Cunninghamia lanceolata genome illuminates the evolutionary dynamics of gymnosperms Authors: Si-Zu Lin, Yu Chen, Chao Wu, Wei-Hong Sun, Zhen Li, Heng-Chi Chen, Jie-Yu Wang, Chang-Mian Ji, Shu-Bin Li, Zhi-Wen Wang, Wen-Chieh Tsai, Xiang-Qing Ma, Si-Ren Lan, Fei-Ping Zhang, Ya-Cong Xie, Lei Yao, Yan Zhang, Meng-Meng Lü, Jia-Jun Zhang, Di-Yang Zhang, Yi-Quan Ye, Xia Yu, Shan-Shan Xu, Zhi-Hui Ma, Guo-Chang Ding, Guang-Qiu Cao, Zong-Ming He, Peng-Fei Wu, Kai-Min Lin, Ai-Qin Liu, Yan-Qing Lin, Shao-Ning Ruan, Bao Liu, Shi-Jiang Cao, Li-Li Zhou, Ming Li, Peng Shuai, Xiao-Long Hou, Yi-Han Wu, Nuo Li Published: 2026 ABSTRACT SummaryCupressaceae , a gymnosperm family, draws attention due to its controversial phylogenetic position. Here, we present a comprehensive genome analysis of Chinese fir (Cunninghamia lanceolata), a Cupressaceae species, to enhance our understanding of gymnosperm evolution. The 11.24 Gb assembled genome, shaped by inefficient long terminal repeat removal, offers insights into its phylogenetic position. Phylogenetic analysis refines gymnosperm relationships between Cycads-Ginkgo and their relation to Gnetales-Pinaceae. Whole-genome duplication (WGD) analysis reveals no evidence for an ancient polyploidization event in the lineage of C. lanceolata, and confirms a seed-plant-shared WGD event. We also explore genomic evidence to explain the population history and adaptability of C. lanceolata, including potential glacial refugia, dispersal centers, and unique sterility. Furthermore, the refined (A)B(C) model for reproductive organ development in C. lanceolata has broader applications across gymnosperms. This study provides a valuable genome sequence and contributes to the understanding of gymnosperm evolution. Introduction Gymnosperms, encompassing extant taxa, such as cycads, Ginkgo, conifers, and gnetophytes1,2,3,4 first emerged during the Late Devonian period, approximately 360 million years ago (Mya).5 Despite their ancient origins, gymnosperms today comprise slightly over 1000 species, significantly fewer than the 354 000 angiosperm species.6 However, fossil records reveal that gymnosperms were once far more diverse, highlighting their historical radiation and extinction, and also indicating that the living gymnosperms represent only a fraction of their past diversity.7,8 This realization presents challenges in accurately reconstructing the phylogenetic relationships of gymnosperms, which is crucial for unraveling the fascinating scientific inquiry into the origin of seed plants. In the traditional plant systematics, gymnosperm classifications are based on morphological characters, such as leaf shape, bract and scale arrangement. Within this framework, the genus Cunninghamia has been classified into the family Taxodiaceae, primarily based on oligo-specific or mono-specific evidence.9,10 The family Taxodiaceae further belongs to the Pinopsida group, alongside Pinaceae and Cupressaceae. Recent studies, however, have classified the genus Cunninghamia into the subfamily Cunninghaminae, which belongs to a well-recognized monophyletic group, i.e., the expanded Cupressaceae, alongside the other six subfamilies, Taiwanioideae, Athrotaxidoideae, Sequoioideae, Taxodiaceae, Callitroideae, and Cupressoideae.1 This consolidation is supported by evidence from various sources, including fossil records, morphological anatomy, chemical structure, and molecular evidence, such as gene fragments or organelle genomes.2,9,10,11,12,13 That said, although taxonomists have classified gymnosperms as Cyadopsida, Ginkgoopsida, and Pinopsida, with Pinopsida including Cupressidae, Pinidae and Gnetidae, the phylogenetic relationships among extant gymnosperms remain elusive.1 Indeed, various hypotheses have been proposed regarding the phylogenetic relationships within gymnosperms, particularly concerning the placements of gnetophytes among the so-called Pinopsida group. For instance, the “Gnepine” hypothesis suggests that gnetophytes are sister to Pinaceae,14,15 whereas the “Gnecup” hypothesis groups gnetophytes with non-Pinaceae conifers (Cupressophytes)16 and the “Gnetifer” hypothesis places gnetophytes as sister to all conifers in the Pinopsida group.17 Additionally, gnetophytes have also been placed as a sister group to all other gymnosperms.18,19 To more accurately infer the phylogenetic relationships among extant gymnosperms, obtaining more complete genomic information is a reliable approach. Such comprehensive genetic data can provide a holistic perspective, enabling scientifically grounded inferences.20,21,22 Until recently, the genomes of several gymnosperm lineages, including conifers, such as Pinaceae: Picea abies,23 Picea glauca,24 Pinus taeda,25,26 Pinus tabuliformis,27 Pinus densiflora,28 Larix kaempferi,29 Taxaceae: Taxus chinensis (a synonym of Taxus wallichiana),30 Ginkgo: Ginkgo biloba,31,32 cycads: Cycas panzhihuaensis,2 gnetophytes:Gnetum montanum33 and Welwitschia mirabilis,34 and Cupressaceae: Metasequoia glyptostroboides,35 Fokienia hodginsii36 have been successfully sequenced. Among the extant Cupressaceae species, the genus Cunninghamia, comprising only two species C. lanceolata and C. konishii, is the sister group to all the other genera in Cupressaceae. This makes Cunninghamia an especially important representative for genomic sequencing, both to improve our understanding of Cupressaceae evolution and to provide a foundational reference for comparative studies across gymnosperms. The family Cupressaceae stands out as the sole gymnosperm lineage with widespread distribution across both the northern and southern hemispheres. C. lanceolata, commonly referred to as “Sha Mu” in China, predominantly thrives in southern China, with additional occurrences noted in northern Vietnam.3 Historical records trace the artificial cultivation of C. lanceolata back over two thousand years, as documented in ancient Chinese texts, such as the “Erya.” Today, owing to its rapid growth, high-quality wood, and resistance to decay, C. lanceolata is the most widely planted timber species in China. Its plantation area exceeds 9.90 million hm2, accounting for approximately one-quarter of the total artificial forest area in China, with a forest stock of 755 million m3, i.e., one-fourth of all man-made forest stock. Given its economic and ecological importance, improving the genetic understanding of C. lanceolata is a key objective for breeders and foresters to enhance timber production and carbon sequestration capabilities. Recently, Shirasawa et al. published a genome assembly of C. lanceolata of Japan that only includes contig-level assembly without annotation. The data were generated employing high-fidelity (HiFi) long-read sequencing technology, which yielded 2,472 assembled contigs spanning a genome size of 12.04 Gb with an N50 value of 11.7 Mb.37 In this study, we present a high-quality chromosomal-level genome assembly of C. lanceolata, which was scaffolded using high-throughput chromosome conformation capture (Hi-C) data. To facilitate genomic comparison, we annotated the C. lanceolata genome assembly of Shirasawa et al. SNP calling was performed using the C. lanceolata genome assembly of Shirasawa et al. and our population resequencing data of C. lanceolata to infer the evolutionary relationships among its populations. Our results provide a valuable resource for exploring the evolutionary history of Cupressaceae and offer a genomic foundation for significant advancements in forestry, breeding, and environmental applications such as carbon sequestration. Results Genome sequencing, assembly, and annotation To construct a chromosome-level genome assembly of C. lanceolata, genomic DNA was extracted from an adult individual. The K-mer analysis indicated that the genome size of C. lanceolata was approximately 10.42 Gb, with a high level of heterozygosity (0.69%) (Figure S1). A de novo assembly of the C. lanceolata genome was performed using PacBio reads (1,113.16 Gb in total) along with Illumina whole-genome sequencing reads (516.42 Gb in total) (Tables S1 and S2), resulting in an assembled genome of 11.24 Gb with a contig N50 value of 2.16 Mb (Table 1; Table S3). C. lanceolata is a diploid plant containing 11 chromosomes (2n = 2× = 22).38 To further refine the assembly, we employed the high-throughput/resolution chromosome conformation capture (Hi-C) technology, assigning 10.89 Gb scaffolds to the 11 C. lanceolata chromosomes (Table 1; Figure S2; Table S4). The lengths of these 11 chromosomes ranged from 0.64 to 1.55 Gb, with a scaffold N50 value of 927.89 Mb (Figure 1A, Table 1; Tables S4 and S5). To verify the completeness of our genome assembly, we mapped all the Illumina reads to the chromosome-scale assembly and obtained a mapping rate of 92.83% (Table S6). A total of 37,225 protein-coding genes were predicted through integrating ab initio predicted protein-coding genes (Table 1), homologous protein alignments, and transcriptome RNA-Seq data integration. Among these, 34,559 (92.84%) protein-coding genes could be functionally annotated (Tables S7 and S8). Additionally, we identified 50 microRNAs (miRNAs), 3,955 transfer RNAs (tRNAs), and 2,930 ribosomal RNAs (rRNAs) (Table S9). The Benchmarking Universal Single-Copy Orthologs (BUSCO, v5.8.2) assessment based on gymnosperm_0db 10 lineage dataset39 revealed that the completeness of genes was 93.64%, and 89.60% in the genome and protein model, respectively (Table 1). These complete BUSCO values were comparable to those reported recently for F. hodginsii,36 indicating a high quality of the C. lanceolata genome.Table 1Assembly and annotation statistics of the genome of C. lanceolataSpeciesC. lanceolataAssessment of genome size (Gb)10.42Contig N50 (bp)2,155,103Total length of Contig (bp)11,242,038,337Scaffold N50 (bp)927,886,105Total length of Scaffold (bp)11,244.041,337Total length of sequences anchored to chromosomes (bp)10,893,531,473Assembled genome BUSCO (%)93.64Number of protein-coding genes37,225Annotated protein BUSCO (%)89.60Total size of TEs (bp)10,377,888,279TE in genome (%)92.31Figure 1Genomic evolutionary history of Cunninghamia lanceolata(A) An overview picture of the C. lanceolata.(B). Genomic landscape of the 11 assembled pseudochromosomes.Track i represents the length of the pseudochromosomes (Mb); tracks ii-iv represent repeat element density, gene density distribution, and GC content, respectively. These metrics were calculated within 300 Mb windows.(C). Phylogenetic tree, gene family expansion and contraction, and divergence time for 19 plant species.The phylogenetic tree was constructed using single-copy genes. The most recent common ancestor (MRCA) was estimated to possess 9205 gene families.Green numbers indicate expanded gene families, while purple numbers denote contracted gene families.Gymnosperms are shaded in rose red, while angiosperms are shaded in purple. The dotted line marks the divergence between angiosperms and gymnosperms.