone crevices in Sichuan [23]. In contrast, E. brevicornu is broadly distributed across the Loess Plateau, Qinling Mountains, and adjacent regions, exhibiting tolerance to strong temperature seasonality and extreme minimum temperatures [12,24,25]. Its wide ecological range provides an excellent model system for studying the role of natural selection in shaping population-level genomic divergence.

Despite advances in taxonomy and systematics, no comprehensive study has addressed whether natural selection underlies population divergence in Epimedium or how adaptive signals correlate with environmental gradients. In this study, we sampled four natural populations of E. brevicornu across heterogeneous elevational gradients and climatic zones and integrated transcriptome data with complete plastid genomes. By reconstruction robust phylogenies, estimating divergence times, and identifying candidate genes under selection, we aim to elucidate how ecological heterogeneity drives adaptive divergence in this species. Our findings enhance phylogenetic resolution in a taxonomically complex genus and provide insights into the molecular mechanisms of environmental adaptation in montane medicinal plants.

2. Materials and Methods

2.1. Sample Collection and Climate Data Acquisition

Gansu represents a core distribution area of E. brevicornu and lies at the transitional zone between the East Asian monsoon region and the continental interior. This region exhibits pronounced environmental heterogeneity, including steep elevational gradients, strong temperature seasonality, extreme minimum temperatures, and variable precipitation regimes, which together provide a valuable framework for investigating local adaptation and ecological differentiation. To capture this variation, we conducted systematic sampling across four ecologically distinct sites in Gansu during July–August 2024. Two populations were located on the western margin of the Loess Plateau, Hezheng County, Linxia City (HZ, 2250 m) and Qilihe District, Lanzhou City (QLH, 2416 m); one in the southeastern mountainous region, Qinzhou District, Tianshui City (TS, 1598 m); and one in the Qinling-Daba Mountains transition zone, Wudu District, Longnan City (WD, 1736 m). From each site, root, stem, and leaf tissues were collected from four healthy mature individuals that were disease-free and representative of local phenotypes. All samples were immediately flash-frozen in liquid nitrogen and stored at −80 °C. Climatic data for each site were obtained from the WorldClim 2.1 global climate database, which provides 19 bioclimatic variables (BIO1-BIO19) for the 1970–2000 period (Table S1) [26]. These variables were used to characterize local environmental heterogeneity and to support subsequent genotype–environment association analyses.

2.2. RNA Extraction, Transcriptome Library Construction, and High-Throughput Sequencing

Total RNA was extracted from each tissue sample using a modified CTAB protocol with PBIOZOL reagent (Beijing, China). RNA concentration and integrity were assessed with a Qubit^® 4.0 fluorometer (MA, USA), and a Qsep400 analyzer (Taiwan, China). To minimize tissue-specific expression bias, equimolar amounts of total RNA from root, stem, and leaf tissues were pooled to construct one representative transcriptome library per population. Polyadenylated mRNA was enriched with Oligo(dT) magnetic beads and fragmented at 94 °C to yield ~200 bp fragments. First- and second-strand cDNA synthesis was followed by end repair, adapter ligation, and purification with AMPure XP beads (CA, USA). Size-selected fragments (250–350 bp) were PCR-amplified to produce strand-specific libraries. Library quality was verified by Qubit quantification and fragment analysis on the Qsep400 platform. Qualified libraries were pooled in equimolar ratios and sequenced on the Illumina NovaSeq 6000 platform (paired-end, 150 bp).

2.3. Data Preprocessing, Transcriptome Assembly, and Functional Annotation

Raw paired-end reads were quality-filtered with Fastp (v0.23.2) [27], removing adapter sequences, reads with >10% ambiguous bases (N), and reads in which >50% of bases had low quality (Q ≤ 20). Clean reads were de novo assembled into primary transcripts using Trinity (v2.15.2) with default parameters [28]. Redundancy was reduced by clustering transcripts with CD-HIT (95% similarity threshold) and grouping isoforms with Corset [29,30]. Transcript abundance was quantified with RSEM, and only unigenes with FPKM ≥ 1 were retained for downstream annotation.

Open reading frames (ORFs) were predicted using TransDecoder (v5.5.0) with a minimum prediction length of 100 amino acids to generate candidate coding sequences (CDS). Functional annotation was performed by aligning unigene sequences against NR, Swiss-Prot, TrEMBL, KEGG, GO, and COG/KOG databases using DIAMOND (v2.1.4.158). Protein domains were further annotated with Pfam using HMMER (v3.4), providing additional functional and structural information.

2.4. Orthologous Gene Identification and Sequence Alignment

To investigate phylogenetic relationships and adaptive evolution in Epimedium, we integrated genomic, transcriptomic, and chloroplast genomic data from 11 species/populations and 3 outgroups. RNA-seq data for 7 species were obtained from the NCBI SRA database (Table S2), and genomic data for E. pubescens (GWHBECS00000000) were retrieved from the National Genomics Data Center (NGDC). All transcriptomes were quality controlled with Fastp, assembled using Trinity, and translated into CDS/protein sequences with TransDecoder. Outgroup genomes and annotations including Aquilegia coerulea (GCA_002738505.1), Coptis chinensis (GCA_015680905.1), and Papaver somniferum (GCA_003573695.1) were downloaded from GenBank.

