Population genomic analysis and genome-wide association mapping of leaf morphology in Phyllanthus emblica L Authors: Qiongling Zhao, Shuman Duan, Shengyou Shi, Yibo Bai, Jie Jin, Wenying Huang, Huiying Luo, Kunjian Qian, Pengfei Wang Published: 2026 ABSTRACT AbstractPhyllanthus emblica L. is an important medicinal and edible fruit tree in tropical and subtropical regions, yet its population genetic structure and the genetic basis of key agronomic traits remain poorly understood. Here, we analyzed 112 accessions representing diverse germplasm collected from China and India. Using ddRAD-seq, we generated 73,405 high-quality SNPs to characterize population structure, genetic diversity, and the genetic architecture of leaf morphology. Population structure and phylogenetic reconstruction separated the accessions into two major genetic groups, which were further resolved into four subpopulations according to source. Genetic diversity varied markedly among groups, with one major source group retaining substantially higher diversity than the others, suggesting that the currently utilized germplasm of P. emblica has experienced uneven genetic utilization and limited systematic improvement. Genome-wide scans based on π ratios and FST identified multiple differentiated genomic regions, containing genes involved in protein turnover, stress response, hormone transport, and cell growth. We further conducted the first genome-wide association study of leaf traits in P. emblica and identified 85 loci significantly associated with leaf morphological variation. Several loci showed strong non-additive effects and clear allele-frequency differences among genetic groups. Candidate genes within these loci were functionally enriched in processes related to leaf development, leaf morphogenesis, auxin transport and hormone signaling, and cell wall biogenesis. Together, these results provide comprehensive insights into the population differentiation and leaf trait variation in P. emblica, and establish a valuable foundation for molecular marker–assisted breeding in this species.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12870-026-09272-5.Keywords: Phyllanthus emblica L., ddRAD sequencing, Population structure, Genetic differentiation, Leaf morphology, Genome-wide association study Introduction Phyllanthus emblica L., a member of the Euphorbiaceae family, is an important medicinal and edible fruit tree with substantial nutritional, medicinal, and ecological value [1]. Its fruits, roots, stems, and leaves contain diverse bioactive compounds, including phenolic acids, tannins, flavonoids, and terpenoids, which underpin antioxidant, hepatoprotective, antitumor, and antiviral activities [2–4]. Natural populations are found in arid and mountainous regions of India, Bangladesh, Southeast Asia, and southwestern China, particularly in Yunnan, Guizhou, and Sichuan provinces. In contrast, cultivated forms are mainly grown in the warmer and more humid regions of southern China, including Guangxi, Guangdong, and Fujian provinces [5]. Despite its ecological and economic importance, the population genetic structure of P. emblica remains poorly understood and has not yet been systematically investigated. Leaves are the primary organs of photosynthesis, capturing solar energy to produce the organic compounds that support plant growth and, ultimately, much of terrestrial life [6, 7]. Leaf shape and size vary widely among and within species, ranging from the broad leaves of poplars to the needle-like leaves of conifers. Such variation reflects long-term adaptation to environmental conditions, especially differences in water availability and light, driven by underlying genetic differences [8]. Examining patterns of leaf variation provides valuable insight into how plants adapt to their environments and how genetic factors shape visible traits [9]. In major crops, leaf morphology is closely associated with key agronomic characteristics, including planting density, tolerance to environmental stress, and photosynthetic performance [10–12]. For P. emblica, leaf traits are also practical because they are easy to quantify accurately in a germplasm orchard and may serve as early indicators of vegetative architecture and growth potential. However, the genetic basis of leaf variation in this species remains unknown. Recent advances in population genomics have made statistical genetics a powerful approach for linking genetic variation to observable traits. Among these approaches, genome-wide association studies (GWAS) have become widely used to uncover the genetic basis of complex traits. GWAS examines natural populations to identify relationships between genetic differences across the genome and variation in specific traits. In rice, a study of 529 