(D). LTR insertion time and frequency of C. lanceolata, Ginkgo biloba, Gnetum montanum, Picea abies, and Pinus taeda. LTR insertion frequencies in C. lanceolata exhibit two distinct peaks, one at approximately 3–7 Mya and another at approximately 14–18 Mya.Only one peak is evident in the other gymnosperms. Table 1Assembly and annotation statistics of the genome of C. lanceolataSpeciesC. lanceolataAssessment of genome size (Gb)10.42Contig N50 (bp)2,155,103Total length of Contig (bp)11,242,038,337Scaffold N50 (bp)927,886,105Total length of Scaffold (bp)11,244.041,337Total length of sequences anchored to chromosomes (bp)10,893,531,473Assembled genome BUSCO (%)93.64Number of protein-coding genes37,225Annotated protein BUSCO (%)89.60Total size of TEs (bp)10,377,888,279TE in genome (%)92.31 Assembly and annotation statistics of the genome of C. lanceolata Genomic evolutionary history of Cunninghamia lanceolata (A) An overview picture of the C. lanceolata. (B). Genomic landscape of the 11 assembled pseudochromosomes. Track i represents the length of the pseudochromosomes (Mb); tracks ii-iv represent repeat element density, gene density distribution, and GC content, respectively. These metrics were calculated within 300 Mb windows. (C). Phylogenetic tree, gene family expansion and contraction, and divergence time for 19 plant species. The phylogenetic tree was constructed using single-copy genes. The most recent common ancestor (MRCA) was estimated to possess 9205 gene families. Green numbers indicate expanded gene families, while purple numbers denote contracted gene families. Gymnosperms are shaded in rose red, while angiosperms are shaded in purple. The dotted line marks the divergence between angiosperms and gymnosperms. (D). LTR insertion time and frequency of C. lanceolata, Ginkgo biloba, Gnetum montanum, Picea abies, and Pinus taeda. LTR insertion frequencies in C. lanceolata exhibit two distinct peaks, one at approximately 3–7 Mya and another at approximately 14–18 Mya. Only one peak is evident in the other gymnosperms. Upon comparing the genomes of 19 plant species, we observed 1,860 expanded gene families within the C. lanceolata genome, with 120 of these families showing significant expansion (p < 0.01) (Figure 1B; Table S10; Figure S3). The Kyoto Encyclopedia of Genes and Genomes (KEGG) and Gene Ontology (GO) enrichment analyses revealed that the significantly expanded gene families were especially enriched in KEGG pathways related to the biosynthesis of secondary metabolites, such as monoterpenoid (map00902) and betalain (map00965) (Table S11), as well as in GO terms associated with lignin biosynthetic/metabolic process (GO:0009809/GO:0009808) and terpene synthase activity (GO:0010333) (Table S12). Lignin biosynthesis involves the hydrolysis of its precursor glucoside by beta-glucosidase, followed by catalysis by the laccase and peroxidase systems. Interestingly, beta-glucosidase (OG0000016) and laccase (OG0000039) exhibited significant expansion, with 58% (28/48) and 51% (20/39) of their copies being tandem repeats, respectively (Tables S13, S14, and S15). Compared with the genome assembly reported by Shirasawa et al. (12.04 Gb with a contig N50 of 11.70 Mb),37 our assembly is of comparable size (11.24 Gb with a contig N50 of 2.16 Mb). However, our genome assembly achieves a higher BUSCO completeness score of 93.64%, versus 89.10% for Shirasawa et al. (Table S16), and has been further anchored to the chromosome level with comprehensive functional annotation. Using NUCmer (v4.0.0rc1),40,41 we aligned the Shirasawa et al. assembly to our chromosomes, revealing per-chromosome coverage ranging from 91.83% to 95.20% (Table S17). Visualization of the alignments confirmed that our chromosomal assemblies were fully covered by the sequences of Shirasawa et al., demonstrating strong concordance between the two assemblies (Figure S4). We performed genome annotation for the C. lanceolata genome of Shirasawa et al. using a combined strategy incorporating de novo prediction, homology-based searches, and transcriptome alignment. A total of 37,944 protein-coding genes were predicted in this annotated genome (Tables S16, S17, and S18), which is highly consistent with the 37,225 protein-coding genes identified in our genome assembly. Subsequently, we conducted a comparative analysis between the two genomes. Specifically, 33,705 genes from our assembly were detected in the genome of the C. lanceolata assembly of Shirasawa et al., and conversely, 33,729 genes from their assembly were identified in our genome. These overlapping genes were defined as the shared gene set between the two accessions (Figure S5). A total of 3,623 genes were uniquely present in our assembly but absent from the genome of the C. lanceolata assembly of Shirasawa et al.; these were classified into three categories: true gene loss (11 genes), highly divergent sequences (433 genes), and sequences with altered start or stop codons (3,179 genes). Functional enrichment analysis revealed that these unique genes were mainly enriched in GO terms related to DNA integration and KEGG pathways associated with base excision repair. In contrast, 4,215 genes specific to the C. lanceolata reported by Shirasawa et al. assembly were missing from our genome, exhibiting a similar classification pattern: 245 true gene losses, 354 highly divergent genes, and 3,616 genes with start/stop codon mutations. Functional annotation showed that these specific genes were enriched in distinct functional pathways, such as the GO term for inorganic molecular transmembrane transporter activity and KEGG pathways related to DNA polymerase function (Figure S6). Genome expansion is associated with long introns and retrotransposons An important feature of the predicted gene structure of C. lanceolata was the abundance of long introns with an average length of 6,482 bp (Table S18). Additionally, repetitive sequences occupy a substantial portion of the C. lanceolata genome, totaling 10.38 Gb and accounting for 92.31% of the assembled genome (Table 1). The proportion is notably higher than that observed in Gnetum (85.93%),34 Taxus (76.09%),30 and Ginkgo (76.58%).35 Among these repetitive sequences, LTRs are most prevalent, accounting for 69.92% of the genome (Ty3/Gypsy 42.38%; Ty1/Copia 23.88%), followed by dictyostelium intermediate repeat sequences (DIRSs, 8.26%) and long interspersed nuclear elements (LINEs, 3.26%) (Table S19). The LTR assembly index (LAI) values, ranging from 16.94 to 20.09 across chromosomes (Table S20) are higher than those in F. hodginsii,36 further indicating the reference-level continuity of our assembled genome and reflecting the substantial LTR contribution to genome structure. We also found that the ratio of solo-LTR to intact-LTR of C. lanceolata is 0.97 (100,395: 104,515), higher than that in Pinaceae species such as P. tabuliformis (0.16), P. abies (0.16), and P. taeda (0.72), but significantly lower than that in most other published gymnosperm genomes, such as T. wallichiana (5.51), Ginkgo (4.26), W. mirabilis (3.87), and Gnetum (2.07).42 Solo-LTRs are thought to arise through excision-based DNA recombination, including between adjacent LTRs of the same element, leading to their removal and genome downsizing.43 Compared with Ginkgo, Gnetum, P. abies, and P. taeda, C. lanceolata has an extra burst of LTR insertion at approximately 3 Mya (Figure 1D), which may induce the extra expansion of the genome, and relatively low LTR clearance efficiency in C. lanceolata leads to the retention of more insertion sequences. Similarly, lower clearance efficiency in P. abies and P. taeda allows them to have larger genomes even if they do not experience additional LTR mass insertion events. However, whether this low LTR clearance mechanism is common in Cupressaceae remains to be analyzed. Longer introns, often found in species with larger genomes,44 such as Ginkgo and various conifers, and other sequenced Cupressaceae species, are thought to arise from stable LTR insertions and subsequent amplification.23,36,45 By analyzing the LTR insertion time, we found that the LTR insertion history in C. lanceolata has two major waves: an initial burst peaking between 18 and 14 Mya, shared with species such as P. abies, P. taeda, G. montanum, and Ginkgo. After this burst, C. lanceolata has a unique secondary wave peaking at 3 Mya (Figures 1C and 1D). This second wave suggests a recent accumulation of LTRs in C. lanceolata, not observed in other gymnosperms. For large gymnosperm genomes that commonly lack effective mechanisms to eliminate LTRs,18,23 the persistent accumulation of LTRs in the C. lanceolata genome likely underlies its large genome size. Together, the proliferation of LTR-RTs and elongation of introns contribute to the genome expansion of C. lanceolata. Phylogenetic position within extant lineages of gymnosperms The present-day gymnosperms represent mere relics of their former diversity, presenting a major challenge in reconstructing evolutionary relationships among extant lineages.46 Utilizing single-copy gene families derived from comparisons across 19 species, including nine gymnosperms, five angiosperms, two ferns, one lycophyte, and two outgroup species (Figures S7A and S7B), we inferred the species phylogeny using concatenated alignments and ASTRAL trees with both nucleotide and amino acid sequences. Among phylogenetic trees (Figures S7A and S7B), the concatenated tree based on amino acid sequences strongly supported G. biloba and Cycas panzhihuaensis clustering together and being sister to all other extant gymnosperms, while T. wallichiana and C. lanceolata formed a sister group of Gnetales (W. mirabilis and G. montanum)–Pinaceae (A. alba, P. abies, and P. taeda) (Figure S7A). Conversely, the phylogenetic trees constructed by the other three methods supported the hypothesis that Gnetales is a sister group of all other extant gymnosperms, with G. biloba-C. panzhihuaensis forming a clade sister to conifers (C. lanceolata, T. wallichiana, and Pinaceae) (Figure S7B). The differences observed between the concatenated trees based on amino acid and the other three phylogenetic trees arise from variations in character state information when using concatenated data, specifically model specification and differences in parallelism/inversion in protein sequences compared to nucleotide coding sequences, which may be affected by substitution saturation. Similar to the angiosperm branches in the inferred phylogeny, the branch lengths of W. mirabilis and G. montanum, constructed based on nucleotide sequences, were significantly longer compared to other gymnosperms (Figure S7B). Consequently, the phylogenetic location of Gnetales as the first diverged gymnosperm branch might be influenced by the long-branch attraction (LBA) phenomenon. Strategies to mitigate LBA artifacts include excluding long-branch taxa and fast-evolving third codon positions, as well as employing inference methods less sensitive to LBA.47 To this end, the concatenated and ASTRAL trees were constructed by selecting the first two codon positions, while the Bayesian phylogenetic tree was constructed using PhyloBayes with the “cat model,” an effective amino acid substitution model for alleviating LBA artifacts. In support of the LBA hypothesis, the results from the Bayesian and concatenated tree approaches favored G. biloba-C. panzhihuaensis as sisters to the remaining extant gymnosperms, whereas the ASTRAL tree clustered G. biloba-C. panzhihuaensis with conifers, including C. lanceolata (Figures S7A and S7B). We further expanded the taxonomy, sampling the phylogenetic analysis to 29 gymnosperms by using 25 transcriptomes and four genomes (Figures S7C and S7D). The inferred phylogenetic trees based on the concatenated alignments of either nucleotide or amino acid sequences supported the first view mentioned above (Figure S7A). In summary, based on the outcomes of various phylogenetic analyses and recent researches,1 we tend to place G. biloba and C. panzhihuaensis as the sister group of conifers. Among conifers, Gnetales-Welwitschia and pine clade are sister groups, forming a sister group to the Cupressaceae plants such as C. lanceolata, C. harringtonia, S. verticillata, and A. cunninghamii. Spermatophyte-wide whole-genome duplication Distribution of synonymous substitutions per synonymous site (KS) for paralogous C. lanceolata genes revealed no clear peaks indicative of an ancient whole-genome duplication (WGD) during its evolutionary history (Figure S8). Similarly, the paralogous KS distributions for G. montanum and T. wallichiana showed no WGD signatures, whereas both C. panzhihuaensis and G. biloba exhibited KS peaks corresponding to a shared WGD event, as previously reported.32 In W. mirabilis, a distinct WGD peak was observed, consistent with a lineage-specific WGD event that occurred after its divergence from G. montanum.34 The absence of a WGD in C. lanceolata is further supported by intragenomic comparisons (Figure S9). When comparing the genome of C. lanceolata with five other chromosome-level