Haplotype-resolved telomere-to-telomere genome assembly of the hybrid Eucalyptus urophylla × E. grandis

Abstract

The hybrid Eucalyptus urophylla × E. grandis integrates strong stress tolerance with rapid growth traits, holding significant value for genetic improvement and industrial application. However, its highly heterozygous and complex genome has hindered the development of high-quality reference assemblies. Here, we report the first haplotype-resolved, gap-free telomere-to-telomere (T2T) genome assembly of E. urophylla × E. grandis, using a combination of Illumina, PacBio HiFi, Oxford Nanopore, and Hi-C sequencing technologies. The two haplotype assemblies span 528.93 Mb and 563.25 Mb, with scaffold N50 values of 50.21 Mb and 55.05 Mb, respectively, and were successfully anchored to 11 pseudochromosomes, including the precise determination of 11 centromeres and 22 telomeres. Quality assessment demonstrated exceptional completeness and accuracy, with both assemblies achieving over 98.3% BUSCO completeness and base-level consensus quality values (QV) exceeding 50. Gene annotation predicted 33,924 and 36,336 protein-coding genes for the respective haplotypes, with functional annotation rates above 97%. This high-quality, haplotype-resolved T2T genome represents a foundational genomic resource for the Eucalyptus genus.

Background & Summary

Eucalyptus, a highly diverse genus within the family Myrtaceae, comprises over one thousand recognized species and plays a pivotal role in forestry economies across temperate and tropical regions worldwide¹. Renowned for its broad environmental adaptability, robust stress tolerance, rapid growth, and superior wood properties, Eucalyptus ranks among the most extensively cultivated commercial hardwood genera globally¹. Beyond its economic value in construction, pulp production, and solid wood processing, certain Eucalyptus species serve as vital sources for high-value natural products, including eucalyptus oil and terpenoids, which have demonstrated substantial potential for applications in the pharmaceutical and chemical industries. To continuously optimize its economic traits, interspecific hybridization has become a central breeding strategy for Eucalyptus genetic improvement. The hybrid Eucalyptus urophylla × E. grandis, developed through controlled crossbreeding using E. urophylla as the maternal line and E. grandis as the paternal line, represents an elite commercial cultivar². This hybrid combines the strong stress tolerance of E. urophylla with the rapid growth characteristics of E. grandis, significantly enhancing cultivation efficiency and timber quality while reducing production costs, thereby establishing itself as a commercially critical forest resource³.

Advancements in sequencing technologies have propelled substantial progress in Eucalyptus genomics. Since the release of the first chromosome-level genome assembly for E. grandis in 2014¹, genome assemblies for over thirty additional Eucalyptus species⁴, including E. globulus⁵, E. rudis⁶, E. camaldulensis⁶, E. regnans⁷, and E. urophylla³, have been completed. However, most of these genomes remain incomplete drafts, frequently exhibiting sequence gaps, misassemblies, and allelic collapses due to high heterozygosity. These limitations severely impede precise identification of structural variations, in-depth investigation of allele-specific expression patterns, and systematic dissection of the genetic basis underlying complex traits. Given the prominent heterosis of the hybridization between E. urophylla and E. grandis and the immense commercial value of the hybrids. Special efforts have been exerted to sequence the hybrids genome. In 2023, Shen et al. reported a reference genome of E. urophylla × E. grandis, which was a consensus genome assembly containing 209 contigs in a total length of 545.75 Mb². For a genome in high-heterozygosity, it is highly desirable to construct the haplogenome assembly to serve as the reference genome. Aided by high-density genetic map, Lötter et al. (2024) constructed fairly discrete haplogenome assemblies for E. urophylla and E. grandis³, with the former consisted of 351 contigs, and the latter consisted of 311 contigs. There remained a total of 1,067 contigs (corresponding to 63.4 Mb) of the E. urophylla and 1,268 contigs (67.8 Mb) of the E. grandis haplogenome assembly could not be anchored. Thus far, constructing haplotype-resolved genomes that achieve true telomere-to-telomere (T2T) completeness without gaps has not been fulfilled in Eucalyptus, which is essential for deepening functional genomics research and advancing precision breeding for the Eucalyptus research community.

This study integrated Illumina short-read sequencing, PacBio HiFi high-fidelity long-read sequencing, Oxford Nanopore ultra-long-read sequencing, and Hi-C chromatin conformation capture technology, combined with a haplotype-phasing strategy, to successfully generate a high-quality, haplotype-phased T2T genome assembly for an elite Eucalyptus cultivar approved by the State Forestry and Grassland administration of China (Registration No. S-SC-EU-001-2011, Fig. 1a). This elite cultivar is a hybrid of E. urophylla × E. grandis, which holds significant commercial value and is widely promoted in the industrial plantations in southern China. The assembled genome consists of two fully phased haplotypes, each comprising 11 gap-free telomere-to-telomere chromosomes, with sizes of 528.93 Mb and 563.25 Mb and scaffold N50 values of 50.21 Mb and 55.05 Mb, respectively. Compared to previously reported genomes for this hybrid², our assembly demonstrates significant improvements in completeness, continuity, and accuracy. The assembled genome provides the most precise genomic resource to date for this key commercial hybrid eucalypt, significantly enriching the genomic repository for the Eucalyptus genus and laying a critical foundation for advanced functional genomics studies, genetic dissection of complex economic traits, and molecular-assisted breeding.