accessions connected genome-wide genetic variation with 840 metabolites, leading to the identification of 36 candidate genes that influence metabolites important for plant physiology and nutrition [13]. Nevertheless, despite the availability of three published P. emblica reference genomes, functional genomics and marker development remain challenging because of its complex polyploid genome, with two subgenomes, 104 chromosomes and a genome size of approximately 1.77 Gb, which complicates read mapping, allele-dosage estimation, and association analysis [14–16]. Encouragingly, statistical models and association methods developed for homologous tetraploid potato provide practical solutions for association mapping in polyploid fruit species [17–19]. In this study, we genotyped 112 P. emblica accessions collected from Yunnan, Guangxi, Guangdong, Fujian and India. We characterized population structure, identified genomic regions showing genetic divergence between different subpopulations, and, for the first time, elucidated the genetic architecture underlying variation in leaf morphology in this species. We further incorporated candidate-gene annotation, functional enrichment analysis, and transcriptional quantification of selected genes to improve the biological interpretation of the association signals. Together, these results provide valuable genomic resources and practical targets for genetic improvement in P. emblica breeding programs. Materials and methods Plant materials and genome sequencing All 112 accessions used in this study were obtained from the Yunnan Innovation Base for P. emblica Germplasm Conservation at the Institute of Tropical Eco-Agriculture, Yunnan Academy of Agricultural Sciences [20]. These materials are maintained under unified orchard management but represent germplasm collected from different geographic origins (Table S1). Fresh young leaves were collected from each accession, immediately frozen in liquid nitrogen, and stored at − 80 °C until DNA extraction. Approximately 0.5 g of leaf tissue per accession was ground into fine powder in liquid nitrogen. Genomic DNA was extracted using a modified CTAB protocol [21, 22]. Briefly, ground tissue was incubated in preheated CTAB extraction buffer at 65 °C for 20 min, followed by chloroform: isoamyl alcohol extraction, RNase treatment, isopropanol precipitation, and ethanol washing. Genomic libraries were constructed using double-digest restriction-site associated DNA sequencing (ddRAD-seq). Briefly, genomic DNA was digested with the restriction enzymes SacI and MseI, ligated to adapters, purified, and size-selected to recover fragments of 300–400 bp. Paired-end sequencing (150 bp) was then carried out on the Illumina HiSeq 2500 platform. Variant calling Raw sequencing reads were processed using fastp v1.1.0 to remove adapter sequences and low-quality bases with default parameters [23]. Read quality was evaluated with FastQC v0.12.0 (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/) before and after filtering. High-quality reads were then aligned to the P. emblica variety ‘Bingtian’ reference genome using BWA-MEM from PRJNA993303 [16, 24]. Prior to alignment, we retained one representative chromosome, namely the longest pseudomolecule, from each homologous chromosome group within each subgenome of the P. emblica reference genome (Table S2). The resulting alignment files were converted to BAM format, sorted by genomic coordinates, and indexed with SAMtools v1.19 [25], and then potential PCR duplicates were identified and marked with the MarkDuplicates module in picard v3.4.0 (https://github.com/broadinstitute/picard). SNP discovery was performed jointly across all 112 accessions. Specifically, individual gVCF files were first generated with GATK v4.6.0 HaplotypeCaller and then merged for joint genotyping with GenotypeGVCFs, with ploidy set to four to match the dosage-based analytical framework used in downstream analyses [26]. A series of stringent quality filters was applied to remove low-confidence variants. SNPs were excluded if they met any of the following criteria: QD < 2.0, MQ < 40.0, FS > 60.0, SOR > 3.0, MQRankSum < − 12.5 or ReadPosRankSum < − 8.0. QD represents variant quality normalized by sequencing depth; MQ denotes mean mapping quality; FS and SOR assess strand bias; MQRankSum and ReadPosRankSum are Wilcoxon rank-sum statistics evaluating differences in mapping quality and read position between reference and alternate alleles, respectively. SNPable (http://lh3lh3.users.sourceforge.net/snpable.shtml) was employed to filter repetitive and ambiguously mapped regions. Variants were retained for subsequent analyses if they had a minor allele frequency greater than 0.05 and a missing rate below 0.2. Functional annotation and