Pangenome analysis of Liriodendron reveals presence/absence variations associated with growth traits

Abstract

Beyond single nucleotide polymorphisms (SNPs), gene presence/absence variation (PAV) plays a crucial role in elucidating species’ genetic diversity, uncovering the genetic basis of key traits, and advancing molecular marker-assisted breeding in plants. In this study, we constructed a pangenome of Liriodendron based on 24 accessions. Comparative analysis with the reference genome revealed 116 Mb of non-reference sequences and obtained 32,773 genes, including 3,558 novel genes. We subsequently employed resequencing data from 247 Liriodendron genotypes to identify PAVs, comprising 13,779 core genes and 18,179 dispensable genes. To further assess PAV applicability, a genome-wide association study (GWAS) was conducted to link gene PAVs with growth traits in hybrid Liriodendron, and identified 14 candidate genes associated with these growth traits above. Additionally, gene PAVs appeared to predominantly contribute to heterosis in growth traits, displaying a dominant expression pattern when comparing leaf, shoot, and phloem tissues of strong and weak heterotic combinations. Additionally, two key candidate genes, Litul.02G164100 and Litul.01G057400, exhibit high parental expression patterns consistent with hybrid vigor in strong heterotic combinations of leaf and shoot tissues. Altogether, this study expands the Liriodendron genomic dataset, identifies candidate genes linked to growth traits, and provides insights into their heterotic mechanisms in hybrid Liriodendron.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12870-025-07109-1.

Introduction

The Liriodendron genus (tulip tree) consists of only two naturally occurring species, Liriodendron chinense and Liriodendron tulipifera, which exhibit significant physiological and morphological differences [1]. Interspecific hybrids of L. chinense × L. tulipifera not only show marked improvements in growth and adaptability—exhibiting exceptionally vigorous vegetative growth, increased biomass accumulation, and enhanced abiotic-stress tolerance compared to either parent [2, 3]—but also benefit from a uniquely tractable experimental system: the genus’s small extant species number, well-curated germplasm collections, and fully sequenced reference genomes [4, 5] facilitate precise genetic and physiological quantification of heterotic traits. Its comparatively rapid juvenile growth, straightforward hand-pollination protocols, and relatively low background genetic variability further streamline both field-based and controlled-environment studies of heterosis [2, 3]. These features collectively render hybrid Liriodendron an ideal platform for dissecting the molecular mechanisms underlying hybrid vigor in tree species.

With recent advancements in sequencing technology, large-scale genomic and transcriptomic data have become increasingly accessible, enabling in-depth analyses of key genetic traits in Liriodendron [6]. Since the release of the L. chinense reference genome, numerous studies have leveraged these resources for genome and transcriptome profiling across Liriodendron spp [6–8]., yielding valuable insights into the genetic basis of important traits and accelerating molecular breeding applications in this genus, providing valuable insights into the genetic basis of important traits and advancing molecular breeding applications for this genus.

However, a single genome provides a limited perspective, focusing on specific gene segments and lacking representation of the full genetic diversity within a population [9]. Resequencing technologies also face inherent limitations in capturing structural variations (SVs) beyond the reference genome. Consequently, constructing a pangenome is a crucial approach for studying SVs. A pangenome encompasses the collective genetic repertoire of all individuals in a population, including not only the reference genome but also the unique genetic information from each individual [9]. Increasingly, studies have shown that the pangenome approach enables a more comprehensive exploration of genetic variation and offers advantages for analyzing specific biological traits. Many pangenomes have been constructed for various horticultural and crop species. Consequently, pangenomes are being developed for numerous species, including tomato [10], grapes [11], Populus [12], soybeans [13], hevea [14], and moso bamboo [15]. For instance, researchers have constructed linear pangenomes and applied PAV-GWAS to uncover candidate genes associated with traits such as fruit flavor, disease resistance, and key agronomic characteristics in horticultural and crop species [9, 10, 16, 17], thereby substantially improving our understanding of the genetic basis underlying these important phenotypes. These studies illustrate the unique advantages of pangenomes for investigating certain biological traits. With declining sequencing costs, large volumes of second-generation sequencing data have become increasingly accessible. In our previous research [3, 6], we used population transcriptome and resequencing data alongside SNP-GWAS, TWAS, and related methods to identify candidate genes associated with leaf morphology and growth traits in Liriodendron. However, GWAS based solely on SNPs cannot capture the full spectrum of genetic variation across the population. In contrast, population analyses based on pangenomes allow for a richer understanding of genetic diversity and reveal phenotype-associated genetic variants. Such insights significantly enhance the interpretability of phenotypic traits in species. In addition to PAV variation, factors such as population structure, genetic heterogeneity, and polygenic effects may also have significant impacts on elucidating the genetic basis of phenotypic variation [18]. Allelic variations, too, play crucial roles in determining species’ phenotypic traits. Therefore, it is essential to comprehensively dissect the genetic architecture of important traits by integrating multiple approaches, such as PAV-GWAS, SNP-GWAS, and TWAS. From this perspective, constructing a pangenome for the Liriodendron genus is both necessary and timely.