Chloroplast genomes and annotated CDS/protein sequences for the 11 species, including P. somniferum (NC_029434.1), A. coerulea (NC_041528.1), C. chinensis (NC_036485.1), E. koreanum (NC_029943.1), E. pubescens (NC_053532.1), E. pseudowushanense (NC_029945.1), E. wushanense (MN857417.1), E. sagittatum (NC_029428.1), E. ilicifolium (OQ366394.1), E. jinchengshanense (NC_058849.1), and E. brevicornu (NC_046776.1) were also obtained from GenBank to compare nuclear and plastid phylogenetic signals.

Orthologous clusters were identified with OrthoFinder [31], aligned at the codon level using PRANK (v.170427) [32], and trimmed with FasParser (v2.13.2) [33]. Three datasets were assembled: (i) a nuclear dataset comprising 4D-sites from 924 orthologous genes across 14 species (69,908 bp) for phylogenetic reconstruction and divergence dating; (ii) a plastid dataset containing 48 orthologous genes across 11 species (55,929 bp) for chloroplast phylogeny; and (iii) an Epimedium population dataset including 3125 codon-aligned nuclear genes across 11 species/populations, used to examine evolutionary rate variation and environmental adaptation in four E. brevicornu populations.

2.5. Phylogenetic Analysis and Divergence Time Estimation

Phylogenetic relationships and evolutionary history in Epimedium were inferred from 4D-site datasets of nuclear and chloroplast genomes. Maximum likelihood (ML) analyses were inferred performed in RAxML (v8.2.12) [34] under the GTR + G4 model, selected with ModelTest-NG (v0.2.0) using the Akaike Information Criterion (AIC). Nodal support was assessed with 1000 ultrafast bootstrap replicates and values were reported as bootstrap percentages (MLB). Bayesian inference (BI) tree was constructed in MrBayes (v3.2.7a) [35], using MCMC chains run for 10 million generations, sampling every 1000 generations, and discarding the first 25% as burn-in. Nodal support in the consensus tree were reported as Bayesian posterior probabilities (BPP).

Divergence times were estimated from the ML topology using both MCMCTree (PAML v4.9) [36] and BEAST (v1.10.5) [37]. MCMCTree employed a strict molecular clock model (clock = 1) 1000 burn-in iterations, followed by sampling every 50 generations to obtain 100,000 posterior samples. BEAST analyses were run for 10 million generations under a strict clock, sampling every 1000 generations with a 1 million generation burn-in. Model setup was performed in BEAUti, and consensus trees were generated with TreeAnnotator. Three critical calibration points from the TimeTree (http://www.timetree.org, accessed on 22 March 2025) were applied in both approaches to constrain divergence estimates: (i) root divergence of P. somniferum vs. other taxa (mean = 107.9 Ma, 95% interval = 102.9–117.2 Ma, SD = 3.65 Ma); (ii) outgroup divergence of Aquilegia + Coptis vs. Epimedium (mean = 89.2 Ma, 95% interval = 84.0–97.1 Ma, SD = 3.34 Ma); and (iii) crown group divergence of Aquilegia vs. Coptis (mean = 49.2 Ma, 95% interval = 25.7–79.6 Ma, SD = 13.75 Ma).

2.6. Lineage-Specific Evolutionary Rate Acceleration Analysis

Lineage-specific rate divergence in Epimedium was examined using 3125 nuclear orthologs under the free-ratio model in codeml (PAML). For each lineage, 150 orthologs were randomly sampled and concatenated 10,000 times to generate background rate distributions, with mean Ka/Ks values representing substitution rates. In the four E. brevicornu populations (HZ, QLH, TS, WD), extreme outliers (Ka/Ks < 0.005 or > 10) were removed, and the 95th percentile of the background Ka/Ks distribution was used as the threshold for accelerated evolution. Genes exceeding this threshold were designated as candidate for rate acceleration. To evaluate population-level divergence in acceleration, a binary gene × population matrix was constructed and summarized in a 2 × 4 contingency table (accelerated vs. non-accelerated genes across populations). Statistical significance was assessed using Fisher’s Exact Test and Pearson’s Chi-squared Test.

2.7. GO Enrichment Analysis of Accelerated-Evolving Genes

Functional divergence underlying adaptive evolution across distinct geographic populations of E. brevicornu was investigated through GO enrichment analyses of accelerated-evolving genes from four populations: HZ (n = 762), QLH (n = 796), TS (n = 767), and WD (n = 630). Analyses were conducted in R (v4.3.2) using clusterProfiler (v4.6.0) [38] with the Benjamini–Hochberg method (FDR < 0.05) [39].

2.8. Correlation Analysis Between Accelerated-Evolving Genes and Environmental Factors

Functional divergence of accelerated-evolving genes was quantified by calculating the proportion of genes associated with significantly enriched GO terms in each population. Redundancy analysis (RDA) was then applied for dimensionality reduction, with the first two axes (PC1, PC2) used to capture major gradients in functional composition. To evaluate environmental influence, Spearman’s rank correlations were computed between RDA axes and environmental variables, including field-measured elevation and 19 bioclimatic indices (BIO1-BIO19) from the WorldClim 2.1, representing key ecological factors such as temperature, precipitation, and seasonality. Two-tailed tests applied with α = 0.05, reporting both correlation coefficient (ρ) and p-value. Given the limited sample size (n = 4 populations), these analyses are exploratory, aiming to highlight potential ecological adaptation trends that warrant validation across broader geographic and population scales.