gymnosperm genomes (G. biloba, C. panzhihuaensis, W. mirabilis, G. montanum, and T. wallichiana), intergenomic collinearity analyses between C. lanceolata and G. biloba, C. panzhihuaensis, and T. wallichiana typically revealed only one orthologous collinear segment in C. lanceolata, consistent with the absence of a lineage-specific WGD in this species. For G. biloba and C. panzhihuaensis, which share a WGD, two orthologous collinear segments would be expected in these species corresponding to one in C. lanceolata; however, due to extensive genomic rearrangements following WGD, collinearity is often reduced.2,32 Although W. mirabilis clearly experienced a lineage-specific WGD after diverging from G. montanum,34 the WGD shared by G. biloba and C. panzhihuaensis has been hypothesized to represent a gymnosperm-wide WGD.32 To test this hypothesis, we performed phylo-KS analysis48 (Figure 2A), inferring species phylogenies in KS units to compare WGD peaks with divergence times (see Methods). This analysis confirmed a lineage-specific WGD in W. mirabilis, whereas the WGD peaks in G. biloba and C. panzhihuaensis mapped slightly before the divergence of extant gymnosperms. Although this may suggest a gymnosperm-wide WGD predating their diversification, the accuracy of KS peak identification48 and the uncertain topology of major gymnosperm lineages limit confident interpretation.Figure 2Whole-genome duplication (WGD) events before the early divergence of seed plants(A). The phylogenetic tree with branch length in KS units inferred in the phylo-KS analysis.46 The three purple squares refer to the three recent WGD events that could be identified in the paralogous KS distributions for W. mirabilis, G. biloba, and C. panzhihuaensis.The positions of squares represent the timing s of the WGD events determined by the half of the KS-peak values inferred in the KS distributions of all paralogous gene pairs and are mapped on each of the three branches starting from the corresponding tips.(B). The species tree used in the statistical gene tree – species tree reconciliation with branch lengths representing absolute age obtained from the time tree.49 Small-scale gene duplication (λ) and loss (μ), and WGD retention (q) were estimated using two models: (1) the critical DL + WGD model (λ = μ, rates vary across branches); and (2) the relaxed DL + WGD model (λ ≠ μ, rates vary across branches).The inferred relative positions of WGDs are marked on their corresponding branches, with solid blue squares referring to a WGD supported by the critical model (q > 0), hollow blue squares referring to WGDs not supported by the critical model (q ≈ 0), and solid orange squares referring to WGDs supported by the relaxed model (q > 0). Duplication and loss rates inferred from the relaxed model were mapped on each branch with a continuous color scale on the left and right tree, respectively.(C). The posterior distribution of the WGD retention rates(q) drawn from the MCMC sampling with the critical model (in blue) and the relaxed model (in orange) in the WHALE analyses.The mean of the retention rates is marked by dotted lines, and the Bayes factors (K) are indicated on each distribution. Whole-genome duplication (WGD) events before the early divergence of seed plants (A). The phylogenetic tree with branch length in KS units inferred in the phylo-KS analysis.46 The three purple squares refer to the three recent WGD events that could be identified in the paralogous KS distributions for W. mirabilis, G. biloba, and C. panzhihuaensis. The positions of squares represent the timing s of the WGD events determined by the half of the KS-peak values inferred in the KS distributions of all paralogous gene pairs and are mapped on each of the three branches starting from the corresponding tips. (B). The species tree used in the statistical gene tree – species tree reconciliation with branch lengths representing absolute age obtained from the time tree.49 Small-scale gene duplication (λ) and loss (μ), and WGD retention (q) were estimated using two models: (1) the critical DL + WGD model (λ = μ, rates vary across branches); and (2) the relaxed DL + WGD model (λ ≠ μ, rates vary across branches). The inferred relative positions of WGDs are marked on their corresponding branches, with solid blue squares referring to a WGD supported by the critical model (q > 0), hollow blue squares referring to WGDs not supported by the critical model (q ≈ 0), and solid orange squares referring to WGDs supported by the relaxed model (q > 0). Duplication and loss rates inferred from the relaxed model were mapped on each branch with a continuous color scale on the left and right tree, respectively. (C). The posterior distribution of the WGD retention rates(q) drawn from the MCMC sampling with the critical model (in blue) and the relaxed model (in orange) in the WHALE analyses. The mean of the retention rates is marked by dotted lines, and the Bayes factors (K) are indicated on each distribution. To further evaluate the existence of a gymnosperm-wide WGD and other potential WGDs before and after gymnosperm divergence, we used WHALE for statistical gene tree-species tree reconciliation.50 To reduce phylogenetic uncertainties, we excluded G. biloba, C. panzhihuaensis, W. mirabilis, and G. montanum (Figures 2B, Figure S10), thereby focusing on the branch leading to the four remaining gymnosperms for testing WGD events after the split from angiosperms, as this branch already includes the stem branch to gymnosperms. In the WHALE analyses, hypothetical WGDs were assigned to the following branches: “WGD 1” on the branch leading to seed plants (spermatophyte-wide); “WGD 2” on the branch leading to the sampled angiosperms; and “WGD 3” on the branch leading to the sampled gymnosperms. Subsequently, we estimated small-scale gene duplication (λ) and loss (μ), and WGD retention (q) using two models: (1) the critical branch-specific DL + WGD model (λ = μ, rates vary across branches); and (2) the relaxed branch-specific DL + WGD model (λ ≠ μ, rates vary across branches)50,51 (Table S21). Guided by Bayes factor analysis, although support for angiosperm- (WGD 2) and gymnosperm-wide (WGD 3) WGDs differed between the two models, both models strongly supported a spermatophyte-wide WGD (WGD 1). Our results further suggest that the WGD shared by G. biloba and C. panzhihuaensis is more likely a lineage-specific event that occurred before their divergence, rather than a genuine gymnosperm-wide event. The discrepancies between model results for angiosperm- and gymnosperm-wide WGDs may stem from sparse taxon sampling or model misspecification. Although the relaxed DL + WGD model is generally considered more biologically realistic,48 limited sampling may obscure the differences between gene duplication and loss rates, leaving the critical model (λ = μ) as a potentially better fit for sparsely sampled data, as observed in other birth-death models with equal birth and death rates for gene family evolution in distantly related species.52 Future studies with denser genome sampling and improved computational approaches will help resolve these outstanding questions. Population history of C. lanceolata Existing fossil records of C. lanceolata are sparse. However, significant findings have contributed to our understanding of its early history through closely related species. C. asiatica, which shows strong morphological similarities to modern C. lanceolata including leaf epidermal structures, branches, leaves, and cones, was found in Northeast China, and confirms that the Cunninghamia genus could date back to the Early Cretaceous.53 Another closely related species, C. protokonishii, has been identified in Late Cretaceous to Miocene strata across eastern Siberia (Russia), Japan, and East China. Cunninghamia appeared later in North America and Europe, primarily in the Early Tertiary.54 This evidence suggests that Cunninghamia originated in East Asia’s circum-Pacific region during the Late Jurassic or Early Cretaceous. It then might spread to North America via the Bering Land Bridge55 and subsequently to Europe through the North Atlantic Land Bridge,56 shaping a broad distribution across the Northern Hemisphere.57 The Quaternary glaciation had a profound impact on plants’ distribution.58,59 As global temperatures dropped, the range of C. lanceolata and other relict species, such as ginkgo,32 contracted significantly southward, with populations persisting only in certain refugia, such as the Sichuan Basin. As the glaciation waned and temperatures gradually rose, C. lanceolata began to disperse from these refugia.60 Fossil evidence from Holocene strata in the Yangtze River, Pearl River, and Min River basins (south of the Qinling Mountains and Huaihe River)57 indicates that the post-glaciation expansion of C. lanceolata did not fully restore its former range, and instead, its current distribution is primarily centered in South China and northern Vietnam. To further explore the population history of C. lanceolata, we collected 128 samples, of which 115 were from natural populations of China and 13 were from Vietnam (see Methods and Table S22). Phylogenetic and principal component analyses, based on 3,419,532 high-quality single-nucleotide polymorphisms (SNPs) (Table S23) generated by the specific locus-amplified fragment (SLAF) method, categorized these 128 samples into three major clades: SCD clade, Middle clade, and Eastern clade (Figures 3A and 3B).Figure 3Phylogeography and population history of Cunninghamia lanceolata(A) Maximum likelihood phylogenetic tree of C. lanceolata populations constructed based on single-nucleotide polymorphism (SNP) loci.(B). Principal component analysis (PCA) of all samples.(C). Population structure of C. lanceolata.(D). Paleo-distribution reconstruction of all samples.(E). Map illustrating the locations of sampled C. lanceolata populations. Phylogeography and population history of Cunninghamia lanceolata (A) Maximum likelihood phylogenetic tree of C. lanceolata populations constructed based on single-nucleotide polymorphism (SNP) loci. (B). Principal component analysis (PCA) of all samples. (C). Population structure of C. lanceolata. (D). Paleo-distribution reconstruction of all samples. (E). Map illustrating the locations of sampled C. lanceolata populations. The SCD clade, located within the Sichuan Basin, identified by the ADMIXTURE and paleo-distribution reconstruction analysis (Figures 3C and 3D), exhibited a unique genetic structure, indicating an ancient and isolated lineage that was effectively protected from glacial impacts by this refugium.61 However, post-glacial migration from this population appears to have been limited, implying that other populations were more responsible for shaping the current distribution of C. lanceolata. While the SCD clade may not have directly contributed to the species’ modern distribution, gene flow between SCD and neighboring populations remains possible. The overlap of the SCD and SXA populations in the paleo-distribution analysis models (Figure 3D), with the SXA population distributed in the northeastern mountains of the Sichuan Basin (Figure 3E), further supports the potential for gene flow. However, more evidence is needed to confirm inter-population exchange and genetic variation within the refugium. Additionally, other refugia with more favorable conditions might also facilitate their spread.62 The Middle clade included the HNX and GZH populations. Interestingly, despite geographic proximity, the two populations show relatively large genetic differences (Figures 3C and E). This divergence could be attributed to the glacial-interglacial cycles that caused repeated range expansion and contraction.63 The Eastern clade was genetically closer to the GZH population (Figure 3E), probably because the Xuefeng and Nanling Mountains hindered the spread of the HNX population. In contrast, the GZH may have migrated across the plains near the Pearl River to eastern China (Figure 3D), as supported by the