Fig. 1.

Morphology of E. urophylla × E. grandis seedlings and assessment of genome size. (a) Phenotypic traits of tissue-cultured seedlings of the elite E. urophylla × E. grandis clone (S-SC-EU-001-2011) two months after pot transplantation. (b) Genome size determination by flow cytometry. The number of detected events and the mean fluorescence intensity for each population are indicated within the plot. The region highlighted by the red box corresponds to the critical peak signal used for genome size estimation. The histogram below shows the distribution of fluorescence intensity, where peak P4 represents the signal from the E. urophylla × E. grandis test sample and peak P5 denotes the signal from the internal reference standard Solanum lycopersicum (tomato). (c) Comparative assessment of genome size estimated from different sequencing data and analytical tools across multiple K-mer lengths. Scatter points with distinct shapes and colors illustrate the estimation results derived from FastK, LVgs_HiFi, Jellyfish, and KMC respectively. The horizontal red dashed line indicates the reference genome size value measured by flow cytometry. (d) GenomeScope genomic profile constructed using 19-mers. The two principal peaks annotated within the plot correspond to the heterozygous K-mer peak (Hetero-peak) and the homozygous K-mer peak (Homo-peak) of the genome.

Methods

Plant materials, DNA and RNA extraction, and multi-platform sequencing

Tissue cultured seedlings of the elite cultivar S-SC-EU-001-2011 were produced by Guangxi Bagui Seedlings High Technology Group Company. Two months after transplanting in pots (Fig. 1a), young leaf tissues of the seedlings were collected for DNA sequencing and RNA sequencing, while stem and bark tissues were collected for RNA sequencing. All collected samples were immediately frozen in liquid nitrogen and stored at −80 °C before use.

Genomic DNA was isolated using an optimized cetyltrimethylammonium bromide (CTAB) method⁸. The integrity and fragment size of the extracted DNA were evaluated using 1% agarose gel electrophoresis. DNA purity was measured with a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific, USA), and the concentration was precisely quantified using a Qubit 3.0 fluorometer (Life Technologies, CA, USA). All DNA samples passed stringent quality control metrics prior to library construction for multi-platform high-throughput sequencing.

For Illumina short-read sequencing, libraries were constructed using the Nextera DNA Flex Library Prep Kit (Illumina, CA, USA) and sequenced on the Illumina NovaSeq X Plus platform employing a paired-end 150 base pair mode. Following quality filtering with fastp v0.21.0⁹, a total of 72.57 Gb of clean data were retained, covering the genome to a depth of approximately 127-fold, with 95.78% of bases reaching Q30 quality (Table 1).

Table 1.

Summary of sequencing data generated for the assembly of Eucalyptus urophylla × E. grandis genome.

Sequencing technology	Data yield (Gb)	Read count	Read length N50 (bp)	Q30 base proportion (%)	Estimated coverage (×)
Illumina	72.57	486,844,056	150 PE	95.78	127
PacBio HiFi	92.45	5,220,227	17,859	NA	162
Hi-C	143.48	974,196,354	150 PE	95.52	251
ONT	106.27	1,231,103	100,000	NA	186
RNA	24.60	82,004,519	150PE	95.92	—

To establish a high-accuracy genomic scaffold, PacBio high-fidelity (HIFI) long-read sequencing technology was employed. Libraries were prepared following the SMRTbell Express Template Prep Kit 2.0 protocol (Pacific Biosciences, CA, USA) and sequenced on the PacBio Revio platform. Raw reads were processed using CCS v6.0.0 with parameters set to a minimum of three passes, a minimum signal-to-noise ratio of 2.5, and a maximum of sixty passes¹⁰. After filtering low-quality subreads, 92.45 Gb qualified HiFi data were retained, containing 5.22 million reads with an N50 length of 17.86 kb and achieving 162-fold genome coverage (Table 1).

Hi-C libraries were constructed using chromosome conformation capture methodology¹¹ and sequenced on the Illumina NovaSeq 6000 platform. Raw Hi-C reads were processed with fastp v0.21.0⁹ and HICUP v0.8.0¹² to remove low-quality sequences and adaptor contamination. A total of 143.48 Gb of high-quality clean Hi-C data were obtained, providing approximately 251-fold genome coverage, with 95.52% of bases achieving a Q30 quality score (Table 1).

To assist in resolving complex genomic regions and optimizing structural assembly, ultra-long-read sequencing was performed using the Oxford Nanopore Technologies (ONT) platform. Libraries were prepared with the SQK-LSK110 kit following standard protocols and sequenced on a PromethION instrument (Nanopore Technologies, Oxford, UK) using an R9.4.1 flow cell. Following filtering and adapter trimming with Filtlong v0.2.1¹³ and Porechop v0.2.4¹⁴, the process generated 106.27 Gb of high-quality ONT reads, characterized by an N50 read length of 100 kb and providing approximately 186-fold genome coverage (Table 1).

For transcriptome sequencing, total RNA was isolated from the designated young leaf, stem, and bark tissues utilizing a magnetic plant RNA extraction kit (TIANGEN Biotech, Beijing, China), and RNA quality was assessed using a NanoDrop spectrophotometer. High-quality RNA was used for library construction and paired-end 150 bp sequencing on the Illumina NovaSeq platform. After quality filtering, at least 6 Gb of clean reads per sample were obtained, generating a total of 24.60 Gb of transcriptome data (Table 1).