Heterosis, or hybrid vigor, is defined by the superior performance of hybrids in yield, quality, and stress tolerance relative to their parents [19]. Despite this, the genomic mechanisms underlying hybrid vigor remain largely unclear, posing challenges for predicting phenotypic trait expression in molecular breeding. Currently, several hypotheses—dominance, over-dominance and epistasis—have been proposed to explain the basis of hybrid vigor, yet none fully account for this phenomenon [20]. Increasing evidence suggests that allele-specific expression (ASE) also plays a significant role in hybrid vigor [21]. Additionally, studies highlight that epigenetic factors are crucial in establishing hybrid vigor in hybrid species [22]. In the case of hybrid Liriodendron, research on hybrid vigor in growth traits has been limited primarily to phenotypic observations and a few molecular marker studies, leaving its underlying molecular mechanisms largely unexplored. Therefore, it is of great significance to explore the mechanism of hybrid vigor in growth traits of hybrid Liriodendron from the level of gene expression.

In our previous study, we utilized 14 parental lines and 233 progenies from 25 hybrid combinations to dissect the genetic basis underlying growth traits, based on dynamic phenotypic data collected over multiple years and SNP analyses. However, the specific role of gene presence/absence variation (gene PAV) in the heterosis of growth trait in these hybrids remains unclear. Therefore, in the present study, we constructed the first linear pangenome of the Liquidambar genus by integrating our previously established hybrid Liriodendron population with Liriodendron accessions from the study by Chen et al. [4], and obtained 116 Mb non-reference sequences and 3,558 novel genes. Based on these data, we constructed a PAV matrix map of each gene among different accessions. In addition, we also performed PAV-based genome-wide association analysis (PAV-GWAS) to identify candidate genes associated with growth traits in hybrid Liriodendron. Additionally, we investigated the potential role of these candidate genes in the development of hybrid vigor and found that most contribute to the hybrid vigor of hybrid Liriodendron through dominant expression patterns at the gene expression level. Finally, Litul.02G164100 and Litul.01G057400 exhibit high parental expression patterns consistent with hybrid vigor. In summary, these studies further enrich the genetic resources of Liriodendron and provide a foundation for gene function research in Liriodendron plants.