2.9. Identification of Candidate Genes Under Positive Selection

Positive selection across extant populations and ancestral lineages of E. brevicornu was assessed using 3125 single-copy orthologous from transcriptome assemblies. Branch-site tests were implemented in codeml (PAML), with seven foreground branches specified: four terminal populations (HZ, QLH, TS, WD) and three ancestral nodes: (HZ + QLH), (TS + WD), and their combined clade ((HZ + QLH) + (TS + WD)). Remaining branches were treated as background. For each gene, likelihood ratio test (LRT) compared Model A (allowing ω > 1 at some codons on the foreground) with null model (Fixes ω = 1 for codons in foreground branches, allowing only neutral or purifying selection). Significance was assessed under a χ² distribution (df = 1), and genes with p < 0.05 were classified as positively selected. For significant genes, codon-level selection was evaluated with the Bayes Empirical Bayes (BEB) method, and sites were considered under strong positive selection if ω > 1 and BEB posterior probability ≥ 0.80. To minimize false positives, all candidates underwent stringent quality control, including validation of transcript structure, inspection of splice junctions and alignment boundaries, filtering of frameshift mutations, and removal of misassembled or chimeric transcripts.

2.10. Protein–Protein Interaction Network Construction and Core Gene Identification

To characterize molecular interaction patterns of positively selected genes (PSGs) in E. brevicornu, we constructed protein–protein interaction (PPI) networks using Arabidopsis thaliana as a reference. PSG from four populations (HZ, QLH, TS, WD) and three internal branches (A0: HZ + QLH; A1: TS + WD; A2: HZ + QLH + TS + WD) were mapped to A. thaliana proteins in UniProt via BLASTP (E-value ≤ 1 × 10⁻⁵, identity ≥ 80%, coverage ≥ 70%). High-confidence interactions were retrieved from BioGRID v4.4.2 [40], and branch-specific PPI networks were reconstructed. Network visualization and topological analyses were performed in Cytoscape v3.10.0, with hub genes identified based on degree, betweenness, and closeness centrality.

3. Results

3.1. Transcriptome Sequencing and De Novo Assembly

To explore gene expression landscapes and adaptive mechanisms in E. brevicornu, transcriptomes were sequenced from pooled root, stem, and leaf tissues of four populations in south-central Gansu Province (HZ, QLH, TS, and WD). Illumina NovaSeq 6000 sequencing generated 83.12–100.71 million raw reads per population (Table S3), of which 79.44–96.01 million high-quality clean reads were retained after stringent filtering. De novo assembly using Trinity, followed by redundancy reduction using CD-HIT and Corset, yielded 30,014–56,274 non-redundant unigenes. The assemblies showed strong integrity, with N50 values ranging from 1466 to 1916 bp, indicating high sequence continuity. ORF prediction identified 22,072–47,266 CDS, and the corresponding peptides were for downstream analyses.

3.2. Transcript Annotation and Functional Characterization

Functional annotation assigned 76.73–89.09% of unigenes across the four populations to at least one major database (Table S4), with WD exhibiting the highest annotation rate (50,137 unigenes). The NR database provided the most comprehensive coverage, with 80.92% (HZ), 79.51% (QLH), 81.73% (TS), and 83.82% (WD) of sequences exhibiting strong homology (E-value < 1 × 10⁻²⁰; mean identity: 76.85–80.15%). Among the annotated genes, 3125 high-confidence single-copy nuclear orthologs were identified, showing near-complete annotation in NR (99.74%), TrEMBL (99.71%), and Pfam (92.22%), thereby reflecting their evolutionary conservation and functional robustness (Figure 1A).

GO annotation revealed extensive functional diversity across cellular components, biological processes, and molecular functions (Figure 1B). Within cellular components, predominant categories included intracellular anatomical structure (n = 2179), organelle (n = 1981), and cytoplasm (n = 1733), reflecting widespread subcellular distribution. Biological process was dominated by metabolic process (n = 1611), regulation of biological process (n = 684), and response to stress (n = 426) underscoring their central roles in metabolism, regulatory control, and environmental adaptation.

Molecular function was strongly represented by nucleic acid binding (n = 554), transferase activity (n = 508), and protein binding (n = 335), reflecting critical involvement in transcriptional regulation, enzymatic modification, and protein–protein interactions. In agreement with these patterns, KOG classification (Figure 2B) revealed enrichment in signal transduction, posttranslational modification and chaperone activity, and translation and ribosome biogenesis, further emphasizing the predominance of genes associated in cellular signaling, protein homeostasis, and adaptive regulation.

3.3. Phylogenomic Reconstruction of Epimedium and Nuclear–Plastid Topological Discordance

Phylogenetic relationships of Epimedium were reconstructed using ML and BI based on 69,908 bp of concatenated 4D-sites from 924 nuclear single-copy genes across 14 taxa (Figure 3). Both ML and BI trees yielded identical topologies, recovering a robustly supported monophyletic Epimedium clade (MLB = 100%; BPP = 1.00) with E. koreanum (sect. Macroceras) as the basal lineage [41].