Genome survey

Flow cytometry was employed as an independent reference standard for genome size estimation. Using tomato (Solanum lycopersicum) as an internal calibration standard with a genome assembly size of 832.8 Mb¹⁵, flow cytometric analysis yielded an estimated genome size of 562.41 Mb for the Eucalyptus urophylla × E. grandis hybrid (calculated as 45,994 × 832.8 / 68,106 Mb) (Fig. 1b). This estimate served as a benchmark for subsequent k-mer-based evaluations.

Genome characteristics were further assessed using k-mer analysis based on Illumina short-read sequencing data. Although Illumina sequencing provides high accuracy, inherent sequencing errors inevitably generate low-frequency k-mers, which typically lead to underestimation of the actual genome size. To mitigate this limitation and identify the most robust analytical strategy, filtered short reads were analyzed across a wide range of k-mer sizes (k = 19, 21, 27, 29, 31, and 51) using multiple k-mer counting tools, including FastK v2.1¹⁶, KMC v3.2.4¹⁷, and Jellyfish v2.2.10¹⁸. The resulting k-mer frequency distributions were modeled using GenomeScope v2.0¹⁹. Comparative analysis demonstrated that FastK produced highly stable estimates across different k values, ranging from 575.34 Mb to 577.01 Mb. These values closely aligned with the flow cytometric estimate, with a discrepancy of less than 10%, indicating high reliability. In contrast, Jellyfish consistently underestimated genome size, yielding predictions between 524.00 Mb and 532.08 Mb, whereas KMC produced unstable results and failed entirely at k = 31 and k = 51.

Given that sequencing accuracy fundamentally constrains the reliability of k-mer-based genome size estimation, orthogonal validation was conducted using corrected HiFi reads. High-accuracy corrected HiFi long-read data were analyzed using LVgs v1.0.0²⁰ and FastK across the same range of k-mer sizes. Estimates derived from HiFi data exhibited strong stability, clustering around 540 Mb, and showed a slight negative correlation with k-mer size, decreasing from 547.40 Mb at k = 19 to 534.14 Mb at k = 51. A comprehensive comparison of genome size estimates derived from different software tools, k-mer sizes, and sequencing data types (short reads and HiFi reads) was visualized using line plots (Fig. 1c).

Across all tested k values, the highest fitting curve resolution was achieved using FastK with a k-mer size of 19. Accordingly, this optimal 19-mer frequency distribution was selected for definitive genome characterization (Fig. 1d). In the sequenced heterozygous diploid genome, the k-mer distribution profile exhibits two distinct peaks corresponding to heterozygous and homozygous k-mers, with the depth of the heterozygous peak being half that of the homozygous peak. Based on the 19-mer model, the haploid genome size of the hybrid was estimated to be 577.01 Mb, with a heterozygosity rate of 2.83% and a repetitive sequence content of 50.8%.

Genome assembly, haplotype phasing, and chromosome anchoring

To construct a high-quality haplotype-resolved reference genome, a stepwise assembly strategy integrating multi-platform sequencing data was employed. Initial contig assemblies were generated using the overlap-layout-consensus (OLC) algorithm implemented in Hifiasm v0.23.0²¹, with both hybrid HiFi and ONT data and HiFi data alone, while NextDenovo was simultaneously used for the independent assembly of ONT data. The genome sizes obtained from different assembly strategies showed notable variation. The assembly relying solely on ONT data yielded a genome size of 841.76 Mb, which was significantly larger than the 582.12 Mb from the hybrid HiFi and ONT assembly and the 611.94 Mb from the HiFi only assembly (Table 2). This difference mainly resulted from the higher error rate of ONT reads, which tended to cause repetitive sequence expansion and redundant assembly. Leveraging the built-in haplotype resolution module of Hifiasm, both hybrid HiFi and ONT assembly and HiFi-only assembly were phased to distinguish and partition allelic sequences derived from distinct parental genomes, yielding two haplotype assemblies designated HaplotypeA and HaplotypeB (Table 2).

Table 2.

Summary statistics of initial contig assemblies generated using three different strategies.

Assembly Strategy	Output Version	Genome size (bp)	Contig N50 (bp)	Number of contigs	Quality Value (QV)	BUSCO (%)
Hifiasm (HiFi + ONT)	Primary contig	582,121,945	54,530,387	391	—	98.1
	Haplotype A	550,326,999	39,246,250	343	47.6809	98.3
	Haplotype B	579,737,664	42,228,237	179	49.3462	98.4
Hifiasm (HiFi only)	Primary contig	611,940,041	42,077,024	1,445	—	98.1
	Haplotype A	585,094,816	21,516,966	893	48.0108	97.3
	Haplotype B	604,118,498	20,052,504	739	47.6066	97.6
NextDenovo (ONT only)	Primary contig	841,761,040	22,990,784	56	38.6239	98.8