Materials and methods

Pangenome construction

Resequencing data for 19 Liriodendron accessions, comprising six L. tulipifera and 13 L. chinense accessions, were sourced from the study by Chen et al. [4]. Additionally, resequencing data for three Liriodendron accessions were obtained from our previous study [3]. In total, Illumina data from 22 Liriodendron accessions and the L. chinense genome (NCBI accession numbers: PRJNA418361 and PRJNA893441) were utilized to construct a linear pangenome of Liriodendron, with L. tulipifera ‘YP108A’ (v1.1) from the Phytozome database (https://phytozome-next.jgi.doe.gov/) serving as the backbone. Considering the overlap between the 20 re-sequenced accessions reported by Chen et al. [4] and those used in our previous studies [3], a total of 22 non-redundant germplasm resources were ultimately selected for the construction of a linear pangenome of Liriodendron. This set included three unique accessions from our previous research [3] that were not present in the study by Chen et al. [4], namely NK (collected from South Carolina, USA), XN (from Xianning City, China), and YY (from Youyang County, China). Quality control of raw data was performed using fastp v0.20.1 [23] to eliminate low-quality reads and splice sequences. The"de novo"strategy was used for linear pangenome construction. First, individual assemblies of the 22 accessions’ genomes were performed using megahit v1.2.9 [24], filtering out contigs shorter than 500 bp. Each assembled contig was then aligned to YP108A with mummer [25], retaining sequences with less than 85% identity as non-reference sequences. Next, non-reference sequences were aligned to the NT database to exclude sequences homologous to microorganisms, animals, and non-plant sources, ensuring no contamination in the assembled sequences. Clean non-reference sequences were then clustered using the cd-hit tool v4.8.1 [26] to remove redundant sequences with a 90% identity threshold. Remaining non-redundant sequences were re-aligned to the reference genome to confirm that no single contig showed strong identity to the reference genome. Further, psvcp v1.01 [27] was employed to identify non-reference sequences in both YP108A and L. chinense (from our previous work [5]) at the chromosome level. Finally, non-redundant, non-reference sequences were merged with the reference genome to generate a comprehensive linear pangenome for Liriodendron.

Repeat sequences within non-reference sequences were annotated using EDTA [28]. Gene structure annotation of these non-reference sequences was performed with BRAKER3 [29], which integrates second-generation transcriptome data and homologous protein sequences to enhance annotation accuracy.

Gene presence/absence variation identification

In order to explore the PAV genes in Liriodendron, the resequencing data of Liriodendron from our previous work [3] were used in this study. In total, Illumina data from 247 accessions—including 14 parental lines and 233 hybrid offspring—were aligned to the linear pangenome. Coding sequence (CDS) coverage was calculated using BWA [30] and mosdepth (v0.3.1) [31]. Genes were considered present if they exhibited > 80% coverage across the gene body; otherwise, they were classified as absent. Gene classification within the pangenome followed the criteria established by Gao et al.: core genes were present in all individuals, softcore genes in at least 99% of individuals, shell genes in 1–99%, and cloud genes in less than 1% of individuals [10].

Confirmation of L. tulipifera species-specific genes

Through PAV gene analysis, we identified L. tulipifera-specific PAV genes. To validate our analysis, we randomly selected four genes specific to L. tulipifera for PCR genotyping. The designed primer sequences are listed in Table S1. The PCR reaction was performed in a 20-µL reaction system that contained 10 µL of 2× Taq Mix, 7 µL of ddH2O, 0.5 µL of each primer (forward and reverse) and 2 µL of DNA.

PAV-GWAS analysis

We performed continuous measurements of growth trait phenotypes, including tree height (H), diameter at breast height (DBH), clear bole height (CBH), and crown length ratio (CLR), in the hybrid Liriodendron population over several years. For this study, PAV-GWAS analysis was conducted using phenotype data collected over the past three years. Detailed information on the experimental population and measurement of phenotypic traits have been described in previous studies [3, 32]. The analysis was carried out with the GAPIT R package [33] using the BLINK model, with the PAV gene matrix as genotype data. The significance threshold was set at 1e-4 to 3.05e-5 (1/32773).

Gene expression analysis and functional annotation

We used the featureCounts program [34] to calculate the gene read count matrix, with transcripts per million (TPM) values representing gene expression levels. First, all transcriptome data from L. chinense and L. tulipifera were aligned to the pangenome, after which featureCounts was applied to quantify gene expression. Differentially expressed genes (DEGs) were identified using the DESeq2 program [35]. Protein-coding genes within the pangenome were annotated using EggNOG-mapper v2.1.8 [36]. For enrichment analysis, gene ontology (GO) terms were explored using the clusterProfiler package in R [37].