Within the sect. Diphyllon, the four populations of E. brevicornu (HZ, QLH, TS, WD) formed distinct, well-supported geographic lineages (MLB = 84–100%; BPP = 1.00), sister to E. pubescens, consistent with recent divergence or incomplete lineage sorting. A separate clade comprising E. ilicifolium, E. sagittatum, E. wushanense, and E. pseudowushanense displayed the topology (E. ilicifolium, (E. sagittatum, (E. wushanense, E. pseudowushanense))). This structure was strongly supported under BI (BPP = 0.99), but weaker under ML (MLB = 47%), likely reflecting rapid radiation or hybridization [42]. E. jinchengshanense and E. wushanense formed a strongly supported sister pair (MLB = 100%; BPP = 1.00), consistent with nuclear genome studies [4].

Chloroplast phylogeny, based on 48 protein-coding genes (55,929 bp) (Figure 4), confirmed the monophyly of Epimedium and basal placement of E. koreanum (MLB = 100%; BPP = 1.00), but showed weak resolution within sect. Diphyllon (MLB = 13–53%). Three key nuclear-plastid conflicts were detected: (i) E. pubescens grouped with (E. sagittatum + E. pseudowushanense) in the nuclear tree but with E. jinchengshanense in the plastid tree; (ii) the branching order of E. ilicifolium was reversed; and (iii) plastid phylogeny lacked hierarchical resolution, displaying polytomies. These discrepancies likely reflect plastid introgression, hybridization, or lineage sorting, compounded by the lower substitution rate of plastid genomes.

3.4. Divergence Time Estimation of Epimedium Lineages

Based on nuclear genome data (Figure 5A,C), the split between E. koreanum and sect. Diphyllon (Node 4, Table S6A) was dated to ~2.00 Ma (MCMCTree) and ~1.70 Ma (BEAST), consistent with previous estimates of 1.80–2.36 Ma [9,12], supporting the establishment of sect. Diphyllon as an independent lineage in the early Pleistocene. The divergence of E. brevicornu from other Chinese lineages (Node 5) was estimated at 0.91–0.78 Ma, slightly younger than 1.05–1.23 Ma reported by Guo et al. (2022) [9], but older than 0.46–0.60 Ma reported by Zhang et al. (2007) [12], suggesting that nuclear genomes provide finer temporal resolution for recent divergence events.

Within sect. Diphyllon, shallow nodes exhibited sequential differentiation. For instance, Node 6 (E. pubescens and relatives) diverged at 0.80–0.69 Ma, while Node 7–10 (E. jinchengshanense to E. pseudowushanense) differentiated stepwise between 0.73 and 0.21 Ma. These estimates agree with Zhang et al. (2023) [4] but improve resolution and accuracy. Notably, we provide the first quantitative divergence times for natural populations of E. brevicornu (QLH, HZ, TS, WD) (Figure 5C; Table S6B). Divergence between QLH and HZ (Node 12) and between TS and WD (Node 13) occurred at 0.53/0.45 Ma and 0.61/0.52 Ma, respectively, while the split between the two subpopulations (Node 11) was dated to 0.67/0.57 Ma. These mid-Pleistocene events coincide with climatic oscillations and montane isolation, suggesting a strong role for environmental change in shaping population structure and adaptation.

By contrast, chloroplast-based divergence time estimates showed systematic biases. Although the plastid tree recovered a topology broadly consistent with the nuclear phylogeny, it underestimated deeper divergence (e.g., Node 2, Ranunculaceae–Berberidaceae, 100.15/74.16 Ma vs. TimeTree reference 84.0–97.1 Ma) while overestimating shallow nodes with inflated confidence intervals. For Node 10, the difference reached 0.13 Ma (0.77 vs. 0.64 Ma), and the 95% CI spanned 1.05 Ma which far broader than nuclear-based estimates. Moreover, Nodes 6–10 displayed overlapping confidence intervals, further reducing resolution. These discrepancies, consistent with prior studies in Epimedium and other taxa [4,8,43], likely reflect the slower evolutionary rate, smaller effective population size, and limited phylogenetic signal of plastid genomes, which constrain their temporal accuracy in recently radiating lineages.

3.5. Molecular Evolutionary Rate Analysis of Chinese Epimedium Lineages

We analyzed 3125 nuclear orthologs using a concatenated random sampling strategy to estimate nonsynonymous-to-synonymous substitution ratios (Ka/Ks, or ω), thereby quantifying lineage-specific selection pressures (Figure 6, Table S7).