prediction of variant effects were performed using SnpEff v5.4 [27]. Population structure analysis Population structure was inferred using ancestry coefficient estimation implemented in the R/LEA v3.20.0 based on sparse nonnegative matrix factorization (snmf) [28, 29]. First, 4,477 unlinked fourfold degenerate sites were extracted, and genotypes were converted to dosage format, coded as 0–4 to represent allele copy number, with missing data coded as 9. The snmf algorithm was performed with ploidy = 4 and 12 independent replicates for each assumed number of genetic clusters (K = 2–10), and other parameters were kept at their default settings unless otherwise specified. Cross-entropy values were used as an auxiliary criterion for model evaluation, whereas biological interpretability of the inferred clusters was used to guide the final presentation of the results. Individual ancestry coefficients were aligned and averaged using CLUMPP v1.1.2 [30]. To further visualize the spatial distribution of ancestry components, we generated a spatial ancestry map using the R/tess3r v1.1.0 [31]. Accessions with an ancestry proportion of at least 0.55 were assigned to a specific subpopulation, whereas those below this threshold were considered admixed. Principal component analysis was performed using the glPCA function in the R/adegenet v2.1.11 [32]. For phylogenetic reconstruction, the same 4,477 fourfold-degenerate SNPs were converted to PHYLIP format using custom Python scripts. Pairwise genetic distances were calculated with the gendist program in PHYLIP v3.5c, and 1,000 unrooted bootstrap trees were generated (https://phylipweb.github.io/phylip/). A neighbor-joining consensus tree was then constructed using the consense module and visualized with the R/ggtree v3.16.3, and the tree was displayed with the India accession placed as a reference root for visualization [33]. Genetic diversity and identification of regions of genetic differentiation Nucleotide diversity (π) and genetic differentiation (FST) between Yunnan and non-Yunnan groups were estimated using the R/PopGenome v2.7.5 [34]. Calculations were performed with sliding windows of 100 kb and a step size of 10 kb. Windows with FST values in the top 5% of the empirical distribution were considered highly differentiated. Extreme π-ratio values were defined using the upper and lower 2.5% tails of the π_Yunnan /π_non-Yunnan and π_non-Yunnan/π_Yunnan distributions. Because the present study did not apply directional haplotype-based tests such as XP-CLR or XP-EHH, these regions were treated as putatively divergent genomic regions rather than definitive selective sweeps. Gene annotation was carried out using eggNOG-mapper v2.1.14 to assign GO terms and KEGG pathways [35]. Functional enrichment analyses were then conducted in R/clusterProfiler v4.16.0 [36]. Phenotype collection and statistical analyses All phenotypic measurements were performed on the same 112 accessions used for ddRAD-seq, which were grown in the same germplasm orchard under unified management. In spring 2020, mature four-year-old P. emblica trees were selected for phenotypic evaluation. Fully developed one-year-old fruiting branches were collected from four canopy orientations (east, west, south, and north) for each tree. Leaves from these branches were pooled, and 30 leaves per accession were randomly selected and scanned using a flatbed scanner. Leaf area, perimeter, length, width, and roundness were measured with Digimizer v6.5.2 (https://www.digimizer.com/). Leaf area was recorded in square centimeters (cm2), leaf perimeter, length, and width were recorded in centimeters (cm), and leaf roundness was treated as a dimensionless shape index. Potential outliers were identified within each accession based on standardized residuals. Specifically, leaf-level measurements with absolute standardized residuals greater than 3 were considered outliers and removed before calculating accession-level arithmetic means. Phenotypic data were subjected to quality control, including the removal of outliers. Trait correlations were assessed using the R/GGally v2.4.0 [37]. Broad-sense heritability (H2) was estimated as H2 = Vg/ (Vg + Ve), where Vg and Ve denote genetic and residual variance, respectively. Principal component analysis of leaf traits was performed with R/FactoMineR v2.12 [38]. Genome-wide association studies Genome-wide association analyses were conducted using SNP dosage data, retaining 73,405 high-quality SNPs with a minor allele frequency greater than 0.05 and a missing rate below 0.2. Kinship and population structure were included as covariates to control for relatedness and stratification. Both additive and complete dominance genetic models were fitted using the R/GWASpoly v2.14 [39]. The kinship