Results

Construction of Liriodendron linear pangenome

A total of 24 accessions were used to construct the Liriodendron pangenome: 22 were from second-generation resequencing data, one was the previously released chromosome-level genome of L. chinense, and L. tulipifera was used as the reference genome. Sequencing data and depth information for the 22 accessions are presented in Table S2. De novo assembly of these 22 accessions yielded a total of 23.55 Gb of contigs, with an average assembly size of 1.12 Gb and an average contig N50 of 4,125 bp (Table S3). After decontamination and redundancy removal, 27.9 Mb of non-reference sequences were obtained, comprising 23,121 contigs. Additionally, alignment of L. chinense to the L. tulipifera genome identified 88 Mb of non-reference sequences and 439 non-reference genes. By merging the reference genome with these non-reference sequences, a combined genome of 1.61 Gb was constructed, including 116 Mb of non-reference sequence space in Liriodendron. Gene structure prediction on the non-reference sequences identified 3,119 protein-coding genes with an average length of 895 bp, which is notably shorter than the average gene length in the reference genome (12,382 bp), suggesting that the second-generation assemblies were more fragmented, resulting in partial or fragmented genes. In summary, this study assembled a Liriodendron pangenome totaling approximately 1.61 Gb and encompassing 32,773 genes (Table 1).

Table 1.

Summary statistics of linear pangenome features in Liriodendron

Genomic feature	De novo	L. tulipifera ‘YP108A’	Final pangenome size
Total contigs size	23.55 Gb contigs	1.50 Gb	1.61 Gb; including 13,779 core genes and 18,179 dispensable genes.
Average genome size of each accession	1.12 Gb	-
Average contig N50	4,125	-
De-duplication	27.9 Mb	88 Mb
Novel genes	3,119	439

PAV analysis

To assess the distribution of annotated genes across different accessions, we employed a ‘map-to-pan’ strategy to identify PAV genes. The read coverage across the gene regions reached 80% and CDS region covered from read depth exceeding two-fold were classified as ‘present’ in an accession; otherwise, they were classified as ‘absent’. The results indicate that, similar to some previous research findings, the constructed Liriodendron pangenome appears to be closed-loop, with most genes included when the random sample size reaches 50 (Fig. 1). Based on gene presence frequencies across accessions, genes were categorized into core genes (13,779, 43.08%), present in all accessions, and dispensable genes (18,179, 56.84%). Dispensable genes were further subdivided into 5,354 softcore genes (present in more than 99% of accessions), 12,724 shell genes (present in 1–99% of accessions), and 101 cloud genes (present in less than 1% of accessions) (Fig. 2). Additionally, the dispensable genes—including softcore, shell, and cloud genes—were all considered PAVs. The proportion of core genes aligns with findings from previous pangenome studies in soybean [38], tomato [10] and pigeon pea [39].

Fig. 1.

Simulation of the increase of the pangenome size and the decrease of the core genome size by iterative randomly sampling accessions in Liriodendron linear pangenome

Fig. 2.

Genome-wide PAVs of Liriodendron. (A) Composition of Liriodendron linear pangenome. (B) The heatmap of variable genes in the linear pangenome; “1” indicates that the gene in the linear pangenome is present in this genotype from hybrid Liriodendron, while “0” signifies its absence

Expression analysis of PAV genes

Additionally, we performed RNA-seq analysis based on previous studies [5, 40] across various tissues to examine gene expression in the Liriodendron pangenome. Results revealed that 21,018 genes in the reference genome and 96 genes in the non-reference genome (Table S4) showed expression levels above 1 TPM in at least one of seven tissues: bracts (BR), leaves (LE), petals (PE), pistils (PI), shoot apices (SA), sepals (SE), and stamens (ST). We also observed that non-reference genes generally exhibited significantly lower expression levels than reference genome genes. To further investigate species-specific PAV genes in L. chinense and L. tulipifera populations and explore their potential connection to unique biological characteristics, we identified a total of 999 L. chinense population-specific genes (Lchi-PAV) (Table S5) and 916 L. tulipifera population-specific genes (Ltu-PAV) (Table S6). Among these, the L. chinense population includes 731 non-reference genes, while the L. tulipifera population has only 11, suggesting that the majority of Lchi-PAV genes derive from non-reference genome assemblies. This finding further highlights the high accuracy of the pangenome and PAV variation map constructed in this study. Furthermore, we found that only 239 and 381 genes within Lchi-PAV and Ltu-PAV, respectively, had expression levels above 1 TPM (Table S5; Table S6). In addition, we randomly selected some Ltu-PAV genes for PCR amplification to verify the accuracy of Ltu-PAV genes identification in L. tulipifera and L. chinense accessions (Fig. 3). This suggests that many of these species-specific genes may not be highly expressed under most conditions but may play crucial roles in certain unique biological functions specific to each species.