At the species level (Figure 6A and Figure S3), the mean Ka/Ks values across E. pubescens, E. jinchengshanense, E. ilicifolium, E. sagittatum, E. wushanense, and E. pseudowushanense ranged from 0.181 ± 0.014 (E. wushanense) to 0.262 ± 0.008 (E. ilicifolium), forming narrow unimodal distributions indicative of strong purifying selection (ω < 1) and relatively uniform evolutionary rates among taxa. By contrast, the four E. brevicornu populations exhibited uniformly lower rates: WD (0.163), HZ (0.170), QLH (0.174), and TS (0.177) (Figure 6A; Table S7). Nonparametric tests (Table S8) confirmed significant pairwise differences (adjusted p < 0.001), except between HZ and TS (p > 0.05), reflecting systematic divergence in population-level evolutionary dynamics. Although, the proportion of PSGs was low (2.5–3.0%, Figure 6B), correlation analyses of Ka and Ks revealed contrasting selection regimes (Figure 6C). WD showed a weak but significant positive correlation (r = 0.10, p = 0.0021), consistent with relaxed constraints, while HZ, QLH, and TS displayed strong negative correlations (r = −0.30 to −0.27, p < 1 × 10⁻¹⁸), suggesting intensified purifying or directional selection. These patterns align with ecological differentiation: relaxed selection in WD’s warm, humid lowlands versus stronger selective pressures in colder or more arid environments such as HZ and TS. Collectively, these findings highlight the pervasive role of purifying selection in Epimedium genomes, while also revealing how local environments shape fine-scale variation in evolutionary rates within E. brevicornu.

3.6. Identification of Accelerated Evolution Candidate Genes and Functional Enrichment Analysis in Four E. brevicornu Geographic Populations

Applying population-specific 95th percentile Ka/Ks thresholds (excluding extreme outliers), we identified 796 candidate accelerated genes in QLH, 762 in HZ, 767 in TS, and 630 in WD. Although no GO or KEGG categories remained significant after FDR correction, consistent enrichment profiles across populations revealed functional divergence in metabolism, stress response, and energy regulation (Figure 7; Tables S9 and S10). Distinct enrichment signatures reflected local ecological pressures. In the humid, low-elevation QLH and WD populations, accelerated genes clustered in steroid biosynthetic pathways (e.g., sterol metabolic process, GO:0016125; steroid biosynthesis, ko00100), implicating secondary metabolite regulation in adaptation. In contrast, the higher-elevation or drier HZ and TS populations exhibited enrichment in carbohydrate metabolism (starch metabolism, GO:0005982; cellular polysaccharide biosynthesis, GO:0033692; starch and sucrose metabolism, ko00500), suggesting enhanced energy storage and drought tolerance. Stress- and defense-related pathways also showed regional specialization. WD displayed signatures in fungal defense (GO:0050832), hormone signaling (GO:0009725), and circadian regulation (ko04712), indicating adaptation to biotic stress. TS was enriched in jasmonic acid signaling and secondary metabolite biosynthesis (ko01110), consistent with chemically mediated defense strategies. Energy regulation and organelle maintenance were particularly prominent in high-altitude populations. HZ showed accelerated evolution in mitochondrial and peroxisomal functions (GO:0007005, GO:0007031) and respiration-related pathways (ko00020, ko00630), while QLH exhibited enrichment in mitochondrial targeting (GO:0006626), ABC transporters (ko02010), and quinone biosynthesis (ko00130), suggesting improved energy efficiency and organelle stability under environmental stress.

3.7. Regional Environmental Correlates of Accelerated Gene Evolution in E. brevicornu Populations

Comparative analyses of the four E. brevicornu populations (HZ, QLH, TS, WD) revealed both shared and population-specific correlations between accelerated-evolving genes and key environmental variables, particularly altitude and precipitation (Table S11). Altitude showed uniformly strong positive correlations (Spearman’s ρ = 1, p = 0) with core cellular functions, including metabolism (GO:0008150), organelle membranes (GO:0031090), and biological regulation (GO:0065007), indicating convergent adaptation to high-elevation stressors such as hypoxia and cold. In contrast, temperature variables (BIO1, BIO5, BIO8–BIO11) exhibited consistent negative correlations (ρ = −1, p = 0) with regulatory processes (GO:0048518, GO:0048522), suggesting reduced investment in energy-intensive regulation under thermal stress.

Precipitation effects were more heterogeneous and population-dependent. The driest month’s precipitation (BIO14) correlated positively with carbohydrate metabolism (GO:0005975) but negatively with stress-buffering processes such as protein folding (GO:0006458), reflecting trade-offs under water limitation. Precipitation seasonality (BIO17–BIO19) was positively linked to nucleic acid binding (GO:0003676) and negatively to ATP-binding (GO:0005524), suggesting shifts in transcriptional and metabolic regulation under variable rainfall regimes. Notably, TS and WD populations displayed unique associations with host-related GO terms (e.g., GO:0033646), potentially reflecting localized adaptation to biotic pressures in humid or transitional habitats. Together, these results indicate that while altitude and temperature exert broad, convergent constraints, precipitation gradients drive fine-scale, population-specific evolutionary trajectories in metabolism, regulation, and biotic interactions.

3.8. PSGs and Functional Adaptations in E. brevicornu Populations

Branch-site analyses of 3125 nuclear orthologs across 11 Epimedium taxa identified PSGs in both extant populations: HZ (11), QLH (20), TS (11), WD (17), and ancestral nodes: HZ_QLH (7), TS_WD (2), and the basal HZ_QLH_TS_WD clade (17) (Figure 8; Table S12), indicating pervasive adaptive selection during both recent and ancestral divergence. Although few GO/KEGG categories remained significant after correction, raw enrichment signals revealed distinct lineage-specific adaptive functions (Tables S13 and S14). In HZ, PSGs were associated with protein transport (importins, trigger factors) and carbohydrate metabolism (glycoside hydrolases), suggesting enhanced organellar trafficking and energy storage. QLH showed signatures in cytoskeletal organization and hormone signaling (auxin pathways, heat shock proteins), consistent with stress integration and developmental plasticity. Their shared ancestor (HZ_QLH) carried signals in cytoskeletal remodeling, trichome development, and TCA cycle genes, pointing to early morphological and metabolic shifts. In TS, PSGs clustered in glycolysis, nitrogen metabolism, and carotenoid biosynthesis, reflecting metabolic optimization and photoprotection under high-altitude stress. WD displayed the broadest enrichment (>60 GO terms), dominated by defense response, salicylic acid signaling, and cell death, highlighting immune- and stress-driven adaptation. The basal clade (HZ_QLH_TS_WD) showed selection on polysaccharide biosynthesis, stress response, and shoot/leaf development, suggesting foundational adjustments in energy metabolism and structural traits during early lineage diversification.