matrix was calculated with the set.K function with the leave-one-chromosome-out option, and the first five principal components were incorporated as fixed effects using set.Params. Genome-wide significance thresholds were determined with the set.threshold function using the default M.eff method, which estimates the effective number of markers by accounting for linkage disequilibrium among SNPs. The resulting model-specific thresholds were approximately 5.99, 5.98, and 5.66 on the -log10 (P) scale for the additive, 1-dom-alt, and 1-dom-ref models, respectively; for consistency of presentation, P < = 1 × 10^-6 was used as the common plotting threshold. Candidate gene identification Candidate genes were identified based on all significant loci detected for the five measured quantitative leaf morphological traits, including leaf area, perimeter, length, width, and roundness. For each significant lead SNP identified under the additive or complete-dominance models, genes overlapping the significant SNP or located within a 200-kb flanking interval were extracted from the reference genome annotation and considered putative candidate genes. Functional annotation was performed using eggNOG-mapper, and GO and KEGG enrichment analyses were conducted with R/clusterProfiler. Transcriptomes of five different tissues from P. emblica were downloaded from PRJDB18024, including stem, leaf, branch, flower and fruit [15]. Adapters and low-quality sequences were removed using fastp, followed by alignment of the clean reads to the reference genome using HISAT2 v2.2.1 [40]. Finally, transcript assembly and gene quantification were performed using StringTie v3.0.3 [41]. Results Population sampling and identification of genetic variations A total of 112 P. emblica accessions were included in this study, comprising one from India, three from Fujian, nine from Guangxi, eleven from Guangdong, and eighty-eight from Yunnan (Fig. 1A, Table S1). This panel captured the major distribution areas represented in the Chinese germplasm collection used here. Population sampling and genotyping. A. Sampling sites of P. emblica in Yunnan, Guangxi, Guangdong and Fujian. Black dots indicate sampling sites, blue lines indicate major rivers, and the background color scale represents elevation in meters. B. The number of SNPs across the whole genome of the P. emblica population calculated in 1 Mb bins. C. Functional annotation categories of SNPs detected in the P. emblica population Genotyping by ddRAD-seq generated an average of approximately 0.57 Gb of raw data per accession, from which 73,405 high-quality single-nucleotide polymorphisms (SNPs) were retained after quality control (Fig. 1B, Table S3). SNPs were unevenly distributed across the genome, and their abundance was not strongly related to chromosome length (R = 0.37, P = 0.06). Chromosome 6B showed the highest SNP density (18.7 SNPs per Mb), whereas chromosome 11B showed the lowest (12.6 SNPs per Mb). Minor allele frequencies ranged from 0.05 to 0.5, with most variants (88.9%) occurring at frequencies below 0.4, whereas polymorphism information content values ranged from 0.09 to 0.375 and displayed a clear bimodal pattern (Fig. S1). Functional annotation indicated that 25.7% of SNPs were located in intergenic regions, 40.3% in regulatory regions, and 34.1% within genes (Fig. 1C). Among variants affecting coding sequences, nonsynonymous substitutions accounted for 11.9%, whereas synonymous substitutions accounted for 10.2%. Population structure in P. emblica population Phylogenetic analysis of 4,477 unlinked fourfold-degenerate SNPs grouped the P. emblica accessions primarily according to their geographic origin (Fig. 2A). All accessions from Guangxi, Fujian, and Guangdong clustered into a single, well-supported clade. Within this clade, accessions from Fujian and Guangdong showed closer genetic relationships than those from Guangxi. Ancestry analysis across K = 2–10 showed a gradual reduction in cross-entropy with increasing K, but K = 2 and K = 4 provided the clearest biologically interpretable subdivision (Fig. 2A, S2, S3, S4). At K = 2, the 112 accessions were clearly divided into Yunnan and non-Yunnan groups. Increasing K to 3 further separated the Yunnan group into two subgroups. At K = 4, the non-Yunnan group split into Guangdong and Guangxi subgroups, although two accessions from Guangxi (Guangjuan1 and Guangjuan6) clustered with the Guangdong subgroup (Fig. 2A). Principal component analysis supported the same structure: the first principal component separated Yunnan and non-Yunnan subgroups, while the second principal component distinguished Guangdong subgroup from Guangxi subgroup (Fig. 2B and C). Together, these analyses resolved the germplasm into two major subgroups: Yunnan and