Fig. 3.

PCR amplification verification of Ltu-PAV genes. The lengths of the marker from top to bottom are 2000 bp, 1000 bp, 750 bp, 500 bp, 250 bp, and 100 bp, respectively

PAV-GWAS analysis of growth traits in hybrid Liriodendron population

In this study, the PAV gene variation map was used as genotype data, while growth trait measurements—including H, DBH, CBH, and CLR—were used as phenotype data for PAV-GWAS analysis. To enhance candidate gene localization accuracy, GWAS was conducted using growth trait data from hybrid Liriodendron offspring collected over three consecutive years. Results identified 4, 8, 2, and 2 significant loci associated with DBH, H, CLR, and CBH, respectively (Table S7). Notably, three significant genes for DBH were consistently identified across all three years of GWAS analysis, and two candidate loci were co-located in two consecutive years. For tree height, a total of eight significant candidate genes were identified, though these loci were derived solely from the 2023 GWAS analysis (Fig. 4). In contrast, the 2021 and 2022 GWAS analyses yielded no significant loci associated with tree height, likely due to challenges in obtaining stable and reliable phenotypic measurements for this trait. In CBH and RCL, two significant candidate genes were consistently co-located, which greatly enhances the precision of gene localization.

Fig. 4.

Genomic loci associated with Tree Height, Diameter at Breast Height, Clear Bole Height, and Crown Length Ratio traits based on PAV genes. In the Manhattan plot, the solid black line represents the GWAS significance threshold -log₁₀(1/32,773) and the dashed line represents a self-defined minimum threshold -log₁₀(1e-4)

To further investigate key candidate genes significantly associated with growth traits, we focused on notable candidate genes highlighted in the significance analysis. For DBH, the candidate gene Litul.08G029000, which encodes a zinc finger BED domain-containing protein (WRKY 117), was associated with DBH-related signals over two consecutive years. This gene, which is present only in L. tulipifera and absent in L. chinense, may be involved in DBH growth. Interestingly, the DBH of L. tulipifera was significantly higher than that of L. chinense, aligning with L. tulipifera’s status as a fast-growing, highly adaptable pioneer species in North America. For H, eight significant loci were identified. Among them, Litul.02G164100 encodes an auxin-responsive protein, directly linked to growth and development processes, and is present in all Liriodendron genotypes (Fig. 3), indicating its involvement throughout hybrid Liriodendron’s growth. In RCL and CBH, two significant loci, Litul.14G076500 and Litul.14G123900, were consistently co-located. These genes encode a putative transcription factor/chromatin remodeling BED-type (Zn) family protein and an F-box protein with interaction domains, respectively (Table S7). Further, Litul.14G076500 was absent only in the MSL accession of Liriodendron, while Litul.14G121900 was present only in the BK accession (Table S6). These apparently species-specific genes may be linked to unique biological traits, potentially playing crucial roles in the growth and development of Liriodendron.

Mining candidate genes related to growth traits in hybrid Liriodendron

To further elucidate the role of candidate genes in the growth and development of hybrid Liriodendron, we examined their expression levels to assess potential links to growth. Based on previous research, we selected one strong and one weak heterotic combination from the Liriodendron population for heterosis analysis. Leaf, shoot, and phloem tissues were collected from each combination and their parent trees for transcriptome sequencing. To explore tissue-specific expression of PAV genes in different hybrid combinations and their relationship with growth vigor, we conducted differential expression analysis (DEG) on offspring from both strong and weak heterotic combinations across different tissues. In leaf tissue, we identified 1,030 DEGs in both combinations, with GO enrichment predominantly in “cell cycle process”, “nuclear division”, “DNA replication” and “mitotic cell cycle process”, suggesting a strong link between these DEGs and growth processes. In shoot tissue, 431 DEGs were identified in both combinations, with GO enrichment in “secondary metabolic process”, “farnesyl diphosphate metabolic process” and “regulation of signaling receptor activity”. In phloem tissue, we found 893 DEGs between the strong and weak heterotic combinations, with GO terms primarily enriched in “plant-type secondary cell wall biogenesis”, “photosynthesis, light harvesting”, “meristem maintenance” and “lignin biosynthetic process” (Fig. 5C and D; Table S8). Interestingly, the number of DEGs in shoot tissue was significantly lower than in leaf and phloem tissues, suggesting that PAV genes associated with growth vigor may play a particularly prominent role in the growth and development of leaf and phloem tissues, thereby contributing to the growth vigor observed in hybrid Liriodendron.