discovery of Holocene fossils and buried wood fossils in the Pearl River basin.62 Unfortunately, we were unable to collect suitable samples from this region, which limits a more precise assessment, and future research is needed to address this gap. In the Eastern clade, the FJM population appears notably ancient, suggesting a potential role as a post-glacial dispersal hub. The neighbor-joining (NJ) tree illustrates that the YNN, TWB, and SXA populations cluster with FJM, whereas JXJ, JXW, and JSZ populations form another subclade (Figure 3A). This pattern implies that the FJM population initially spread along the southeastern coastline, establishing a coastal subclade, followed by northward dispersal to form an inland subclade, consistent with south-to-north post-glacial warming. However, the clustering of SXA population clusters within this clade with TWB and YNN, and its geographic location, suggest that it may be more appropriately placed within the Middle clade. Additional sampling would help clarify the evolutionary relationships among these populations. Furthermore, we investigated the phylogenetic evolutionary status of the C. lanceolata assembly of Shirasawa et al., and revealed that it exhibits a close genetic relationship with the TWB and YNN populations (Figure S11). This finding suggests that C. lanceolata in Japan comes from China with a dispersal route via the island of Taiwan. Formation of astringent seeds and adaptability to aluminum-rich soils Unlike angiosperms, gymnosperms lack double fertilization. Consequently, the tissue surrounding the gymnospermous embryo is not the true endosperm but a relatively simple structure called the megagametophyte.64 However, in conifer species, the megagametophytes can serve as nutrient storage organ, although not as extensively as the endosperm.65 A unique abortive phenomenon occurs in C. lanceolata, in which the nutrients in the megagametophyte are replaced by astringent-tasting secondary metabolites, resulting in abortive seeds referred to as “astringent seeds.”66,67 This presents a challenge in distinguishing abortive seeds from viable seeds through simple methods such as visual observation or buoyancy testing. For further investigation, we dissected and compared the morphological differences between the astringent and normal viable seeds (Method S1). Our findings revealed that embryo abortion typically initiates approximately 95–105 days after pollination (Figure 4A). During this stage, the nucellus undergoes partial decomposition, leading to the formation of a cavity between the embryo and the nucellus. However, these conditions are transient and do not persist until seed maturity. Instead, metabolites accumulate in the original location of the nucellus cells, gradually filling the cavity and giving the impression of seed growth (Figure 4B). By the time seeds reach maturity, it becomes impossible to distinguish between the two types based solely on appearance. Unfortunately, in C. lanceolata seed plantations, the proportion of astringent seeds often exceeds 30%,68 significantly impeding the selection of high-quality seeds.Figure 4Astringent seed (AS) formation(A) Phenotypic characteristics depict the development of germinating seeds (GS) and AS.The marked times represent the corresponding days after pollination. Length of the scale bar represents 1 mm.(B) Microscopic observation of AS and GS at different development stages.Key: mi, micropyle; em, embryo; ho, homozygous female gametophyte; ac, astringent compound. Microscopic scanning revealed that GS displays a complete embryo structure during development, whereas exhibits aborted embryos, broken homozygous female gametophytes, and accumulated secondary metabolites.The marked times represent the corresponding days after pollination. Length of the scale bar represents 1000 μm.(C) Proposed putative mechanism of astringent seed formation. The formation of astringent seeds is closely linked to the disruption of key hormonal signaling pathways involved in embryonic development. Abnormal expression of genes in the brassinosteroid (BR) and abscisic acid (ABA) biosynthetic pathways is observed during astringent seed formation.As the embryo degenerates, tissue cells begin to disintegrate, triggering stress response mechanisms. This includes the activation of the jasmonic acid (JA) pathway and the flavonoid biosynthesis pathway, which collectively result in the accumulation of astringent compounds.(D) Phenotypic differences in T1 generation transgenic A. thaliana seeds.Compared to the wild type (WT), seeds introduced and over-expressed with the ClLAR and ClANR genes displayed a darker color, indicating an increased accumulation of flavonoids.Seeds introduced with the ClDET2 and ClBZR1 genes exhibited a higher proportion of seeds with abnormal sizes, suggesting possible disruptions in embryo development. Length of the scale bar represents 200 μm.(E) Seed germination rates in each transgenic line. The error bars represent the standard deviation (SD). Astringent seed (AS) formation (A) Phenotypic characteristics depict the development of germinating seeds (GS) and AS. The marked times represent the corresponding days after pollination. Length of the scale bar represents 1 mm. (B) Microscopic observation of AS and GS at different development stages. Key: mi, micropyle; em, embryo; ho, homozygous female gametophyte; ac, astringent compound. Microscopic scanning revealed that GS displays a complete embryo structure during development, whereas exhibits aborted embryos, broken homozygous female gametophytes, and accumulated secondary metabolites. The marked times represent the corresponding days after pollination. Length of the scale bar represents 1000 μm. (C) Proposed putative mechanism of astringent seed formation. The formation of astringent seeds is closely linked to the disruption of key hormonal signaling pathways involved in embryonic development. Abnormal expression of genes in the brassinosteroid (BR) and abscisic acid (ABA) biosynthetic pathways is observed during astringent seed formation. As the embryo degenerates, tissue cells begin to disintegrate, triggering stress response mechanisms. This includes the activation of the jasmonic acid (JA) pathway and the flavonoid biosynthesis pathway, which collectively result in the accumulation of astringent compounds. (D) Phenotypic differences in T1 generation transgenic A. thaliana seeds. Compared to the wild type (WT), seeds introduced and over-expressed with the ClLAR and ClANR genes displayed a darker color, indicating an increased accumulation of flavonoids. Seeds introduced with the ClDET2 and ClBZR1 genes exhibited a higher proportion of seeds with abnormal sizes, suggesting possible disruptions in embryo development. Length of the scale bar represents 200 μm. (E) Seed germination rates in each transgenic line. The error bars represent the standard deviation (SD). Subsequently, we delved into the primary components of secondary metabolites in astringent seeds and their molecular mechanisms of synthesis. Metabolome analysis revealed that the megagametophytes primarily contained flavonoids and their derivatives, with a significant increase in the content of flavonoid metabolites observed from 105 days after pollination (Method S2 and Table S24). Among the 142 genes associated with flavonoid biosynthesis identified in the C. lanceolata genome, a considerable proportion were found duplicated (Table S25). Notable examples include flavanone-3-hydroxylase (F3H), flavonoid 3′-hydroxylase (F3’H), flavonoid 3′,5′-hydroxylase (F3′5′H), and flavonol synthase (FLS) (Figures S12 and S13). Additionally, 29 genes showed upregulation during the mid to late stages (Figure S14), corresponding with the observed sequence of morphological changes. We further identified 771 embryo-defective genes (EMBs)69 (Table S26), 109 of which exhibited significant differential expression between astringent and germinated seeds (Figure S15). Abnormal levels of endogenous hormones associated with embryonic development and changes in related gene expression were also observed. These phenotypic, metabolic, and transcriptomic findings suggest that the formation of astringent seeds may result from two interconnected processes. First, the altered accumulation and expression of endogenous hormones, such as brassinosteroids (BR) and abscisic acid (ABA), alongside genes involved in embryonic development, disrupt normal embryo formation. Second, embryo degeneration triggers cell disintegration, causing signaling molecules, such as α-linoleic acid, to activate the jasmonic acid (JA) signaling pathway. Additionally, the potential accumulation of reactive oxygen species (ROS)70 within the tissue further stimulates flavonoid biosynthesis, which, through antioxidant responses,71,72,73 leads to the gradual accumulation of astringent compounds (Figure 4C). We then randomly selected four relevant genes for transgenic verification in A. thaliana (method S3). The results showed that, compared to the wild type (WT), the T1 generation seeds exhibited a darker color following the introduction of ClLAR and ClANR, two genes involved in flavonoid biosynthesis. In contrast, the introduction of ClDET2 and ClBZR1, genes associated with BR synthesis and signaling, led to an increase in the number of seeds with abnormal sizes (Figure 4D). Furthermore, the transgenic lines exhibited a significant reduction in seed germination rates compared to WT (Figure 4E). We identified that all four genes regulating the astringent seed trait are present in the genome reported by Shirasawa et al. Notably, the homologous gene of Cl08429.1 (Cl08429.1-homo) harbors a G-to-T substitution at the 667th nucleotide of the CDS region, which introduces a premature stop codon (Table S27; Figure S16). It suggests that this gene in the genome assembly by Shirasawa et al. has been pseudogenized. We find it perplexing that a plant such as C. lanceolata, which has undergone millions of years of evolution, would develop the trait of producing astringent seeds. On the surface, it appears to invest energy into “growing” seeds without providing any immediate benefits. Therefore, we scrutinized its habitat and proposed a potential underlying hypothesis. C. lanceolata is primarily distributed in southern China, where the predominant soil type is acidic red soil (Figure S17). Soil acidification leads to the release of significant amounts of active aluminum, resulting in a higher aluminum content in the soil, which adversely affects plant root growth.74,75,76 Mu et al.77 observed that aluminum stress significantly increased active Al3+ ion content, malondialdehyde (MDA) levels, and superoxide dismutase (SOD) activity in C. lanceolata seedlings, while inhibiting root elongation. However, the introduction of a low concentration of astringent seed water extract alleviated these effects, suggesting that astringent seeds may protect germinating seeds by releasing secondary metabolites that mitigate aluminum toxicity. This indicates that astringent seeds represent an evolutionary adaptation to high-aluminum environments. Certain plants have the ability to sequester activated aluminum ions within the vacuoles of mesophyll cells as aluminum ion-organic acid complexes, thus reducing aluminum toxicity to roots.78,79 Unlike most plants, whose leaves promptly shed upon wilting, the leaves of C. lanceolata remain on branches for extended periods. These leaves serve as reservoirs for absorbed reactive aluminum, impeding the flux and circulation of reactive aluminum between plants and soil.80 Consequently, this further reduces the detrimental effects of reactive aluminum in the soil (Method S4 and Figure S18). In summary, the development of astringent seeds and the persistence of dead branches in C. lanceolata represent specific adaptations to its environment, enabling the plant to thrive in high-aluminum environments. Evolution of reproductive organs Gymnosperms possess both male and female reproductive