3.9. Protein Interaction Networks Reveal Core Regulators of Adaptive Evolution in E. brevicornu

To explore the cooperative mechanisms of PSGs in E. brevicornu, we constructed protein–protein interaction (PPI) networks for four natural populations (HZ, QLH, TS, WD) and three inferred ancestral branches (HZ_QLH, TS_WD, HZ_QLH_TS_WD) (Figure 9). Despite population-specific divergence in PSG repertoires, the network exhibited high connectivity and clear modularity, suggesting that adaptive evolution operates through coordinated regulation and complex molecular cascades.

Two ancestral PSGs, OG1168 (pre-mRNA splicing factor SPF27 homolog) and OG1988 (peptidyl-prolyl cis-trans isomerase CYP71), emerged as central hubs, bridging multiple population-specific modules. SPF27, a core component of the PRP19–CDC5L spliceosome complex, is known to regulate mRNA splicing, post-transcriptional control, and stress responses, with Arabidopsis homologs involved in photomorphogenesis and development [44,45]. OG1988 encodes a PPIase with dual roles in protein folding and spliceosome stability, paralleling the Arabidopsis homolog PPIH [46]. These ancestral hubs linked to population-specific modules via OG2807 (Heat shock 70 kDa protein 17, degree = 3) and OG1139 (chloroplastic glutamate–cysteine ligase, degree = 4). OG2807, under strong positive selection in TS, is a key thermotolerance protein [47], while OG1139, selected in HZ, regulates glutathione biosynthesis and antioxidant defense [48]. Additional population-specific hub genes highlighted distinct adaptive pathways. In QLH, OG423 (serine/threonine kinase) and OG2025 (40S ribosomal protein S3a) formed core nodes, implicating rapid signal transduction and translational regulation in stress adaptation [49,50]. OG2807 also played a central role in this population, suggesting recurrent thermal selective pressures. In TS, OG2051 (RNA-binding NOB1-like protein) was a hub involved in ribosome biogenesis and rRNA processing [51], consistent with enhanced post-transcriptional control under high-altitude stress. In WD, OG2372 (DEAD-box RNA helicase 28) occupied the most central position, reflecting strong reliance on RNA metabolism and gene expression remodeling in humid lowland environments [52]. Together, these results reveal that while ancestral hubs provide conserved regulatory backbones, each population has recruited distinct hub genes to fine-tune metabolic, regulatory, and stress-response processes, underscoring the interplay of conserved and lineage-specific regulators in E. brevicornu adaptation.

4. Discussion

4.1. Systematic Classification and Evolutionary Dynamics of Sect. Diphyllon

The Chinese Epimedium clade (sect. Diphyllon) constitutes the evolutionary and biogeographic center of the genus, comprising approximately 50 endemic species that dominate the temperate montane forests of central and southern China, with disjunct populations extending into adjacent regions of East Asia [1]. Stearn’s classical taxonomy (2002), based on leaf architecture, spur morphology, and indumentum traits, provided the first systematic framework. However, microenvironmental heterogeneity has driven pronounced phenotypic plasticity and ecological continuity in these characters [4,9,10,53], resulting in blurred species delimitations and enduring taxonomic challenges within the group. Molecular markers were subsequently introduced to address these challenges. Early phylogenies based on nuclear ITS and single plastid genes (e.g., ndhF, matK) improved resolution but often yielded unresolved polytomies due to limited nuclear loci and the slow evolutionary rate of plastid genomes [9,54]. Subsequent phylogenomic analyses utilizing complete chloroplast genomes significantly improved clade resolution, as demonstrated by Guo et al. (2022) [9], who identified four well-defined principal lineages among 32 species. However, these studies still yielded limited support for deep phylogenetic relationships and produced divergence time estimates potentially biased toward younger ages [8,9]. These limitations highlight the inherent constraints of plastid-only approaches in complex plant radiations. In contrast, nuclear genomic data offer superior phylogenetic resolution due to their biparental inheritance patterns and rich repertoire of single-copy nuclear genes [55]. Comparative studies, such as Zhou et al. in Indigofera, demonstrated that transcriptomic datasets outperform chloroplast sequences in both resolution and temporal precision for rapidly radiating clades [55]. Similarly, recent nuclear-based analyses of Epimedium emphasized whole-genome evidence to reconcile inconsistencies across plastid phylogenies [4]. Our transcriptome-derived nuclear phylogeny further underscores these advantages: it yields superior node support, resolves topological conflicts, and narrows divergence-time intervals compared with chloroplast-based trees [4,56]. Although informative, chloroplast phylogenies often conflict with nuclear trees, likely due to plastid capture, maternal bias, or ancient hybridization [8,9,11,12,54].