Fig. 5.

DEGs and PAV-GWAS candidate genes on the expression levels of offspring from dominant and non-dominant combination in different tissues. (A-B) Expression heatmap of genes related to hybrid vigor in growth traits; Note: HP, advantage male parent; LP, non-advantage male parent; MP, female parent; HF, advantage combination offspring; LF, non-advantage combination offspring. (C) The top 10 items with the most significant GO enrichment analysis results for DEGs between strong and weak heterotic combinations in leaf tissue. (D) The top 10 items with the most significant GO enrichment analysis results for DEGs between strong and weak heterotic combinations in phloem tissue

Our previous research [2, 3] suggests that non-additive expression patterns play a key role in the growth of hybrid Liriodendron. To further investigate the dominant and over-dominant expression of PAV genes in leaf, shoot, and phloem tissues, we classified these expression levels into four dominant patterns and six over-dominant patterns. Dominant expression patterns included high-parent expression (III and IV; Hybrid expression levels biased toward the more highly expressed parent) and low-parent expression (V and VI; Hybrid expression levels biased toward the less highly expressed parent). Over-dominant expression patterns included expressions above the low-parent (VII, VIII and IX; Hybrid expression levels significantly lower than those of both parents) and above the high-parent (X, XI and XII; Hybrid expression levels significantly higher than those of both parents). In leaf tissue, dominant expression genes in weak heterotic combinations were notably higher than in strong heterotic combinations, whereas no significant differences were observed in the other two tissues (Table S9). This suggests that, in leaf tissue, PAV genes linked to growth vigor may negatively regulate growth-related genes through increased dominant expression in weak heterotic combinations, allowing strong heterotic combinations to exhibit greater heterosis in growth traits. Additionally, leaf tissue exhibited significantly more over-dominant expression genes compared to shoot and phloem tissues. This finding further indicates that growth vigor may be closely linked to the photosynthetic processes in leaf tissues, where over-dominant expression may play an essential role in growth and development (Table S9).

To investigate candidate genes associated with growth vigor and their roles in heterosis via gene expression changes, we analyzed expression patterns of GWAS-identified candidate genes. Notably, Litul.02G164100 and Litul.01G057400 (encode terpene synthase) exhibited high-parent dominant expression in the strong heterotic combinations of leaf and shoot tissues, respectively, thereby contributing to growth vigor. Furthermore, Litul.01G057400 displayed differential expression levels in dominant versus non-dominant shoot tissue combinations, suggesting that these genes may play critical roles in growth vigor of hybrid Liriodendron (Fig. 5A and B). These findings underscore the significance of non-additive expression patterns of PAV genes, especially in dominant gene expression, in the development of growth vigor in hybrid Liriodendron.