non-Yunnan, and four subpopulations: Yunnan1, Yunnan2, Guangdong and Guangxi (Table S4, S5). The single Indian accession was assigned to the Yunnan group at K = 2 but did not exceed the assignment threshold for any subpopulation at K = 4 and was therefore treated as admixed. The three Fujian accessions were assigned to the non-Yunnan group at K = 2 and further clustered with the Guangdong subpopulation at K = 4. Population structure of the P. emblica population. A. Neighbor-joining phylogeny and ancestry components of the whole population. The ancestry barplots show K = 2 to K = 5. Accessions with ancestry coefficients ≥ 0.55 were assigned to the corresponding subpopulation, whereas accessions below this threshold were considered admixed. B. PCA plot of the whole population using the first two principal components. Yunnan-mix represents individuals that are assigned to the Yunnan subgroup but the primary ancestral components across the four subpopulations do not exceed the predefined threshold. Different colors and symbols indicate inferred genetic subpopulations. C. PCA plot of the whole population using the second and the third principal components Genetic differentiation in P. emblica population Genetic diversity differed clearly between the Yunnan and non-Yunnan subgroups. The Yunnan group showed higher nucleotide diversity (3.1 × 10⁻⁵) than the non-Yunnan group (2.7 × 10⁻⁵), and genetic differentiation between the two subgroups was moderate (FST = 0.18) (Fig. S5). Among the four subpopulations, Guangxi had the lowest diversity (2.1 × 10⁻⁵), followed by Guangdong (2.6 × 10⁻⁵), Yunnan1 (3.0 × 10⁻⁵), and Yunnan2 (3.1 × 10⁻⁵). The two subpopulations from Yunnan were only weakly differentiated (FST = 0.04), whereas both showed strong differentiation from the subpopulations from non-Yunnan (FST > 0.2). In addition, Guangdong and Guangxi were highly differentiated from each other (FST = 0.25). Genome-wide comparisons between Yunnan and non-Yunnan groups identified 267 divergent genomic regions containing 658 genes using nucleotide-diversity ratios and FST (Fig. 3A and B, Table S6, S7). These genes were enriched in functions related to protein turnover, stress response, hormone transport, and cell growth (Fig. 3C). Pathway analysis indicated that many of these genes were involved in core biological processes, including DNA replication, carbohydrate metabolism, and hormone signaling (Fig. 3D). Further comparison among the four subpopulations revealed 600 divergent regions encompassing 1,489 genes (Table S8, S9). These genes were mainly associated with redox processes, environmental sensing, cell wall–related metabolism, and amino acid modification (Fig. S6). Enriched pathways included DNA replication, MAPK signaling, and carbohydrate metabolism (Fig. S7). Genetic differentiation between Yunnan and non-Yunnan subgroups. A. Genome-wide FST value between Yunnan and non-Yunnan subgroups. B. Genome-wide π ratio between Yunnan and non-Yunnan subgroups. C. GO enrichment analysis of 658 genes in divergent genomic regions. D. KEGG enrichment analysis of 658 genes in divergent genomic regions Cluster analysis of leaf traits in P. emblica population Analysis of five leaf traits across the 112 accessions showed that all traits had skewed distributions (Fig. S8). Strong positive correlations were observed among all leaf characteristics. The strongest relationship was between leaf area and leaf width (R = 0.98, P < 0.01), whereas the weakest was between leaf length and leaf roundness (R = 0.53, P < 0.01). Estimates of broad-sense heritability varied widely, ranging from 0.31 to 0.94, with leaf area showing the highest heritability and leaf roundness the lowest. Significant differentiation in leaf morphology was observed between Yunnan and non-Yunnan subgroups. The non-Yunnan group consistently displayed increased leaf size, characterized by greater leaf area, length, and width, compared with Yunnan individuals (Fig. S9). Among non-Yunnan groups, the Guangdong subgroup exhibited the most pronounced expansion in leaf area and width, exceeding values observed in both Yunnan1 and Yunnan2 subpopulations (Fig. S10). Principal component analysis captured most of the variation in leaf morphology, with the first two components explaining 98.8% of the total variance (Fig. 4A, S11). PC1 alone accounted for 93.5% and was positively associated with all traits, reflecting overall leaf size. Based on phylogenetic relationships and trait patterns, the accessions were classified into three leaf types (Fig. 4B). Accessions from non-Yunnan groups were characterized by large leaves (Juanye9), while Yunnan accessions were divided into medium-leaf (Daguo) and small-leaf types (Yuganzi173) (Fig. 4C). Phenotypic classification of leaf traits in