organs, typically in the form of cones, whereas angiosperms usually feature reproductive organs enclosed within flowers, including sepals, petals, stamens (male reproductive organs), and carpels (female reproductive organs).81 Based on studies of homozygous mutants, the ABC model has been proposed to elucidate how distinct floral organs acquire their specific “identities” during development.82 Subsequently, the ABC model was expanded into the “ABCDE model,” wherein class A + E genes specify sepals, A + B + E genes specify petals, B + C + E genes specify stamens, C + E genes specify carpels, and C + D + E genes specify ovules.83,84,85,86,87 Among the five classes of floral homeotic genes, classes A and E originated from the duplication of a common ancestor shared by gymnosperms and angiosperms (A), whereas the C and D functions of angiosperms were derived from a combined C/D function provided by AG-like genes (C) in extant gymnosperms and stem-group seed plants.88,89,90,91,92 Considering these insights, the overarching ABCDE model was simplified to the (A)B(C) model applicable to all seed plants.88,93 MADS-box transcription factors regulate floral organ morphogenesis and flowering time, especially within the AP1/FUL (class A), AP3/PI/AGL32 (class B), AG/STK (class C/D), and SEP (class E) subfamilies.94,95 To elucidate the phylogenetic relationships among floral homeotic genes, we identified MADS-box genes in the C. lanceolata genome and constructed a phylogenetic tree based on the MADS-box genes from four gymnosperm and 10 angiosperm species (Figure S19). A total of 50 MADS-box genes were identified, including 32 type II genes (27 MIKCc-type genes and five MIKC∗-type genes) and 18 type I genes (11 Mα genes and seven Mγ genes) (Table S28). The MIKCc-type genes were classified as AGL6 (two members), AGL32 (seven members), AG (six members), GMADS (two members), TM8 (six members), SOC1 (one member), and SVP (three members) (Table S28). Phylogenetic reconstruction of the MIKc-type MADS-box gene family revealed distinct clades corresponding to classes A(AP1/FUL), B(AP3/PI/AGL32), C + D (AG/STK), and E (SEP) of floral homeotic genes (Figure S19). Notably, the A(AP1/FUL), E(SEP), and AGL6 subfamilies shared a common ancestor, while the AP1-like and SEP-like genes were absent in gymnosperms. Furthermore, the AGL6 subfamily in gymnosperms exhibited division into two clades, one unique to gymnosperms and the other clustering with angiosperms. This suggests the presence of orthologues of floral homeotic class A and E genes in extant gymnosperms and angiosperms, with AGL6-like genes potentially playing crucial roles in gymnosperm reproductive organs. Experimental evidence across different species has confirmed that AGL6-like genes exert an E function.96,97,98 Additionally, the AG-like (class C) and STK-like (class D) genes clustered together in an evolutionary branch, with gymnosperm AG-like genes positioned at the base (Figure S19), indicating a potential role in specifying reproductive organ identity in gymnosperms. Taken together, most floral homeotic genes in extant gymnosperms exhibit close relationships or are putative orthologues, raising questions about the roles of genes, such as class B (AGL32-like genes), class C + D (AG-like genes), and class A + E (AGL6-like genes), in the development of gymnosperm reproductive organs (Figure 5A).Figure 5Analysis of reproductive organ genes of Cunninghamia lanceolata(A). Homologous genes regulating reproductive organ development in angiosperms and gymnosperms.The (A) class genes in the gymnosperm reproductive organ development model (A)B(C) are homologous to the A and E class genes in the angiosperm flower development ABCE model.The (C) class genes are homologous to the C and D class genes (Figure S8).(B). Expression profile of MIKCc-type genes. HFG, homozygous female gametophyte.(C) Angiosperm flower model.(D) Regulation model of female (left) and male (right) organs in C. lanceolata. Analysis of reproductive organ genes of Cunninghamia lanceolata (A). Homologous genes regulating reproductive organ development in angiosperms and gymnosperms. The (A) class genes in the gymnosperm reproductive organ development model (A)B(C) are homologous to the A and E class genes in the angiosperm flower development ABCE model. The (C) class genes are homologous to the C and D class genes (Figure S8). (B). Expression profile of MIKCc-type genes. HFG, homozygous female gametophyte. (C) Angiosperm flower model. (D) Regulation model of female (left) and male (right) organs in C. lanceolata. C. lanceolata features conical male cones clustered at the apex of its branches, while its female cones typically serve as the female reproductive unit, with multiple spirally arranged fertile seed scales that are axillary to non-fertile bract scales. Subsequently, we analyzed the expression patterns of floral homeotic genes in the reproductive organs of C. lanceolata to elucidate their role in determining reproductive organ characteristics. Previous studies have indicated that AG-like genes play a crucial role in distinguishing reproductive organs from non-reproductive organs, while the differential expression of AGL32-like genes is associated with the differentiation between male and female reproductive organs.82,99,100 Our expression analyses (Figure 5B) revealed two distinct expression patterns for AG-like (C class) genes in C. lanceolata. Specifically, the AG-like genes Cl10262, Cl36126, and Cl26543 exhibited expression in reproductive organs, with their highest levels observed in the seed scales, whereas Cl13850, Cl35439, and Cl26063 were solely expressed in the roots. Regarding AGL32-like genes, all except Cl27678 exhibited expression in reproductive organs, with peak expression in male cones and notable Cl14691 expression in non-fertile bract scales. The expression patterns of class B and C gene orthologues in C. lanceolata closely resemble those observed in angiosperm class B and C genes (Figures 5C and 5D). Specifically, class C genes (AG-like genes Cl10262, Cl36126, and Cl26543) are expressed in both male and female reproductive organs, especially in the seed scales, whereas class B genes (AGL32-like gene Cl14691) are mainly expressed in male reproductive organs and bract scales. In summary, classes B+(C) specify male cones and bract scales, while class C specifies seed scales. In line with previous findings,96,101 we identified two distinct expression patterns for AGL6-like genes. Specifically, Cl29520 was exclusively expressed in reproductive organs (male cones, female cones, seed scales, and bract scales), whereas Cl35065 was widely expressed in both reproductive and vegetative organs (Figure 5B). The expression of the AGL6-like gene Cl29520 overlapped with that of the AGL32-like (Cl14691) and AG-like (Cl10262, Cl36126, and Cl26543) genes, suggesting its involvement not only in the differentiation of reproductive meristematic tissues but also in specifying reproductive organ identity alongside AGL32-like and AG-like genes. The proposed gymnosperm (A)B(C) model suggests that the B-(A)-(C) genes determine male cones, whereas the (A)-(C) genes specify female cone organs.88 Combining the results of the phylogenetic and expression analyses, we further refined the gymnosperm (A)B(C) model: B-(A)-(C) specifies bract scales and male cones, while (A)–(C) specifies seed scales (Figure 5C and 5D). Discussion Gymnosperms, which appeared approximately 360 Mya. However, the present-day gymnosperms represent only a fraction of their diversity of the past, posing a significant challenge in reconstructing the evolutionary relationship among extant lineages (such as Gnetales, Cycads, Ginkgo, and conifers), especially because of limited genomic data on the Cupressaceae family of conifers. This study explores the chromosome-level genomic profile of the Cupressaceae species C. lanceolata, prevalent in acidic, aluminum-rich soils of southern China. The final assembled genome of C. lanceolata stood at 11.24 Gb, with repetitive sequences accounting for 10.38 Gb, predominantly comprised of LTR elements. The larger genome size of C. lanceolata can be attributed to longer introns, specific LTR accumulation, and the lack of an efficient LTR elimination mechanism. Phylogenomic analyses confirmed the clustering of Ginkgo and Cycads as a sister group to all other extant gymnosperms. The Taxaceae (T. wallichiana)–Cupressaceae (C. lanceolata) clade forms a sister group to Gnetales (W. mirabilis and G. montanum)–Pinaceae (A. alba, P. abies, and P. taeda). The WGD analysis showed that seed plant ancestors diverged into angiosperms and gymnosperms following a seed plant-specific WGD event. Moreover, the ancestor of the extant gymnosperms underwent a gymnosperm-specific WGD event. No evidence supports the occurrence of an ancient polyploidization Event in the lineage of C. lanceolata. Our study, encompassing 128 samples of C. lanceolata from 10 populations in China and Vietnam, identified a potential refugium in the Sichuan Basin (SCD population). The FJM population likely served as the origin of contemporary C. lanceolata populations. Furthermore, we found that certain embryos of C. lanceolata begin to abort 95–105 days after pollination, being replaced by flavonoids and forming astringent seeds (non-fertile) that are visually indistinguishable from healthy seeds. By constructing a phylogenetic tree from the MADS-box genes of 14 seed plants and analyzing the expression profiles of the MADS-box genes in C. lanceolata nutrient and reproductive organs, we employed the (A)B(C) model to elucidate the development of reproductive organs in C. lanceolata. This study provides a valuable genome sequence and contributes to the understanding of gymnosperm evolution. Limitations of this study In the presented article, we report the genome of C. lanceolata, a key species in the Cupressaceae family, and expand on the evolution of gymnosperms. While we believe this work is significant, it has several limitations. The techniques used, including sequencing, assembly, and annotation, are somewhat outdated, as the project began earlier. With current technological advancements, we believe the quality of the assembly could have been greatly improved. Secondly, we did not explore certain genomic aspects of C. lanceolata, such as methylation sites, which are essential for understanding traits such as wood formation, growth habits, reproductive development, and stress resistance. Additionally, our sampling has limitations. For instance, our population structure analysis did not include samples from regions such as Guangdong and Guangxi. Although we surveyed these areas, the strong human influence on the cultivation and distribution of C. lanceolata seeds has complicated our understanding of its natural distribution. Therefore, we only collected samples from primary forests, aiming for those older than 50 years. Had we been more persistent in acquiring samples from these regions, our data might have been more comprehensive. Of course, there are other limitations beyond those mentioned here, and we hope future research will address these gaps. Resource availability Lead contact Further information and requests for resources should be directed to and will be fulfilled by the Lead Contact, Zhong-Jian Liu (zjliu@fafu.edu.cn). Materials availability This study did not generate new unique reagents. Data and code availability •The data supporting this work are available within the paper and supplementary files.•PacBio whole-genome sequencing data, Illumina data, and genome assembly sequences have been deposited to the NCBI Sequence Read Archive (SRA) as Bioproject PRJNA668674.•This paper does not report original code.•Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. •The data supporting this work are available within the paper and supplementary files. The data supporting this work are available within the paper and supplementary files. •PacBio whole-genome sequencing data, Illumina data, and genome assembly sequences have been deposited to the NCBI Sequence Read Archive (SRA) as Bioproject PRJNA668674. PacBio whole-genome sequencing data, Illumina data, and genome assembly sequences have been deposited to the NCBI Sequence Read Archive (SRA) as Bioproject PRJNA668674. •This paper does not report original code. This paper does not report original code. •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Acknowledgments We acknowledge the support received through the Innovative Research and Constructive Platform Funds of Fujian Agriculture and Forestry University (118-612014032 and KLE18010A), awarded to S-Z. L.; the 10.13039/501100012166National Key Research and Development Program of China (2021YFD2201302), and 10.13039/501100001809National Natural Science Foundation of China (no. 32572020), awarded to Y. C; and Forestry Peak Discioline Construction Project of 10.13039/501100008766Fujian Agriculture and Forestry University (72202200205), awarded to Z.-J. L.; Y.V.d.P. acknowledges funding from the European Research Council (ERC) under the European Union’s 10.13039/501100007601Horizon 2020 research and innovation program (no. 833522) and from 10.13039/501100004385Ghent University (Methusalem funding, 10.13039/501100007229BOF.MET.2021.0005.01) ; Z.L. acknowledges funding from the Junior Research Project of 10.13039/501100003130FWO (G0ADO25N) and the Special Research Grant from 10.13039/501100004385Ghent University (10.13039/501100007229BOF.BAF.2024.0889.01). Author contributions S.-Z.L. and Z.-J.L. managed the project; , Y.V.d.P., R.M., H.-K.Z., S.-Z.L., Y.C., W.-H.S., C.W., and Z.L. planned and coordinated the project; S.-Z.L., Z.-J.L., Y.C., W.-H.S., C.W., and Z.L. wrote the manuscript; W.-H.S., Y.C., C.W., Y.-C.X., L.Y., Y.Z., M.-M.L., W.-C.T. and Y.-C.X. collected and sequenced the plant material; H.-K.Z., Q.-G. Z., W.-H.S., C.W., L.X., D.-K.L., D.-Q.C., and L.N. assembled and annotated the genome; S.-Z.L., Z.L., H.-C.C., J.-Y.W., Z.-W.W., and Z.-W.L., performed gene family clustering and comparative phylogenomics; Y.C., D.-Y.Z., X.Y., D.-K.L., G.-Z.C., J.H., M.-Z.H., X.Z., W.-Y.Z., F.-L.W., Y. L., Q. Z., and X. Z., executed transcriptome sequencing and analysis; W.-H.S., W.-C.T., Y.-C.X., S.-S.X., A.-Q.L., Y.-Q.L., S.-N.R., and B.L. conducted the evolution of reproductive organs analysis; C.W., X.-Q., M., F.-P., Z., Y., L.Y., Y.Z., M.-M.L., J.-J.Z., Z.-M.H., P.-F.W., and K.-M.L., conducted the formation of astringent seeds analysis. S.-Z.L., Z.-J.L., X.-Y.C., Y.C, C.-M.J., S.-R.L., S.-B.L., Y.-Q.Y., Z.-H.M., G.-C.D., G.-Q.C., S.X., and J.Z., conducted the evolution of reproductive organs analysis. All authors read and approved the manuscript. Declaration of interests The authors declare no competing interests. STAR★Methods Key resources table REAGENT or RESOURCESOURCEIDENTIFIERBacterial and virus strainsE. coli JM109TaKaRaATCC: 53323Agrobacterium GV3101Lab ownedDSM: 12364Biological samplesC.lanceolata: leafThis workN/AC.lanceolata: astringent seedThis workN/AC.lanceolata: germinating seedThis workN/AC.lanceolata: phloemThis workN/AC.lanceolata: stemThis workN/AC.lanceolata: rootThis workN/AC.lanceolata: homozygous female gametophytesThis workN/AC.lanceolata: male conesThis workN/AC.lanceolata: female conesThis workN/AC.lanceolata: seed scalesThis workN/AC.lanceolata: bract scalesThis workN/AA.thaliana: seedThis workN/ACritical commercial assaysP6-C4Pacific Biosciences(PacBio)N/AHiSeq X Reagent KitIlluminaN/ANovaSeq 6000 Reagent KitIlluminaN/ADeposited datawhole-genome sequencing data, Illumina data, and genome assembly sequencesThis workNCBI BioProject: PRJNA668674OligonucleotidesANR3301-F: 5’TGACCTCGAGACTAGTATGAGTTGCACTAAGAAGGT3’This workN/AANR3301-R: 5’TGTAGTCCATACTAGTGTTACCGACATCATTAGAGC3’This workN/ALAR3301-F: 5’TGACCTCGAGACTAGTATGGCCTGTGCTCCCAAAGT3’This workN/ALAR3301-R: 5’TGTAGTCCATACTAGTAAGGTACTGGTTGAAAAAAT3’This workN/ADET23301-F: 5’TGACCTCGAGACTAGTATGGCGCATGTTCTGCAACA3’This workN/ADET23301-R: 5’TGTAGTCCATACTAGTGAATATGAAGGGGAAAAGAG3’This workN/ABZR13301-F: 5’TGACCTCGAGACTAGTATGTCCACCCGCTCCGTAAT3’This workN/ABZR13301-R: 5’TGTAGTCCATACTAGTGGGACAAAAATGTTGAGACA3’This workN/ARecombinant DNApCambia3301Lab ownedRRID: Addgene_210757Software and algorithmsFASTX Toolkit (ver. 0.0.11)Hannon Labhttps://github.com/agordon/fastx_toolkitGenomeScopeRanallo-Benavidez et al.102http://genomescope.orgBwa softwareLi et al.103http://maq.sourceforge.netCanu v1.5Koren et al.104https://github.com/marbl/canuWTDBG2Ruan et al.105https://github.com/ruanjue/wtdbgPilon v1.22Walker106https://github.com/broadinstitute/piloBUSCO v5.8.2Manni et al.107https://gitlab.com/ezlab/buscoHiC-Pro v2.8.1Servant et al.108https://github.com/nicmoya/HiC-ProLACHESISBurton et al.109https://github.com/shendurelab/LACHESISGenScan v3.1Burge et al.110https://pbil.univ-lyon1.fr/members/duret/cours/INSA/exercise4/pgscan.htmlAugustu v3.1Stanke et al.111http://augustus.gobics.deGlimmer HMM v3.0.4Majoros et al.112https://ccb.jhu.edu/software/glimmerhmm/Gene ID v1.4Alioto et al.113https://github.com/guigolab/geneidSNAP v 2006-07-28Johnson et al.114http://homepage.mac.com/iankorfGeMoMa v1.3.1Keilwagen et al.115https://www.jstacs.de/index.php/GeMoMaHisat v2.0.4Kim et al.116https://ccb.jhu.edu/software/hisat/index.shtmlStringTie v1.2.3Pertea et al.117https://ccb.jhu.edu/software/stringtie/TransDecoder v2.0Haas, BJhttps://github.com/TransDecoder/TransDecoderGeneMarkS-T v5.1Tang et al.118https://exon.gatech.edu/PASA v2.0.2Haas et al.119https://github.com/PASApipelineEVM v1.1.1Haas et al.120https://github.com/EVidenceModelerBLAST v2.2.31Altschul et al.121https://blast.ncbi.nlm.nih.gov/Blast.cgi?CMD=Web&PAGE_TYPE=BlastHomeInfernal v1.1.1Nawrocki et al.122http://infernal.janelia.org/tRNAscan-SE 1.3.1Lowe et al.123https://trna.ucsc.edu/tRNAscan-SE/GenBlastA v1.0.4She et al.124http://genome.sfu.ca/projects/genBlastA/GeneWise v2.4.1Birney et al.125https://www.ebi.ac.uk/∼birney/wise2/LTR_FINDER v. 1.06Zhao et al.126http://tlife.fudan.edu.cn/ltr_finder/PILER-DF v 2.4Edgar et al.127https://www.drive5.com/piler/PASTEClassifier v1.0Hoede et al.128http://urgi.versailles.inra.fr/Tools/PASTEClassifierRepeatMasker v. 4.0.5Tarailo-Graovac et al.129http://www.repeatmasker.org/RepeatModeler/LTR retrieverOu et al.130https://github.com/oushujun/LTR_retrieverOrthoMCL v1.4Li et al.131http://orthomcl.org/orthomcl/MUSCLE v3.8.31Edgar et al.132http://www.drive5.com/muscle/TrimalCapella-Gutierrez et al.133https://github.com/inab/trimalRAxMLStamatakis134https://github.com/stamatak/standard-RAxMLPAML4.9Yang135http://abacus.gene.ucl.ac.uk/software/paml.htmlCAFÉ 4.2De Bie et al.52https://github.com/hahnlab/CAFEWgd v1.1.2Chen et al.136https://github.com/heche-psb/wgdDiamond v2.1.11Buchfink et al.137https://github.com/bbuchfink/diamondMCLvan Dongen138http://www.micans.org/mcl/i-adhore v3.0.01Proost et al.139http://bioinformatics.psb.ugent.be/webtools/i-adhore/JCVITang et al.140https://github.com/tanghaibao/jcviOrthoFinder v2.3.3Emms et al.141https://github.com/davidemms/OrthoFinderPRANKLöytynoja et al.142http://wasabiapp.org/software/prank/trimAICapella-Gutiérrez et al.143https://vicfero.github.io/trimal/WHALEZwaenepoel et al.50https://github.com/arzwa/Whale.jlMrbayes v.3.2.6Huelsenbeck et al.49https://github.com/NBISweden/MrBayesALEobserveSzöllősi et al.144https://github.com/ssolo/ALETimeTreeKumar et al.145https://timetree.org/MEGAXKumar et al.146https://www.megasoftware.net/ADMIXTURE v 1.3.0Alexander et al.147https://dalexander.github.io/admixture/SMARTLetunic et al.148http://smart.embl-heidelberg.de/MAFFTKatoh et al.149https://mafft.cbrc.jp/alignment/software/FastTree v2.1.10Price et al.150https://github.com/morgannprice/fasttreeFigtree v1.4.4Andrew Rambauthttp://tree.bio.ed.ac.uk/software/figtree/ REAGENT or RESOURCESOURCEIDENTIFIERBacterial and virus strainsE. coli JM109TaKaRaATCC: 53323Agrobacterium GV3101Lab ownedDSM: 12364Biological samplesC.lanceolata: leafThis workN/AC.lanceolata: astringent seedThis workN/AC.lanceolata: germinating seedThis workN/AC.lanceolata: phloemThis workN/AC.lanceolata: stemThis workN/AC.lanceolata: rootThis workN/AC.lanceolata: homozygous female gametophytesThis workN/AC.lanceolata: male conesThis workN/AC.lanceolata: female conesThis workN/AC.lanceolata: seed scalesThis workN/AC.lanceolata: bract scalesThis workN/AA.thaliana: seedThis workN/ACritical commercial assaysP6-C4Pacific Biosciences(PacBio)N/AHiSeq X Reagent KitIlluminaN/ANovaSeq 6000 Reagent KitIlluminaN/ADeposited datawhole-genome sequencing data, Illumina data, and genome assembly sequencesThis workNCBI BioProject: PRJNA668674OligonucleotidesANR3301-F: 5’TGACCTCGAGACTAGTATGAGTTGCACTAAGAAGGT3’This workN/AANR3301-R: 5’TGTAGTCCATACTAGTGTTACCGACATCATTAGAGC3’This workN/ALAR3301-F: 5’TGACCTCGAGACTAGTATGGCCTGTGCTCCCAAAGT3’This workN/ALAR3301-R: 5’TGTAGTCCATACTAGTAAGGTACTGGTTGAAAAAAT3’This workN/ADET23301-F: 5’TGACCTCGAGACTAGTATGGCGCATGTTCTGCAACA3’This workN/ADET23301-R: 5’TGTAGTCCATACTAGTGAATATGAAGGGGAAAAGAG3’This workN/ABZR13301-F: 5’TGACCTCGAGACTAGTATGTCCACCCGCTCCGTAAT3’This workN/ABZR13301-R: 5’TGTAGTCCATACTAGTGGGACAAAAATGTTGAGACA3’This workN/ARecombinant DNApCambia3301Lab ownedRRID: Addgene_210757Software and algorithmsFASTX Toolkit (ver. 0.0.11)Hannon Labhttps://github.com/agordon/fastx_toolkitGenomeScopeRanallo-Benavidez et al.102http://genomescope.orgBwa softwareLi et al.103http://maq.sourceforge.netCanu v1.5Koren et al.104https://github.com/marbl/canuWTDBG2Ruan et al.105https://github.com/ruanjue/wtdbgPilon v1.22Walker106https://github.com/broadinstitute/piloBUSCO v5.8.2Manni et al.107https://gitlab.com/ezlab/buscoHiC-Pro v2.8.1Servant et al.108https://github.com/nicmoya/HiC-ProLACHESISBurton et al.109https://github.com/shendurelab/LACHESISGenScan v3.1Burge et al.110https://pbil.univ-lyon1.fr/members/duret/cours/INSA/exercise4/pgscan.htmlAugustu v3.1Stanke et al.111http://augustus.gobics.deGlimmer HMM v3.0.4Majoros et al.112https://ccb.jhu.edu/software/glimmerhmm/Gene ID v1.4Alioto et al.113https://github.com/guigolab/geneidSNAP v 2006-07-28Johnson et al.114http://homepage.mac.com/iankorfGeMoMa v1.3.1Keilwagen et al.115https://www.jstacs.de/index.php/GeMoMaHisat v2.0.4Kim et al.116https://ccb.jhu.edu/software/hisat/index.shtmlStringTie v1.2.3Pertea et al.117https://ccb.jhu.edu/software/stringtie/TransDecoder v2.0Haas, BJhttps://github.com/TransDecoder/TransDecoderGeneMarkS-T v5.1Tang et al.118https://exon.gatech.edu/PASA v2.0.2Haas et al.119https://github.com/PASApipelineEVM v1.1.1Haas et al.120https://github.com/EVidenceModelerBLAST v2.2.31Altschul et al.121https://blast.ncbi.nlm.nih.gov/Blast.cgi?CMD=Web&PAGE_TYPE=BlastHomeInfernal v1.1.1Nawrocki et al.122http://infernal.janelia.org/tRNAscan-SE 1.3.1Lowe et al.123https://trna.ucsc.edu/tRNAscan-SE/GenBlastA v1.0.4She et al.124http://genome.sfu.ca/projects/genBlastA/GeneWise v2.4.1Birney et al.125https://www.ebi.ac.uk/∼birney/wise2/LTR_FINDER v. 1.06Zhao et al.126http://tlife.fudan.edu.cn/ltr_finder/PILER-DF v 2.4Edgar et al.127https://www.drive5.com/piler/PASTEClassifier v1.0Hoede et al.128http://urgi.versailles.inra.fr/Tools/PASTEClassifierRepeatMasker v. 4.0.5Tarailo-Graovac et al.129http://www.repeatmasker.org/RepeatModeler/LTR retrieverOu et al.130https://github.com/oushujun/LTR_retrieverOrthoMCL v1.4Li et al.131http://orthomcl.org/orthomcl/MUSCLE v3.8.31Edgar et al.132http://www.drive5.com/muscle/TrimalCapella-Gutierrez et al.133https://github.com/inab/trimalRAxMLStamatakis134https://github.com/stamatak/standard-RAxMLPAML4.9Yang135http://abacus.gene.ucl.ac.uk/software/paml.htmlCAFÉ 4.2De Bie et al.52https://github.com/hahnlab/CAFEWgd v1.1.2Chen et al.136https://github.com/heche-psb/wgdDiamond v2.1.11Buchfink et al.137https://github.com/bbuchfink/diamondMCLvan Dongen138http://www.micans.org/mcl/i-adhore v3.0.01Proost et al.139http://bioinformatics.psb.ugent.be/webtools/i-adhore/JCVITang et al.140https://github.com/tanghaibao/jcviOrthoFinder v2.3.3Emms et al.141https://github.com/davidemms/OrthoFinderPRANKLöytynoja et al.142http://wasabiapp.org/software/prank/trimAICapella-Gutiérrez