Discussion

The identification of genetic variations within genomes has become a major focus in genomics research, given its importance in discovering functional genes for key traits and understanding biodiversity. Currently, various methods are available for constructing pangenomes, such as “de novo,” “map-to-pan,” and graph pangenomes [41]. Among these, linear pan-genomics approaches, specifically the first two, are widely used for their advantages in handling large sample sizes at low cost with easy accessibility. The “map-to-pan” method involves aligning original second-generation reads with a reference genome to capture non-reference reads, followed by de novo assembly to obtain non-reference sequences [41]. This approach is particularly suited for constructing pangenomes in large populations due to its efficiency and simplicity. In this study, we used a de novo method to construct a linear pangenome for Liriodendron. Each sample underwent de novo assembly individually, and resulting contigs were aligned to the reference genome to identify non-reference sequences. This process yielded a total of 116 Mb of non-reference sequences and identified 3,558 full-length genes that were either incompletely assembled in the reference genome, likely due to limitations inherent in assembling second-generation sequencing data, or were truncated in the lineage of the L. tulipifera reference tree. Genes annotated from these fragmented sequences assembled using second-generation data tend to be incomplete, which is reflected by the higher number of gene annotations found on shorter sequence lengths. Such incomplete gene models may impact subsequent gene functional studies. Unfortunately, this inherent limitation is difficult to overcome in the linear pangenome constructed in the current study. In future work, we plan to address these issues by obtaining high-quality genome maps from multiple Liriodendron species using PacBio HiFi and Hi-C technologies, or even by constructing a telomere-to-telomere (T2T) genome with Oxford Nanopore sequencing to build a graphical pangenome. These approaches would greatly resolve the aforementioned issues and enhance the reliability of downstream analyses. To analyze the distribution of all genes within the Liriodendron pangenome and to identify PAV information, we constructed a PAV matrix across 233 hybrid Liriodendron populations. Among the identified genes, 43.08% were classified as core genes, while 56.92% were non-core genes (Here, the majority refer to PAVs). This pangenome provides a valuable genetic resource for subsequent research into gene function and diversity in Liriodendron.

Based on the linear pangenome constructed in this study and prior transcriptome data, we observed that PAV genes expressed in certain tissue-specific contexts generally exhibit low expression levels. This may stem from incomplete assembly or annotation of the Liriodendron reference genomes rather than true gene absence. Although our tissue-specific RNA-seq analyses revealed that PAV-assigned genes tend to have very low expression levels, both the L. chinense annotation [4] and the re-annotation of L. chinense [5] were based largely on short-read mRNA sequencing and Iso-Seq from a limited set of tissues. Such approaches, while superior to ab initio predictions, can still miss low-abundance transcripts and collapse duplicated loci, leading to false PAV calls. To enhance future pangenome studies, reference assemblies should incorporate deep, multi-tissue transcriptome characterization and annotation pipelines optimized for lowly expressed and paralogous genes. In addition, assembly quality assessment must evolve beyond single-copy gene metrics to include benchmarks that capture the presence and integrity of dispensable gene families essential for understanding intraspecific diversification.

Moreover, no structural-variant analyses were conducted in this study from the perspective of growth traits heterosis. We acknowledge that identification of structural variants (SVs), a standard and illuminating component of pangenome analyses, was not undertaken in the present work. Although we noted earlier that SV discovery is commonplace in pangenome studies, future research should incorporate comprehensive SV detection to refine PAV assignments and to uncover potential false positives. In particular, mapping SVs across multiple assemblies will help determine whether ‘softcore’ dispensable genes are truly absent in some accessions or simply excluded from certain assemblies or annotations due to structural rearrangements. Integrating SV data thus promises to distinguish genuine gene absence from technical artifacts and to resolve the remaining question of whether some softcore genes are misclassified as dispensable.

Although PAV-GWAS partially compensates for the shortcomings of SNP-GWAS, the phenotypic variance explained by either type of genetic variant is affected by the quality of the genome assembly. In contrast, k-mer-based GWAS analysis has a distinct advantage in handling missing genomic fragments, thereby enhancing the accuracy of gene mapping [42]. Consequently, comprehensively dissecting the genetic basis of important traits remains a significant challenge for researchers, particularly when presence–absence variation (PAV) can underlie large-effect phenotypes that SNP-based analyses may miss. In this study, we conducted a PAV-GWAS and identified robust candidate genes associated with H, DBH, CLR and CBH. By capturing gene–trait associations driven by PAV-GWAS complements traditional SNP-GWAS and uncovers additional loci that contribute to heterosis and adaptive variation. These PAV-derived candidates not only validate and extend SNP-based findings but also enrich the pool of genetic resources available for molecular breeding of Liriodendron, enabling targeted improvement of growth and form traits in this genus.