Hi-C data were subsequently used for chromosome-level scaffolding. The hybrid HiFi and ONT assembly, which exhibited the highest contig N50 and sequence completeness, was selected as the optimal backbone for downstream decontamination and scaffolding. The quality-controlled Hi-C reads were aligned to the two haplotype contigs using Juicer v1.6²² to construct a genome-wide chromatin interaction matrix. The contigs were then clustered using ALLHiC v0.9.8²³, followed by ordering and orientation based on Hi-C interaction signals through the 3D-DNA v180419²⁴ pipeline, ultimately yielding 11 pseudochromosomes. The assembly was visualized and manually curated using Juicebox v1.11.08²⁵ to improve the accuracy of chromosome structure. Finally, SubPhaser v1.2.6²⁶ was applied to partition and assign the final haplotypes based on specific k-mers and collinear orthologous gene pairs. Compared with the consensus assembly of E. urophylla × E. grandis produced by Shen et al.², both haplotypes exhibited high assembly continuity, with scaffold N50 values of 50.21 Mb for HaplotypeA and 55.05 Mb for HaplotypeB (Fig. 2, Table 3).

Fig. 2.

Genomic features of the haplotype-resolved telomere-to-telomere (T2T) assembly of the Eucalyptus urophylla × E. grandis hybrid. From the outermost to the innermost layers, the circular tracks represent (I) 11 pairs of homologous pseudochromosomes with an outermost scale in megabases (Mb), where white dots indicate centromere positions and black triangles denote telomere positions; (II) gene density; (III) GC content; (IV) repeat sequence density; (V) Gypsy retrotransposon density; (VI) Copia retrotransposon density; and (VII) genomic collinearity between the two haplotypes.

Table 3.

Comparison of genome assembly and annotation statistics for Eucalyptus urophylla × E. grandis.

Feature	Previous Assembly	Current Assembly
	(Shen et al.2)	HaplotypeA	HaplotypeB
Assembly size (Mb)	545.75	528.93	563.25
Number of contigs	—	343	179
Contig N50 (Mb)	39.94	39.25	42.23
Number of scaffolds	209	11	12
Scaffold N50 (Mb)	51.62	50.21	55.05
Anchoring rate to chromosomes (%)	98.29	100	99.99
Number of telomeric ends	—	22	22
Number of centromeres	—	11	11
GC content (%)	39.89	39.48	39.49
Assembly BUSCO completeness (%)	98.4	98.5	98.3
Repeat content (%)	58.56	54.31	54.34
LTR retrotransposon content (%)	39.4	34.27	33.69
Number of protein-coding genes	34,502	33,924	36,336
Functionally annotated genes (%)	90.55	97.14	97.06

T2T gap filling and sequence polishing

To achieve telomere-to-telomere (T2T) level genome completeness, systematic gap-filling and sequence polishing were performed on the chromosome-level scaffolds. Gap filling was performed following a predefined priority, sequentially utilizing ONT ultra-long reads and PacBio HiFi reads. Reads were aligned to the gap regions using Winnowmap2²⁷, and the longest aligned sequences were selected for gap closure. Following gap filling, Racon v1.6.0²⁸ was used to perform multiple rounds of sequence consensus correction by aligning PacBio HiFi reads to the updated genome assembly. To avoid introducing potential phasing errors, the–haplotype parameter was enabled during Racon polishing to preserve the original haplotype phasing information. Haplotype phasing results before and after Racon polishing were further compared using WhatsHap v1.7²⁹, yielding a phasing concordance greater than 99.5% and confirming that the Racon polishing procedure did not compromise the quality of the final haplotype assemblies. This refinement significantly enhanced genome completeness and accuracy, resulting in gap-free T2T assemblies for both haplotypes. The final assembled genome sizes were 528.93 Mb for HaplotypeA and 563.25 Mb for HaplotypeB (Table 4).

Table 4.

Chromosome-level statistics of the haplotype-resolved telomere-to-telomere (T2T) Eucalyptus urophylla × E. grandis genome assembly.

Chromosome	Length (bp)	Centromere position (bp)	Centromere size (bp)	Telomere 1 position (bp)	Telomere 2 position (bp)
chr1A	39,556,722	6,293,879–6,656,227	362,349	1–13,356	39,555,071–39,556,719
chr1B	44,363,340	6,725,643–7,160,356	434,714	17–13,705	44,344,859–44,363,339
chr2A	48,589,335	17,383,181–17,910,671	527,491	5–19,281	48,559,382–48,575,803
chr2B	58,651,396	15,848,574–16,269,209	420,636	3–12,586	58,633,042–58,651,571
chr3A	61,503,901	30,944,388–31,233,457	289,070	57–16,646	61,486,055–61,503,897
chr3B	59,206,540	30,147,411–30,575,812	428,402	1–7,807	59,191,769–59,206,540
chr4A	38,264,619	35,270,480–35,501,456	230,977	35–19,288	38,257,006–38,264,614
chr4B	40,056,144	36,980,182–37,102,392	122,211	91–11,461	40,037,517–40,056,138
chr5A	55,807,066	36,213,228–36,406,015	192,788	2–9,243	55,790,702–55,807,059
chr5B	63,809,895	36,431,329–36,771,357	340,029	6–16,317	63,808,692–63,811,187
chr6A	53,905,989	34,331,735–34,746,702	414,968	48–18,861	53,896,306–53,905,704
chr6B	55,046,264	34,250,611–34,760,645	510,035	1–17,777	55,043,921–55,045,852
chr7A	50,210,886	21,501,120–21,649,199	148,080	6–17,198	50,194,529–50,204,724
chr7B	54,748,201	23,935,623–24,223,781	288,159	5–1,561	54,735,089–54,748,183
chr8A	64,075,537	25,125,242–25,279,334	154,093	6–4,099	64,070,147–64,075,273
chr8B	67,674,454	27,103,098–27,526,228	423,131	4–19,904	67,670,654–67,674,440
chr9A	39,242,806	28,457,907–28,691,282	233,376	7–13,231	39,235,212–39,242,795
chr9B	40,136,365	28,346,304–28,601,657	255,354	13–12,119	40,129,368–40,136,353
chr10A	37,738,620	15,030,060–15,366,108	336,049	3–12,496	37,725,548–37,738,403
chr10B	37,506,972	17,588,609–18,034,391	445,783	275–4,666	37,487,632–37,506,971
chr11A	40,036,767	36,220,838–36,421,027	200,190	7–12,749	40,018,513–40,036,766
chr11B	42,055,098	37,923,602–38,270,002	346,401	1–566	42,040,434–42,055,096