Collectively, these findings demonstrate that although plastid genomes advance baseline systematic resolution, nuclear genomic data provide stronger phylogenetic stability, finer temporal accuracy, and more reliable evidence for disentangling the intricate evolutionary history of Epimedium. Importantly, such plastid–nuclear discordance also bears practical implications. For taxonomy, plastid-only phylogenies risk producing misleading species delimitations in groups with frequent hybridization and reticulate evolution, underscoring the need for integrative approaches that combine nuclear genomic evidence with morphology and ecology to establish a stable classification framework. For conservation, discordant signals highlight the existence of cryptic lineages and hybrid populations that may not be reflected in traditional taxonomy, and recognizing these evolutionarily significant units is essential to avoid underestimating genetic diversity and to design more effective conservation strategies.

4.2. Ecological and Historical Interpretation of Divergence Times in Chinese Epimedium Lineages

Early phylogenetic studies based on AFLP, ITS, and plastid markers (e.g., trnK–matK, atpB–rbcL) suggested that Epimedium originated in the Late Miocene–Early Pliocene (9.7–7.4 Ma), with the Chinese sect. Diphyllon radiating rapidly during the Pleistocene (2.6–0.4 Ma) [11,12]. Nuclear phylogenomics refined this timeline, dating the split between E. koreanum and the Chinese clade to ~2.0 Ma (95% HPD: 1.92–2.08 Ma), coincident with intensified climatic oscillations of the Early Pleistocene and the final phase of Qinghai–Tibet Plateau uplift (1.7–3.6 Ma) [4,9,12]. These events reshaped East Asian monsoon regimes, fragmented habitats, and promoted allopatric speciation, particularly within the subtropical latitudinal belt (21–34° N), the core diversification zone of sect. Diphyllon [4]. The crown diversification of sect. Diphyllon was estimated at ~0.91 Ma (95% HPD: 0.87–0.94 Ma), followed by a burst of speciation between 0.80 and 0.62 Ma and more recent divergence at ~0.21 Ma (95% HPD: 0.19–0.24 Ma), giving rise to closely related taxa such as E. wushanense and E. pseudowushanense. Within E. brevicornu, four natural populations clustered into two lineages—QLH + HZ (~0.53 Ma, 95% HPD: 0.57–0.50 Ma) and TS + WD (~0.61 Ma, 95% HPD: 0.64–0.58 Ma), derived from a common ancestor at ~0.67 Ma (95% HPD: 0.70–0.64 Ma). This chronology coincides with the most severe glaciations of the Qinghai–Tibet Plateau (~0.8–0.6 Ma), implicating Pleistocene glaciation as a primary driver of diversification [9,57]. We also note that absolute divergence times are inherently sensitive to the choice and placement of calibration points, with deeper nodes showing greater variance, whereas relative branching patterns and more recent splits remain comparatively robust. Accordingly, the ages reported here should be viewed as approximate, providing a relative framework for diversification while underscoring the need for additional reliable fossil evidence within Berberidaceae to further refine absolute timescales.

These divergence times correspond with the broader genus-level diversification window of Epimedium (~2.0–0.5 Ma) [4,9,12] and fit regional biogeographic models that attribute rapid speciation in temperate Chinese flora to Quaternary climatic oscillations [58]. Climatic and topographic data suggest dual mechanisms of divergence: geographic isolation in high-altitude habitats (e.g., HZ, QLH) and ecological adaptation in refugial lowlands (e.g., WD, TS). WD (eastern Sichuan Basin, 1736 m, 9.37 °C, 689 mm precipitation) and TS (northern Qinba Mountains, 1598 m, 7.75 °C, 652 mm) likely functioned as humid refugia, supporting early isolation and independent evolutionary trajectories [59,60]. As emphasized by Wen et al., Quaternary climatic fluctuations repeatedly reshaped dispersal routes and genetic structures of montane plants across China [58]. Integrating divergence times, climatic history, and geographic distribution, we infer that Epimedium diversification followed a molecular-ecological evolutionary model of “geographic isolation → local refugiation → rapid divergence → climatic adaptation” [61].

4.3. Ecological Drivers of Accelerated Population Evolution: Divergent Selective Pressures Shape Gene Function in E. brevicornu