This study used a pangenome approach to identify candidate genes related to the growth traits of hybrid Liriodendron, further research is needed to clarify the roles of these genes in the development of growth vigor. Previous studies have shown that non-additive effects play a crucial role in hybrid vigor. Here, 14 candidate genes were identified through PAV-GWAS, yet it remains to be determined if these genes contribute to Liriodendron’s hybrid vigor by influencing gene expression through PAV mutations. In prior research, we developed dominant and non-dominant hybrid Liriodendron combinations based on growth performance and performed transcriptome sequencing of various tissues to analyze the mechanisms of hybrid vigor at the gene expression level. Building on this, we analyzed the expression patterns of all PAV genes in leaf, shoot, and phloem tissues, finding that non-additive expression patterns—primarily dominant expression—were prevalent in PAV genes. Notably, two candidate genes for growth traits exhibited dominant expression in the strong heterotic combinations of leaf and shoot, suggesting a possible role in hybrid vigor. Among them, Litul.02G164100, which encodes auxin-responsive proteins directly involved in growth, may be critical for hybrid vigor in Liriodendron. Another gene encodes terpene synthase, which might also contribute significantly to growth vigor in hybrid Liriodendron. These key genes warrant further functional validation in future studies. While this study identified growth-related candidate genes through PAV-GWAS and explored their roles in hybrid vigor by examining gene expression, dominant expression patterns were found to be the primary mode. However, numerous studies suggest that allele-specific expression (ASE), influenced by methylation and transposable elements (TEs), plays a significant role in hybrid vigor [22, 43]. Due to the lack of allelic information for the hybrid Liriodendron genome in this study, it was challenging to explore hybrid vigor mechanisms from the perspective of ASE. Therefore, future studies could benefit from constructing a haplotype-resolved genome for hybrid Liriodendron to investigate hybrid vigor mechanisms from the perspectives of ASE and single-parent expression (SPE), along with the effects of TEs, SVs, and methylation on ASE and SPE. This would offer a more comprehensive understanding of the intrinsic mechanisms of growth traits hybrid vigor in Liriodendron. In summary, this study constructed a linear pangenome for Liriodendron, identifying candidate genes associated with growth traits in hybrid Liriodendron and exploring their potential roles in hybrid vigor. This work provides valuable genetic resources for Liriodendron gene function studies and serves as a reference for investigating hybrid vigor in other forest tree species.

Conclusion

In this study, we constructed the first linear pangenome for Liriodendron using the high-quality L. tulipifera reference genome from the Phytozome database and resequencing data from 247 genotypes. Moreover, we also identified two key candidate genes encoding an auxin-related protein and a terpene synthase, respectively, which may contribute to hybrid vigor in growth through dominant expression patterns. In summary, this study provides a foundation for further exploration of the molecular mechanisms underlying growth traits in hybrid Liriodendron. It also illustrates the utility of PAVs in identifying candidate genes and unraveling the regulatory network mechanisms governing growth and development in hybrid Liriodendron.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Hainan Wu completed the data analysis and wrote the paper. Siqi Chen, Jing Wang, Yaxian Zong, Lichun Yang aided the data analysis and experimental work. Chunfa Tong guided the data analysis. Huogen Li conceived this project and revised the paper. All authors contributed to the paper and approved the submitted version.

Funding

This study was supported by funds from the National Natural Science Foundation of China (32371910), National Key Research and Development Program (2022YFD2200104) and the Priority Academic Program Development of Jiangsu Higher Education Institutions (PAPD).

Data availability

We have uploaded the linear pan-genome sequence (fasta format), the corresponding gene annotation files (gff3 format), as well as protein sequences to the Figshare repository (https://doi.org/10.6084/m9.figshare.28740278). The whole‑genome resequencing data used for the PAV‑GWAS in this study are available from NCBI under accession number PRJNA893441. Additionally, all analysis pipeline used in this manuscript are available on the Figshare (https://doi.org/10.6084/m9.figshare.29484641).We appreciate early access to the Liriodendron tulipifera genome assembly and annotation generated by the U.S. Department of Energy Joint Genome Institute (https://ror.org/04xm1d337; proposal:10.46936/10.25585/60001405), a DOE Office of Science User Facility, supported by the Office of Science of the U.S. Department of Energy operated under Contract No. DE-AC02-05CH11231.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.