P. emblica population. A. Principal component analysis based on five leaf morphological traits measured in the P. emblica population. B. Hierarchical clustering of the accessions based on phenotypic similarity of the measured leaf traits. C. Representative scanned leaves of three germplasm accessions corresponding to the three leaf-type categories GWAS identified loci associated with leaf morphological traits in P. emblica population Using SNP dosage data, we performed a genome-wide association study to identify genomic regions influencing five leaf traits. The analysis detected a total of 85 significant loci, including 42 with additive effects and 43 with non-additive effects (Table S10). Significant loci were distributed across 25 of the 26 retained reference chromosomes, with no detectable association observed on Chr3B. Chromosome 6 A harbored the maximum count of eight significant loci, while chromosomes 2B, 9B, and 10 A each exhibited only one significant locus. One significant locus, Chr6A_16142276, showed effects on both leaf area and leaf width, indicating that a single genomic region influences multiple aspects of leaf size. Two lead SNPs, Chr7B_3294862 and Chr8A_9202560, overlapped with high-FST regions defined as the top 5% of the empirical FST distribution between the Yunnan and non-Yunnan groups. For leaf area, the additive model identified 13 significant loci, whereas the non-additive model identified 23 loci (Fig. 5A-D). These included eight loci showing dominance of the reference allele and fifteen showing dominance of the alternative allele. At the additive locus Chr10B_7933775, two genotypes were observed. Accessions carrying the GGTT genotype had significantly larger leaves and were found only in non-Yunnan groups, where this genotype accounted for 60% of the Guangdong subgroup (Fig. 5E and F, S12). At the non-additive locus Chr13B_18722001, four genotypes were detected with significant differences in leaf area. The GGGG genotype was associated with the smallest leaves and was common in Yunnan accessions (79.5%). In contrast, the AAAG and AAGG genotypes were linked to the largest leaf area. The AAAG genotype was rare in Yunnan accessions (2.4%) but predominated in the Guangxi subgroup (83.3%), indicating pronounced allele-frequency differentiation associated with leaf-size variation between subgroups (Fig. 5G and H, S13). Genome-wide association study of leaf area in the P. emblica population. A. Manhattan plot of leaf area under the additive model, the red dashed line represents threshold 1e− 6. B. The QQ plot of leaf area under the additive model. C. Manhattan plot of leaf area under the non-additive model, the red dashed line represents threshold 1e− 6. D. The QQ plot of leaf area under the non-additive model. E. Leaf area among different genotypes at Chr10B-7,933,775. Leaf area is shown in cm². Different letters indicate significant differences according to Duncan’s multiple range test at P < 0.05. F. Genotype distribution in two subgroups at Chr10B-7,933,775. G. Leaf area among different genotypes at Chr13B-18,722,001. Different letters indicate significant differences according to Duncan’s multiple range test at P < 0.05. H. Genotype distribution in two subgroups at Chr13B-18,722,001 Candidate genes associated with leaf morphological trait-related loci in P. emblica population A total of 1,446 putative candidate genes were identified within the candidate intervals associated with the five measured quantitative leaf morphological traits (Table S11). Functional annotation indicated that these genes were represented in GO terms related to leaf development and morphogenesis, including leaf development, regulation of leaf development, and leaf morphogenesis, as well as in terms associated with auxin transport, phytohormone responses, and cell wall biosynthesis (Fig. 6A). KEGG analysis further suggested the involvement of plant hormone signal transduction and carbohydrate/cell wall-related metabolic pathways (Fig. 6B). These enriched terms contain a total of 76 candidate genes potentially associated with leaf traits. Category–gene network analysis showed that several genes were shared among multiple functional terms, implying coordinated regulation of leaf morphological traits (Fig. 6C and D). Tissue expression profiles across five organs revealed substantial expression divergence among 76 candidate genes, providing additional support for their potential roles in the regulation of leaf development and leaf shape variation in P. emblica (Fig. 6E). Functional annotation and expression profiles of candidate genes associated with leaf morphology in the P. emblica population. A. GO enrichment of the candidate genes identified within the leaf morphology-associated genomic intervals. B. KEGG