et al.143https://vicfero.github.io/trimal/WHALEZwaenepoel et al.50https://github.com/arzwa/Whale.jlMrbayes v.3.2.6Huelsenbeck et al.49https://github.com/NBISweden/MrBayesALEobserveSzöllősi et al.144https://github.com/ssolo/ALETimeTreeKumar et al.145https://timetree.org/MEGAXKumar et al.146https://www.megasoftware.net/ADMIXTURE v 1.3.0Alexander et al.147https://dalexander.github.io/admixture/SMARTLetunic et al.148http://smart.embl-heidelberg.de/MAFFTKatoh et al.149https://mafft.cbrc.jp/alignment/software/FastTree v2.1.10Price et al.150https://github.com/morgannprice/fasttreeFigtree v1.4.4Andrew Rambauthttp://tree.bio.ed.ac.uk/software/figtree/ Experimental model and study participant details Plant materials and growing conditions All genome sequencing materials used in this study were collected in April 2016 from a 10-year-old C. lanceolata plant cultivated in the third-generation seed orchard of C. lanceolata at the Youxi National Forest Farm in Fujian Province, China. The geographical coordinates of the orchard are 25°50 ' – 26°26' N, 117°48 ' – 118°39' E. Method details DNA preparation and sequencing All genome sequencing materials used in this study were sourced from an adult C. lanceolata plant cultivated in the third-generation seed orchard of C. lanceolata at the Youxi National Forest Farm in Fujian Province, China. A sodium dodecyl sulphate-based lysis method was used to extract total genomic DNA from young leaves for Illumina and PacBio sequencing. For Illumina sequencing, the DNA was ultrasonicated to a fragment size of 270 bp, followed by library preparation using an Ultra DNA Library Prep Kit (NEB, 240 County Road, Ipswich, USA) according to the manufacturer’s instructions. Library sequencing was performed using a Hiseq4000 platform, generating paired-end sequencing reads. These reads were then trimmed using Fastq _quality_trimmer in the FASTX Toolkit ver. 0.0.11 (https://github.com/agordon/fastx_toolkit) with default parameters. For PacBio sequencing, DNA was interrupted using g-TUBE (Covaris), and the SMRTbell template preparation, involving DNA concentration, damage repair, end repair, hairpin adapter ligation, and template purification, was performed using the AMPure PB Magnetic Beads (Pacific Biosciences, 1305 O'Brien Drive, Menlo Park, USA). Subsequently, The the PacBio Sequel platform was employed to perform 20-kb single-molecule real-time DNA sequencing. Genome size and heterozygosity estimation To estimate the genome size and heterozygosity of C. lanceolata, we constructed eight 270-bp paired-end libraries and generated a K-mer distribution map. As shown in Figure S1, the average K-mer depth corresponding to the main peak was 39, from which we inferred the genome size based on K-mer number/K-mer depth. In addition, the K-mer depth of the small peak to the left of the main peak was 20. We estimated the C. lanceolata genome size to be 10.42 Gb using the GenomeScope.102 For heterozygosity estimation, we employed the combined SNP calling results obtained upon comparing the 217 Gb Illumina data with the assembled C. lanceolata genome using the Bwa software.103 This allowed us to determine the number of heterozygous SNPs in the genome and calculate the rate of heterozygosity. In total, we obtained 77 334 119 SNPs, with 77 050 655 being heterozygous and 283,464 being homozygous. Consequently, the heterozygosity of the Chinese fir genome was estimated to be 0.69%. Genome assembly The assembly of the C. lanceolata genome involved three main steps. Initially, Canu v1.5 (available at https://github.com/marbl/canu)104 was utilised to correct errors in the clean PacBio data. Canu selects longer seed reads (genomeSize = 1000000000’ and ‘corOutCoverage = 50’), detects clean-read overlaps using the high-sensitive overlapper MHAP (map-2.1.2, option ‘corMhapSensitivity = low/normal/high’), and performs an error correction through the falcon_sense method (option ‘correctedErrorRate = 0.025’). Subsequently, error-corrected reads were trimmed using unsupported bases and hairpin adapters to obtain the longest supported range using default parameters. Finally, Canu generated a draft assembly using the longest 80 coverage-trimmed reads. Additionally, WTDBG2 (https://github.com/ruanjue/wtdbg)105 was employed to construct draft assembly reads. WTDBG2 initially generated a draft assembly with the parameter ‘wtdbg -i pbreads.fasta -t 64 -H -k 21 -S 1.02 -e 3 -o wtdbg’ and then utilized error-corrected reads from Canu to enhance draft assembly performance. The consensus draft assembly results were obtained with the parameter ‘wtdbg-cns -t 64 -i wtdbg.ctg.lay -o wtdbg.ctg.lay.fa -k 15’, followed by three rounds of polishing using Illumina data through Pilon v1.22 (https://github.com/broadinstitute/pilon).106 The first polishing step adopted a quiver/arrow algorithm using SMS data with 40 threads, while the second polishing adopted the Pilon algorithm using Illumina data with the parameters ‘--mindepth 10 --changes --threads 4 --fix bases’. The completeness of the C. lanceolata genome assembly was evaluated against the embryophyta_odb10 lineage (n = 1375) using BUSCO v5.8.2.107 Additionally, BWA was used to compare the short sequences obtained through Illumina sequencing with the assembled genome to assess its integrity. Hi-C library construction and chromosome assembly Hi-C fragment libraries, ranging from 300 to 700 bp in insert size, were constructed as described by Rao et al.151 These libraries were then sequencing using Illumina high-throughput sequencing technology, with sequencing read lengths set to PE150. Upon filtering out raw-read adapter sequences and low-quality PE reads, a total of 638.73 Gb clean Hi-C reads were obtained, representing 62.23× coverage of the estimated genome size (Table S29. The clean Hi-C reads were initially truncated at putative Hi-C junctions and were subsequently aligned to the assembly results using the BWA. Only uniquely aligned read pairs with a mapping quality of >20 were retained for further analysis, while invalid read pairs, including the Da gluing-end and self-cycle, re-ligation, and dumped products, were filtered out using HiC-Pro v2.8.1,108 resulting in 91.98% of valid interaction pairs. These pairs were then used for clustering or sorting and orienting scaffolds onto chromosomes through LACHESIS109 (Table S30). The final pseudochromosomes were constructed manually. To assess the accuracy of the Hi-C assembly, an interaction heatmap of the Hi-C assembly chromosomes was generated. Gene prediction and annotation Three independent methods were employed to predict protein-coding genes: de novo, homology-based, and transcriptome-based prediction. For de novo gene prediction, GenScan v3.1,110 Augustu v3.1,111 Glimmer HMM v3.0.4,112 Gene ID v1.4,113 and SNAP v 2006-07-28 (http://homepage.mac.com/iankorf)114 were utilized with default parameters. Homologous proteins from six known whole-genome sequences (Arabidopsis, Ginkgo, G. montanum, P. abies, Populus trichocarpa, and P. taeda) were aligned to the C. lanceolata genome sequence using GeMoMa v1.3.1115 with default parameters. Transcriptome data assembly was performed using Hisat v2.0.4116 and StringTie v1.2.3117 with default parameters, followed by gene prediction using TransDecoder v2.0 (https://github.com/TransDecoder/TransDecoder) and GeneMarkS-T v5.1118 with default parameters. PASA v2.0.2119 with default parameters was employed to predict unigene sequences assembled from transcriptome data and full-length transcripts assembled from full-length transcriptome data. The results from all methods were combined and refined using EVM v1.1.1120 and PASA v2.0.2. The completeness of the annotated C. lanceolata genome was assessed using BUSCO v5.8.2. Additionally, protein-coding gene annotation was performed using BLAST v2.2.31 (1e−5)121 against the NR, EuKaryotic Orthologous Groups (KOG), GO, Translated European Molecular Biology Laboratory, and KEGG databases. Identification of non-coding RNA, pseudogene, and repetitive sequences Non-coding RNAs encompass RNA types with known functions, such as microRNAs (miRNAs), ribosomal RNA (rRNA), and transfer RNA (tRNA). The Rfam v12.1152 and miRBase (v21) databases, along with Infernal v1.1.1 (http://infernal.janelia.org/)122 (1e−5), were used to predict rRNA and miRNA, respectively. The tRNAs were predicted using tRNAscan-SE 1.3.1123 with the option ‘-E -H’. Pseudogene homologue sequences were BLASTed using GenBlastA v1.0.4 (-e 1e−5),124 and non-mature termination codes and frameshift mutations were identified using GeneWise v2.4.1 (-both -pseudo).125 Repeat sequences in C. lanceolata genome were predicted as follows: Initially, a de novo repeat library was constructed using LTR_FINDER v. 1.06 (http://tlife.fudan.edu.cn/ltr_finder/),126 Repeat Scount v1.0.5, and PILER-DF v 2.4127 with default parameters. The database was classified using PASTEClassifier v1.0128 before being combined to build a new repeat database using the RepBase v19.06153 (http://www.girinst.org/repbase). RepeatMasker v. 4.0.5 (http://www.repeatmasker.org/RepeatModeler/)129 was used to align sequences and to screen repeats, including simple repeats, satellites, and low-complexity repeats, using the set parameter ‘-nolow -no_is -norna -engine wublast -qq -frag 20000’. The timing of LTR insertion was estimated using an LTR retriever.130 Building gene families with OrthoMCL The amino acid and nucleotide sequences of 19 representative species were obtained from various sources: Arabidopsis thaliana, Oryza sativa, Selaginella moellendorffii, and Physcomitrella patens from Phytozome (https://phytozome.jgi.doe.gov/)154; Cinnamomum kanehirae, P. taeda, Anthoceros angustus, G. montanum, W. mirabilis, C. panzhihuaensis, and T. wallichiana from the National Center for Biotechnology Information (NCBI; https://www.ncbi.nlm.nih.gov/genome); A. alba and P. abies from the Plant Genome Integrative Explorer Resource (http://plantgenie.org/)155; Ginkgo from GigaDB156; Amborella from Ensembl plants; Nymphaea tetragona from Genome Warehouse (https://bigd.big.ac.cn/gwh/)157; and Salvinia cucullata and Azolla filiculoides from FernBase (www.fernbase.org).158 Gene families or orthologous groups of these species and C. lanceolata were constructed using OrthoMCL v1.4 (http://orthomcl.org/orthomcl/).131 In addition, KEGG and GO enrichment analyses were performed on the unique gene families identified in the C. lanceolata genome. Phylogenetic reconstruction Orthological analysis revealed the absence of single-copy gene families in 19 species. Therefore, we extracted genes from the pan-single-copy gene family (i.e., the single-copy gene family found in at least 50% of the species) to construct a phylogenetic tree. Initially, we identified 27 pan-single-copy gene families. Subsequently, BLASTP was employed to align multiple-copy gene families in a particular species with single-copy gene families in other species. The gene with the best alignment score was selected as the single-copy gene of the species. A total of 58 single-copy gene families were identified. The amino acid sequences of these single-copy orthologues were aligned using MUSCLE v3.8.31 (http://www.drive5.com/muscle/).132 Subsequently, the amino acid alignment was converted to nucleic acid multiple sequence alignment based on codon correspondence. After filtering the multiple sequence alignment results using Trimal,133 phylogenetic trees were constructed using RAxML134 via concatenation and ASTRAL methods based on nucleic acid and amino acid sequences, respectively. For nucleic acid and amino acid sequences, the parameter was set to -m GTRGAMMA and -m PROTGAMMAJTT, respectively. Additionally, to mitigate LBA artifacts, we selected the first two bases of the codon for nucleotide construction in concatenated and astral trees. Subsequently, we constructed the Bayesian phylogenetic tree using PhyloBayes with the ‘cat model’ to effectively avoid LBA artifacts. Estimation of divergence time The divergence time of each