Telomere and centromere identification

To validate the completeness of the T2T assemblies, precise identification of telomeric and centromeric regions was conducted for each chromosome. Telomeres were identified based on the presence of canonical plant telomeric repeat sequences (CCCTAAA/TTTAGGG) enriched at chromosomal termini. A reference database was first constructed using known telomeric sequences, and Nucmer v3.1³⁰ was used to align these sequences to both ends of each pseudochromosome for telomere detection and verification.

Centromere identification was based on structural sequence features characterized by highly enriched tandem repeats and minimal gene density^31,32. Candidate centromeric regions on each chromosome were predicted using the CentroMiner module implemented in QuarTeT v1.2.1³³. Tandem repeat monomers were first identified genome-wide using TRF v4.09.1³⁴ with parameters “2 7 7 80 10 50 500 -d -h”, and candidate centromere-associated repeats were subsequently filtered based on repeat periodicity and copy number characteristics. Candidate monomers were then clustered and dereplicate on each chromosome using cd-hit v4.8.1³⁵ with parameters “-c 0.8 -n 5 -d 0 -g 1 -aS 0.9”, generating representative monomers. These representative sequences were aligned to their corresponding chromosomes, and candidate regions were ranked based on total alignment length and retrotransposon content. Tandem repeat and gene coverage were calculated using BEDTools v2.30.0³⁶ in 50 kb sliding windows. Final centromeric regions were defined based on the co-occurrence of tandem repeat enrichment and minimal gene density. Ultimately, 22 telomeres and 11 centromeric regions were identified across the 11 chromosomes of both haplotypes, supporting the high structural integrity of the chromosome-level assembly (Fig. 3, Table 4).

Fig. 3.

Genome-wide distribution of centromeres, telomeres, and gene density in the haplotype-resolved T2T E. urophylla × E. grandis genome. Each chromosome is illustrated as an idiogram. Telomeres and centromeres are indicated by green triangles and orange squares, respectively. The gene density heatmap was calculated using a 50 kb sliding window, with red and blue denoting regions of high and low gene density, respectively.

Repetitive sequence annotation

To identify repetitive sequences in the HaplotypeA and HaplotypeB genomes, an integrated strategy combining de novo prediction and homology-based annotation was employed. Initial de novo repeat prediction for each haplotype genome was performed using RepeatModeler v2.0.5³⁷ with the parameters “-database mydb -threads 16” to construct the preliminary repeat library. Long terminal repeat retrotransposons (LTRs) were independently identified using LTR_FINDER v1.0.7³⁸ (-threads 16 -harvest_out -size 1000000 -time 300) and LTRharvest v1.6.5³⁹ (-minlenltr 100 -maxlenltr 7000 -mintsd 4 -maxtsd 6 -motif TGCA -motifmis 1 -similar 85 -vic 10 -seed 20 -seqids yes). The LTR predictions generated by the two approaches were subsequently integrated and deduplicated using LTR_retriever v2.9.8⁴⁰ with the parameters “-threads 16 -noanno” to generate a nonredundant LTR sequence set.

The resulting LTR library was merged with the de novo repeat library to construct a comprehensive de novo repeat library. Unclassified sequences within the library were further classified using TEclass v2.1.3⁴¹. The de novo repeat library and the RepBase database⁴² were subsequently combined to generate a composite repeat library, which was used for repeat annotation based on sequence similarity and protein homology using RepeatMasker v4.1.5 and RepeatProteinMask v4.1.5, respectively. Finally, the annotation results derived from sequence similarity and protein homology analyses were integrated and deduplicated to generate a unified genome-wide repeat annotation set.

The annotation results showed that repetitive sequences accounted for 54.31% and 54.34% of the HaplotypeA and HaplotypeB genomes, respectively, representing the predominant genomic component in both haplotypes. LTR retrotransposons were the most abundant and occupied approximately one-third of the total genome length. Within the LTR subclasses, both the abundance and total sequence length of LTR-Copia were substantially higher than those of LTR-Gypsy, with the former being approximately 1.5-fold greater than the latter (Table 5).

Table 5.

Summary of annotated repetitive elements in the HaplotypeA and HaplotypeB genome assemblies.