enrichment of the candidate genes identified within the leaf morphology-associated genomic intervals. C. GO category–gene network of selected leaf morphology-related terms. D. KEGG category–gene network of selected pathways. E. Expression profiles of the 76 candidate genes across five tissues based on TPM values Discussion As species expand their geographic range, genetic diversity often increases and population structure becomes more pronounced. In rice, this process has been associated with divergence between major subspecies and the formation of geographically differentiated groups [42]. P. emblica is a long-lived woody species distributed throughout tropical and subtropical regions. Previous analyses of P. emblica populations in China revealed two genetically distinct groups, with populations from drier environments showing higher genetic diversity than those from more humid regions [5]. This pattern suggests that environmental conditions and cultivation history have jointly shaped its genetic structure. In this study, 112 P. emblica accessions were genotyped using ddRAD sequencing. The bimodal distribution of polymorphism information content values reflects a mixture of highly variable loci and nearly fixed sites, indicating substantial genetic variation and emerging population structure (Fig. S1). Integrating phylogenetic reconstruction, ancestry inference, and principal component analysis, we identified two major subgroups, Yunnan and non-Yunnan, which could be further divided into four geographically defined subpopulations. These results provide robust evidence that the extant P. emblica germplasm in China has already undergone clear regional differentiation, and they establish an essential population-genetic framework for interpreting subsequent analyses of genomic divergence and leaf-trait variation. Patterns of genetic diversity in cultivated plants are shaped not only by breeding history but also by population size, dispersal routes, and the intensity of human selection. In annual crops and in some intensively improved perennial fruit species, cultivated populations often show reduced diversity relative to their wild relatives, whereas in other perennial crops diversity can be maintained or even increased through recurrent hybridization and long-term clonal selection [43–48]. In the present study, P. emblica displayed a distinct pattern: non-Yunnan group exhibited markedly lower genetic diversity than the Yunnan group. This result suggests that the currently utilized germplasm outside Yunnan likely originated from a relatively narrow genetic base, potentially reflecting founder effects, restricted germplasm exchange, and the still limited extent of systematic breeding in this species. Thus, the genetic structure of P. emblica appears to have been shaped more by dispersal history and incomplete genetic improvement than by intensive modern breeding. Consistent with this pattern, genomic regions identified by the π ratio and FST analyses were enriched for functional categories associated with regulatory and metabolic processes rather than a few isolated loci. GO enrichment highlighted pathways related to RNA processing and RNA binding, ubiquitin-dependent protein turnover, and auxin efflux transmembrane transporter activity, together with categories associated with cell cycle regulation and cell growth. KEGG enrichment further pointed to DNA replication, transcription factor-related functions, starch and sucrose metabolism, glycosyltransferases, protein kinases, diterpenoid biosynthesis, and plant hormone signal transduction. These pathways collectively suggest that population differentiation in P. emblica may involve coordinated changes in growth regulation, developmental timing, carbohydrate metabolism, hormone-mediated signaling, and specialized metabolite biosynthesis. Therefore, the differentiated genomic regions identified here are more appropriately interpreted as candidate regions underlying population divergence and human-mediated utilization, and they provide a functional framework for future validation and molecular improvement in P. emblica. However, these candidate genes remain putative and require further validation through expression profiling, allelic analysis, and functional studies before causal roles can be established. Current germplasm utilization and variety improvement in P. emblica still rely largely on phenotypic selection from existing germplasm resources, with limited application of modern molecular breeding approaches. As a result, variety improvement is slow, and the pool of available cultivars is much smaller than that of other tropical fruit crops. Although hybridization and seedling selection are standard methods in fruit breeding, their efficiency is constrained