The correlation between accelerated gene evolution and environmental gradients in four E. brevicornu populations highlights how abiotic pressures sculpt genome function. Altitude consistently acted as a dominant driver, showing positive associations with metabolism (GO:0008150), organelle membranes (GO:0031090), and protein complexes (GO:0032991). This pattern aligns with evidence that high elevations, marked by hypoxia, UV radiation, and cold stress, impose convergent selection on mitochondrial and cellular homeostasis pathways to sustain energy efficiency and structural stability [62]. By contrast, thermal variables (BIO1, BIO5) were negatively correlated with regulatory GO terms (GO:0048518, GO:0048522), consistent with reduced reliance on costly transcriptional regulators in warmer climates. Comparative studies show that plants in high-temperature habitats downregulate stress-response regulators such as HSPs, MYB, ARF, and CAMTA families, reallocating resources toward metabolic efficiency [63,64]. Precipitation exerted more complex, population-specific pressures. The driest-month rainfall (BIO14) was positively associated with carbohydrate metabolism (GO:0005975) and mitochondrial electron transport (GO:0006123), but negatively with protein folding (GO:0006458) and ion homeostasis (GO:0050801), indicating trade-offs that favor energy production under drought stress. Similar reprogramming has been observed in drought-adapted taxa, e.g., Nicotiana maintains mitochondrial ETC function under water deficit [65], while Quercus ilex and Triticum boeoticum suppress glycolysis but enhance TCA and ATP synthesis [66], and Pinus massoniana redirects carbon flux toward osmoprotective trehalose under heat and drought [67]. Precipitation seasonality (BIO17–BIO19) was linked to selection on nucleic acid binding (GO:0003676, GO:0003723) and ATP-binding (GO:0005524) functions, suggesting that climatic variability drives transcriptomic reorganization and energy reallocation. Seasonal field transcriptomics in Arabidopsis halleri similarly revealed rainfall-driven oscillations in transcription factor and ATP-binding gene expression [68,69].

Finally, biotic interactions emerged in TS and WD, where host–microbe interface genes (GO:0033646) were enriched under BIO14. This resonates with findings in Arabidopsis, where variation in specialized metabolites shapes microbial hub abundance, linking ecological pressures to immune-metabolic selection [70]. Overall, altitude and temperature act as uniform ecological filters imposing convergent constraints, whereas precipitation introduces fine-scale, population-specific selective regimes, thereby generating divergent evolutionary trajectories in E. brevicornu. This dual pattern highlights the need to integrate macroclimatic and microhabitat drivers when dissecting evolutionary dynamics in montane plant systems.

4.4. Positively Selected Genes and Phylogenetic Architecture: Ancestral Node Regulatory Network Remodeling and Population Specialization Trends

Analysis of PSGs and PPI networks revealed a dual adaptive mode in E. brevicornu, integrating ancestral regulatory remodeling with population-level specialization. Branch-site analyses identified 11–20 PSGs in extant populations and 17 at the basal node HZ_QLH_TS_WD, marking a pivotal evolutionary transition. These ancestral PSGs, enriched in RNA splicing (OG1168, SPF27), protein folding (OG1988, CYP71), jasmonate signaling (OG2580, MES5), and carbohydrate metabolism (OG2390, GH9A1), formed central hubs in the PPI network. Their connectivity indicates that spliceosome regulation and protein homeostasis served as core modules conferring transcriptional flexibility and metabolic resilience during mid-Quaternary climatic oscillations (~0.53–0.67 Ma), consistent with plant adaptation via splicing, hormone integration, and metabolic optimization [71,72,73,74,75].

By contrast, extant populations exhibited strong local-environment-driven specialization. In TS (1598 m), PSGs included OG2807 (HSP70, thermotolerance) and OG2051 (NOB1-like RNA-binding protein, ribosome biogenesis), reflecting adaptation to heat and light fluctuations [76,77]. HZ (2250 m), situated in a cold–arid climate, showed selection on OG1139 (glutamate–cysteine ligase), enhancing ROS scavenging and cellular homeostasis [78]. QLH (2416 m), in a high-altitude arid ecotone, exhibited PSGs such as OG423 (serine/threonine kinase, stress signaling) and OG2025 (ribosomal protein S3a), likely reinforcing stress signaling and translational efficiency [79,80]. WD (1736 m), in a humid, pathogen-rich environment, revealed selection on OG2372 (DEAD-box RNA helicase 28), mediating RNA-based regulation of defense responses [81]. The PPI network exhibited a modular yet interconnected architecture, linking ancestral hub nodes with population-specific regulators, consistent with the “trunk–branch” adaptive framework widely observed in montane plants [82,83,84]. This dual strategy, ancestral regulatory rewiring combined with population-level specialization, provides a molecular basis for understanding adaptive evolution in E. brevicornu and highlights key candidate genes for future medicinal and breeding studies.

5. Conclusions

This study integrates nuclear and chloroplast genomic data to resolve the phylogeny, divergence history, and adaptive evolution of Epimedium, focusing on four natural populations of E. brevicornu. Nuclear phylogenomics revealed mid-Quaternary population splits (~0.61–0.45 Ma), driven by climatic oscillations and hydro-thermal gradients. Lineage-specific adaptation was shaped by distinct environmental niches, with accelerated evolution in metabolic, signaling, and defense pathways. PPI networks uncovered a dual adaptive mechanism: ancestral regulatory rewiring alongside population-level specialization. Key candidate genes for breeding and conservation include stress-responsive (HSP70), jasmonate-related (methyl-jasmonate esterase), and flavonoid-modifying (UDP-glycosyltransferases) genes, alongside regulatory hubs (spliceosome components, PPIase, DEAD-box helicase) governing environmental plasticity. These findings not only elucidate the role of hydro-thermal gradients in molecular divergence, but also offer practical genomic tools for enhancing stress resilience and medicinal compound biosynthesis in Epimedium. Future work should prioritize functional validation, genotype–phenotype associations, and marker-assisted breeding to optimize conservation and agricultural applications. By bridging evolutionary insights with applied genomics, this study advances both the understanding and utilization of Epimedium’s adaptive potential in a changing climate.