Repeat Class	HaplotypeA		HaplotypeB
	Length (bp)	Genome (%)	Length (bp)	Genome (%)
LTRs (Long Terminal Repeats)	181,248,152	34.27	189,772,776	33.69
LTR/Copia	60,194,626	11.38	64,837,761	11.51
LTR/Gypsy	41,053,418	7.76	40,880,416	7.26
DNA Transposons	74,995,017	14.18	83,425,259	14.81
LINEs (Long Interspersed Nuclear Elements)	23,144,985	4.38	24,861,150	4.41
SINEs (Short Interspersed Nuclear Elements)	1,522,374	0.29	1,590,178	0.28
Unknown	24,258,422	4.59	25,367,903	4.5
Total Annotated Repeats	287,280,918	54.31	306,096,424	54.34

Non-coding RNA annotation

Non-coding RNAs were predicted based on their characteristic sequences and structures. Transfer RNAs (tRNAs) were predicted using tRNAscan-SE v2.0.12⁴³ leveraging their conserved secondary structure features. Ribosomal RNAs (rRNAs) were annotated using RNAmmer v1.2⁴⁴. Small nuclear RNAs (snRNAs) and microRNAs (miRNAs) were identified through homology searches using INFERNAL v1.1.4⁴⁵ to align covariance models from the Rfam database version 14.9⁴⁶.

The results showed that the composition and number of non-coding RNAs were largely consistent between the two haplotype genomes. A total of 7003 non-coding RNAs (247 miRNAs, 480 tRNAs, 5850 rRNAs and 426 snRNAs) and 6967 non-coding RNAs (242 miRNAs, 500 tRNAs, 5789 rRNAs and 436 snRNAs) were identified in the HaplotypeA and HaplotypeB genomes respectively (Table 6).

Table 6.

Summary of non-coding RNA (ncRNA) annotations in the HaplotypeA and HaplotypeB genomes.

Haplotype	ncRNA Type	Number	Average length (bp)	Total length (bp)	Genome (%)
HaplotypeA	miRNA	247	125	30,799	0.0058
	tRNA	480	75	35,898	0.0068
	rRNA	5,850	181	1,059,684	0.2003
	snRNA	426	120	51,101	0.0097
	Total	7,003			0.2226
HaplotypeB	miRNA	242	127	30,773	0.0055
	tRNA	500	75	37,461	0.0067
	rRNA	5,789	174	1,007,014	0.1788
	snRNA	436	117	50,848	0.009
	Total	6,967			0.1990

Structural annotation of protein-coding genes

The structural annotation of protein-coding genes employed an integrated strategy combining ab initio prediction, homology-based prediction, and transcriptome-based prediction. Multiple ab initio gene prediction tools were utilized including Augustus v3.5.0⁴⁷, Genscan v1.0⁴⁸, and GlimmerHMM v3.0.4⁴⁹. For homology-based prediction, protein sequences from closely related species such as Arabidopsis thaliana⁵⁰, E. grandis¹, and Populus trichocarpa⁵¹ were aligned to the genome using Miniprot v0.13⁵² to identify potential homologous gene models. In the transcriptome-based approach, clean reads obtained from the collected leaf, stem and bark tissues were aligned to the genome using Hisat2 v2.2.1⁵³. The alignment results were then used for transcript reconstruction with StringTie v2.2.1⁵⁴, and open reading frames were predicted from the reconstructed transcripts using TransDecoder v5.7.0⁵⁵. Ultimately, all prediction evidence was subjected to weighted integration and optimization using MAKER v3.01.03⁵⁶ and EVidenceModeler v2.1.0⁵⁷ to generate a high-quality, non-redundant gene set.

The structural annotation analysis revealed similar gene features between the two haplotypes (Table 7). A total of 33,924 and 36,336 protein-coding genes were predicted for HaplotypeA and HaplotypeB respectively, corresponding to gene densities of 64.14 and 64.51 genes per megabase. In HaplotypeA, the average gene length was 4,134 bp, with an average coding sequence length of 1,214 bp. Each gene contained an average of 5.07 exons, with mean exon length of 314.3 bp. HaplotypeB exhibited comparable gene structure parameters, with an average gene length of 4,208 bp, an average coding sequence length of 1,192 bp, 4.92 exons per gene, and average exon length of 315.9 bp.

Table 7.

Summary of gene structure annotation statistics for the HaplotypeA and HaplotypeB genomes.

Haplotype	Method	Number of genes	Mean gene length (bp)	Mean CDS length (bp)	Exons per gene (mean)	Mean exon length (bp)
HaplotypeA	Ab initio prediction	33,967	2,990.03	1,039.83	4.65	223.61
	Homology-based	45,962	4,823.75	1,354.42	6.1	221.95
	Transcriptome-based	23,868	5,290.15	1,216.76	5.65	342.94
	Integration	29,134	4,316.97	1,178.83	5.30	308.48
	Final gene set	33,924	4,134.12	1,213.68	5.07	314.31
HaplotypeB	Ab initio prediction	36,477	2,778.84	1,003.53	4.38	229.09
	Homology-based	46,433	5,051.76	1,363.45	6.13	222.44
	Transcriptome-based	24,167	5,333.13	1,220.62	5.63	343.63
	Integration	33,844	5,098.0	1,098.89	4.88	303.52
	Final gene set	36,336	4,208.46	1,192.11	4.92	315.93

Functional annotation of protein-coding genes

To elucidate the potential biological functions of the predicted protein-coding genes, a comprehensive functional annotation was performed. Protein sequence alignments were conducted against the UniProt, NCBI non-redundant (NR), Gene Ontology (GO), and Kyoto Encyclopedia of Genes and Genomes (KEGG) public databases^58–61 using DIAMOND v2.1.8⁶². In addition, InterProScan v5.55-88.0⁶³ was applied to scan InterPro, Pfam, and other domain databases^64,65 to identify conserved functional domains and motifs. The hmmscan program within the HMMER v3.3.2⁶⁶ was utilized to search the Eukaryotic Orthologous Groups (KOG) database⁶⁷ for the identification of orthologous relationships and potential functional classifications.