tree node was inferred using the MCMCtree tool from the PAML4.9 package135 The analysis utilized a correlated molecular clock and the JC69 model, with other settings left as default). Nucleic acid replacement was modeled using the GTR model and the molecular clock followed an independent rate model. The Markov Chain Monte Carlo (MCMC) process included 100 000 burn-in iterations, followed by 1 000 000 sampling iterations, with one sample collected every 100 iterations). The phylogeny was calibrated using various fossil records or molecular divergence estimates, with soft bounds placed at the split nodes of several key species pairs, including G. montanum–Ginkgo (230–282 Ma), A. thaliana–G. montanum (289–330 Ma), A. angustus–G. montanum (392–422 Ma), and P. patens–G. montanum (450–514 Ma). Gene family expansion and contraction Based on the phylogenetic tree, gene family expansion and loss rates were inferred using CAFÉ 4.2 (https://github. com/hahnlab/CAFE).52 Functional enrichment analysis was conducted on genes belonging to significantly expanded and contracted gene families in C. lanceolata. However, owing to the limited number of genes showing significant contraction and lack of enriched functions, the functional annotation results of these genes are listed in Tables S11 and S12. Analysis of whole-genome duplications (WGDs) and intra- and inter-genomic comparisons We used synonymous substitutions per synonymous site (Ks) distribution analysis to identify WGD events in the gymnosperm genomes with the package of wgd v1.1.2.136 Simply speaking, Diamond v2.1.11137 was employed for self-alignment of the protein sequences of the genomes of these species, with subsequent extraction of the mutual optimal alignment from the results. MCL 22–282138 was executed to cluster all alignment results into gene families. Codeml in the PAML package135 was used to calculate the Ks values between gene pairs in each gene family. Additionally, i-adhore v3.0.01139 (default parameter) was utilized to identify collinear segments within paranomes and to filter out all paralogous gene pairs that not located in colinear segments. Finally, the Ks distribution of all paralogous gene pairs, and of all anchor pairs, were obtained by the internal function of wgd v1.1.2. For intra- and inter-genomic comparisons, JCVI140 with default parameters was employed to conduct a collinear analysis and visualization of the gymnosperms’ genomes, including G. biloba, C. panzhihuaensis, C. lanceolata, T. wallichiana, W. mirabilis, and G. montanum. Then, we performed the phylo-Ks analysis to place the identified WGD events in G. biloba, C. panzhihuaensis, and W. mirabilis on a species tree with branch lengths in the Ks unit.46 Concisely, OrthoFinder v2.3.3141 was used to infer orthologous gene families using default parameters for 10 selected seed plants genomes, including Selaginella moellendorffii, Adiantum capillus, Azolla filiculoides, Cunninghamia lanceolata, Taxus wallichiana, Gnetum montanum, Welwitschia mirabilis, Cycas panzhihuaensis, Ginkgo_biloba, and Amborella trichopoda. All single-copy orthologous amino acid gene families were obtained from the output of OrthoFinder. PRANK142 with default parameters was used to perform multiple sequences alignment (MSA) for each gene family with protein sequences. Then, trimAI143 with ‘-backtrans -automated1’ parameters were employed to map each nucleotide data to corresponding amino acid MSA matrix and simultaneously trim it to obtain final codon-level MSA matrix for each single-copy gene families. All codon-level MSAs were concatenated followed by a Codeml analysis to infer the branch lengths in the Ks unit under the free ratio model. For the statistical gene tree – species tree reconciliation, WHALE50 was used to test the validity of the previously proposed ancient WGD events.159,160 OrthoFinder v2.3.3141 was used to infer orthologous gene families using default parameters for 11 selected seed plants genomes, including Selaginella moellendorffii, Adiantum capillus, Azolla filiculoides, Cunninghamia lanceolata, Taxus wallichiana, Picea abies, Pinus taeda, Amborella trichopoda, Nymphaea colorata, Arabidopsis thaliana, Oryza sativa. Gene families lacking genes from both clades at the root or exceeding twice the median of the square root of the family size based on a Poisson outlier criterion, were filtered out. An amino acid multiple sequence alignment (MSA) for each gene family was obtained using PRANK.142 The resulting MSAs were then used as input for Markov Chain Monte Carlo (MCMC) analysis in mrbayes v.3.2.649 to sample from the posterior probability distribution. ‘Aamodelpr’ was set as ‘fixed (LG)’, and the rates were set as gamma-distributed variations approximated using four categories. The sampling frequency was set to 100, and 1 100 000 generations were run to obtain 11 000 posterior samples. Subsequently, ALEobserve144 from ALE v1.0 was used to construct the conditional clade distribution (CCD) containing marginal clade frequencies, with a burn-in of 1000 based on 11 000 posterior samples for each gene family. The topology of the species tree was set as shown in Figure 2B, and divergence times were retrieved from TimeTree.145 The duplication-loss DL + WGD model under the critical and relaxed branch-specific model was employed to infer corresponding WGD retention rates (q) with the three hypothetical WGDs under the using Bayesian scheme, with WGD1 before the divergence of the seed plants, WGD2 before the divergence of gymnosperms, and WGD3 before the divergence of angiosperms. In the critical branch-specific DL + WGD model, several priors were specified, with η representing the parameter of the geometric prior distribution on the number of genes at the root set to follow a truncated univariate Beta distribution with shape parameters (3,1) within the interval [0.01, 0.99]. The prior r, denoting the mean of the branch rate distribution was set to follow a flat distribution, while the prior σ, representing the deviation of the branch rate distribution, followed an exponential distribution within a scale of 0.1. Additionally, λ, denoting the duplication rate of each branch, was set to follow a multivariate normal distribution for each branch, with the loss rate μ equal to λ. Conversely, in the relaxed branch-specific model, λ and μ were considered independent, with the rate variation parameter τ set to follow an exponential distribution with a scale of 1. In the model estimating the branch-specific duplication and loss rates, λ and μ were set to follow a normal distribution with a mean of 0 and a standard deviation of 5 in log scale for each branch, independently. All branch lengths were set to 1, and no WGD nodes were considered. The Bayes Factor was calculated using the “bfact.jl” script within the public Github repository of WHALE to measure the strength of evidence favoring the assumed WGD models using the Savage-Dickey density ratio. Transcriptomic data and analysis The materials used for the transcriptome sequencing were obtained from four sources. Firstly, young tissues from both vegetative organs (leaf, phloem, stem, and root) and reproductive organs (homozygous female gametophytes, male cones, female cones, seed scales, and bract scales) were obtained as the initial genome sequencing material. Secondly, germinating and astringent seeds were collected at four distinct time points (105, 115, 125, and 135 days) (Method S1). For library construction, a total of 1.5 μg RNA was prepared, and libraries were constructed using the NEBNextR UltraTM Directional RNA Library Prep Kit for IlluminaR (NEB). Index codes were incorporated to attribute sequences to each sample. The index-coded samples were then clustered using an acBot Cluster Generation System with a TruSeq PE Cluster Kitv3-cBot-HS (Illumina) according to the manufacturer’s instructions. Following cluster generation, the libraries were sequenced on an Illumina HiSeq platform, resulting in the generation of paired-end reads. Raw data in FASTQ format were first processed using in-house Perl scripts. This step involved obtaining clean data by eliminating reads containing adapters, poly-N, and low-quality reads from the raw data. StringTie (1.3.1) was used to calculate the Fragments Per Kilobase of exon model per Million mapped fragments (FPKMs) of coding genes in each sample. Gene fragments per kilobase of transcript per million mapped reads (FPKMs) were computed by aggregating the FPKMs of the transcripts in each gene group. Demographic history analysis Considering the extensive artificial introduction and transfer of germplasm resources across C. lanceolata provinces in China over decades, the genetic structure of commercial forests across different regions may have been strongly influenced by artificial intervention; thus, they cannot accurately represent the phylogeographical characteristics of C. lanceolata. Therefore, we focused on identifying communities of C. lanceolata in natural forests located within their distribution areas. We specifically targeted individual plants aged over 50 years as samples. A total of 128 individuals from 10 natural C. lanceolata populations were selected to represent the majority of known C. lanceolata localities (Table S22). DNA was extracted from young leaves of each individual collected for SLAF sequencing. Library construction followed the methodology outlined by Sun et al.161 Genomic DNA was digested into fragments of approximately 330 bp using the EcoRV-HF restriction enzyme. Thereafter, fragment ends were repaired and ligated with indexed paired-end adapters to obtain adapter-modified ends. PCR was performed to amplify the target fragments, and fragments of appropriate sizes were selected based on agarose gel electrophoresis. Finally, the pooled libraries were sequenced using an Illumina HiSeq X-ten paired-end sequencer, according to the manufacturer’s protocol (Illumina, San Diego, CA, USA). The reads of each sample were aligned to the reference genome using Bwa software. Reads in which both ends were aligned with the reference genome successfully and uniquely were considered reliable and were used to define SLAF tags. We used GATK v3.8 to obtain SNPs based on the corresponding genome-wide SLAF tags. A total of 3,419,532 genome-wide SNPs were obtained, and 187,611 genome-wide SNPs were quality-filtered with a minor allele frequency of >0.05, and no more than 20% missing data. Finally, we performed population evolution analysis. A neighbour-joining tree was constructed using MEGAX146 and a p-distance model with genome-wide SNPs. A bootstrap consensus tree was constructed using 1000 replicates. Principal component analysis (PCA) of genome-wide SNPs was performed using EIGENSOFT software version 6.0, and the first three eigenvectors were plotted in two or three dimensions. The software ADMIXTURE v 1.3.0,147 which is based on the likelihood model embedded in STRUCTURE software, was applied to infer historical lineages that show clusters of similar genotypes. The membership of each genotype was run for a range of genetic clusters from K = 1–10 using the admixture model. MADS-box gene family analysis HMM profiles of MADS (PF00319) genes were obtained from Pfam (https://pfam.xfam.org/).162 The MADS-box candidate protein was searched separately using HMMER 3.2.1 (http://hmmer.org)(with default parameters) and BLASTP (E-value of e−5). Subsequently, the domains of all MADS-box candidate gene sequences were identified using SMART (http://smart. embl-heidelberg.de/).148 MADS-box classification was based on sequence similarity searches of MADS-box genes identified in Arabidopsis and Amborella. All candidate MADS-box genes were aligned using MAFFT.149 A phylogenetic tree was constructed using FastTree v2.1.10150 and edited using Figtree v1.4.4(http://tree.bio.ed.ac.uk/software/figtree/). Quantification and statistical analysis All details of the statistics applied in this study are provided alongside the respective analysis in the method details section.