in perennial species by long juvenile phases, which require considerable time and resources before fruiting. Consequently, contemporary breeding strategies emphasize the development of molecular markers that enable early selection at the seedling stage [49]. Leaf morphology is a key developmental trait with broad effects on plant performance. In Populus deltoides, leaf structure influences chlorophyll content, leaf thickness, and stomatal density, thereby affecting nitrogen use, gas exchange, and photosynthetic efficiency [50–52]. In litchi, economically important traits such as dwarf growth habit, which facilitates high-density and mechanized cultivation, are closely associated with leaf shape [53–55]. Similar relationships between leaf size, photosynthetic capacity, and yield have been documented in major cereal crops, including rice and maize [10–12]. This study represents the first genome-wide association analysis of leaf traits in P. emblica. Because ddRAD-seq is a reduced-representation sequencing approach, marker density and locus-specific sequencing coverage are lower and more uneven than in whole-genome resequencing data. Such features may contribute to genotype uncertainty, missingness, and mild deviations in QQ plots; therefore, the GWAS signals reported here should be interpreted as candidate loci rather than confirmed causal variants. We identified multiple loci associated with leaf morphology, several of which also showed marked allele-frequency differences between Yunnan and non-Yunnan subgroups (Fig. 5F and H). Notably, the candidate genes within these loci were functionally enriched in processes related to leaf development, leaf morphogenesis, shoot system development, auxin transport and hormone responses, and cell wall biogenesis. In particular, candidate genes associated with auxin polar transport, response to auxin, response to brassinosteroid, response to cytokinin, cellulose synthase activity, plant-type primary cell wall biogenesis, and xylan biosynthetic process provide plausible biological links to variation in leaf size and shape. These results suggest that leaf morphology in P. emblica is likely regulated through the combined action of developmental programs, phytohormone signaling, and cell wall remodeling. Collectively, these loci provide a valuable foundation for developing molecular markers and offer practical targets for improving leaf traits and accelerating breeding progress in P. emblica. Conclusion In conclusion, this study used ddRAD-seq to characterize population structure, genetic differentiation, and leaf morphological variation in 112 P. emblica accessions. Population genomic analyses resolved the germplasm into Yunnan and non-Yunnan groups and revealed clear regional genetic differentiation. GWAS identified 85 loci associated with five quantitative leaf morphological traits, and candidate genes within these loci were enriched in pathways related to leaf development, hormone signaling, auxin transport, and cell wall biogenesis. Although these loci and genes remain putative and require further functional validation, the results provide useful genomic resources and candidate targets for molecular marker-assisted breeding and germplasm improvement in P. emblica. Supplementary Information Supplementary Material 1. Supplementary Material 2. Acknowledgements The computations in this paper were run on the bioinformatics computing platform of the National Key Laboratory of Crop Genetic Improvement, Huazhong Agricultural University. Authors’ contributions S. S. and P. W. conceived, designed the research and edited the manuscript. Q. Z. performed the experiments and drafted the manuscript. S. D. and Y. B. performed part of the experiments. J. J., W. H., H. L. and K. Q. contributed valuable suggestions in the manuscript. All authors reviewed the manuscript. Funding This work was supported by funds from the Yunnan Science and Technology Vice President (202405AI350322, 202405AI350367), Yunnan Innovation Guidance and Tech-based Enterprise Cultivation Program (202404BQ040141), Yunnan Rural Revitalization Construction Project of Yuanmou Science and Technology Innovation County (202304BT090025-04) and the Yunnan Academy of Agricultural Sciences Basic Research Project (2024KYZX-16, 2024KYZX-15). Data availability All raw reads generated for the individuals in the study have been deposited in the National Genomics Data Center under BioProject PRJCA054869. Declarations Ethics approval and consent to participate Not applicable. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Material 1. Supplementary Material 2. Data Availability Statement All raw reads generated for the individuals in the study have been deposited in the National Genomics Data Center under BioProject PRJCA054869.