The annotation results indicated that 32,954 genes (97.14%) in HaplotypeA and 35,267 genes (97.06%) in HaplotypeB were functionally annotated in at least one database (Table 8). An analysis of the overlap among database annotations revealed that approximately 30% of the genes in both haplotypes were concurrently annotated by all seven databases (Fig. 4).

Table 8.

Summary of gene structure annotation statistics for the HaplotypeA and HaplotypeB genomes.

Annotation Category	HaplotypeA count	HaplotypeA (%)	HaplotypeB count	HaplotypeB (%)
Annotated in at least one database	32,954	97.14%	35,267	97.06%
NR (NCBI)	32,076	94.55%	34,197	94.11%
UniPro	31,485	92.81%	33,528	92.27%
GO	25,923	76.41%	27,742	76.35%
KEGG	22,519	66.38%	24,014	66.09%
KOG	12,031	35.46%	12,805	35.24%
Pfam	24,012	70.78%	25,201	69.36%
Interpro	31,941	94.15%	34,152	93.99%

Fig. 4.

Venn diagrams of gene function annotations in HaplotypeA and HaplotypeB. Panels (a,b) illustrate the overlap of predicted gene annotations across seven public databases in HaplotypeA and HaplotypeB, respectively. These databases include the NCBI non-redundant protein database (NR), UniProt, Kyoto Encyclopedia of Genes and Genomes (KEGG), Eukaryotic Orthologous Groups (KOG), Gene Ontology (GO), InterPro, and Pfam. The numbers within each region indicate the count of genes annotated by the corresponding database or combination of databases. The central red region represents the number of genes co-annotated by all seven databases.

Data Records

All sequencing datasets generated in this study have been deposited in the NCBI Sequence Read Archive (SRA) under BioProject accession number PRJNA1290067. The Illumina paired-end genomic sequencing reads, PacBio HiFi long reads, Hi-C sequencing data, and Oxford Nanopore Technologies (ONT) ultra-long reads are available under accession numbers SRR34490132⁶⁸, SRR34490133⁶⁹, SRR34490134⁷⁰, and SRR34490135⁷¹, respectively. Transcriptome sequencing datasets derived from stem, leaf, and bark tissues are available under accession numbers SRR38044340⁷², SRR37991218⁷³, and SRR37991220⁷⁴, respectively. The Figshare repository⁷⁵ contains the haplotype-resolved telomere-to-telomere genome assemblies for the two haplotypes in FASTA format, together with the corresponding genome annotation datasets in GFF3 format, including protein-coding gene annotations, repetitive sequence annotations, and non-coding RNA annotations.

Technical Validation

To rigorously evaluate the quality of the haplotype-resolved T2T genome assemblies, a series of complementary validation strategies were employed. First, the contiguity, completeness, and accuracy of the genome assemblies were evaluated. HaplotypeA and HaplotypeB exhibited exceptional long-range contiguity with scaffold N50 values of 50.21 Mb and 55.05 Mb respectively. Assessment of conserved genes using BUSCO v5.3.0⁷⁶ with the embryophyta_odb10 database revealed high assembly completeness, with 98.5% and 98.3% of complete BUSCO genes present in HaplotypeA and HaplotypeB assemblies respectively (Table 1). Consensus base accuracy was evaluated with Merqury v1.3⁷⁷ based on k-mer spectrum analysis, yielding high quality value (QV) scores of 50.02 for HaplotypeA and 50.42 for HaplotypeB (Table 9), indicating extremely low base-level error rates and high assembly accuracy.

Table 9.

Summary of assembly quality and orthology-based annotation assessment for the two haplotype genomes.

Haplotype	QV	k-mer comp. (%)	Assembly size (Mb)	Scaffold N50 (Mb)	Hap-specific genes	Complete HOGs (%)	Lineage-consistent (%)
HaplotypeA	50.02	99.39	528.93	50,21	851	96.72	86.50
HaplotypeB	50.42	99.98	563.27	55,05	2,079	96.74	85.81

Furthermore, an orthology-based evaluation of annotation completeness and consistency was conducted using OrthoFinder v2.5.4⁷⁸ and OMArk v0.3.1⁷⁹ on the protein-coding genes of the two haplotypes. OrthoFinder analysis identified 24,200 shared orthogroups between HaplotypeA and HaplotypeB, comprising 30,202 and 32,385 genes, respectively, together with 851 and 2,079 haplotype-specific genes (Table 9). Comparison with the previously published Eucalyptus urophylla × E. grandis genome² revealed 3,330 and 4,383 newly annotated genes in HaplotypeA and HaplotypeB, respectively, whereas 2,074 genes from the reference genome lacked corresponding orthogroups in the current annotations. OMArk analysis based on the conserved hierarchical orthologous groups of the malvids clade showed that annotation completeness exceeded 96% in both haplotypes, with missing proportions of only 3.28% and 3.26%, respectively. In addition, 86.50% and 85.81% of the protein-coding genes were classified as lineage-consistent, and no detectable contamination signal was observed (Table 9). These results collectively support the completeness, accuracy, and biological reliability of the current assemblies and gene annotations.

Second, clean reads derived from Illumina, ONT, and PacBio HiFi platforms were aligned back to each haplotype genome to validate assembly consistency and base accuracy. The alignment analysis was performed using BWA v0.7.17⁸⁰ and Minimap2 v2.17⁸¹, with mapping rates and coverage calculated using SAMtools v1.9⁸². Alignment results showed that the mapping rates for all three sequencing data exceeded 99.86%, with genomic coverage above 99.6% across various depth thresholds (Table 10). Sequencing depth analysis indicated average depths ranging from 125 × to 180.21 × for the different data types. Notably, ONT long reads achieved nearly 100% coverage on both haplotypes, supporting the high fidelity of the assemblies (Table 10).

Table 10.

Alignment statistics of Illumina, PacBio HiFi, and ONT reads against haplotype-resolved assemblies.

Sequencing technology	Reference genome	Mapping Rate (%)	Mean depth (×)	Genome Coverage (%)	Coverage ≥ 4 × (%)	Coverage ≥ 10 × (%)	Coverage ≥ 20 × (%)
Illumina	HaplotypeA	99.86	130.5	99.60	99.38	99.23	98.87
	HaplotypeB	99.88	125.0	99.63	99.38	99.23	98.87
PacBio HiFi	HaplotypeA	100.00	161.9	99.99	99.99	99.97	99.92
	HaplotypeB	100.00	153.1	100.00	99.99	99.97	99.82
	HaplotypeA	100.00	180.2	100.00	100.00	100.00	99.99
ONT	HaplotypeB	100.00	170.2	99.99	99.99	99.98	99.97

Finally, Hi-C data were utilized to construct chromatin interaction heatmaps to examine the chromosome-level assembly accuracy and haplotype phasing reliability. Hi-C contact matrices generated with HiCExplorer v3.6⁸³ revealed strong interaction signals between homologous chromosomes and higher intra-haplotype contact intensities compared to inter-haplotype signals (Fig. 5a), supporting accurate haplotype phasing and chromosomal anchoring. Whole-genome synteny analysis was performed to assess assembly accuracy and characterize structural divergence between the two haplotypes. MCScanX v1.0.0⁸⁴ was used to identify collinear gene blocks within and between haplotypes (Fig. 5b), revealing highly conserved gene order patterns in both haplotypes. SyRI v1.1.1⁸⁵ was used to compare syntenic relationships between the previous assembly and HaplotypeA and HaplotypeB, identifying insertions, deletions, and inversions (Fig. 5c). Gene and repeat density analyses using 100 kb sliding windows revealed a genome-wide complementary distribution, with non-syntenic regions showing reduced gene density and elevated repeat accumulation (Fig. 5c). A whole-genome dot plot generated using Minimap2 v2.17⁸¹ provided a visualization of sequence synteny and major structural variation sites between the two haplotypes (Fig. 5d). These results were highly consistent with synteny-based analyses, strongly supporting the accuracy of the chromosome-level assemblies of both haplotypes.

Fig. 5.

Quality assessment of the chromosome-scale haplotype-resolved genome assembly for E. urophylla × E. grandis. (a) Hi-C interaction heatmap illustrating chromatin interaction patterns across chromosomes in the two haplotypes. (b) Gene collinearity blocks within and between haplotypes are shown in different colors. (c) Synteny and structural variations between the reference genome and the two haplotypes, together with gene and repeat density distributions calculated in 100-kb windows. (d) Whole-genome dot plot showing sequence collinearity and major structural variations between the two haplotypes.

Author contributions

S.W. and Y.C. conceived and designed the research. D.Z. and Z.P. collected the samples. H.Y., S.W. and X.Y. analyzed the data and drafted the manuscript. S.W. revised this paper. All authors reviewed and approved the final manuscript.

Data availability

All raw sequencing data generated in this study have been deposited in the NCBI Sequence Read Archive (SRA) under BioProject accession number PRJNA1290067. Illumina paired-end reads, PacBio HiFi reads, Hi-C sequencing data, Oxford Nanopore Technologies (ONT) ultra-long reads, and transcriptome sequencing data are publicly available under accession numbers SRR34490132, SRR34490133, SRR34490134, SRR34490135, SRR38044340, SRR37991218, and SRR37991220. The haplotype-resolved telomere-to-telomere genome assemblies and associated annotation files, including gene, repetitive sequence, and non-coding RNA annotations in FASTA and GFF3 formats, have been deposited in the Figshare repository at https://doi.org/10.6084/m9.figshare.32231856.

Code availability

In this study, no custom code or script was developed or used. All software and pipelines were strictly carried out in accordance with the manuals and protocols of the published bioinformatics tools. The specific versions of the software and detailed parameters had been described in the methods. If the relevant parameters of the software were not mentioned, used its default parameters.

Competing interests

The authors declare no competing interests.